thesis_sahena - University of South Wales DSpace

Reliable Broadband Satellite-integrated
Network Design through Propagation and
Networking Solutions
By
Sahena Begum
Submitted in partial fulfilment for the degree of Doctor of
Philosophy
University of Glamorgan
May, 2009
Acknowledgements:
May I take this opportunity to express my gratitude to my Director of Studies, Dr.
Ifiok Otung, for his scholastic guidance, constant support and care throughout the
years. I sincerely appreciate the valuable time and close attention he has given me
during the course of this study. If it were not his vision and expertise, this endurance
would have been prolonged. I also wish to thank Prof. Khalid Al-Begain, my second
supervisor.
I would like to thank all my colleagues at the University for their time and company.
My heartiest thanks to my friends back home, here and abroad for their love, support
and encouragement.
My special “Thank You” to all my family for their love.
ii
Abstract:
Satellites will play an indispensable role in the deployment of commercial networks to
meet an increasing demand for supporting multimedia services at high data rates.
Next generation satellite systems, operating at high frequency bands offer large
bandwidth and are able to provide broadband services. To interface satellite links
with existing terrestrial networks for providing communication access to a variety of
users directly, several performance issues need to be addressed. Current thesis
presents a technically viable satellite-integrated network model that is efficient in
carrying broadband services to users over a wide scattered area.
Accurate prediction of attenuation level is necessary for a reliable network model to
operate with required service availability. Long term rainfall data has been analysed
to characterise attenuation level at a selected region such as Dhaka. It is shown that
rainfall is highly seasonal and attenuation level is quite high during monsoon.
However, the seasonal behaviour of rainfall can be exploited to improve the link
availability. Radar and rain gauge measurements at Sparsholt are also used to find
rain cell size distribution, which is an important factor in site diversity
implementation to combat severe rain fade. It is found that convective rain cell has
extension in the region of 10 km.
The network model is designed with dimensioning the effective bandwidth to support a
number of users over the satellite link by taking into account the multimedia traffic
characteristics. Concatenated coding, a robust coding scheme is implemented to
improve the link quality at a level required to deliver broadband services. The ITU-T
performance objectives of 7.5×10 8 for CLR and 1.4×10 6 for CER over satellite
links are met at a required Eb/No of 2.95 dB and 2.88 dB respectively. Different
enhancement mechanisms for optimum TCP performance are implemented to combat
the large propagation delay associated with a satellite link. It is revealed through the
simulation that TCP performance over a satellite link is as efficient as terrestrial links
with these enhancement mechanisms.
Finally, the overall performance of the designed network is evaluated through link
budget analysis and simulation. An innovative downlink power control strategy has
been implemented to maintain the link during the rainiest months. The interference
level due to high power satellite transmission in the designed system is also
calculated to protect other existing communication links sharing the same frequency
bands.
A feasible broadband network designed with characterising propagation as well as
networking issues will efficiently deliver broadband communication services to a
large population promptly and in a cost-effective manner. Such a network solution
will be in the realm of current R & D towards broadband satellite networks.
iii
Table of Contents
Acknowledgements ........................................................................................................ii
Abstract .........................................................................................................................iii
Table of Contents .......................................................................................................... iv
List of Figures ..............................................................................................................vii
List of Tables ................................................................................................................ ix
Acronyms and abbreviations.......................................................................................... x
1
Introduction .......................................................................................................... 1
1.1
Introduction: ................................................................................................... 2
1.2
Research Motivation: ..................................................................................... 4
1.3
Overview - contents of thesis:........................................................................ 7
1.4
Summary of novel contribution: .................................................................. 11
2
Radio Wave Propagation .................................................................................. 14
2.1
Introduction: ................................................................................................. 15
2.2
Earth’s atmosphere: ..................................................................................... 17
2.3
Interaction mechanism: ................................................................................ 18
2.3.1
Absorption- .......................................................................................... 19
2.3.2
Scattering- ............................................................................................ 19
2.3.3
Refraction- ........................................................................................... 19
2.3.4
Diffraction- .......................................................................................... 19
2.3.5
Multipath- ............................................................................................ 19
2.3.6
Scintillation-......................................................................................... 20
2.3.7
Fading- ................................................................................................. 20
2.3.8
Frequency Dispersion-......................................................................... 20
2.4
Propagation factors (above 3 GHz): ............................................................ 20
2.4.1
Gaseous attenuation- ........................................................................... 21
2.4.2
Hydrometeor attenuation- .................................................................... 21
2.4.3
Depolarisation- .................................................................................... 21
2.4.4
Radio noise- ......................................................................................... 22
2.4.5
Angle of arrival variations- ................................................................. 22
2.4.6
Bandwidth coherence- ......................................................................... 22
2.4.7
Antenna gain degradation- .................................................................. 23
2.4.8
Tropospheric scintillation-................................................................... 23
2.5
Propagation factors (up to 3 GHz): .............................................................. 23
2.5.1
Ionospheric scintillation- ..................................................................... 24
2.5.2
Polarisation rotation- .......................................................................... 24
2.5.3
Group delay- ........................................................................................ 24
2.5.4
Multipath fading and scintillation- ...................................................... 25
2.6
Propagation impairments in an earth-space link: ......................................... 25
2.6.1
Free space attenuation- ....................................................................... 25
2.6.2
Phenomenon associated with refractive index- ................................... 26
2.6.3
Attenuation by atmospheric gases- ...................................................... 26
2.6.4
Hydrometeor attenuation- .................................................................... 27
2.6.5
Fog and cloud attenuation – ................................................................ 27
2.6.6
Rain attenuation- ................................................................................. 28
2.7
Conclusions: ................................................................................................. 34
3
Characterisation of Rain Attenuation in Bangladesh ..................................... 36
3.1
Introduction: ................................................................................................. 37
3.2
Rainfall characterisation: ............................................................................. 38
3.2.1
Seasonal precipitation trends: ............................................................. 38
iv
4
5
6
7
3.2.2
Annual precipitation trend (Dhaka): ................................................... 41
3.3
Annual and monthly rainfall distribution:.................................................... 45
3.4
Rain rate distribution: .................................................................................. 49
3.4.1
Seasonal variations in rain rate distribution: ...................................... 54
3.5
Rain attenuation: .......................................................................................... 56
3.6
Risk estimation in rain attenuation prediction: ............................................ 62
3.7
Conclusions: ................................................................................................. 66
Rain Cell Size Distribution ................................................................................ 70
4.1
Introduction: ................................................................................................. 71
4.2
Rain cell translation velocity: ...................................................................... 73
4.3
Wind field from Doppler analysis: .............................................................. 76
4.4
Comparison of cell translation speed, Doppler speed and wind speed: ....... 79
4.5
Rain cell size distribution: ........................................................................... 81
4.6
Conclusions: ................................................................................................. 85
Traffic Analysis and Network Design .............................................................. 89
5.1
Introduction: ................................................................................................. 90
5.2
Telecommunication infrastructure in Bangladesh: ...................................... 90
5.3
Dimensioning traffic path: ........................................................................... 93
5.3.1
Traffic load or intensity: ...................................................................... 94
5.3.2
Grade of Service (GoS):....................................................................... 96
5.3.3
Traffic model:....................................................................................... 97
5.3.3.1 Capacity by Erlang B traffic model: ................................................ 98
5.3.3.2 Capacity by Erlang C traffic model: ................................................. 99
5.4
Characteristics of Ethernet data: ................................................................ 101
5.4.1
Self-similarity:.................................................................................... 101
5.4.2
Properties of self-similarity: .............................................................. 102
5.4.3
The Hurst parameter- The measure of self-similarity: ...................... 105
5.5
Dimensioning Ethernet traffic by Closed-Queuing Network (CQN) model:
……………………………………………………………………………105
5.6
Satellite-integrated Network model: .......................................................... 109
5.7
Conclusions: ............................................................................................... 111
Performance of ATM over Satellite Links..................................................... 115
6.1
Introduction: ............................................................................................... 116
6.2
ATM layer and QoS parameters: ............................................................... 118
6.3
Concatenated coding with Reed-Solomon and convolution coding: ......... 121
6.4
Parameters affected by coding scheme in satellite-integrated networks: .. 124
6.5
Performance of ATM with concatenated coding scheme: ......................... 126
6.6
Performance of TCP in terms of segment error ratio with concatenated
coding scheme:....................................................................................................... 128
6.7
Conclusions: ............................................................................................... 130
TCP/IP Performance Evaluation over Satellite Links ................................. 134
7.1
Introduction: ............................................................................................... 135
7.2
TCP overview: ........................................................................................... 136
7.2.1
TCP vanilla or Tahoe: ....................................................................... 137
7.2.2
TCP Reno: .......................................................................................... 138
7.2.3
TCP New Reno:.................................................................................. 139
7.2.4
SACK: ................................................................................................ 140
7.3
TCP throughput:......................................................................................... 140
7.4
Enhancing TCP over satellite channels: .................................................... 142
7.4.1
Path MTU discovery: ......................................................................... 142
v
7.4.2
TCP window scaling: ......................................................................... 142
7.4.3
Selective Acknowledgements: ............................................................ 144
7.4.4
Forward error correction: ................................................................. 144
7.4.5
Split TCP connections: ...................................................................... 145
7.4.6
Multiple data connections (XFTP): ................................................... 146
7.5
MATLAB simulation environment for TCP protocol: .............................. 146
7.6
Variation of TCP throughput with propagation delay: .............................. 151
7.7
TCP (Reno) throughput performance over satellite link: .......................... 152
7.8
Conclusions: ............................................................................................... 162
8
Link Budget Analysis and Simulation............................................................ 166
8.1
Introduction: ............................................................................................... 167
8.2
Link budget set up:..................................................................................... 168
8.3
Link budget analysis: ................................................................................. 169
8.4
Time series simulation: .............................................................................. 174
8.5
Conclusions: ............................................................................................... 179
9
Conclusions and Further Work ...................................................................... 181
9.1
Conclusions: ............................................................................................... 182
9.2
Future prospects: ........................................................................................ 186
Appendix-A................................................................................................................ 187
Appendix-B ................................................................................................................ 191
Appendix-C ................................................................................................................ 195
vi
List of Figures
Figure 1.1: Comparison of predicted attenuation by SAM and ITU-R model (using
ITU-R rain rate for both models) ................................................................................... 5
Figure 1.2: Block diagram of thesis layout .................................................................... 8
Figure 2.1: Schematic representation of the propagation of a transverse
electromagnetic wave................................................................................................... 15
Figure 2.2: Atmospheric layers .................................................................................... 17
Figure 3.1: Rain gauge stations over Bangladesh ........................................................ 39
Figure 3.2: Seasonal rainfall trend in different stations over Bangladesh (1961-2003)
...................................................................................................................................... 40
Figure 3.3: Annual rainfall time series in Dhaka (1961-2003) .................................... 41
Figure 3.4: Annual rainfall trend in Dhaka (a) 1961-1969 (b) 1970-1975 (c) 1976-81
(d) 1982-88 (e) 1989-94 (f) 1995-03 ........................................................................... 42
Figure 3.5: Power spectral density of annual rainfall in Dhaka (1961-2003) .............. 43
Figure 3.6: Annual rainfall distribution in Dhaka ........................................................ 46
Figure 3.7: Rainfall distribution for the month July in Dhaka ..................................... 47
Figure 3.8: Rainfall distribution for the month January in Dhaka ............................... 48
Figure 3.9: Measured annual rainfall statistics in comparison with predicted rain rate
by Rice-Holmberg model for Sparsholt, UK and Surabaya, Indonesia ....................... 50
Figure 3.10: Rain rates for different stations (Bangladesh) by Rice-Holmberg model
...................................................................................................................................... 53
Figure 3.11: Seasonal variations of rainfall rates by Rice-Holmberg model in Dhaka55
Figure 3.12: Predicted rain attenuation using R-H and ITU rain rate for Dhaka ......... 57
Figure 3.13: Approximation of attenuation reduction factor using Sparsholt data ...... 58
Figure 3.14: Comparison of predicted attenuation applying reduction factor (Dhaka)
...................................................................................................................................... 60
Figure 3.15: Attenuation for various levels of availabilities as a function of the month
...................................................................................................................................... 62
Figure 3.16: Rain attenuation risk estimation for Bangladesh ..................................... 64
Figure 3.17: Rain attenuation variability from three standard deviations bound for
Bangladesh ................................................................................................................... 65
Figure 4.1: PPI of reflectivity field, dBz (colorbar on right) on 23 March,2004 ......... 73
Figure 4.2: Identified rain cell on PPI on 23 March, 2004 (a) Cell in the previous scan;
(b) Correlated cell in the next scan .............................................................................. 75
Figure 4.3: Motion vector on an azimuth plane ........................................................... 77
Figure 4.4: Doppler velocity from event on 23 March, 2004 at a range 58 km .......... 78
Figure 4.5: Comparison of wind speed, Doppler wind speed and cell translation speed
...................................................................................................................................... 80
Figure 4.6: Cumulative distribution of rain cell sizes obtained from rain gauge data . 83
Figure 4.7: Cumulative distribution of rain cell sizes obtained from radar observation
...................................................................................................................................... 84
Figure 5.1: Microwave coverage of Bangladesh (source-BTTB annual report, 2001) 91
Figure 5.2: Traffic flow model..................................................................................... 94
Figure 5.3: Closed-queueing network (CQN) model ................................................. 106
Figure 5.4: A satellite-integrated network model ...................................................... 110
Figure 6.1: An ATM cell ........................................................................................... 118
Figure 6.2: ATM HEC operation at receiver ............................................................. 120
Figure 6.3: Block diagram of a concatenated coding scheme ................................... 121
Figure 6.4: Array representation of symbol interleaver ............................................. 123
vii
Figure 6.5: Performance of concatenated coding scheme on AWGN channels ........ 123
Figure 6.6: CLR and CER vs. Eb /No (filled circles are extrapolated values)........... 127
Figure 6.7: TCP Segment Error Ratio vs. Eb /No (filled circles are extrapolated
values) ........................................................................................................................ 129
Figure 7.1: The chronology of slow start and congestion avoidance algorithm of TCP
.................................................................................................................................... 147
Figure 7.2: Illustration of RTT calculation ................................................................ 148
Figure 7.3: Comparison of results from MATLAB with experiment ........................ 150
Figure 7.4: TCP throughput at different propagation delay ....................................... 151
Figure 7.5: Throughput for different file sizes without loss (Maximum TCP window
64 Kbytes) .................................................................................................................. 152
Figure 7.6: Window dynamics for file size of 200 Kbytes (Maximum TCP window 64
Kbytes) ....................................................................................................................... 153
Figure 7.7: Window dynamics for file size of 10 Mbytes (Maximum TCP window 64
Kbytes) ....................................................................................................................... 154
Figure 7.8: Throughput for a 10 Mbytes file transfer for different TCP window sizes
(without loss) ............................................................................................................. 155
Figure 7.9: Throughput for a 10 Mbytes file transfer for different TCP window sizes
(with loss and without loss) ....................................................................................... 156
Figure 7.10: Throughput for a 10 Mbytes file transfer for different TCP window sizes
with large segment size (no loss) ............................................................................... 157
Figure 7.11: Throughput for a 100 Mbytes file transfer for different TCP window
sizes with large segment size (no loss) ...................................................................... 158
Figure 7.12: Comparison of percentages of throughput with different window sizes
over terrestrial and satellite links ............................................................................... 159
Figure 7.13: Comparison of percentages of throughput with different window sizes
and file sizes over satellite links ................................................................................ 160
Figure 7.14: Comparison of percentages of throughput over satellite links with loss
and without loss ......................................................................................................... 161
Figure 8.1: Comparison of Attenuation level for Sparsholt and Dhaka .................... 174
Figure 8.2: Attenuation time series for Sparsholt and Dhaka .................................... 175
Figure 8.3: Instantaneous link behaviour with rain event for Dhaka (Zonal beam) .. 176
Figure 8.4: Instantaneous link behaviour with rain event for Dhaka (Spot beam) .... 177
viii
List of Tables
Table 2.1: Propagation concerns for satellite communication systems ....................... 33
Table 3.1: Statistical parameters for different periods of Annual rainfall in Dhaka .... 43
Table 3.2: Comparison of ITU-R and R-H derived rain rates against measurement at
Sparsholt (UK) ............................................................................................................. 51
Table 3.3: Comparison of ITU-R and R-H derived rain rates against measurement at
Surabaya (Indonesia) ................................................................................................... 51
Table 3.4: Rice-Holmberg parameters for different stations in Bangladesh ............... 52
Table 3.5: Comparison of the 0.01% rain rates of ITU-R and R-H computations ...... 54
Table 4.1: Statistics of rain cells from rain gauge and radar data ................................ 82
Table 5.1: Telecommunication indicator 2007 ............................................................ 92
Table 5.2: Traffic model characteristics ...................................................................... 97
Table 6.1: Effects of bit errors on ATM cell ............................................................. 119
Table 6.2: ATM performance objectives for satellites (class 1 services) .................. 128
Table 8.1: Earth stations parameters .......................................................................... 168
Table 8.2: Satellite parameters ................................................................................... 169
Table 8.3: Most significant link budget parameters (clear air) for zonal beam ......... 170
Table 8.4: Most significant link budget parameters (under rain) for zonal beam ...... 171
Table 8.5: Most significant link budget parameters (under rain) for spot beam ........ 172
ix
Acronyms and abbreviations
4G – Fourth-Generation Wireless
4KSS – 4K Slow Start
AAL – ATM Adaptation Layer
ACK – Acknowledgements
ARQ – Automatic Repeat Request
ATM – Asynchronous Transfer Mode
AWGN – Additive White Gaussian Noise
BDP – Bandwidth Delay Product
BER – Bit Error Ratio
B-ISDN – Broadband- Integrated Service Digital Network
BPSK – Binary Phase Shift Keying
BSS – Broadcasting Satellite Service
BTTB – Bangladesh Telegraph and Telephone Board
CAMRa – Chilbolton Advanced Meteorological Radar
CBR – Constant Bit Rate
CER – Cell Error Ratio
CLR – Cell Loss Ratio
CMR – Cell Misinsertion Rate
CQN – Closed Queuing Network
CTD – Cell Transfer Delay
DSL – Digital Subscriber Line
DSD – Drop Size Distribution
EIRP – Effective Isotropically Radiated Power
ENSO – El-Nino Southern Oscillation
FEC – Forward Error Correction
FFT – Fast Fourier Transform
FS – Fixed Service
FSS – Fixed Satellite Service
ftp – file transfer protocol
GEO – Geostationary Earth Orbit
GoS – Grade of Service
HEC – Header Error Control
http – hyper text transfer protocol
ICT – Information and Communication Technology
IETF – Internet Engineering Task Force
IP – Internet Protocol
IS – Infinite Server
IT – Information Technology
ITU-R – International Telecommunication Union- Radio Recommendation
LAN – Local Area Network
MTU – Maximum Transmission Unit
PAWS – Protection Against Wrapped Sequence
PCM – Pulse-code Modulation
PER – Packet Error Ratio
PD – Propagation Delay
PS – Processor Sharing
QoS – Quality of Service
QPSK – Quaternary Phase Shift Keying
x
RCVWND – Receive Window
RFC – Request for Comments
R-H – Rice- Holmberg
RS – Reed-Solomon
RTO – Retransmission Time-out
RTT – Round Trip Time
RTTM – Round Trip Time Measurement
SACK – Selective Acknowledgements
SAM – Simple Attenuation Model
SECR – Severely Errored Cell Ratio
SER – Segment Errror Ratio
SMTP – Simple Mail Transfer Protocol
ssthresh – slow start threshold
TCP – Transmission Control Protocol
TDM – Time Division Multiplex
TDMA – Time Division Multiple Access
TRMM – Tropical Rainfall Measuring Mission
UDP – User Datagram Protocol
UMTS – Universal Mobile Telecommunication Systems
VHF – Very High Frequency
VSAT – Very Small Aperture Terminal
WAN – Wide Area Network
WiMAX – Worldwide Inter-operability for Microwave Access
www – world wide web
xi
Chapter 1: Introduction
1 CHAPTER 1
Introduction
1
Chapter 1: Introduction
1.1
Introduction:
Satellites form an essential part of telecommunications systems worldwide, carrying
large amounts of data and telephone traffic in addition to television signals from its
inception. Satellites have the advantage of offering a number of exclusive features not
readily available by other means of communication [1] [2]

Minimum extra transportation cost associated with the “last-mile”

Cost effectiveness on thin routes.

Satellites can be used in remote areas as gap fillers and in areas where
terrestrial coverage is inadequate or unreliable to off-load traffic

More robustness to meteorological conditions

Scalable architecture i.e. a new user can join a satellite communication
system by acquiring the necessary tools
Optical fibre systems tend to concentrate around developed areas giving further
impetus for growth to an already developed region whereas satellites give an equal
coverage to all regions because of their inherent broadcasting capability. Thus for
example, the advantages of broadband could reach the remotest regions instantly
using a satellite system whereas it is likely to be a long time before optical fibre links
to the region can be economically justified.
In recent years the increased demand for electronic connectivity across the globe has
been driving newer technologies such as Universal Mobile Telecommunication
Systems (UMTS), Worldwide Interoperability for Microwave Access (WiMAX),
fourth-generation wireless (4G) etc to meet the requirements of multimedia
2
Chapter 1: Introduction
communications. Due to its unique broadcasting capability and large global coverage,
satellite systems will play a significant role in tandem with these newer technologies,
either on its own or as a complementary means of communication. In this context, the
next generation satellite systems often termed “broadband satellite networks”,
“multimedia satellite networks” or “satellite-fibre networks” are being envisaged to
support various data communication services such as private intranet, digital video
broadcasting and internet.
Many of the applications under the broadband umbrella are Transmission Control
Protocol/Internet Protocol (TCP/IP) based high availability services that require
higher speed and higher quality of data transmissions with more efficiency and
reliability than existing services such as voice, broadcasting etc. Higher frequency
bands such as Ku-band (14/12 GHz) and Ka-band (30/20 GHz) proved attractive to
provide these services although these bands are more susceptible to rain fading [3].
TCP protocol does not perform efficiently in a satellite environment due to errors
induced and large propagation delay imposed by the channel. Thus providing
broadband services at required availability and optimum fidelity require newer
solution at the corresponding layer of the protocol stack or cross-layer protocol
optimisation [4]. Interoperating satellite communication networks with terrestrial
networks where the major technologies are Internet Protocol (IP) and Asynchronous
Transfer Mode (ATM) require propagation as well as networking solutions which are
technically challenging to meet the different Quality of Service (QoS) requirements at
relevant layers. Numerous mitigation techniques have been investigated for the
efficient transport of ATM and TCP/IP applications over the satellite network at
different functional layers and these are often inter-related [2] [4].
3
Chapter 1: Introduction
In this thesis the main goal was to design a reliable satellite-integrated network by
finding suitable solution to combat inherent characteristics of satellite link such as
fading and the large propagation delay to deliver the broadband traffic at the required
QoS. A tropical region such as Bangladesh has been chosen to realise the satelliteintegrated network, where rainfall level is generally high and where most of the city
areas have well developed communication networks whereas rural areas are cut-off
from the mainstream communication infrastructure. A feasible broadband network
will allow multimedia services to a large population promptly and in a cost-effective
manner, especially for users separated by long distances, including rural areas and
remote islands.
1.2
Research Motivation:
Following the challenging issues outlined above of the satellite-integrated network,
prime consideration for the system design was given to improving the rain attenuation
prediction model based on local meteorological measurements, and optimising the
satellite link quality as well as TCP/IP performance evaluation in delivering
broadband services at the desired quality to end users.
Accurate prediction of the rain attenuation level is crucial in the system design
procedure such as choice of services at the selected frequency and service availability.
There are several rain attenuation prediction models [5] [6] [7] [8] available to
quantify the rain attenuation level, which were developed for different climatic
conditions. Figure 1.1 shows the predicted attenuation by Simple Attenuation Model
(SAM) and International Telecommunication Union (ITU-R) model using ITU-R rain
4
Chapter 1: Introduction
rate for Bangladesh at 18.7 GHz frequency band and 46.10 path elevation angle. As
can be seen, SAM predicts significantly different attenuation level compared to ITUR model, which is usually applied for benchmarking the attenuation level for
commercially designed systems. For a more reliable prediction, a novel rain
attenuation prediction model has been developed over the region based on the local
meteorological measurements.
Figure 1.1: Comparison of predicted attenuation by SAM and ITU-R model
(using ITU-R rain rate for both models)
As the rain structure in the slant path is directly related to rain rate, rainfall volume
and local climate [9], the statistical knowledge of the structure of rain is important in
site diversity implementation. Rain gauge records and radar data for a large number of
years are used to find the rain cell size distribution over Sparsholt, UK, a temperate
climate. Convective rain cells extend over more limited regions in tropical regions, a
typical cell size ranging from 2-5 km and often lasting for only 10- 20 minutes [10]
5
Chapter 1: Introduction
[11]. The information of intense rain cell extension over UK could still be applicable
over Bangladesh in setting-up the diversity stations as the simultaneous occurrence of
rain within this close proximity will be very low.
A critical point in designing a satellite-integrated network is to dimension the required
bandwidth to carry the multimedia traffic. Different traffic models are analysed and
Closed Queuing Network Model (CQN) is applied to calculate the bandwidth for
multimedia traffic.
Different error control schemes can be found in the research literature to carry
efficiently either the ATM or TCP/IP traffic over the error prone satellite links [2]
[12]. A robust concatenated coding scheme has been implemented which is efficient
in delivering the QoS parameters at ATM layer as well as in reducing the Packet Error
Ratio (PER) at TCP layer required for optimum TCP performance.
The above concatenated coding scheme presents TCP with a more reliable satellite
channel. Nevertheless, TCP performance is still degraded as its feedback triggered
algorithm suffers severely due to a large propagation delay. To address these
deficiencies and optimise TCP performance on the designed satellite network,
selected parameters such as window scaling and Path MTU (Maximum Transmission
Unit) discovery have been tuned which are currently standardised by Internet
Engineering Task Force (IETF) [13].
The interest in extending broadband services over satellite links have triggered a wide
spectrum of research topics concerning different protocol layers- application,
6
Chapter 1: Introduction
transport, data link and physical link. In conjunction with protocol layers, different
performance attributes such as router buffer size, queuing discipline, link quality and
network availability have also been addressed for satellite-multimedia networks.
However, the vast majority of studies concentrated on protocol performance. This
thesis presents several novel approaches namely improving the system availability
and link quality, dimensioning the required capacity of the link to carry the traffic and
tuning the TCP parameters for optimum network performance. The novel network is
constructed with the optimised parameters to deliver high availability broadband
services at the required quality.
1.3
Overview - contents of thesis:
Following the introductory chapter the thesis is divided into two major sections,
namely Propagation and Networking, containing different items of work undertaken
in the research leading to a reliable satellite-integrated network. The initial design
starts with addressing physical layer performance analysis and characterising rain
attenuation level over the region, which is the primary parameter affecting network
availability. A robust coding scheme is implemented at the physical layer to optimise
the QoS parameters at ATM layer. End-to-end performance enhancement technique
was addressed at the TCP layer for efficient data transfer in the network. System level
analysis of the network was then performed by incorporating individual parameters
from different layers. The structure of the thesis is depicted in Figure 1.2.
7
Chapter 1: Introduction
Chapter 1
Introduction
Propagation aspects
Chapter 2
Radio wave
propagation
Chapter 3
Rain attenuation
characterisation
Networking aspects
Chapter 4
Rain cell size
distribution
Chapter 5
Network
design
Chapter 6
ATM
performance
Chapter 7
TCP/IP
performance
Chapter 8
Link budget and simulation
Chapter 9
Conclusions and further work
Figure 1.2: Block diagram of thesis layout
Chapter 2 looks into the fundamentals of radio wave propagation to provide a general
understanding of the topics before attempting to develop a solution. The Earth’s
atmosphere, interaction mechanism, propagation factors and different propagation
impairments are all described. A detailed derivation of rain attenuation computation
method is presented, rain being the biggest obstacle to the design of earth-space
communication links at Ku-band frequencies and above.
Chapter 3 describes the characterisation of rain attenuation over Bangladesh, the main
propagation topic undertaken in the research, using 40 years of rainfall data. This will
be incorporated in the link budget analysis to work out the availability of the network.
8
Chapter 1: Introduction
The analyses include periodicity of annual rainfall, rainfall distribution, seasonal
variation of rainfall etc which are vital to be considered in improving the system
availability.
Chapter 4 discusses the rain cell size distribution derived from long term rain gauge
and radar measurements at Sparsholt, UK. By tracking the rain cell in consecutive
radar scans, cell translation velocity was obtained, which was then applied in the
synthetic storm technique to find the cell size distribution.
Chapter 5 presents a novel satellite-integrated network based on dimensioning the
effective bandwidth. Different conventional traffic models are investigated detailing
their application for different types of traffic. Based on the characteristics of
multimedia traffic, CQN model was chosen for calculating the required bandwidth.
Chapter 6 provides the error control mechanisms to improve the link quality at a
desired BER required to deliver ATM QoS parameters as well as to reduce PER for
optimum TCP performance. Concatenated coding scheme with Reed Solomon (RS)
outer code and ½ rate convolution inner code with block interleaving of depth five in
between is implemented to improve the link quality.
Chapter 7 summarises the two performance enhancing mechanisms standardised by
IETF that can be deployed end-to-end, namely window scaling and Path MTU
discovery. The algorithms of TCP Reno are implemented in MATLAB to analyse the
impact of performance enhancing schemes.
9
Chapter 1: Introduction
Chapter 8 gives the link budgets, methodology and final results of the simulation
scenario. Seasonal variability of rainfall, spot beam configuration and attenuation time
series generated from Sparsholt event are incorporated in the link budget analysis and
simulation.
Chapter 9 finally concludes the thesis summarising the work as a whole and
highlights directions that could lead to further research in the relevant area.
10
Chapter 1: Introduction
1.4
Summary of novel contribution:
The novel contribution of the thesis is summarised as follows
An improved rain attenuation prediction model. This work was published
in Radio Science [14].

Rain cell size distribution in the UK. This work was published in
EuCAP2006 [15] and in Radio Science [16].

Determination of cell translation velocity by tracking intense rain cells in
successive radar PPI (Plan Position Indicator) scans.

Novel satellite-integrated network with effective link bandwidth.

Optimisation of ATM QoS parameters and TCP Segment Error Ratio
(SER) with concatenated coding scheme. The last two works will be
submitted to International Journal of Satellite Communications and
Networking [17] for publication.

Optimisation of TCP throughput over satellite links.
11
Chapter 1: Introduction
References:
1. M. Riccharia, Satellite Communication Systems, McMillan, 1995.
2. A. Jamalipour, M. Marchese, H. S. Cruickshank, J. Neale and S. N. Verma,
“Guest Editorial- Broadband IP Networks via Satellites- Part 1, IEEE Journal
On selected Areas in Communication, Vol. 22, No. 2, 2004, pp. 213-217.
3. D. Panagopoulos, M. Arapoglou and P.G. Cottis, “Satellite Communications at
Ku, Ka and V bands, Propagation Impairments and Mitigation Techniques”, IEEE
Communications Surveys and Tutorials, Vol.6, No. 3, Third Quarter 2004.
4. G. Giambene and S. Kota, “Cross-layer Protocol Optimization for Satellite Communications Networks: A Survey”, International Journal of Satellite
Communication, 24, 2006, pp. 323-314.
5. ITU-R, “Propagation data and prediction methods required for the design of
Earth-space telecommunication systems”, Rec. ITU-R P.618-8, 2003.
6. W.L. Stutzman and W.K. Dishman, “A Simple Model for The Estimation of
Rain-induced Attenuation along Earth-to-Space Paths at Millimeter
Wavelengths”, Radio Science, Vol. 17, 1982, pp. 1465-1475.
7. M. J. Leitao and P. A. Watson, “Method for Prediction of Attenuation on Earthspace Links based on Radar Measurements of the Physical Structure of Rainfall”,
IEE proceedings, Vol.133, No.4, July 1986, pp. 429-440.
8. R. K. Crane, “Prediction of Attenuation by Rain”, IEEE Transactions on
Communications, Vol. Com-28, No.9, September 1980, pp. 1717-1733.
9. G.H. Bryant, I. Adimula, C. Riva and G. Brussaard, “Rain Attenuation Statistics
from Rain Cell Diameters and Heights”, International Journal of Satellite
Communication, 19, 2001, pp. 263-283.
10. H. E. Green, “Propagation Impairments on Ka-Band SATCOM Links in
12
Chapter 1: Introduction
Tropical And Equatorial regions”, IEEE Antennas and Propagation Magazine,
Vol. 46, No. 2, 2004, pp.31-45.
11. Q. W. Pan and G. H. Bryant, “Results of 12 GHz Propagation Measurements in
Lae (PNG)”, Electronics Letter, 28, 1992, pp. 2022-2024.
12. I. F. Akyildiz and S. Jeong, “Satellite ATM Networks: A Survey”, IEEE ComMunications Magazine, July 1997, pp. 30-43.
13. M. Allman, D. Glover, J. Griner, K. Scott, J. Touch and D. Tran, “Ongoing TCP
Research Related to Satellites”, RFC 2760, 1998.
14. S. Begum and I. E. Otung, “Characterisation of Rain Attenuation in
Bangladesh and Application to Satellite Link Design”, Radio Science, Vol. 43,
RS1008, doi:10.1029/2007RS003634, 20008.
15. S. Begum, C. Nagaraja and I. E. Otung, “Analysis of Rain Cell Size Distribution
for Application in Site Diversity”, European Conference on Antennas and
Propagation, Nice, France, 6-10 November, 2006.
16. S. Begum and I. E. Otung, “Rain Cell Size Distribution Inferred from Rain
Gauge and Radar Data in the UK”, Radio Science, Vol. 44,
RS2015, doi:10.1029/2008RS003984, 20009.
17. S. Begum and I. E. Otung, “Design and Performance Analysis of a Satellite-,
integrated Network”, International Journal of Satellite Communications and
Networking, submitted, 2009.
13
Chapter 2: Radio wave propagation
2 CHAPTER 2
Radio Wave Propagation
14
Chapter 2: Radio wave propagation
2.1
Introduction:
An electromagnetic wave, referred to as a radio wave at radio frequencies, is
characterised by variations of its electric and magnetic fields (Figure 2.1). The
oscillating motion of the field intensities vibrating at a particular point in space at a
frequency f excites similar vibrations at neighbouring points and the wave is said to
travel or to propagate. Mathematically,
S = E×H
(2.1)
where S is the Poynting vector representing the direction of propagation and power
per unit area of the wave.
E
x
Poynting vector
Ex
z
y
Hy
Wavelength
H
Figure 2.1: Schematic representation of the propagation of a transverse
electromagnetic wave
The wavelength λ of the electromagnetic wave is the spatial separation of two
successive oscillations, which is the distance the wave travels during one cycle of
15
Chapter 2: Radio wave propagation
oscillation. Number of oscillations per unit time is termed as the frequency of the
electromagnetic wave. The frequency and the wavelength in free space are related by
λ = c/f
where c is the propagation or phase velocity of light in vacuum given as
c=
1
 o o
, εo is the permittivity which is 10 9 /36 π Farad per metre
μo is the permeability which is 4π × 10 7 Henry per metre
resulting c = 3×10 8 m/s.
A radio signal will be degraded depending on frequency, temperature, pressure and
water vapour concentration while propagating through the earth’s atmosphere. It is
important to characterise the signal interaction with the atmosphere at different
frequencies to calculate attenuation level to design the communication link [1].
This chapter presents the earth’s atmosphere, interaction mechanism of radio waves
with the atmosphere and propagation factors. Different propagation impairments in an
earth-space link are also presented with detailed description of rain attenuation in
particular which is the main impairment factor that needs to be considered in this
thesis to design an earth-space communication link.
16
Chapter 2: Radio wave propagation
2.2
Earth’s atmosphere:
A signal travelling between an earth station and a satellite must pass through the earth
atmosphere which is divided into categories as in Figure 2.2. The troposphere is the
region of the atmosphere adjacent to the earth and extending upwards to about 10 km.
It is in this region the clouds are formed. The ionosphere extends from about 50 km to
roughly 2000 km above the surface. Due to the radiations from the sun, the
ionosphere takes on a stratified character called the D, E and F layers. The radiations
are mainly ultraviolet rays, γ rays and cosmic particles such as electrons and protons.
Figure 2.2: Atmospheric layers
17
Chapter 2: Radio wave propagation
The D layer is found occasionally at a height of 50 to 100 km in the daytime and is of
little importance. The E layer is a relatively permanent layer at about 100 km. During
the day the ionic density in this layer is strongest and may almost vanish at night due
to the recombination of ions. The F layer is also more or less permanent at about 300
km. In the daytime, it divides into the F 1 and F 2 layers and is subject to erratic
variations. Apart from seasonal variations, sun-spot activity causes further magnetic
storms and consequent radio fadeouts [1].
It is the ionised layers high up in the ionised atmosphere and the moist, turbulent
layers way down in the lower reaches of the neutral atmosphere which are the
principal factors in radio wave propagation. The various regions in the ionosphere act
as reflectors or absorbers to radio waves at frequencies below about 30 MHz, and
space communications is not possible. As the frequency is increased, the reflection
properties of the E and F layers are reduced and the signal will propagate through.
Radio waves above about 30 MHz will propagate through the ionosphere with
degradation of varying degrees depending on the frequency, geographic location and
time of the day. As the frequency of the wave increases, ionospheric effects become
less significant and above about 3 GHz the ionosphere is essentially transparent to
space communications [1] [2].
2.3
Interaction mechanism:
The interactions of the radio wave with the atmosphere occur through the following
mechanism [1] [3] [4]:
18
Chapter 2: Radio wave propagation
2.3.1
Absorption-
Absorption is a reduction in the amplitude of a radio wave caused by an irreversible
transfer of energy from the radio wave to matter in the propagation path.
2.3.2
Scattering-
Scattering is a process in which the energy of a radio wave is dispersed in direction
due to interaction with inhomogeneities in the propagation medium.
2.3.3
Refraction-
This is a change in the direction of propagation of a radio wave resulting from the
spatial variation of refractive index of the medium.
2.3.4
Diffraction-
A change in the direction of propagation of a radio wave resulting from the presence
of an obstacle, a restricted aperture, or other object in the medium is referred to as a
diffraction. This phenomenon can be explained by physical optics rather than
geometrical optics.
2.3.5
Multipath-
The propagation condition that results in a transmitted radio wave reaching the
receiving antenna by two or more propagation paths is called multipath. This effect
can arise due to refractive index irregularities in the troposphere or ionosphere, or
from structural and terrain scattering on the earth’s surface.
19
Chapter 2: Radio wave propagation
2.3.6
Scintillation-
Owing to the presence of irregularities inside the medium, an electromagnetic wave
undergoes rapid variations in its amplitude, phase and directions of arrival. These
variations are called scintillations and are characterised by their depth, period and
speed of variation.
2.3.7
Fading-
Fading is the variation of the amplitude of a radio wave caused by changes in the
transmission path with time. The terms fading and scintillation are often used
interchangeably; however, fading is usually used to describe slower time variations,
on the order of seconds or minutes, while scintillation refers to more rapid variations,
on the order of fractions of a second in duration.
2.3.8
Frequency Dispersion-
Frequency dispersion is a change in the frequency and phase components across the
bandwidth of a radio wave, caused by a dispersive medium. A dispersive medium is
one whose constitutive components (permittivity, permeability and conductivity)
depend on frequency (temporal dispersion) or wave direction (spatial dispersion).
2.4
Propagation factors (above 3 GHz):
Generally, radio waves with frequency 3 GHz and above are affected through the
following propagation factors primarily produced in the troposphere [1] [2] [3]:
20
Chapter 2: Radio wave propagation
2.4.1
Gaseous attenuation-
Gaseous attenuation is an absorption process that results in a reduction in signal
amplitude caused mainly by oxygen and water vapour. Gaseous attenuation increases
with increasing frequency and is dependent on atmospheric temperature, pressure and
humidity.
2.4.2
Hydrometeor attenuation-
Condensation of atmospheric water vapour gives rise to hydrometeors such as rain,
clouds, fog, snow, ice which attenuate radio waves through absorptive and scattering
effects. Rain attenuation is the dominant impairment in the space communication
systems operating at 10 GHz and above. Cloud and fog attenuations are much less
severe than rain attenuation; however, they must be considered in link design,
particularly for frequencies above 15 GHz. Dry snow and ice particle attenuation is
usually so low that it is unobservable on space communications link operating below
30 GHz.
2.4.3
Depolarisation-
Depolarisation is a phenomenon whereby all or part of a radio wave emitted with a
given polarisation no longer has any determined polarisation after propagation. Rain
and ice depolarisation can be a problem in the frequency bands above about 12 GHz,
particularly for “frequency reuse” communication links which employ dual
independent orthogonal polarised channels in the same frequency band to increase
channel capacity. Depolarisation caused by multipath propagation is generally limited
21
Chapter 2: Radio wave propagation
to very low elevation angle space communication and is dependent on the polarisation
characteristics of the receiving antenna.
2.4.4
Radio noise-
Radio noise is the presence of undesired signals or power in the frequency band of a
communication link generated by natural or man-made sources. It can degrade the
noise characteristics of receiver systems and affect antenna design or system
performance. Atmospheric gases (oxygen, water vapour), rain, clouds and surface
emissions are all natural noise sources and effective in the frequency range of 1 GHz
and above. Man-made sources include- other space or terrestrial communication links,
electrical equipment and radar systems. Extraterrestrial cosmic noise must only be
considered for frequencies below about 1 GHz,.
2.4.5
Angle of arrival variations-
Angle of arrival variations are a refraction process caused by refractive index changes
in the transmission path. It is only observable with large aperture antennas (10 metres
or more) and at frequencies well above 10 GHz. The angle of arrival change results in
an apparent shift in the location of satellite position and can be compensated for by repointing the antenna.
2.4.6
Bandwidth coherence-
Coherence bandwidth is a statistical measure of the range of frequencies over which
the channel can be considered "flat", or in other words the approximate maximum
bandwidth or frequency interval over which two frequencies of a signal are likely to
22
Chapter 2: Radio wave propagation
experience comparable or correlated amplitude fading. If the multipath time delay
spread equals D seconds, then the coherence bandwidth Wc in hertz is given
approximately by the equation: Wc = 1/2πD. The coherence bandwidth for typical
space communication frequencies is one or more gigahertz, and is not expected to be a
severe problem.
2.4.7
Antenna gain degradation-
Amplitude and phase fluctuations induced by the atmosphere can produce
perturbations across the antenna aperture, resulting in a reduction of total power
available at the antenna feed. The resulting effect is termed as antenna gain
degradation. This effect can be produced by intense rain; however it is observable
with very large aperture antennas at frequencies above about 30 GHz and for very
long path lengths through the rain, i.e. low elevation angles.
2.4.8
Tropospheric scintillation-
Refractive index irregularities due to high humidity gradients and temperature
inversion layers in the first few kilometres of altitude cause tropospheric scintillation.
The effects are seasonally dependent, vary day to day, and with local climate.
2.5
Propagation factors (up to 3 GHz):
The major propagation factors which affect space communication at frequencies
above ionospheric penetration frequency and up to about 3 GHz are as follows:
23
Chapter 2: Radio wave propagation
2.5.1
Ionospheric scintillation-
Ionospheric scintillation is produced by electron density irregularities near the altitude
of maximum electron density, the F region, at approximately 200-400 km in altitude.
These conditions are most prevalent in the equatorial regions at high latitude location
and during periods of high sunspot activity. Ionospheric scintillations have been
observed at frequencies from 20 MHz through 6 GHz, with the bulk of data being
amplitude scintillation observations in the VHF (30-300 MHz) bands. Scintillations
can be very severe in the frequency bands below 300 MHz and often are the limiting
factor for reliable communications system performance in the VHF bands.
2.5.2
Polarisation rotation-
In the presence of earth’s magnetic field, the ionosphere exhibits birefringence,
splitting the incident wave into ordinary and extraordinary components. Since these
waves propagate with different phase velocities, the resultant plane of polarisation of
the combined wave rotates as it propagates. This effect is known as Faraday rotation.
2.5.3
Group delay-
Due to the presence of free electrons in the propagation path the propagation time
from satellite to the Earth is longer than the time calculated in free space. The wave
propagates with a group velocity lower than the speed of light in vacuum. The
resulting delay, known as the group delay or group propagation time, induces an error
in the estimation of the distance from the satellite. The effect can be extremely critical
for radio-navigation or satellite ranging links which require an accurate knowledge of
range and propagation time for reliable performance. Group delay will be about 25
24
Chapter 2: Radio wave propagation
microseconds at 100 MHz for an earth-space path at a 30 degree elevation angle and
is approximately proportional to the reciprocal of the frequency squared.
2.5.4
Multipath fading and scintillation-
Multipath fading and scintillation are the variations in the amplitude and phase of a
radio wave, caused by terrain and surface roughness conditions. This problem is
important in terrestrial communications and must also be considered for earth-space
transmissions at low elevation angles, and for VHF mobile satellite links.
2.6
Propagation impairments in an earth-space link:
Besides free space attenuation, propagation impairments on earth-space links mainly
involve effect of irregularities of refractive index inside the troposphere and the
ionosphere, absorption due to atmospheric gases, namely oxygen and water vapour,
and attenuation caused by hydrometeors like clouds, rain, fog, snow and ice [1] [5].
2.6.1
Free space attenuation-
Radio waves experience transmission loss due to the dispersion of energy which takes
place as the wave travels away from the transmitter. This free space attenuation A0 is
given by
A 0 = 20log 10 (
4d

) dB
(2.2)
25
Chapter 2: Radio wave propagation
where λ is the wavelength and d is the distance travelled between the transmitter and
receiver.
2.6.2
Phenomenon associated with refractive index-
The refractive index of the media (troposphere, ionosphere etc.) varies along the
direction of propagation of a radio wave. Accordingly, an electromagnetic wave will
follow a curvilinear trajectory and the wave is said to be refracted. The curvature of
the trajectory is proportional to the refractivity gradient. From this phenomenon
originate a number of different other effects such as lengthening of the path, changes
in the propagation velocity or the angle of arrival, scintillation etc.
2.6.3
Attenuation by atmospheric gases-
The transmission attenuation caused by atmospheric gases results from the molecular
resonance of oxygen and water vapour. An oxygen molecule has a single permanent
magnetic moment. At certain frequencies, its coupling with the magnetic field of an
incident electromagnetic wave causes resonance absorption. In particular, at
frequencies around 60 GHz a coupling occurs between the intrinsic moment of the
electron, its spin and the rotational energy of the molecule, generating a series of
absorption lines quite close to each other in the spectrum. These absorption lines
come to merge, thus forming a single and broad absorption band. The water vapour
molecule behaves like an electric dipole. The interaction of such a molecule with an
incident wave disorientates the molecule by generating an additional internal potential
energy. The maximum attenuation reached in the band around 22 GHz is due to the
26
Chapter 2: Radio wave propagation
resonance of the water molecule which starts to rotate while absorbing a high
proportion of the incident electromagnetic energy.
For evaluating the attenuation due to atmospheric gases it is required to take into
account the contribution of all the absorption lines of oxygen and water vapour and
the continuous spectrum of the absorption due to water and ice. There are several
models [6][7][8] to find the attenuation level by atmospheric gases. As a numerical
application of the ITU-R model, the specific attenuation due to atmospheric gases in
the case of an average atmosphere (7.5 g/m 3 ) was found to be equal to approximately
0.2 dB/km and 15 dB/km at 20 and 60 GHz respectively [2].
2.6.4
Hydrometeor attenuation-
Hydrometeor refers to the condensed form of water vapour such as rain, hail, ice, fog,
cloud or snow etc. Hydrometeors attenuate the signal in two ways: the energy
absorption by Joule effect and the wave diffusion induced by the particles.
2.6.5
Fog and cloud attenuation –
Attenuation due to clouds and fog is determined on the basis of the total water content
per unit volume. The liquid water concentration is typically equal to approximately
0.05 g/m 3 inside a moderate fog (visibility of the order of 300 m) and of 0.5
g/m 3 inside a thick fog (visibility of the order of 50 m). In the case of clouds and fogs
consisting entirely of very small droplets with diameter less than 0.01 cm on average,
the Rayleigh approximation is valid at frequencies lower than 200 GHz. Attenuation
27
Chapter 2: Radio wave propagation
can therefore be expressed as a function of the total water content per unit volume
(g/m 3 ). Under these assumptions, the specific attenuation in clouds or fogs is
 C  Kl M
(dB/km)
where γ C is the specific attenuation inside the cloud in dB/km, K l is the specific
attenuation in dB/km per g/m 3 and M is the concentration of liquid water in clouds or
in fog in g/m 3 .
In the case of a moderate fog (0.05 g/m 3 ), the orders of magnitude for attenuation are
0.002 and 0.1 dB/km at the 20 and 60 GHz frequency ranges respectively, while for a
thick fog (0.5 g/m 3 ) they reach 0.02 and 1 dB/km respectively.
2.6.6
Rain attenuation-
Due to the condensation of water vapour, the diameter of the droplets from which
cloud forms increases, either by coalescence or by the absorption of the water vapour
around them. With the increase of size, the fall speed also increases and precipitation
occurs either in the form of drizzle if the diameter of the droplets lies between 0.1 and
0.5 mm or in the form of rain, if the droplets are of larger dimensions. If the
temperature falls below 0 0 C, hydrometeors occur in solid form (snow or hailstones).
Rain drops both absorb and scatter energy. Rain also causes depolarisation, rapid
amplitude and phase fluctuations, antenna gain degradation and bandwidth coherence
reduction. The signal loss by rain in an earth-space link is an integral of all the
28
Chapter 2: Radio wave propagation
individual increments of attenuation caused by the drops encountered along the path.
This is the physical approach to predicting rain attenuation. Prediction of rain
attenuation relies more particularly on precipitation intensities (rain rate) expressed in
mm/h. Rainfall is highly variable in time and space. In the mid latitudes stratiform
rain can span diameters up to several hundreds of kilometres with vertical heights of 4
to 6 km. Convective rains, often associated with thunderstorm events, are of much
smaller horizontal extents, usually only several kilometres but can extend to much
greater vertical heights because of convective upwelling [6].
The classical development for the determination of the radio wave attenuation due to
rain assumes the following:

The intensity of the wave decays exponentially as it propagates the volume
of the rain.

The water drops are spherical, and

The contributions of each drop to attenuation are additive and independent
of the other.
To determine the specific attenuation let us consider a plane wave of transmitter
power P t incident on a volume of uniformly distributed spherical water drops, all of
radius ‘r’, extending over length L. The received power P r will be
Pr  Pt e  kL
(2.3)
29
Chapter 2: Radio wave propagation
where k is the attenuation coefficient for the rain volume, expressed in units of
reciprocal length.
The attenuation of the wave is expressed as a positive decibel value given by
A  10 log 10
Pt
Pr
Converting the log to the base e
A  10 log 10 e kL
= 4.343kL
(2.4)
The attenuation coefficient k is expressed as
k  Qt
(2.5)
Where ρ is the drop density i.e. no of drops per unit volume, Q t is the attenuation
cross-section of the drop radius r, expressed in units of area, which is the sum of a
scattering cross-section Q s and an absorption cross-section Q a .
The attenuation cross-section is a function of the drop radius r, the wavelength of the
radio wave, and the complex refractive index m of the water drop. That is
Qt  Qt (r ,  , m)
30
Chapter 2: Radio wave propagation
The drops in a real rain are not all of uniform radius and the attenuation coefficient
must be determined by integration over all drop sizes. The size distribution of
raindrops represents the number of raindrops with an equivalent radius between r and
r+dr per unit volume (m 3 ), and is written in the form n(r)dr. In terms of these, the
attenuation coefficient is
k   Qt (r ,  , m)n(r )dr
(2.6)
The specific attenuation in decibels per kilometre is, with L = 1 km
  4.343 Qt (r ,  , m)n(r )dr
(2.7)
The above equation emphasizes the dependence of rain attenuation on drop size, drop
size distribution, rain rate and attenuation cross-section. The first three parameters are
characteristics of the rain structure only, and it is through the attenuation cross-section
that the frequency and temperature dependence of rain attenuation is included.
Considering the type of rain and the rain regime of the region, the drop size
distribution have been found to be well approximated by an exponential of the form
n ( r )  N o e  D
where N 0 and Λ are experimentally determined constants, whose values depend on the
nature of the rain under consideration. Equation (2.4) can therefore be written as
31
Chapter 2: Radio wave propagation
  4.343N 0  Qt (r ,  , m)e  r dr
(2.8)
where Q t is found by employing the classical scattering theory of Mie for a plane
wave radiation upon an absorbing sphere as
Qt 
2
2

 (2n  1) Re[ a
n 1
n
 bn ]
a n and b n are the Mie scattering coefficients, which are complex functions of m, r
and λ.
By introducing the frequency and temperature dependent coefficients which
approximately represent the complex behaviour of the complete representation of the
specific attenuation in equation (2.3), the relationship between rain rate, as measured
on the earth’s surface, and specific attenuation can be approximated by
  aR b
(dB/km)
(2.9)
The total rain attenuation for an earth-space slant path is thus obtained by integrating
the specific attenuation over the total path L
L
A(dB)   dx
(2.10)
0
32
Chapter 2: Radio wave propagation
where the integration is taken over the extent of the rain volume in the direction of
propagation [1].
The significant atmospheric impairments encountered by a radio signal along the path
between an earth station and a satellite are summarized in Table 2.1 [7].
Table 2.1: Propagation concerns for satellite communication systems
Propagation
impairment
Attenuation and sky
Physical cause
Prime importance
Atmospheric gases,
Frequencies above about 10
noise
cloud, rain
GHz
Signal depolarization
Rain, ice crystals
Dual-polarization systems at
C and Ku bands (depends on
system configuration)
Refraction, atmospheric
Atmospheric gases
multipath
Communication and
tracking at low elevation
angles
Signal scintillations
Tropospheric and
Tropospheric at frequencies
ionospheric refractivity
above 10 GHz and low-
fluctuations
elevation angles;
ionospheric at frequencies
below 10 GHz
Reflection multipath,
Earth’s surface, objects
blockage
on surface
Propagation delays,
Troposphere, ionosphere
variations
Mobile satellite services
Precise timing and location
systems; time division
multiple access system
Intersymbol interference
Ducting, scatter,
Mainly C band, rain scatter
diffraction
may be significant at higher
frequencies
33
Chapter 2: Radio wave propagation
As rain is highly variable in space and time, the value of the total attenuation can be
determined only from the knowledge of the characteristics at each point of the path of
the raindrops. While measurements of rain attenuation have been realised, either with
satellite beacons or with radiometers, these measurements are temporally and spatially
scattered and are severely limited as regards frequency. They cannot therefore be
readily generalised at all the places around the globe. This has led to the development
of several different models based on physical processes and on meteorological data,
more particularly on the cumulative distribution of rain intensities, for the evaluation
of the link margins to be applied in the deployment of telecommunication systems [2].
2.7
Conclusions:
Different interaction mechanisms and their effects on radio signal have been discussed
and the range of tropospheric and ionospheric effects to be expected on earth-space
link operating at different frequencies outlined. The discussion emphasised in greater
detail the computation of rain attenuation since rain is the most significant impairment
factor at frequencies above 10 GHz.
The next chapter focuses on the work carried out during the course of this research to
obtain an improved location-specific estimation of rain attenuation for Bangladesh
and other regions of similar climate.
34
Chapter 2: Radio wave propagation
References:
1. L. J. Ippolito, Radiowave Propagation in Satellite Communications,
Van Nostrand Reinhold Company Inc. 1986.
2. H. Sizun, Radiowave Propagation for Telecommunication
Applications, Springer, 2005.
3. J.E. Allnut, Satellite-to-ground Radiowave Propagation, Peter Peregrinus Ltd,
1989.
4. M.P.M. Hall,L.W.Barclay and M.T. Hewit, Propagation of Radio waves, The
Institution of Electrical Engineers, 1996.
5. T. Pratt, C. Bostian and J. E. Allnut, Satellite Communications, John Wiley and
Sons Inc., 2000.
6. H. J. Liebe, G. A. Hufford and M. G. Cotton, “Propagation Modelling of Moist
Air and Suspended Water/ice Particles at Frequencies below 1000 GHz”, AGRAD
52nd specialist meeting of the EM wave propagation panel, Palma de Maiorca.
7. Salonen et al., “Study of Propagation Phenomena for Low Availabilities”,
ESA/ESTEC contract 8025/88/NL/PR, Final report.
8. ITU-R, “Attenuation by atmospheric gases”, Rec. ITU-R P.676-7, 2007.
9. L. J. Ippolito, “Radio Propagation for Space Communications System”,
Proceedings of the IEEE, Vol.69, No.6, June 1981, pp. 697-727.
10. D Roddy, Satellite Communiactions, McGraw-Hill, 1989.
35
Chapter 3: Characterisation of rain attenuation in Bangladesh
3 CHAPTER 3
Characterisation of Rain Attenuation in Bangladesh
36
Chapter 3: Characterisation of rain attenuation in Bangladesh
3.1
Introduction:
For a reliable quantification of rain attenuation in an earth-space link, determination
of the temporal, year-to-year variation of rainfall is required as the distribution of
rainfall is not uniform over the year. Bangladesh, a tropical region experiences high
annual rainfall. Propagation impairments, especially rain attenuation have a
significant impact on radio communication links in tropical regions [1]. However, rain
is not always present so that propagation impairments have a significant impact only
for a certain percentage of the time during a year. The different available empirical
models for rain attenuation prediction produce different estimates of the long-term
mean fade probability since these models are derived from a few measurement points
around the world over limited time periods whereas rain is a natural phenomenon
which varies from location-to-location and from year-to-year.
Bangladesh has a tropical monsoon climate, the main seasons being Winter (Nov Feb), Summer (Mar - Jun) and Monsoon (Jul - Oct). Most of the rainfall occurs during
monsoon. To design earth-satellite communication link for a tropical region,
modelling of rain fading is central to accurate calculations of signal power budget for
links operating above 10 GHz.
This chapter reports on the characterisation of rain attenuation using daily
accumulated rainfall data collected over the period 1961 to 2003 from different rain
gauge stations of the Bangladesh Meteorological Department. The data were analysed
to obtain a reliable mean annual rainfall depth. Eight rain gauge stations were selected
in different parts of the country (Figure 3.1). The rain gauges are manual-type 203
mm-diameter gauges used for daily rainfall measurement. Two percent of the rainfall
37
Chapter 3: Characterisation of rain attenuation in Bangladesh
data were missing in four of the selected stations. In this case, missing data gaps were
filled by linear interpolation. The Rice-Holmberg model is then applied to find the
rain rate distribution from annual rainfall depth. Ten second integrated rainfall data
covering a one year period from February 2000 to January 2001 and recorded in
Sparsholt, UK by a drop counting rain gauge were employed to validate the
cumulative rain rate distribution for Bangladesh derived using the Rice-Holmberg
procedure.
Rain attenuation level over the region is predicted using the derived rain rate. An
improved rain attenuation prediction model is then devised based on the predicted and
measured attenuation at Sparsholt, UK. The uncertainty in predicted attenuation is
characterised by an ad hoc model.
3.2
Rainfall characterisation:
The attenuation and rain rate processes are stochastic with a physical cause and effect
relationship between rain rate and attenuation [2]. To characterise the rain attenuation
over a certain region, it is worthwhile to examine the pattern and periodicity of
rainfall in that region which might reveal any significant geographical as well as
seasonal variations within the region.
3.2.1
Seasonal precipitation trends:
The seasonal trend of rainfall can be assessed from the monthly precipitation data. For
this analysis, the mean monthly precipitation, M m (mm) for m = 1,2,3,…,12 obtained
38
Chapter 3: Characterisation of rain attenuation in Bangladesh
over N = 43 years of data is determined. Also, the mean annual precipitation M (mm)
is calculated for the same data. We can define a ratio
Figure 3.1: Rain gauge stations over Bangladesh
(source: hydro.iis.u-tokyo.ac.jp/GAME-T/GAIN-T/map/raingauge_bmd.html)
39
Chapter 3: Characterisation of rain attenuation in Bangladesh
rm =
Mm
M
(3.1)
which represents the fraction of precipitation during an average year that can be
expected in the m th month.
Here, data for 8 stations (Figure 3.1) are analysed. The plot of r m in Figure 3.2 shows
that the seasonal precipitation has similar trend for an average year in all parts of the
country though some of the stations have higher rainfall than the others. In particular,
the south-east part of the country has the highest rainfall, whereas the central part has
the lowest rainfall and the northern part has rainfall levels in between.
Figure 3.2: Seasonal rainfall trend in different stations over Bangladesh (19612003)
40
Chapter 3: Characterisation of rain attenuation in Bangladesh
3.2.2
Annual precipitation trend (Dhaka):
The time series of annual rainfall of Dhaka is plotted for this analysis. From the year
1961 to 1980, the annual rainfall ranges between 1500 mm and 2400 mm. But from
1980 to 2003 the annual rainfall range is much wider, from as high as 3000 mm to as
low as 1169 mm (Figure 3.3). A moving average processing with window length 5
years was applied to smooth the annual rainfall time series in order to reveal the long
term rainfall trend in Bangladesh. This is shown in Figure 3.3.
Figure 3.3: Annual rainfall time series in Dhaka (1961-2003)
It is interesting to observe that the rainfall trend exhibits an oscillatory behaviour
superimposed on an overall small positive slope of 0.68 mm/year (Figure 3.3). This
means that the annual rainfall is increasing by 0.68 mm per year. However, over
smaller interval of 5 to 8 years, the annual rainfall trends vary between positive and
41
Chapter 3: Characterisation of rain attenuation in Bangladesh
negative slopes. More specifically, from 1961 to 1969, the annual rainfall has a
negative slope. From 1970 to 1975, the trend has a positive slope. From 1976 to 1981,
the trend is negative and from 1982 to 1988, it is positive. From 1989 to 1994 the
annual rainfall shows positive trend and from 1994 to 2004, the trend is negative
(Figure 3.4).
Figure 3.4: Annual rainfall trend in Dhaka (a) 1961-1969 (b) 1970-1975 (c) 197681 (d) 1982-88 (e) 1989-94 (f) 1995-03
For these periods we can calculate the mean and variance in order to examine the
variability of annual rainfall. As can be seen from the Table 3.1, the mean rainfall
over different periods are different over the years with the highest value in the period
1982-1988. The coefficient of variation and the standard deviation are higher in the
period of 1989-1994, which are 34.64 and 699.03 respectively. That is the annual
rainfall is more variable in that period.
42
Chapter 3: Characterisation of rain attenuation in Bangladesh
Table 3.1: Statistical parameters for different periods of Annual rainfall in
Dhaka
Parameter
1961-69
1970-75
1976-81
1982-88
1989-94
1995-03
Overall
(1961-2003)
Mean(mm)
1954
2047.2
2045
2347
2018
1968.8
2055.7
Standard
227.46
247.18
209.35
389.86
699.03
261.61
365.34
11.64
12.07
10.24
16.61
34.64
13.29
17.77
Deviation
Coefficient
of variation
(std/mean)×100
Figure 3.5: Power spectral density of annual rainfall in Dhaka (1961-2003)
43
Chapter 3: Characterisation of rain attenuation in Bangladesh
It will be interesting to see the hidden periodicity in the observational record or time
series. Fourier analysis allows us to analyse various frequency components. To find
the periodicity of rainfall data in hand, we analyse the time series of annual rainfall by
applying Fourier transform. As the data is of short period, noticeable spectral leakage
occurs due to abrupt truncation of time series. There are different windowing
techniques that smooth the sharp transitions and reduce spectral leakage. This is
achieved by making the values of the signal nearly the same at the two ends of the
measurement interval. This value is usually zero or close to zero [3]. Different
windowing techniques such as Hamming, Kaiser, Hanning were applied to reduce the
sidelobe. Hanning window reduces spectral leakage significantly. The dominant
period was found to be 3.31 years, which is reciprocal of the horizontal axis point in
Figure 3.5 corresponding to a peak in the power spectral density. That is, the rainfall
pattern (observed over 43 years) repeats about every 3 to 4 years. This means that a
peak in rainfall occurs on average every 3-4 years, and similarly for a trough.
Although rainfall is a continuous unpredictable time series, 43 years of observation
period can be considered long enough to reflect its inherent characteristics such as
periodicity.
Whether the observed periodicity of annual rainfall is related with the sunspot activity
or the El-Nino Southern Oscillation (ENSO) could not be assessed in the present
thesis, although a study of Tropical Rainfall Measuring Mission (TRMM)
observations by Haddad et al. [4] appears to indicate that El-Nino is a significant
driver of year-to-year annual variability of global rainfall. However, TRMM was
launched only in 1997 and the measurement duration is therefore insufficient to give a
reliable indication of annual rainfall periodicity.
44
Chapter 3: Characterisation of rain attenuation in Bangladesh
3.3
Annual and monthly rainfall distribution:
To describe the random variation that occurs in the data from many physical
processes, the Gaussian (normal) distribution is most often assumed. The bell-shaped
distribution can be described by two parameters [5], namely the arithmetic mean, μ
and the standard deviation, σ. The data sets are therefore commonly described by the
expression μ± σ.
According to the central limit theorem, the probabilities approach a normal density
function for a large number of sample data. In its simplest form, this mathematical
law states that the sum of many (n) independent, identically distributed random
variables is, in the limit n→∞, normally distributed. The equation of the normal
probability density is
1 x
(
1
f ( x;  , ) 
e2
2
2

)2
,  n  
(3.2)
where,  and  are distribution parameters.
In our analysis, the population of annual rainfall have a normal frequency curve. As
an averaged quantity, the distribution of annual rainfall is predicted by the central
limit theorem, assuming sufficiently light tails to be Gaussian. It may therefore be
desirable to use a goodness-of-fit test to establish whether or not the annual rainfall
data are drawn from a normal distribution. To compare an observed frequency
distribution with the corresponding values of theoretical distribution, the statistics for
goodness-of-fit test is
45
Chapter 3: Characterisation of rain attenuation in Bangladesh
( o i  ei ) 2
 
ei
i 1
k
2
(3.3)
where o i and e i are the observed and expected frequencies. The sampling distribution
of this statistics is approximately the chi-square distribution with k-m degrees of
freedom, where k is the number of terms in the formula for  2 and m is the number
of quantities, obtained from the observed data that are needed to calculate the
expected frequencies. The interval of the class value i.e. bin size plays crucial role in
deriving the distribution. An algorithm was developed to find optimum bin size (14
bins of 133 mm annual rainfall) needed for the distribution to best fit into the
population frequency curve.
Figure 3.6: Annual rainfall distribution in Dhaka
46
Chapter 3: Characterisation of rain attenuation in Bangladesh
We found that the annual rainfall distribution follows Normal distribution with mean
μ= 2055.7 and standard deviation σ= 365.34 at the 6% significance level (Figure 3.6).
This means that there is a 6% chance of rejecting the null hypothesis that the
measured data are normally distributed, when in fact the measured data have the
normal distribution. Recall that for the normal distribution 68% of the samples (in this
case, observed annual rainfall) will lie within one standard deviation of the above
mean, i.e., within the range μ± σ, 95% will lie within μ± 2σ and 99.7% within μ± 3σ.
The year-to-year variation of annual rainfall is seen in Figure 3.3.
Figure 3.7: Rainfall distribution for the month July in Dhaka
The monsoon rainfall was also found to be normally distributed. Likewise the
distributions for different months were examined and the months of April, May, June,
July, August, September and October were found to have normally distributed
rainfall. Figure 3.7 shows the rainfall distribution for the month July.
47
Chapter 3: Characterisation of rain attenuation in Bangladesh
The occurrence of rare events usually leads to an exponential distribution given by
f ( x) 
1

exp(
x

)
for x>0, β>0
(3.4)
which has mean μ = β and variance σ 2 = β 2 .
Figure 3.8: Rainfall distribution for the month January in Dhaka
We found that the monthly rainfall for November, December, January, February and
March (months during which there is usually little rain) followed exponential
distribution. Figure 3.8 shows the monthly rainfall distribution for January, which
passed the goodness-of-fit test at the 6% significance level.
48
Chapter 3: Characterisation of rain attenuation in Bangladesh
3.4
Rain rate distribution:
The probability of occurrence of rain of a specified intensity at a specified
geographical location is represented by a distribution with parameters that may
change from one month to the next but be stationary from one year to the next. A
record of rain rate collected as a function of time (a time series) can be used to
compile a histogram of the occurrences of rain rates of different intensities. The
histogram can be summed to generate an ordered, cumulative distribution of the
measured occurrences of the different intensities, which can be used as an estimate of
the probability distribution for the rain process [6].
Since only daily accumulated rainfall data were available over Bangladesh, the RiceHolmberg model was applied to obtain the distribution of rain rate. The parameters
used are average annual rainfall depth M and thunderstorm ratio β, which is the
fraction of total annual rainfall that is of convective type. The monthly rainfall data
were used to determine M whereas β is taken from a map accompanying the RiceHolmberg model [7].
Rain rate distribution obtained by the Rice-Holmberg model is compared against
measured data to see how accurately the model can estimate the rain rate distribution.
For southern England, fine resolution like 10 s rain rate data is available. The 10 s rain
rate data from February 2000 to January 2001 is converted into 1-min rain rate as:
R1 min 
1 5
 R10 (k )
6 k 0
(3.5)
49
Chapter 3: Characterisation of rain attenuation in Bangladesh
where R 10 (k) means measured instantaneous rain rate by the rain gauge sampled at 10
seconds. By using the average annual rainfall depth of 682 mm and thunderstorm ratio
of 0.1 in the southern England, rain rate distribution is obtained by applying the RiceHolmberg model. From the plot (Figure 3.9), comparison of measured rain rate and
estimated rain rate for different percentages of an average year is tabulated in the
Table 3.2 [8].
Figure 3.9: Measured annual rainfall statistics in comparison with predicted rain
rate by Rice-Holmberg model for Sparsholt, UK and Surabaya, Indonesia
50
Chapter 3: Characterisation of rain attenuation in Bangladesh
Table 3.2: Comparison of ITU-R and R-H derived rain rates against
measurement at Sparsholt (UK)
% of time
ITU-R
R-H derived
Measured rain rate
% error
rain rate(mm/h)
rain rate(mm/h)
(mm/h)
(Measured Vs R-H)
.001
59.51
102
77
32.47
.01
21.29
28.7
29
1.03
.1
6.81
10.7
10
7
The analysis shows Rice-Holmberg model provide a good estimate of the rain rate
distribution in the 0.01-0.1 percent range. However, for smaller percentages of the
year, Rice-Holmberg model overestimates the rain rate. This result is consistent with
the different studies [9] [10] [11] that averaging affects only the higher rain rates at
the smaller percentages of the year.
Table 3.3: Comparison of ITU-R and R-H derived rain rates against
measurement at Surabaya (Indonesia)
% of time
ITU-R
R-H derived
Measured rain rate
% error
rain rate(mm/h)
rain rate(mm/h)
(mm/h)
(Measured Vs R-H)
.001
164.8
205.9
175
17.14
.01
100.2
129
120
7.5
.1
37.58
52.1
50
4.2
51
Chapter 3: Characterisation of rain attenuation in Bangladesh
Surabaya, Indonesia has similar climatic conditions to Bangladesh. For Surabaya, the
mean annual rainfall depth is 2000 mm and thunderstom ratio is 0.7 [12]. Rain rate
distribution was derived using these values in the Rice-Holmberg model (Figure 9).
The measured rain rates (based on 1-minute rainfalls) at different time percentages
were read from Surayana, 2005 [13]. Again we see from Table 3.3, that the RiceHolmberg model provides a good estimate of rain rate distribution in the 0.01-0.1
percent range.
The Rice-Holmberg model is therefore applicable to produce the rain rate distribution
of a region from its accumulated rainfall data. Different stations were chosen scattered
over Bangladesh to find the rain rate distribution by applying Rice-Holmberg model.
The parameters of Rice-Holmberg model for the different stations are given in Table
3.4.
Table 3.4: Rice-Holmberg parameters for different stations in Bangladesh
Parameters
Dhaka
Chittagong
Rangpur
Bogra
Ishurdi
Barisal
CoxsBazar
Srimangal
M (mm)
2055
2696
2087
1635
1353
1988
3483
2013
β
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
52
Chapter 3: Characterisation of rain attenuation in Bangladesh
Figure 3.10: Rain rates for different stations (Bangladesh) by Rice-Holmberg
model
Figure 3.10 shows the rain rate distribution obtained by Rice-Holmberg model for
different stations and rain rate distribution from ITU. Table 3.5 compares the rain rate
distribution for different stations at 0.01% of time. It is of interest to quantify the
year-to-year variation of the Rice-Holmberg-derived distribution which is
characteristic of an average year. This may be accomplished in part by assessing the
year-to-year variation of M and . The former has a normal distribution (as earlier
shown), whereas the latter has a value of β = 0.5 (as read from [7]). However, due to a
lack of local measurements, the variability of β could not be assessed.
53
Chapter 3: Characterisation of rain attenuation in Bangladesh
Table 3.5: Comparison of the 0.01% rain rates of ITU-R and R-H computations
Station
Latitude
Longitude
ITU(mm/h)
R-H(mm/h)
% error
(DegreeN) (DegreeE)
Dhaka
23.723
90.4086
109.73
119
8.45
Chittagong
22.27
91.82
128.36
127.4
0.75
Rangpur
25.73
89.23
95.45
119.3
24.99
Bogra
24.85
89.37
101.06
111
9.84
Ishurdi
24.13
89.05
99.15
104.7
5.6
Barisal
22.75
90.33
113.7
118
3.78
CoxsBazar
21.43
91.93
125
137
9.6
Srimangal
24.3
91.73
100.13
117.5
17.35
Analysis of M shows that the year-to-year annual rainfall M is normally distributed
i.e. the long term mean annual rainfall lies within the 95% confidence bounds for the
true mean. According to the world map of β [7], the value is 0.5 i.e. 50% rain is of
convective type for Bangladesh. However, due to the lack of local measurement,
corresponding estimate of the accuracy of β was not obtained.
3.4.1
Seasonal variations in rain rate distribution:
Average annual rainfall depth, M partially reflects annual variation in rainfall as longterm data (i.e. at least 11 year period) is used. But within the annual variations there
are seasonal as well as diurnal variations. To characterise diurnal variations in rain
54
Chapter 3: Characterisation of rain attenuation in Bangladesh
rate distribution, fine resolution like 1min data are required, which are not available
over Bangladesh.
However, from the knowledge of seasonal trend of rainfall, it is expected to have a
higher rain rate distribution during monsoon season and a comparatively lower
distribution for winter. In Bangladesh, most rainfall occurs during monsoon season.
July is the rainiest month, whereas January is the driest month (Figure 3.2).
Figure 3.11: Seasonal variations of rainfall rates by Rice-Holmberg model in
Dhaka
To compare the rain rate statistics of different months, Rice-Holmberg model is
applied to the monthly average by combining 43 years of monthly data such as
January, July etc. Figure 3.11 shows the rainfall distributions for the months of
January and July. At 0.01% of time rain rate exceeded is 144 mm/h for July compared
55
Chapter 3: Characterisation of rain attenuation in Bangladesh
to only 14.2 mm/h for January. In fact July singularly provides the worst-month
statistics in Bangladesh as its rain rate value exceeds that of every other month at all
percentage points. To provide a service such as Broadcasting Satellite Service (BSS)
at the required availability (usually 1% of the worst month), worst-month statistics
has to be considered.
3.5
Rain attenuation:
On the basis of the rain rate analysis, slant path rain attenuation is predicted using the
18.7 GHz frequency band for a path elevation of 46.1 0 . The station (Dhaka) location
is 23.72 0 north latitude and 90.41 0 east longitude at an altitude of 6.5 m above sea
level. Rain attenuation statistics is predicted for this slant path using ITU-R model,
Rec. 618-8 [14], which employs the rain rate at 0.01% of an average year. At 0.01%
of time, the predicted attenuation using ITU rain rate is 37 dB exceeded for 0.01% of
an average year. However, using Rice-Holmberg derived rain rate, the exceeded
attenuation is 39 dB for the same time percentage (Figure 3.12).
56
Chapter 3: Characterisation of rain attenuation in Bangladesh
Figure 3.12: Predicted rain attenuation using R-H and ITU rain rate for Dhaka
As the empirical models for rain attenuation are derived from a few measurement
points around the world over limited time periods, variability is inherent in the
estimation of rain attenuation for a particular location. Rain attenuation can vary from
location-to-location and from year-to-year. According to the Sparsholt measurement
at 18.7 GHz from April 1997 to July 2001, the year-to-year measured attenuation
statistics are different though the concurrent year-to-year rain statistics are similar
[15]. The year-to-year variation could be due to the way intense rain is distributed
along the slant path.
57
Chapter 3: Characterisation of rain attenuation in Bangladesh
Figure 3.13: Approximation of attenuation reduction factor using Sparsholt data
It was important to obtain a measure of how closely the rain attenuation distribution
predicted using Rice-Holmberg (R-H) derived rain rates matches with a direct
measurement of rain attenuation distribution using a satellite beacon. We determined
this measure, referred to as a reduction factor r, at Sparsholt (UK) where both
distributions are available, and defined as
r = Predicted attenuation / Measured attenuation
58
Chapter 3: Characterisation of rain attenuation in Bangladesh
This was found to be well approximated (standard error of 1.55%) by (Figure 3.13)
 3.928558  10 3
r  0.26478 exp 
 12.955 p  ln
p



5.233166  10 4
p  0.29566 
 7.688 p 
p


(3.6)
where p is the annual time percentage (0.002 ≤ p ≤0.05).
This factor was then applied to improve the R-H based prediction of rain attenuation
in Bangladesh. The assumption made here is that the R-H model performs similarly in
both climates. This is a reasonable assumption given that the R-H model has
comparable global applicability (e.g. see Figure 9) and has hitherto not been shown to
perform significantly differently in any region. As a result we obtain the attenuation
distribution in Figure 3.14 (dashed line), which is compared with attenuation
distributions predicted using R-H and ITU-R rain rates, and shows lower attenuations
(by up to ~18%) at small time percentages (< 0.01%).
It is worth mentioning that at frequencies below around 30 GHz rain type plays a
significant role in determining the size of attenuation. For smaller percentages of the
year, the attenuation results mainly from thunderstorm rain. It is likely that the ITU-R
rain attenuation model (used for the two solid curves of Figure 3.14) somewhat overestimates the attenuation for Bangladesh at these small percentages of the year. This
may be due to the fact that the models for rain drop size distribution (DSD) in tropical
regions are quite different from those adopted by ITU-R [16]. An intense rain rate
could result in a lower than predicted rain attenuation if the DSD consists of
59
Chapter 3: Characterisation of rain attenuation in Bangladesh
predominantly larger drops. Maitra found that the prevailing DSD in a tropical region
can cause significant variation in rain attenuation distribution [17].
Figure 3.14: Comparison of predicted attenuation applying reduction factor
(Dhaka)
To further check the validity of the above improved attenuation prediction for
Bangladesh, comparison was made with measured attenuation values at 12 GHz
reported in the literature for a similar climatic region (tropical- monsoon characterised
by heavy rainfall), namely Bandung, Indonesia [13]. The measured attenuation at
0.01% of the time was 17 dB, which scales to 35 dB at 18.7 GHz, using the ITU-R
frequency scaling procedure [14]. This value is comparable to the 34 dB attenuation
predicted for Dhaka by the improved model.
For some specific applications such as BSS, it is often necessary to determine worst
month statistics from annual statistics, since an annual distribution may be the only
60
Chapter 3: Characterisation of rain attenuation in Bangladesh
statistics available. To relate worst month and annual statistics, the ITU-R model [18]
provides the following expression
pa  0.30 pw
1.15
(3.7)
where p a is annual exceedance probability and p w is worst month exceedance
probability. According to the above relationship, 1% of the worst month value
corresponds to 0.3% of the annual value. That is, to determine the margin necessary to
maintain an outage no greater than 1% of the worst month, a 0.3% outage of
attenuation margin should be read from the available annual distribution plot. Figure
3.14 shows that for the link under consideration this margin would be 9 dB.
Monthly attenuation values were predicted by using 0.01% rain rate for each month
calculated as discussed in section 3.1. Figure 3.15 gives a plot of the rain fade margins
required to meet various annual availabilities (99 - 99.99%) at each month, assuming
the rainfall pattern of that month was characteristic of the whole year. There is clearly
a strong seasonal dependence. For example a margin of ~40 dB is required to achieve
99.99% availability during the monsoon months, compared to only ~11 dB in January.
Thus year-round operation of high-availability links will not be feasible in
Bangladesh at Ku-band frequencies and above without recourse to uneconomic
system design and/or sophisticated fade mitigation techniques. However the results of
Figure 3.15 indicate that these frequencies can be exploited in Bangladesh to provide
high-availability communication links to support non-monsoon seasonal applications
such as social or sporting activities.
61
Chapter 3: Characterisation of rain attenuation in Bangladesh
Figure 3.15: Attenuation for various levels of availabilities as a function of the
month
3.6
Risk estimation in rain attenuation prediction:
The occurrence of attenuation by rain is a random process and each single-year
distribution function is a sample from that process [2]. Based on the predicted rain
rate, rain attenuation is estimated for Dhaka, which reflects the attenuation levels at
the required time percentages for the Earth-space link design. But there is always
inherent statistical uncertainty in the attenuation prediction due to random behaviour
of rain process.
To estimate the risk to be associated with a rain attenuation prediction model, an ad
hoc model is applied as by Crane [2]. As an ad hoc procedure for the estimation of the
62
Chapter 3: Characterisation of rain attenuation in Bangladesh
uncertainty associated with an attenuation prediction, the observed deviation from
model predictions are used to estimate the distributions of deviations to be expected
for a path.
Risk, R, is often computed as the probability of a failure, F N (in prediction),
occurring at least once in a period of N years. If the object of an attenuation prediction
is the fade margin required to prevent a failure for all but one in N years, then the
return period P needed for the calculation of the attenuation cumulative distribution
function for design is found from:
R = F N = 1-(1- F 1 ) N
F 1 =1-(1-R) 1 / N
(3.8)
P=1/ F 1
where F 1 is single year probability of failure. The attenuation for the specified risk,
A R , is calculated from
A R = A exp(σS m )
(3.9)
where A is the attenuation values, σ is the standard deviation relative to a single year
probability of failure found from F 1 by look up in a normal probability table and S m
is 0.23 for the year-to-year variation of path attenuation at a single location and 0.29
for the combined year-to-year and location-to-location variations for path attenuation
within climate zone [2].
63
Chapter 3: Characterisation of rain attenuation in Bangladesh
Figure 3.16: Rain attenuation risk estimation for Bangladesh
Figure 3.16 shows the upper bound (5%) and lower bound (95%) of rain attenuation
obtained by applying an ad hoc model on the predicted rain attenuation after applying
reduction factor. According to the graph, median value of the rain attenuation for
0.01% of an average year exceeded is 34 dB. However, 5% of observed years of
measurement will have an attenuation in excess of 56 dB at this time percentage,
whereas for 95% of observed years the exceeded attenuation will be 21 dB. There is a
risk of one failure in N = 5 years of attenuation observation (23 years return period).
The predicted rain rate by Rice-Holmberg model is 119 mm/h at 0.01% time.
Applying the ad hoc model, the upper bound of rain rate is found to be 186 mm/h and
the lower limit is 76 mm/h at the same percentage of time. These rain rates correspond
64
Chapter 3: Characterisation of rain attenuation in Bangladesh
to attenuation of 50 and 30 dB, respectively. The slight difference in the variability is
due to the fact that the occurrence of year-to-year rain rate and attenuation is not
similar.
Figure 3.17: Rain attenuation variability from three standard deviations bound
for Bangladesh
In our analysis annual rainfall is normally distributed. Using the mean rainfall at three
standard deviations bound, the attenuation is predicted to assess the variability of
attenuation (Figure 3.17). According to the graph, the attenuation is 34 dB in the
lower bound and 41 dB in the upper bound at 0.01% time. Although the annual
rainfall is normally distributed, the rain rate which causes the attenuation is lognormally distributed [2]. As a result, the variability of mean annual rainfall does not
reflect the true variability in attenuation.
65
Chapter 3: Characterisation of rain attenuation in Bangladesh
3.7
Conclusions:
Accurate prediction of rain attenuation is crucial in the planning of a reliable
communication system at any location especially in tropical regions like Bangladesh.
The analysis of rainfall pattern reveals that the rainfall has similar pattern in different
parts of the country and is highly seasonal. Most of the rainfall occurs during the
monsoon (Jul-Oct) period, whereas in winter (Nov-Feb) there is very little rain.
Efficient utilisation of satellite communication system resources can be improved by
using seasonal information along with annual statistics in the link design process. For
example the transmit power of earth station and/or satellite (if on-board processing)
can be set at two different levels, one for the rainy months, and a much lower setting
for the dry winter months. This would be a simple seasonal link budget approach, as
opposed to a more complex fade mitigation strategy through power control which
necessitates some provision for monitoring changes in link attenuation.
The Rice-Holmberg model was used to convert local measurement of annual rainfall
into rain rate distribution. Based on this R-H derived rain rate, the predicted
attenuation was 39 dB at 0.01% time whereas based on ITU-R rain rate, the rain
attenuation was 37 dB. By applying the improved rain attenuation model presented,
the predicted rain attenuation for Dhaka was found to be 34 dB for the same time
percentage.
Typically, power margins of 5-10 dB at C-band and 10-15 dB at Ku, Ka-band can be
relatively easily achieved by available power margins (obtained with reasonably sized
antennas and with RF transmit power). However, our study suggests that rain
attenuation will exceed available power margins for tropical regions like Bangladesh
during a significant percentage of the year. Additional methods such as sophisticated
66
Chapter 3: Characterisation of rain attenuation in Bangladesh
fade mitigation strategies, and innovative solutions such as seasonal link budgets or
seasonal communication applications need to be considered in order to achieve
acceptable link availabilities at affordable costs.
Rain attenuation level together with the seasonal characteristics of rainfall will be
incorporated in the link budget analysis of the communication network in chapter 8.
67
Chapter 3: Characterisation of rain attenuation in Bangladesh
References:
1. G.H. Bryant, I. Adimula, C. Riva and G.Brussard, “Rain Attenuation Statistics
from Rain Cell Diameters and Heights”, International Journal of Satellite
Communications, Vol.19, 2001, pp. 263-283.
2. R. K.Crane, “Estimating Risk for Earth-Satellite Attenuation Prediction”,
Proceedings of the IEEE, Vol.81, No.6, June 1993, pp. 905-913.
3. C. Schuler and M. Chugani, Digital Signal Processing: A Hands-on
approach, McGraw-Hill, New York, 2005.
4. Z. S. Haddad, J. P. Meagher, R. F. Adler, E. A. Smith, I. Eastwood and
S. L. Durden, “Global Variability of Precipitation according to the Tropical
Rainfall Measuring Mission”, Journal of Geophysical Research, Vol. 109, 2004.
5. L. L. Lapin, Probability and statistics for modern engineering, 1990.
6. R. K. Crane, “Prediction of Attenuation by Rain”, IEEE transactions on
communication, Vol. Com-28, No.9, Sept. 1980, pp. 1717-1733.
7. P. L. Rice and N. R. Holmberg, “Cumulative Time Statistics of SurfacePoint Rainfall Rates”, IEEE transactions on communication, Vol. Com-21, No.10,
Oct. 1973, pp. 1131-1136.
8. ITU-R, “Characteristics of precipitation for propagation modelling””, Rec. ITU-R
P.837-4, 2003.
9. R. R. Rogers, “Statistical Rainstorm Models: Their Theoretical and Physical
Foundations”, IEEE transactions on Antennas Propagation, Vol. AP-24, 1976,
pp. 547-566.
10. R. K. Crane, “Prediction of the Effects of Rain on Satellite Communication
Systems”, Proceedings IEEE, vol. 65, 1977, pp. 456-474.
11. D. M. A. Jones and A. L. Sims, “Climatology of Instantaneous Rainfall Rates”,
68
Chapter 3: Characterisation of rain attenuation in Bangladesh
Journal of Applied Meteorology, vol. 17, 1978, 1135-1140.
12. A. Dissanayake, J. Allnutt, and F. Haidara, “A Prediction Model that
Combines Rain Attenuation and Other Propagation Impairments along EarthSatellite Paths”, Online Journal of Communications, Issue No. 2, 2002, pp. 1-37.
13. J. Surayana, S. Utoro, K. Tanaka, K. Igarashi, and M. Iida, “Study of
Prediction Models Compared with the Measurement Results of Rainfall Rate and
Ku-band Rain Attenuation at Indonesian Tropical Cities”, ICICS, 2005.
14. ITU-R, “Propagation data and prediction methods required for the design of
Earth-space telecommunication systems”, Rec. ITU-R P.618-8, 2003.
15. S. Ventouras, C. L. Wrench, “Measured Slant Path Attenuation and Rainfall
Statistics in Southern England in Relation to ITU-R Predictions”, COST Action
280, 1st International Workshop, July 2002.
16. G. O. Ajayi, and R. L. Olsen, “Modelling of a tropical raindrop size
distribution for microwave and millimetre wave applications”, Radio Science,
Vol. 20, 1985, 193-202.
17. A. Maitra, “Rain Attenuation Modeling From Measurements of Rain Drop
Size Distribution in the Indian Region”, IEEE Antennas and Wireless
Propagation Letters, Vol. 3, 2004, pp. 180-181.
18. ITU-R, “Conversion of annual statistics to worst-month statistics”, Rec. ITU-R
P.841-4, 2003.
69
Chapter 4: Rain cell size distribution
4 CHAPTER 4
Rain Cell Size Distribution
70
Chapter 4: Rain cell size distribution
4.1
Introduction:
Knowledge of rain cell size distribution is relevant for the modelling of earth-space
propagation in radio communication. To determine the spatial structure of rain cells,
long term rain rate time series can be processed by applying the synthetic storm
technique assuming some known value of storm translation speed. The underlying
hypothesis is that rain patterns move along a line with a constant speed and that
advection is the predominant mechanism to account for the spatial variability of rainrate. The hypothesis holds when a statistical description of rain structure is required,
rather than the exact space distribution of rain. Furthermore, as rainfall patterns move
over a rain gauge, it is possible to estimate the horizontal extent of rain cells from the
duration of various rain rate thresholds as recorded by the rain gauge if a mean
advection velocity of rain cells is assumed.
The weather radar is powerful in registering three-dimensional rain (reflectivity)
fields in time and space. Since it can provide a way to sample the spatial structure of
the rain fields with the spatial continuity, the radar data over a limited period of time
(like one rainy season) contain quite complete characterisation of rain structures and
types for the covered region [1].
Radar scans send out electromagnetic radiation that strikes hydrometeors in the
atmosphere, a part of which reflect back toward the radar. This backscattered energy
carries the characteristics of the reflecting bodies such as hydrometeor types, the size
of the hydrometeor and absorption qualities of the hydrometeors. Reflectivity fields of
precipitation from high resolution scanning radar show structures of various sizes
often embedded in each other. Although precipitation intensities are highly variable in
71
Chapter 4: Rain cell size distribution
space and time they manifest some organisation as higher intensity cores are clustered
in features of various scales of cells measuring a few square kilometres most of the
time. If a cell is sufficiently long-lived, it is possible to track their movement through
successive scans separated in time.
Scanning Doppler radar being coherent can measure the phase of the returning signal,
thus enabling the determination of the radial velocity component of the targets. Wind
movement is in general three-dimensional and varies over time and space. Complete
characterisation of the wind movement requires simultaneous measurements using
multiple Doppler radars. However, making some simplifying assumptions about the
structure of the observed wind field, a single Doppler radar measurement will suffice
to extract the wind field [2].
The main topic of this chapter is the rain cell size distribution obtained from rain
gauge and radar data. Reflectivity, dBZ (mm 6 mm 3 ) and Doppler radial velocity
(m/s) available in 300 range gates, each of 300 m length at 0.25 s interval from
Chilbolton Advanced Meteorological Radar (CAMRa) situated at Chilbolton
(51.1445 0 N, 1.4370 0 W), UK were used for 7 events on 26 September, 08 October
and 12 May in 2001, 01 July, 22 September and 31 October in 2003 and on 23 March
in 2004. The reflectivity data is calibrated and noise free. The unambiguous velocity
measured by Chilbolton radar is ±15 m/s and velocity measured beyond these ranges
is unfolded [3]. Rain gauge data by a drop counting rain gauge recorded at Sparsholt,
7.8 km from Chilbolton were used for the whole year of 2000, 2003, 2005 and
January 2004 to May 2004. Wind measurements at Brize Norton, 63 km from the
radar location, were also used for the corresponding periods of radar observation.
72
Chapter 4: Rain cell size distribution
4.2
Rain cell translation velocity:
Radar can provide information on the horizontal structure of storm (rain) cell from
PPI measurements of the radar reflectivity factor (Figure 4.1) through constant
elevation slices. Different definitions of rain cell are found in the scientific literature
with different meanings [4]. In Crane [5] rain cell refers to a volume in which
convective phenomena take place. In other approach, a rain cell is considered to be an
entity constituted by an area inside of which the rain rate (or the radar reflectivity) is
equal to or higher than a specified threshold value. This definition implies that the cell
is continuous and that along the contour that bounds it, the rain rate is at the threshold
value. The area where rain rate falls below threshold is ignored [6].
Figure 4.1: PPI of reflectivity field, dBz (colorbar on right) on 23 March,2004
A rain cell defined by the latter approach can be identified in consecutive PPI scans
separated by a fixed time interval. To track the movement of the rain cell a correlation
technique is employed, which partitions the reflectivity fields of each scan into blocks
and determines by trial and error the displacement of the previous scan that maximises
73
Chapter 4: Rain cell size distribution
its correlation with the next scan. Dividing this displacement by the inter-scan interval
yields the cell translation velocity. Cross-correlation techniques treat the data as a
two-dimensional field from which the movement of features may be inferred [7] [8]
[9] [10].
A correct match of each cell to its corresponding appearance in a subsequent scan is
complicated by the physical changes that rain cells exhibit between radar scans. For
example a rain cell might decay, grow, merge or split between observations.
Moreover, a rain cell might move independently in directions that differ from that of
the whole radar image. The movement of a rain cell can be either a propagation,
whereby a portion of the cell movement arises from growth on new echoes, or a
translation, which is the motion of the cell centroid not resulting from propagation
[11].
To identify a rain cell within the precipitation areas, a threshold reflectivity value of
35 dBZ was used [6]. When tracking a previous rain cell (at time t1) based on a
current scan (at time t2), difficulties arise as new cells can appear and existing cells
can split, merge or disappear. There are five possible scenarios [12]:
1. A cell at t2 has no predecessor at t1, which means a new cell came into existence.
2. A cell at t1 has no successor at t2, which means an existing cell disappeared.
3. A cell at t1 has exactly one successor at t2. This is cell translation.
4. A cell at t1 has more than one successor at t2 (i.e. the cell split into several parts).
5. A cell at t2 has more than one predecessor at t1 (i.e. several cells merged into
one).
74
Chapter 4: Rain cell size distribution
The above correlation analysis was performed only to find cell translation, ignoring
cell growth and cell decay (Appendix-A). A frame is taken centred at peak intensity
around an identified cell in the previous image (at time t1). The choice of the size of
this frame is guided by two considerations: A frame that is too small in size will
contain too few data points for the correlation coefficients to be stable, whereas a
frame that is too large will only give a general mean flow on a broad spatial scale. In
our study, a 5 km×5 km frame size, partitioned into 25×25 blocks, was considered a
good compromise in reliably tracking a cell. Each block was assigned a value of zero
if its reflectivity fell below threshold and a value of 1 otherwise. The search for
maximum correlation (i.e. approximate match) between a frame from the previous
scan and a neighbourhood in the current scan extends from zero displacement up to
an excursion limit set by a maximum translation speed of 30 m/s.
Figure 4.2: Identified rain cell on PPI on 23 March, 2004 (a) Cell in the previous
scan; (b) Correlated cell in the next scan
75
Chapter 4: Rain cell size distribution
As an example, Figure 4.2 shows two successive scans of a rain event on 23 March
2004 featuring isolated and intense rain cells (maximum reflectivity 51 dBZ). The
time difference between the successive scans is from 2 min 15 s to 3 min. The block
at the centre of a frame in the LHS scan (at time say t1) is marked, and has coordinates say (x1, y1). At a later time t2 corresponding to the RHS scan, the location of
this block will be at point (x2, y2) within the RHS scan which gives the highest
correlation between the frames centred at (x1, y1) and (x2, y2). The translation speed v
and direction of motion  (counter-clockwise from East) are then obtained as
v
 x2  x1    y2  y1  t2  t1  ;
2
2
  tan1  y2  y1   x2  x1 
(4.1)
Translation speeds and directions were similarly computed for all other events.
4.3
Wind field from Doppler analysis:
Browning and Wexler [13] showed that if a wind field varies almost linearly then the
velocity components of a wind field can be approximated by a Taylor series
expansion limited to first derivatives. In this way, the velocity field inside an observed
domain is described by the sum of the value at the centre and the gradient terms. As
the radar senses only radial velocity, the tangential component can be determined by
analysing measured radial velocity values at various points along a circle
corresponding to different azimuth β and range r values within a scan as in Figure 4.3.
Under these assumptions, the radial velocity V r (β) can be seen as a periodic function
with base period 2π (Appendix-B) that can be written in the form of a Fourier series
expansion [2]:
76
Chapter 4: Rain cell size distribution
Figure 4.3: Motion vector on an azimuth plane

Vr (  )  a 0   (a n cos n  bn sin n )
(4.2)
n 1
The first three coefficients are given by,
a0 
1
u v
r cos 2  (  )  w sin 
2
x y
(4.3)
a1  v0 cos 
(4.4)
b1  u0 cos 
(4.5)
where, u 0 and v 0 are the horizontal components of velocity, α is the elevation angle.
77
Chapter 4: Rain cell size distribution
Figure 4.4: Doppler velocity from event on 23 March, 2004 at a range 58 km
However, if there is no scatterer at a particular point on the circle or the scatterer
motion at this point is perpendicular to the radar beam, then the radial velocity will be
zero. Such points were filled by cubic interpolation. The data was smoothed (Figure
4.4) using an ideal low pass filter of cut-off 50 cycles/deg and an FFT was applied to
obtain the coefficients that yielded the horizontal wind speed as
a1  b1
2
VH  u o  vo 
2
2
2
cos 
(4.6)
The horizontal divergence is given by
u v
2


(a0  w sin  )
x y r cos 2 
(4.7)
78
Chapter 4: Rain cell size distribution
The average vertical velocity can be calculated from the continuity equation:
u v wa
 
0
x y z
(4.8)
Integrating between levels z and z 0
z
wa ( z )  wa ( z 0 )    (
z0
u v
 )dz
x y
(4.9)
Considering vertical velocity, wa at surface (level z 0 ) to be zero then vertical velocity
at any level z will be
wa ( z )  (
u v
 )z
x y
(4.10)
Vertical motion at top of any layer is proportional to layer mean divergence. If
divergence is negative, then vertical motion is positive, which implies rising motion
i.e. updraft. If divergence is positive, then vertical motion is negative, which implies
sinking motion i.e. downdraft. The vertical velocity ranges from 5 ~ 26 cm/s in the
events studied. The detail analysis of vertical velocity is not presented for brevity.
4.4
Comparison of cell translation speed, Doppler speed and wind speed:
Figure 4.5 shows comparisons of the radar Doppler-derived wind speed, the
correlation-based storm translation speed, and the ground measured wind speed at
Brize Norton for corresponding events or observation intervals (numbered from 1 to
79
Chapter 4: Rain cell size distribution
14 along the x-axis). Rain cells are expected to move with the wind speed at 700 mb
pressure level at 2 to 3 km above ground [14]. The wind speed measured at ground
level was increased by a factor of 1.5 to account for the variation of wind speed with
height above ground [15].
Figure 4.5: Comparison of wind speed, Doppler wind speed and cell translation
speed
Although the computed translation speed of rain cells shows broad agreement with
elevated wind speed, their detail variation is different. The differences could be due to
the fact that although rain cell movement is driven by winds, the rain cell pattern
observed on the ground at a given region may reflect the speed and direction of the
generating precipitation at a higher altitude [16]. There is however a considerable
difference between cell translation speed and Doppler derived wind speed. It is worth
noting that the Doppler method described above gives a reliable measure of only the
radial velocity component of a large rain volume, and the steps employed to derive
80
Chapter 4: Rain cell size distribution
horizontal translation speed from this radial measurement assumes a linear flow field
which is not the case in localized convective events.
4.5
Rain cell size distribution:
Rain gauges provide a record of the amount of rainfall over time, or rain rate time
series, at different locations. Since rain is a moving entity, this time series can be
converted into a spatial series by employing some known value of rain translation
speed. This is the so called synthetic storm technique [17]. Rain rate measurements at
10 s integration time were converted into 1-min rain rate time series before
determining the event durations for threshold rain rates from 5 to 50 mm/h.
Multiplying these durations by the average rain cell translation speed of 10.1 m/s
yielded an equivalent distance span and hence size of the corresponding rain cell.
Rain cell sizes were also deduced from radar measurements comprising 66 PPI scans
in 7 rain events within a 100 km range of Hampshire. The radar reflectivity field z
(dBZ) was converted into rain rate R (mm/h) through the relation z = 200R 1.6 , which
assumes a Marshall-Palmer rain drop size distribution. The “contour” and “polyarea”
functions of MATLAB have been tested and found to be very accurate in obtaining
the vertices and area of a polygon. So, these functions were employed to compute the
area of rain cells of defined thresholds, and this area was then converted to cell size
(i.e. diameter) in km by equating with a circular shaped region.
The numbers of rain cells of various rain rate thresholds obtained using rain gauge
data are much lower than the numbers from radar data (Table 4.1). This is because the
81
Chapter 4: Rain cell size distribution
rain-gauge-based estimate employs the average rain rate in 10 s intervals at a single
location whereas radar captures instantaneous rain rate values at multiple locations.
The radar measurement employed covered one rain event in summer, four events in
autumn and two in spring, and may therefore not have sufficient seasonal variety for
the results to be reliably representative of an average year. The rain gauge data on the
Table 4.1: Statistics of rain cells from rain gauge and radar data
Rain rate threshold
No. of cell from Rain gauge
No. of cell from Radar
≥ 5 mm/h
6432
9719
≥ 10 mm/h
4066
5033
≥ 20 mm/h
217
3448
≥ 30 mm/h
130
2493
other hand covered all seasons of a 3-year period. The smallest cell size determined
using rain gauge data was 0.61 km, so cells of size less than 0.61 km were not
considered in the radar data analysis.
82
Chapter 4: Rain cell size distribution
Figure 4.6: Cumulative distribution of rain cell sizes obtained from rain gauge
data
Figures 4.6 and 4.7 show the cumulative distributions of rain cell sizes derived from
rain gauge and radar measurements respectively, for rain rate thresholds from 5 to 50
mm/h. The lower rain rate cells (of thresholds 5 to 10 mm/h) extend up to 65 km in
diameter for results derived from rain gauge measurements using the synthetic storm
technique. Intense rain cells (thresholds  20 mm/h) on the other hand have
extensions around 5 to 10 km, which is in agreement with the results found in earlier
studies [14] [17]. It can be seen that the two methods yield comparable results for
intense rain cells, whereas there is considerable difference for the lower rate cells
where radar observations show cell sizes up to 25 km. The families of curves for
intense cells are similar in overall appearance to those reported by Yau and Rogers
[14]. It should be noted that the accuracy of the synthetic storm technique depends on
83
Chapter 4: Rain cell size distribution
Figure 4.7: Cumulative distribution of rain cell sizes obtained from radar
observation
the use of correct values of storm translation speed for the different rain patterns. In
our analysis, the same translation speed determined for the intense rain cells
associated with highly mobile convective rain was applied to the lower rate cells of
stratiform rain. This is likely to have resulted in overestimation of the spatial extent of
weaker rain cells. Rain rates lower than 5 mm/h were not considered since the
assumption of a circular cell shape no longer holds, as revealed by the radar
observations. Very intense rain cells above 50 mm/h appeared on the radar scans as
small dots whose areas could not be reliably computed, so rain intensities above 50
mm/h were excluded from the analysis. Furthermore, the sample size of intense rain
cells observed using both rain gauge and radar was quite small (between 2 and 6)
making their distributions unstable [17]. In particular, distributions below 0.02% were
84
Chapter 4: Rain cell size distribution
not possible, and cell sizes indicated in Figures 4.6 & 4.7 at small percentages (<~
0.2%) may not accurately represent those in an average year.
The largest cell sizes determined from radar data (Figure 4.7) were around 25 to 26
km for a 5 mm/h rain rate threshold. This compares well with other rain cell size
distribution studies where cell sizes larger than 20 km are considered to be cell
clusters controlled by air motions of scales larger than the rain cell [6] [18] [19].
4.6
Conclusions:
Rain cell translation speed obtained using correlation analysis of consecutive radar
PPI scans has been compared with Doppler derived wind speed and scaled ground
measured wind speed. Distributions of rain cell sizes determined from both radar and
rain gauge measurements were also presented. These results show that intense rain
cells have sizes generally less than 10 km. The results are similar to the values found
in Yau and Rogers [14] and Matricciani and Bonati [17]. As further highlighted in
chapter 8, one application of this study might be in site diversity as a rain fade
mitigation technique in satellite communications, where a spacing of about 10 km
between diversity stations could provide significant protection against rain-induced
link outage, considering that at this spacing both stations will normally not
concurrently lie within one intense rain cell. Thus in the event of a large attenuation of
the satellite-to-earth radio signal due to intense rain on the primary link, the downlink
can be automatically temporarily re-routed through the secondary or diversity station.
In this way diversity gains of up to about 10 dB can be realised [20], the exact value
depending on climate, link frequency and path elevation angle.
85
Chapter 4: Rain cell size distribution
Rain cell size distributions from rain gauge data are found in the scientific literature
where 700 mb wind speed is employed to obtain cell size distribution. In this study,
intense rain cells were tracked to obtain their translation speed. A mean rain cell
translation speed of 10.1 m/s was then employed, which is a novel approach to
determine cell size distribution. The results showed close agreement with earlier
findings for intense rain cells. However, the diameters of light rain cells were  65 km
which appeared to be significantly overestimated when compared to radar-derived
estimates  25 km. Better agreement between rain gauge and radar estimates of cell
sizes for all regimes of rain intensity was obtained by using the computed average cell
translation speed of 10.1 m/s for intense cells and a trial translation speed of 4 m/s for
light rain ( 10 mm/h). There were only a few trackable cells in the radar
measurements available, so further cell tracking analysis as described in this chapter is
required along with higher level wind speed measurements in order to confirm the
translation speed of light rain for application in the synthetic storm technique.
86
Chapter 4: Rain cell size distribution
References:
1. R. K. Crane, Electromagnic Wave Propagation through Rain, John Wiley and
Sons Inc. 1996.
2. H. Sauvageot, Radar Meteorology, Artech House, 1992.
3. www.met.rdg.ac.uk/radar/doc/camra.html
4. C. Capsoni, F. Fedi, C. Magistroni, A. Paraboni and A. Pawlina, “Data and
Theory for a New Model of the Horizontal Structure of Rain Cells for Propagation
Applications”, Radio Science, Vol. 22, No. 3, 1987, pp. 395-404.
5. R. K. Crane, “Automatic Cell Detection and Tracking”, IEEE Transactions on
Geoscience Electronics, Vol.GE17, No.4, October, 1979, 250-262.
6. H. Sauvageot, F. Mesnard and R. S. Tenorio, “The Relation between the Area
Average Rain Rate and the Rain Cell Size Distribution Parameters”, Journal of
the Atmospheric Sciences, Vol. 56, January, 1999, pp. 57-70.
7. M. Dixon and G. Wiener, “TITAN- Thunderstom Identification, Tracking,
Analysis and Nowcasting- A Radar-based Methodology”, Journal of Atmospheric
and Oceanic Technology, Vol. 10, No.6, December, 1993, pp. 785-797.
8. R.E. Rinehart and E. T. Garvey, “Three Dimensional Storm Motion Detection by
Conventional Weather Radar”, Nature, Vol. 273, 25 May, 1978, pp 287-289.
9. M. E. Weber, M. L. Stone, “Low Altitude Wind Shear Detection Using
Airport Surveillance Radars”, Lincoln Laboratory, MIT.
10. E. S. Chornboy, A. M. Matlin, and J. P. Morgan, “Automated Storm
Tracking for Terminal Air Trafic Control”, The Lincoln Laboratory Journal,
Vol.7, No. 2, 1994, pp. 427-448.
11. L. J. Battan, Radar Observation of the Atmosphere, The University of
Chicago Press, 1973.
87
Chapter 4: Rain cell size distribution
12. H. Hinterberger, B. Bauer-Messmer, “Discrete Object detection and
Motion Registration Based on a Data Management Approach”, Proceedings of
SSDBM’98, July 1-3, 1998, Capri, Italy.
13. K. A. Browning and R. Wexler, “The Determination of Kinematics Properties
of a Wind Field using Doppler Radar”, Journal of Applied Meteorology, Vol. 7,
1968, pp. 105-113.
14. M. K. Yau and R. R. Rogers, “An Inversion Problem on Inferring the Size
Distribution of Precipitation Areas from Raingauge Measurements”, Journal of
Atmospheric Sciences, Vol. 41, February, 1984, pp. 439-447.
15. T. C. Tozer, and D. Grace, “High-altitude platforms for wireless
communications”, Electronics & Communication Engineering Journal, 2001, pp.
127-137.
16. J. S. Marshall, “Precipitation Trajectories and Patterns”, Journal of Meteorology,
Vol. 10, February, 1953, pp. 25-29.
17. E. Matricciani, A. P. Bonati, “Rain Cell Size Statistics Inferred
from Long Term Point Rain Rate: Model and Results”, Proc. Third Ka band
utilization conference, Sorrento, Italy, September, 1997.
18. J. Goldhirsh, and B. Musiani, “Rain Cell Size Statistics Derived from Radar
Observations at Wallops Island, Virginia”, IEEE Trans. Geosci. Remote Sens.,
Vol. 24, 1986, pp. 947-954.
19. T. G. Konrad, “Statistical Models of Summer Rainshowers Derived from
Fine-scale Radar Observations”, J. of Appl. Meteorol, Vol. 7, 1978, pp. 171-188.
20. S. A.Callaghan, , B. Boyes, A. Couchman, J. Waight, C. J. Walden, and S.
Ventouras , “An Investigation of Site Diversity and Comparison with ITU-R
Recommendations”, Radio Science, Vol. 43, 2008.
88
Chapter 5: Traffic analysis and network design
5 CHAPTER 5
Traffic Analysis and Network Design
89
Chapter 5: Traffic analysis and network design
5.1
Introduction:
Most of the cities around the world now have well established B-ISDN networks, but
often these networks do not extend beyond city areas as it is not cost-effective to lay
the terrestrial network where initial user numbers may not be large. Broadcasting
capabilities of the satellite network makes it appealing to integrate with the existing
terrestrial network to take the broadband facilities to the remote areas [1]. However,
regional communication infrastructure should be taken into consideration for the
system to be both technically and economically viable.
Suitability of satellite-integrated network is emphasized based on the current
telecommunication infrastructure of Bangladesh. In telecommunication system design
it is important to determine the number of trunks required on a route or connection
between exchanges. Different traffic models could be applied to dimension the route
in the network. However, it is vital to consider the traffic characteristics of multimedia
communication. A novel satellite-integrated network is proposed based on the
dimensioning of the effective bandwidth for multimedia traffic.
5.2
Telecommunication infrastructure in Bangladesh:
The communication link of Bangladesh consists of

Satellite

Microwave links

Optical fibre links

Cellular coverage
90
Chapter 5: Traffic analysis and network design
Figure 5.1: Microwave coverage of Bangladesh (source-BTTB annual report,
2001)
Inter-city connectivity is provided through microwave link and international gateway
connectivity is obtained through satellite stations (Figure 5.1). Microwave links are
the major communication backbone of the country with data rate from 34-155 Mbps.
Bangladesh Telegraph and Telephone Board (BTTB) has recently migrated its
backbone link from microwave to optical fibre along the national railway network and
the national highway route that increased the total backbone bandwidth to 644 Mbps
[2].
91
Chapter 5: Traffic analysis and network design
Table 5.1: Telecommunication indicator 2007
Fixed Telephone Density
Mobile Telephone Density
Internet
(per 100 inhabitants)
(per 100 inhabitants)
(per100 inhabitants)
0.75
21.66
0.32
The figures (Table 5.1) from International Telecommunication Union (ITU) in 2007
indicate that cellular coverage increased significantly in recent years though the
internet users still remain low [3]. This may be due to the fact that readily available
Information Technology (IT) infrastructure is concentrated primarily in major cities
and towns. Moreover, the capacity of the telecom backbone of the country, which
consists of microwave radio link, is not enough to support broadband IT services
throughout the country.
Although the broadband infrastructure are available in most of the city areas, rural
areas are cut-off from the mainstream communication network as laying broadband
infrastructure is not cost effective in thin route communication. A new customer in
remote location can join the satellite network system simply by acquiring the right
equipment. There is no need for expensive and labour intensive underground cabling
to access the high-speed communication service. Owing to its geographical location,
Bangladesh experiences frequent natural disasters such as cyclone, flood etc. with a
consequent disruption of wireline broadband link. Deployment of satellite
communication system can be an obvious choice in these situations for its scalability,
inherent broadcasting capability and more robustness to meteorological conditions.
92
Chapter 5: Traffic analysis and network design
Due to the flat-terrain characteristics of Bangladesh, wireless technology can be
introduced by a handful of towers say 15 towers (of height ~ 24 m, where height, h =
d2 /8R, d is the line of sight distance and R is the effective radius of the Earth) with 50
km radius (with a line of sight distance of 30-40 km) to cover the entire 144000
km 2 country. However, for the area where the potential users are sparse, construction
of many towers and base stations will not be economically feasible solution. Satellite
communication is increasingly being recognised as the most cost-effective and
efficient method of providing the Information and Communication Technology (ICT)
connectivity, particularly in the areas where little or no terrestrial infrastructure is
available.
5.3
Dimensioning traffic path:
Traffic is defined as the amount of information or the number of messages over a
traffic path during a given period of time. A traffic path is a channel, time slot,
frequency band, line, trunk switch or circuit over which individual communications
pass in sequence [4]. A simple model of information flow or calling process between
sources and destinations through a traffic path is shown in Figure 5.2 where blocking
of information is likely to occur in the local switches or within the network resulting
in some traffic being carried, delayed, retried, or lost [5], as defined below

Carried traffic- Carried traffic is the volume of traffic actually carried to its
destination.

Offered traffic- Offered traffic is the volume of traffic offered to a switch.
93
Chapter 5: Traffic analysis and network design

Blocked traffic- It represents the amount of traffic that cannot be carried to its
destination on the first attempt as the switch may be busy with other traffic or
calls.

Lost traffic- Offered traffic minus carried traffic equals lost calls. A call is lost
usually because it meets blockage or congestion at that switch.

Retried traffic- Retried traffic are the blocked calls that could be retried at a later
time at the discretion of the callers.

Delayed traffic- Some or all blocked calls may be delayed in the network until
facilities are available to handle them.
Queue
Offered traffic
Carried traffic
Sources
Destinations
Delayed traffic
Retried traffic
Lost traffic
Figure 5.2: Traffic flow model
To dimension a traffic path i.e. to find the capacity of path, we must know the traffic
intensity representative of the normal busy hour [4] [6] as well as the Grade of
Service (GoS) or probability of blockage.
5.3.1
Traffic load or intensity:
Traffic load is the volume of traffic presented to a trunk group during the busy hour,
measured in Erlang, which can be expressed as
94
Chapter 5: Traffic analysis and network design
A = C×h
(5.1)
where C designates the number of calls originated during the period of one hour (call
arrival rate) and h is the average holding time, usually given in hours. Erlang is a
dimensionless unit because we are multiplying calls/hour by hour/call.
To illustrate the traffic load, let us assume that 10 calls were made each of which had
the durations of 62, 74, 84, 90, 94, 70, 96, 48, 64 and 126 s during 90 min period. We
can find the traffic intensity as follows:
Call arrival rate = no of calls/ time period in hours
= 10/90min = 10/1.5hour = 6.7 calls/hour
Average holding time = sum of the individual call durations/total number of calls
= (62+74+84+90+94+70+96+48+64+126)s/10
= 80.8 s
This means that for the line in question the traffic intensity is
A = C×h =
6.7  80.8
= 0.15 Erlang
3600
By definition, a traffic flow of one Erlang is equivalent to a circuit being occupied for
a complete hour. Erlang is a dimensionless unit of traffic intensity named after the
Danish mathematician A. K. Erlang.
95
Chapter 5: Traffic analysis and network design
5.3.2
Grade of Service (GoS):
During the busy period we can expect moments of congestion such that additional call
attempts will meet blockage. A switch is dimensioned to handle the busy hour traffic
with a certain Grade of Service (GoS) that expresses the probability of meeting
blockage during the busy hour. A busy hour GoS of 0.01 implies that one call in 100
will not be successful at the first attempt, owing to the network congestion.
This probability depends on a number of factors, the most important of which are [4]:
1. The distribution in time and duration of offered traffic (e.g. random or periodic
arrival and constant or exponentially distributed holding time)
2. The number of traffic sources [limited or high(infinite)]
3. The availability of trunks in a group to traffic sources (full or restricted
availability)
4. The manner in which lost calls are handled.
In conventional telephone traffic theory, three methods are considered for handling or
dispensing of lost calls:

Lost calls held (LCH)- The LCH concept assumes that the telephone user will
immediately reattempt the call on receipt of a congestion signal and continue
to redial.

Lost calls cleared (LCC)- The LCC concept, which is primarily used in
Europe or those countries that have adopted European practice, assumes that
96
Chapter 5: Traffic analysis and network design
the user will hang up and wait some time interval before reattempting if the
user hears the congestion signal on the first attempt.

Lost calls delayed (LCD)- The LCD concept assumes that the user is
automatically put in queue i.e. in a waiting line or pool.
5.3.3
Traffic model:
Dimensioning a route involves determining the optimum number of circuits to serve
the route. To determine the number of circuits based on the busy hour traffic load,
several factors need to be considered such as call arrivals and holding time
distributions, number of traffic sources, availability (full or limited) and handling of
lost calls. Based on these factors several traffic models are available to express grade
of service or the probability of finding ‘x’ channels busy (Table 5.2).
Table 5.2: Traffic model characteristics
Traffic model
Sources
Arrival pattern Blocked
Call Holding times
disposition
Poisson
Infinite
Random
Held
Exponential
Erlang B
Infinite
Random
Cleared
Exponential
Extended Erlang B
Infinite
Random
Retried
Exponential
Erlang C
Infinite
Random
Delayed
Exponential
Engset
Finite
Smooth
Cleared
Exponential
97
Chapter 5: Traffic analysis and network design
5.3.3.1
Capacity by Erlang B traffic model:
Let us assume that a network is operating with the following assumptions

Traffic at each node has a random arrival pattern

Hold times are exponential

Blocked calls are cleared from the system

There are an infinite number of callers
Under these assumptions we can use the Erlang B formula given as [7]
ac
B (c, a )  c c! k
a

k  0 k!
(5.2)
where B(c,a) is the probability of blocking the call
a is the traffic load
c is the number of circuits
If a route carried 16.68 Erlang of traffic then at 0.001 blocking (B) 30 trunks or
circuits would be required by applying equation (5.2). If the grade of service were
reduced to 0.05, the 30 trunks could carry 24.80 Erlangs of traffic. Since a circuit or
trunk is a single-channel Pulse-code modulation (PCM) signal with bit rate of 64
kbps, the bandwidth for 30 trunks is ~ 1.8 Mbps.
98
Chapter 5: Traffic analysis and network design
5.3.3.2
Capacity by Erlang C traffic model:
For data transmission the probability of delay, the most common index of grade of
service for waiting systems when dealing with full availability and a Poissonian call
arrival process (i.e. random arrivals), is calculated using the Erlang C formula, which
assumes an infinitely long queue length [7]. Let us assume that a network is
transferring data at 400 packets per second with 200 bytes per packet under the
following assumptions

Traffic at each node has a random arrival pattern

Hold times are exponential

Blocked calls are delayed in the system

There are an infinite number of sources
Erlang C model is designed around queuing theory with the above assumptions given
as [7]
acc
c!(c  a )
B (c, a )  c 1 k
a
acc


c!(c  a )
k  0 k!
(5.3)
where B(c,a) is the probability of delaying the call or transfer
a is the traffic load
c is the number of circuits
To evaluate the capacity by the Erlang C model, the required parameters are the
number of calls or packets in the busy hour, the average call length or packet size and
99
Chapter 5: Traffic analysis and network design
the expected amount of delay in seconds. As circuit capacity is 64 kbps i.e. the
amount of data necessary to keep the circuit busy for 1 second, the required number of
circuits to keep the delay factor under 10 ms to transfer data at 400 packets per
seconds can be found as follows:
Traffic intensity, a = (400×200×8)/64000 = 10 Erlang
Average packet transmission time = (200×8)/64000 = 0.025 seconds
Delay factor = 10/25 = 0.4
At 0.4 delay factor (B), 10 Erlangs of traffic require approximately 12 circuits
requiring ~ 0.77 Mbps.
Erlang B model is widely used for voice traffic around the world except United States
where Poisson model is favoured [4]. One popular application of Erlang B or Poisson
model until now is to calculate the capacity for setting up the call centre. Data
services such as electronic mail generate traffic that is not greatly different from voice
traffic to which Erlang B formula can be applied [5]. Other data services such as batch
transfers between mainframes represent traffic that can be queued where Erlang C
formula can be applied [5]. Besides these conventional traffic models there are few
stochastic models for packet traffic such as Markov-Modulated Poisson processes [8],
packet-train models [9] and fluid flow models [10].
Although no traffic model can accurately capture real life situations of traffic
behaviour such as call arrival patterns, holding times of the calls, number of source in
the route etc., these models reflect to the best the average in each situation.
100
Chapter 5: Traffic analysis and network design
5.4
Characteristics of Ethernet data:
The conventional telephone traffic models assume the nature of aggregate traffic as
“Poisson-like”, namely that aggregate traffic becomes smooth (less bursty) as the
number of traffic sources increases. However, contrary to this, Ethernet traffic is
statistically self-similar i.e. aggregating streams of such traffic intensifies the selfsimilarity (burstiness) instead of smoothing it. None of the commonly used traffic
models is able to capture the self-similar behaviour which has serious implications for
the design, control and analysis of high-speed, cell-based networks [11]. Appropriate
traffic models are critical for optimal resource allocation such as bandwidth
assignment, determination of buffer sizes at the switch etc. in order to provide desired
Quality of Services (QoS) for the nertwork [12].
5.4.1
Self-similarity:
Measurements show that Ethernet traffic seems to look the same in the large (min, h)
as in the small (s, ms) timescales. The structural similarity across a wide range of
timescales is termed as self-similar phenomenon. Self-similarity is the property
associated with “fractals” which are objects whose appearances are unchanged
regardless of the scale at which they are viewed. In self-similar traffic, traffic burst
consists of bursty subperiods separated by less bursty subperiods [11].
The burstiness measure is defined as the ratio of the peak spread of the trace against
its average spread. Mathematically,
101
Chapter 5: Traffic analysis and network design
b
p
(5.4)
2
where p is the peak rate of the trace
λ is the average rate of the traffic trace
σ 2 is the variance of the trace
For bursty traffic peaks are very large compared to average i.e. the distribution is
heavy-tailed. In other words, the superposition of file transfers in such distribution is
long-range dependent [13] [14].
5.4.2
Properties of self-similarity:
Let us assume
X = (X t , t = 0.1.2…..)
as a covariance stationary stochastic process with mean μ, variance σ 2 and
autocorrelation function r(k), k≥0.
In particular, we assume that X has an
autocorrelation function of the form
r(k) ~ k   L(t), as k→∞
(5.5)
where 0<β<1 and L is slowly varying at infinity i.e.
102
Chapter 5: Traffic analysis and network design
lim t 
L(tx)
 1 , for all x>0
L(t )
For each m = 1.2.3….., let
X ( m)  ( X k
( m)
: k  1.2.3......)
denote the new covariance stationary time series (with corresponding autocorrelation
function r (m ) ) obtained by averaging the original series X over non-overlapping
blocks of size m. That is, for each m = 1.2.3……, X (m ) is given by
Xk
(m)

1
( X km  m 1  .....  X km ), k  1
m
The process X is called exactly second-order self-similar with self-similarity
parameter H = 1- β/2 if for all m = 1.2…..,
var( X ( m) )   2 m  
And
r ( m) (k )  r (k ), k  0
(5.6)
X is called (asymptotically) second-order self-similar with self-similarity parameter H
= 1- β/2 if for all k large enough
r ( m) (k )  r (k ), as m →∞
(5.7)
103
Chapter 5: Traffic analysis and network design
with r(k) given by (5.5). In other words, X is exactly or asymptotically second-order
self-similar if the corresponding aggregated processes X (m ) are the same as X or
become indistinguishable from X, at least with respect to their autocorrelation
functions.
Mathematically, self-similarity manifests itself in a number of equivalent ways-

The variance of the sample mean decreases more slowly than the reciprocal of
the sample size (slowly decaying variances), i.e. var(X (m ) )~ a 2 m   , as m→∞,
with 0<β<1 (a 2 , a 3 ,…denote positive constants)

The autocorrelations decay hyperbolically rather than exponentially fast,
implying a non-summable autocorrelation function
 r (k )  
(long-range
k
dependence), i.e. r(k) satisfies relation (5.5) and

The spectral density f(.) obeys a power-law near the origin (1/f – noise), i.e.
f(λ)~ a 3 λ  , as λ→0, with 0<γ<1 and γ = 1- β
Intuitively, the most striking feature of (exactly or asymptotically) second-order selfsimilar processes is that their aggregated processes X (m ) posses a nondegenerate
correlation structure, as m→∞ [11].
104
Chapter 5: Traffic analysis and network design
5.4.3
The Hurst parameter- The measure of self-similarity:
The Hurst parameter, H is a measure of the level of self-similarity of a time series. In
order to determine if a given series exhibits self-similarity, a method is needed to
estimate H for a given series. There are three approaches to estimate H
1. Analysis of the variances of the aggregated processes X (m )
2. Analysis of the rescaled range (R/S) statistic for different block sizes
3. A Whittle estimator
For self-similar processes H takes values from 0.5 to 1 [12].
5.5
Dimensioning Ethernet traffic by Closed-Queuing Network (CQN) model:
Elastic data applications can adapt to time-varying available bandwidth via a feedback control such as the transmission control protocol (TCP) or the available bit rate
(ABR) transfer capability in ATM. Let us assume that the closed-loop controls for the
elastic data applications are performing well and the key attribute of a wellperforming control is that it maintains some bytes in queue at the bottleneck link with
minimal packet loss. We can focus on a single bottleneck target link and a simple
product-form, closed-queuing network (CQN) model in heavy traffic. An important
practical feature of these CQN models is their insensitivity: The distribution of the
underlying random variables influences the performance only via the mean of the
distribution [15].
105
Chapter 5: Traffic analysis and network design
Infinite-Server(IS) node
represents the N sources
and the remaining
network components
besides the target link
Processor-Sharing(PS)
node represents the
network node’s output
port and buffer at the
target bottleneck link
Figure 5.3: Closed-queueing network (CQN) model
The parameters of the CQN model are (Figure 5.3):
1. The number of sources, N
2. The mean service at the Processor-Sharing (PS) node, denoted  1 . The mean
service time at the PS node represents the mean time to transmit a file on the
target link given no other files present.
3. The mean time in the Infinite-Server (IS) node, denoted  1 . The mean time in
the IS node represents the mean time between initiation of file transfers by a
source, in the hypothetical case that the target link imposes negligible
constraint on the transfer.
In elastic data applications, the user, and hence the network designer, is concerned
with the delay in transferring a file. Since the file sizes vary greatly, a single delay
objective, such as 100 ms, for all file is not sensible. Rather, the delay objective
should be normalised by the file size, which yields a performance objective in units of
106
Chapter 5: Traffic analysis and network design
s/bit. More conveniently, the reciprocal, in bits/s, is considered as the performance
objective in terms of the bandwidth that an arbitrary active source obtains. Let B s
denote the bandwidth that an arbitrary source obtains in steady state. This bandwidth
per source is defined as [15]
Bs 
B
Q̂1
where B is the link bandwidth and Q̂1 is the conditional number of jobs in the
Processor Sharing (PS) node given that the PS node is not empty. The performance
criteria on the mean and on the tail probability of B s :
E[ Bs ]  b
(5.8)
Or
Pr( Bs  b)  
(5.9)
for given b and α, where typical values for b in the range of 10 4 - 10 6 bits/s and α in
the range of 0.01-0.1.
To derive engineering rules for dimensioning the bandwidth, the dimensioning
problem is simply stated as [15]
Minimize B such that the chosen
performance criteria (5.8) or (5.9) is satisfied
(5.10)
107
Chapter 5: Traffic analysis and network design
An approximate solution to (5.10) is
B = h. N
(5.11)
Given the mean performance criteria (5.8) where [15]
1 1
h  (  ) 1
b f
λf is the throughput a source would obtain assuming the target link is not constraining
the flow and is determined by the constraints of other network components and
include the idle times at the source whereas b, the per-flow bandwidth objective
applies only during active periods of file transfers. Effective rate, h, from each source
is the harmonic mean of two rates that are indeed naturally associated with elastic data
and will occur under respective limiting network conditions.
A service provider could reasonably choose a bandwidth objective that is equal to or
greater than λf. When b is larger than λf, the service provider can realize significant
savings in the engineered bandwidth, B as compared with the full allocation. In the
satellite-integrated network where users (or sources) will be interested in broadband
communication (such as voice, data file, www page, video), we need to know the
mean idle time and mean file sizes to dimension the effective bandwidth. Irrespective
of user data rate, the mean file size and mean idle time are constant in the network
node i.e. TCP/IP or ATM switches under the condition that the control loop is
performing well. CQN model gives per-flow or per-connection bandwidth. Assuming
108
Chapter 5: Traffic analysis and network design
mean idle time, λ 1 = 5 seconds and mean file size, f = 200 Kbytes (including
overheads), we can calculate the required bandwidth as follows:
1 1
h  (  ) 1
b f
(
1
1 1

)
320 320
(assuming b = λ f)
= 160 kbps
Bandwidth, B = N × h = 50×160 kbps = 8 Mbps.
This is the bandwidth that will be required on the link to support 50 users in the
network. CQN model gives simple and robust engineering rules for dimensioning
bandwidth for elastic data traffic for a single bottleneck link. The robustness of the
dimensioning rules follows from the insensitivity property of the CQN model,
whereby the distribution of the underlying random variables is pertinent only via the
mean, and of particular interest, the mean of the file sizes and not their heavy-tail
characteristics. Despite the simplicity, CQN model accurately predicts the distribution
for number of active sources at the bottleneck link, given the condition that the
feedback control is performing well.
5.6
Satellite-integrated Network model:
The satellite-integrated network is as shown in Figure 5.4. The user group in the
remote, isolated terrestrial network communicate via a local exchange and will be
connected with the existing mainstream network through satellite link. The users in
109
Chapter 5: Traffic analysis and network design
mainstream network communicate via the backbone segment (say fibre optic link or
DSL etc.) where ATM is the core technology. The cell streams from the terrestrial
network reach the sat-ATM interface which provides some essential functions such as
channel coding, modulation, conversion of data into satellite frame before
transmission on the RF link. In the receiving end, the interface performs the reverse
functions such as conversion of satellite frame to data, demodulation and decoding
etc. before forwarding to the destination.
ISDN
CBR
ISDN/ATM
UBR
Users/
Sources
IP/ATM
Sat
Mod
em
Sat
Mod
em
ATM/ISDN
Internet
ATM/IP
ABR
VBR
ATM
B-ISDN
MPEG/ATM
Local exchange
Figure 5.4: A satellite-integrated network model
The sat-ATM interface performs two key roles in the network model. Firstly, to
mitigate the low BER performance requirement for the ATM QoS parameters robust
error-control coding scheme is implemented in the interface. Secondly, delay sensitive
services such as voice, interactive video are given priority than the non-sensitive ones
such as data, image transfer etc. to overcome the inherent large propagation delay of
110
Chapter 5: Traffic analysis and network design
the satellite link. Priority based delay is not evaluated in this study. However, mean
Cell Transfer Delay (CTD) for Constant Bit Rate (CBR) services is slightly higher
than 5 ms. Assuming 250 ms of end-to-end delay and 5 ms of queuing delay, the total
delay is well within the ITU-T objective of 320 ms for networks with a geostationary
satellite link [16].
In our model, the satellite is a bent-pipe Geostationary Earth Orbit (GEO) satellite
with the assumption of an Additive White Gaussian Noise (AWGN) channel model
for the satellite link which is appropriate whenever the occurrence of bit errors is
random and independent. This assumption holds well under clear air condition. When
the link condition degrades due to severe weather phenomena such as rain attenuation,
the link could see bursts of error. But under this severe condition, the link would
degrade sufficiently to result in an outage and the loss of communication.
5.7
Conclusions:
With a view to extending broadband facilities to the remote areas, a satelliteintegrated network is designed by taking into account broadband traffic
characteristics. Different parameters such as traffic, traffic models, traffic
characteristics, self-similarity of multimedia traffic etc. are discussed. ClosedQueuing Network (CQN) model which was derived for dimensioning bandwidth for
elastic traffic is considered to be more relevant to calculate bandwidth. Applying
CQN model, the required bandwidth is 8 Mbps to support 50 users in the network
[page 109].
111
Chapter 5: Traffic analysis and network design
Satellite communication systems have been deploying to provide fixed circuit data,
voice, broadcasting services as well as Internet access to home and small corporate
users through Very Small Aperture Terminal (VSAT). The capacity on the link could
be either service specific or could be calculated by conventional traffic models. For
example, audio channel requires 64 to 128 kbps, data is transmitted at 20 to 512 kbps
(low to high speed), video conferencing at ~5 Mbps and TV distribution at 20- 50
Mbps. However, the conventional method is not applicable to calculate capacity in
satellite-integrated network where multimedia data follow self-similar traffic
characteristics.
Evaluation of ATM performance parameters such as cell loss ratio (CLR) and cell
error ratio (CER) and TCP/IP throughput performance in the satellite-integrated
network, will be found in the next two chapters.
Overall performance of network model through link budget analysis incorporating
link bandwidth from this chapter and simulation of the network considering
propagation aspect such as rain attenuation level from chapter 3 is presented in
chapter 8.
112
Chapter 5: Traffic analysis and network design
References:
1. M. Riccharia, Satellite Communication Systems, McMillan, 1995.
2. www.bttb.net
3. www.itu.int/ITU-D/ict/statistics
4. R.L. Freeman, Fundamentals of Telecommunications, Wiley interscience, 2005.
5. B. E. Carne, Telecommunications Primer: Signals, Building blocks and Networks,
Prentice-Hall Inc. 1995.
6. ITU-T, “Traffic Intensity Measurement Principles”, Recommendation E.500,
1998.
7. R.L. Freeman, Reference Manual for Telecommunication Engineering, 3rd ed.,
Section 1, Wiley, New York, 1994.
8. H. Heffes and D.M. Lucantoni, “A Markov Modulated Characterization of
Packetized Voice and Data Traffic and Related Statistical Multiplexer
Performance”, IEEE J. Select. Areas Commun., vol. SAC-4, 1986, pp. 856-868.
9. R. Jain and S.A. Routhier, “Packet Trains: Measurements and a New Model for
Computer Network Traffic”, IEEE J. Select. Areas Commun., vol. SAC-4, 1986,
pp. 986-995.
10. D. Anick, D. Mitra and M. M. Sondhi, “Stochastic Theory of a Data Handling
System with Multiple Sources”, Bell System Tech. J., vol. 61, 1982, pp. 18711894.
11. W.E. Leland, M.S. Taqqu, W. Willinger and D. V. Wilson, “ On the Self-Similar
Nature of Ethernet Traffic (extended version)”, IEEE/ACM Transactions on
Networking, Vol. 2, No. 1, 1994, pp. 1-15.
12. Z. Sahinoglu and S. Tekinay, “On Multimedia Networks: Self-Similar Traffic and
Network Performance”, IEEE Communications Magazine, 1999, pp.48-52.
113
Chapter 5: Traffic analysis and network design
13. V. Paxson and S. Floyd, “Wide-area Traffic: The Failure of Poisson Modelling”,
IEEE/ACM Transactions on Networking, Vol. 3, No. 1, 1995, pp. 226-244.
14. W. Willinger, M.S. Taqqu, R. Sherman and D. V. Wilson, “Self-similarity through
High Variability: Statistical Analysis of Ethernet LAN Traffic at the Source
Level”, IEEE/ACM Transactions on Networking, Vol. 5, No. 1, 1997, pp. 71-86.
15. A. W. Berger, and Y. Kogan, “Dimensioning Bandwidth for Elastic Traffic in
High-Speed Data Networks”, IEEE/ACM Transactions on Networking, Vol. 8,
No. 5, 2000, pp. 643-654.
16. S. Nawrot, H. Elsayed, T. Kapoor, K. Shuaib, M. Lee and T. Saadawi,
“Performance of a Hybrid Terrestrial/Satellite ATM Network- Experimental
Results for CBR Traffic CTD and CDV QoS Parameters”, IEEE Communication,
1998.
114
Chapter 6: Performance of ATM over satellite links
6 CHAPTER 6
Performance of ATM over Satellite Links
115
Chapter 6: Performance of ATM over satellite links
6.1
Introduction:
For broadband communication in most terrestrial networks ATM (Asynchronous
Transfer Mode) is used to carry signals such as voice, video, data etc. ATM has been
commercially deployed as the core technology for Broadband-Integrated Services
Digital Network (B-ISDN) in many local and wide area networks where physical
media of communication is mostly coaxial and fibre optic cables, which has excellent
error characteristics. However, there is a shortage of broadband terrestrial connections
in many areas, particularly in more remote or rural areas where terrestrial lines are
expensive to install and operate. To bring the services of ATM to remote or isolated
areas ATM networks can be extended to include the satellite links. The performance
of ATM is affected by two fundamental constraints of satellite links namely error
bursts and large propagation delay. The large propagation delay of a satellite link does
not influence the ATM layer itself; applications based on ATM and higher layer
protocol performance are affected by large delay. Unlike the terrestrial links for which
ATM was initially developed, satellite links present a much higher bit error ratio
(BER).
A satellite channel is often modelled as an Additive White Gaussian Noise (AWGN)
channel, which is reasonable when operating with a geostationary satellite and fixed
user antennas. Such a channel produces random single bit errors and the error rate
depends on the received carrier-to-noise ratio, which for digital communications can
be expressed as the ratio of the bit energy to the noise spectral density, E b /N o . To
operate at very low carrier-to-noise ratios, forward error correction (FEC) is often
used in satellite modems. On average, coding reduces the bit error ratio or alternately
decreases the transmission power needed to achieve a certain quality of service for a
116
Chapter 6: Performance of ATM over satellite links
given carrier-to-noise ratio. However, burst errors may be generated in the decoding
process. Burst errors considerably degrade the performance of ATM which was
designed to be robust with respect to random single bit errors [1].
To translate ATM layer performance measures such as cell loss ratio (CLR) and cell
error ratio (CER) to the physical layer performance required from a satellite link,
burst error characteristics must be considered. The ATM performance parameters are
related to the link bit error ratio and are also dependent on the bit error distribution.
The bursty nature of bit errors resulting from the FEC scheme in satellite links can be
handled by header bit interleaving and cell interleaving [2] [3] [4] [5] to deliver the
services at the required ATM Quality of Service (QoS). However, these methods by
themselves will not help in reducing the BER performance requirements for data
services with high throughput requirements. This is because even a single error will
lead to packet dropping. BER performance could be improved at a desired level by
concatenated coding scheme with Reed-Solomon outer code and convolution inner
code to provide users with the high throughput needed for high speed data protocols
such as Transmission Control Protocol (TCP). Besides improved BER performance,
concatenated coding scheme has a lower interleaving delay compared to header bit
and cell interleaving scheme.
This chapter presents the work done in applying a robust concatenated coding scheme
to improve ATM performance over satellite links. TCP Segment Error Ratio (SER) is
also evaluated under the same coding scheme.
117
Chapter 6: Performance of ATM over satellite links
6.2
ATM layer and QoS parameters:
The ATM layer is a cell based layer that is independent of the physical media and
application. An ATM cell consists of a 5 byte header and 48 byte payload, resulting in
a total length of 53 bytes. The header is again subdivided into an addressing and
signalling field and an 8 bit header error control (HEC) field. The HEC is used to
protect the header from transmission errors.
Header
5 bytes
Payload
48 bytes
Figure 6.1: An ATM cell
It is specified by the generator polynomial
g ( x)  x 8  x 2  x  1
(6.1)
This HEC can correct single bit errors. ATM, unlike IP networks has built-in
mechanisms for providing different QoS to different type of traffic. Guaranteed QoS
of ATM differs for various applications. For instance, voice and video are sensitive to
cell delay and cell delay variation, but able to tolerate some loss of cells. Data on the
other hand is not affected by cell delay but cannot tolerate cell loss.
118
Chapter 6: Performance of ATM over satellite links
Table 6.1: Effects of bit errors on ATM cell
Error hits Error is
Consequence of error
Header
Header
Valid cell
Cell misinsertion
Header
Payload
Payload
Corrected
Wrongly
corrected
Detected
At least one bit
At least two bits
Cell loss
Cell error
Severe cell error
Relevant QoS
parameter at ATM
layer
cell misinsertion
rate(CMR)
cell loss ratio (CLR)
cell error ratio (CER)
severely errored cell
ratio (SECR)
An error burst has several effects on the ATM cell as in Table 6.1 [1] and these
parameters such as CLR, CER are used in the measurement of ATM performance.
The entire ATM cell will be lost if the error in the header is detected but is not
corrected. At the ATM layer only the header is protected. The basic receiver
operations in the ATM layer switches between two modes as depicted in Figure 6.2
[6]:

Correction mode- The receiver remains in the correction mode in the absence of
detected errors in the header. If a single bit error is detected in the header, this
error is corrected and the receiver switches to the detection mode.

Detection mode- If multiple errors are detected the cell is discarded and the
receiver switches to the detection mode. When operating in the detection mode,
all cells with detected errors in the header are discarded by the receiver. If the
header is received with no detected errors, the cell is accepted and the receiver
switches to the correction mode.
119
Chapter 6: Performance of ATM over satellite links
Multibit error detected
(cell discarded)
No error detected
(no action)
No error detected
(no action)
Detection
mode
Correction
mode
Error detected
(cell discarded)
Single bit error detected
(correction)
Figure 6.2: ATM HEC operation at receiver
In correction mode the probability of cell loss is that of more than one error occurring
in the header, whereas in the detection mode, this probability is that of one or more
errors occurring in the header. For simplicity, cell loss was obtained in our simulation
by assuming the receiver to be in the detection mode, i.e. cell loss results if one or
more errors occur in the header.
The payload or the information field is not protected at ATM layer, but an error
detection mechanism exists at the ATM Adaptation Layer (AAL) for the payload. If
the payload contains at least one bit error this is called an ATM cell error. However, if
the multiple bits are in error then the ATM cell is said to be severely errored. In our
simulation, cell error is declared if one or more bit errors affect the 384- bit payload.
The AAL layer does not provide error correction but relies on the error correction
capability of the higher layers such as TCP/IP in the protocol stack model.
120
Chapter 6: Performance of ATM over satellite links
6.3
Concatenated coding with Reed-Solomon and convolution coding:
The block diagram of concatenated coding scheme is shown in Figure 6.3 where
Reed-Solomon (RS) code serves as an outer code with block interleaving and the
convolution code is used as an inner code. The RS codes are linear block codes with
an alphabet size of 2 m , where m is the number of bits in a symbol. An RS code is
specified as RS (n,k), which means that k m-bit symbols of information are added to r
= n-k parity symbols to form an n-symbol codeword. The codeword length is n = 2 m 1 symbols. If an RS code has r redundant symbols, the code is able to correct any
pattern of t symbol errors, where r = 2t. Reed-Solomon codes are particularly
advantageous when used in the presence of error bursts as present at the output of a
Viterbi decoder. This is because an error burst of length b can corrupt s i (the integer
part of b/m) or s i +1 symbols at maximum. This makes the concatenation of RS (outer
codes) with convolution codes (inner codes) a very powerful FEC technique that
offers high reliability at modest complexity. The performance of such concatenated
codes is enhanced by inserting a symbol interleaver/de-interleaver, which reduces the
correlation of the errored symbols. However, in the case of rare occurrence of very
long error burst that exceeds the correction capability of RS decoder, the whole
decoding process fails.
Outer code
Interleaver
Inner code
Modulator
Channel
Outer
decoder
Deinterleaver
Inner decoder
Demodulator
Figure 6.3: Block diagram of a concatenated coding scheme
121
Chapter 6: Performance of ATM over satellite links
The outer RS code has m = 8 bits per symbol and codeword length n = 255 bytes. The
message portion is k = 223 bytes. Parity portion is n-k = 32 bytes with correctable
errors t = ½(n-k) = 16 bytes. With this particular RS code, the overhead associated
with the parity symbols is only about 15 percent. The inner code is a rate ½
convolution code with constraint length of 7, decoded by an 8-level soft decision
Viterbi decoder. The Reed-Solomon code is decoded by a Massey-Berlekamp
algorithm. An interleaving depth of 5 is employed in the interleaver [7] [8] [9].
Block interleaving is an efficient technique to combat the effect of burst errors which
involves rearranging symbols in a sequence in a predefined manner from two or more
codewords before transmission on the channel (Figure 6.4). The number of codewords
that are interleaved is referred to as the depth of the interleaver, λ. If the interleaver
has sufficient depth the fading processes that affect the successive symbols belonging
to the same codeword will be uncorrelated as any burst of less than λ symbol errors
results in isolated errors at the deinterleaver output and isolated errors are separated
by at least 2 m -1 symbols [8]. Interleaving does not decrease the long-term bit error
ratio but it is successful in decreasing the number of errors in each codeword, so that
the FEC algorithm is able to correct the erroneous symbols in it after deinterleaving.
As a result, bit error ratio drops significantly at the output of the RS decoder. As,
burst of symbol errors in a codeword is distributed by interleaver and correcting
capability of RS code is t bytes, FEC and interleaving can be effective as long as tλ
exceeds b, the average burst length. For satellite links with convolution code, b
typically lies in the range of ~5 bytes [1]. In such an environment, interleaving depth
of 5 is considered to be sufficient to distribute error burst into independent errored
symbols.
122
Chapter 6: Performance of ATM over satellite links
Figure 6.4: Array representation of symbol interleaver
Figure 6.5: Performance of concatenated coding scheme on AWGN channels
Figure 6.5 shows the bit error performance of the convolution, RS and concatenated
(RS+convolution) coding schemes in an AWGN channel. In terms of system
performance, the concatenated coding scheme provides a coding gain of 7.9 dB at an
123
Chapter 6: Performance of ATM over satellite links
error ratio of 10 6 whereas convolution code and RS code alone provide gains of 5.3
dB and 4.3 dB respectively at the same BER level. As can be seen from simulation
results, RS code outperforms convolution code at higher Eb/No values beyond 6 dB.
6.4
Parameters affected by coding scheme in satellite-integrated networks:
In satellite communications antenna size and power are restricted to minimise earth
station cost. Also, available bandwidth is limited and usually high-rate codes (i.e.
codes with less redundancy) are preferred. However, there is always a trade-off
between coding gain and bandwidth. Coding scheme is usually chosen based on the
Bit Error Ratio (BER) performance to deliver a particular service at a desired level.
Besides BER performance and complexity in implementation, the following points
need to be considered for concatenated coding scheme:

To maximise the use of satellite capacity the code rate of the coding
scheme should be as high as possible while still achieving good protection
at acceptable reduction in information bit rate. If the time for transmission
is to be the same for the coded message as for the uncoded message, the
bandwidth has to be increased to accommodate the higher bit rate. The
required bandwidth is directly proportional to bit rate and it has to be
increased by a factor of 1/coderate. For example, the Reed-Solomon and
convolution concatenated coding scheme has rate of 0.44. If the link
message rate is 8 Mbps then the transmission rate becomes
8 Mbps × 1/.44 = 18.16 Mbps
And the required bandwidth is
124
Chapter 6: Performance of ATM over satellite links
1  
BIF  
 Rb , where α is roll-off factor and Rb is the bit rate
 2 
 1  .2 
=
  18.16 Mbps
 2 
= 10.9 MHz
The extra bandwidth required for concatenated code is only 13.5% compared to the bandwidth required for rate ½ convolution code only.

To randomise the bursty errors symbol interleaving is used. This method
requires no additional bandwidth overhead. However, one disadvantage of
interleaving is that it introduces additional delay which is inversely
proportional to the data rate. This extra delay results as deinterleaving can
be started only after all the interleaved data is received. In the concatenated
coding scheme the symbol interleaver has a depth of 5 codewords and the
codeword length is 255 symbols. Thus if the data rate is 8 Mbps then the
interleaving delay can be calculated as
 255  5  8 
D  2
 = 2.5 ms
6
 8  10 
Considering that the one way satellite propagation delay is around 290 ms
and interactive voice can tolerate upto 400 ms of one-way delay, this interleaving delay is negligible.
125
Chapter 6: Performance of ATM over satellite links
6.5
Performance of ATM with concatenated coding scheme:
In the transmitting side, the outer code is RS (255,223) whose input block comprises 4
ATM cells from the incoming data stream appended with 11 bytes of zeros, and
whose Galois field GF (2 8 ) is calculated using the irreducible polynomial P(x) as
P(x) = 1  x 2  x 3  x 4  x 8
(6.2)
Five consecutive outer codes thus formed are stored row-wise in a buffer as a 5 ×255
code array (Figure 6.4). The code array is then interleaved by being read out columnwise and the binary sequence, regarded as an information sequence, is then encoded
by the inner convolution code encoder.
At the receiving end, the received sequence is first decoded by a Viterbi decoder. The
decoded information bits are then grouped into m-bit symbols in GF (2 8 ) and stored
in the deinterleaver buffer as 5 ×255 code array. The array in the deinterleaving buffer
is read out row by row and each row is decoded by the outer code. The extra bytes of
zeros are then discarded to yield the original ATM cell stream.
126
Chapter 6: Performance of ATM over satellite links
Figure 6.6: CLR and CER vs. Eb /No (filled circles are extrapolated values)
Figure 6.6 shows the result of CLR and CER in terms of required energy per bit to
noise power per unit bandwidth (Eb/No) obtained from MATLAB simulation. As can
be seen from the graph to meet the ITU-T performance objectives of 7.5×10 8 [10] for
CLR (Table 2) over satellite links, the required Eb/No is 2.95 dB. On the otherhand, at
Eb/No of 2.88 dB, the ITU-T performance objective of 1.4×10 6 is met for the CER.
For a given BER, the CER is higher than the CLR since there is a higher probability
that an error burst will impact the 48 byte payload than the 5 byte cell header.
According to the simulation result ATM cell loss ratio is varying proportionally with
the bit error ratio which is expected for burst error environment [11]. The relationship
can be written as
CLR = k×BER
(6.3)
127
Chapter 6: Performance of ATM over satellite links
where k is 15.65 from our simulation. It is worth mentioning that the interleaving
scheme randomizes the error burst thereby decreasing the average burst length which
again increases the CLR.
Table 6.2: ATM performance objectives for satellites (class 1 services)
ATM parameters
ITU objective, end-to-end
ITU objective, satellite
CLR
3×10 7
7.5×10 8
CER
4×10 6
1.4×10 6
6.6
Performance of TCP in terms of segment error ratio with concatenated
coding scheme:
The Transmission Control Protocol (TCP) is a connection oriented, end-to-end
protocol designed for moderate speed terrestrial networks with excellent error
performance where propagation delay is negligible. TCP protocol is not efficient on
geostationary satellite links, which features a high bandwidth delay product together
with a non-negligible random loss of packets. Moreover, the AAL layer (i.e. AAL5
and AAL3/4) for data transfer performs error detection only and relies on the error
correction capability of the transport layer protocol (i.e. TCP). The error correction
strategy taken by TCP is generally to retransmit the complete packet even if the
received packet has only a single bit or a single cell in error. One way of increasing
the efficiency of the TCP protocol on satellite links is to drastically reduce error rate
by using the concatenated coding scheme described above. This eliminates the need
for frequent retransmission of packets and hence boosts throughput efficiency.
128
Chapter 6: Performance of ATM over satellite links
In the transmit side, 128 TCP packets of length 512 bytes from the incoming data
stream construct an IP packet (64 Kbytes) from which 48 payload bytes are taken to
add 5 bytes header to form an ATM cell. Outer RS (255,223) code is then formed as
detailed in Section 6.5. In the receive side ATM cell stream are obtained through the
same scheme as in Section 6.5 and header bytes are discarded to form IP pack from
which original TCP packs are then retained. Header bits are not considered in the TCP
and IP packets as header performs the error detection only.
Figure 6.7: TCP Segment Error ratio vs. Eb /No (filled circles are extrapolated
values)
SER performance of TCP as a function of Eb/No is shown in Figure 6.7. Usually, a
BER of 10 8 or lower is needed for successful TCP transfer [12]. This low BER
corresponds to a SER of 10 6 (post-FEC) according to the simulation, so
129
Chapter 6: Performance of ATM over satellite links
implementation of this coding scheme ensure that most packet losses seen by TCP are
in fact due to congestion for the TCP protocol to invoke congestion control
mechanism.
The errors at bit level will propagate to the packet level as follows:
SER = 1- (1- BER)Segment length
(6.4)
which is true for uncorrelated errors [11]. Our simulation result for SER is lower than
the above relationship for a given BER due to coding scheme implemented and bursty
errors. For example, at BER of 4.33×10-5, equation (6.4) gives SER of 0.02, whereas
SER from simulation yields SER of 0.005. The SER varies proportionally with BER,
the proportionality constant being 124.89. It can be gleaned from Figure 6.7, that the
TCP segment error ratio is higher than the ATM cell error ratio. This is expected since
the TCP segment is of longer length than ATM cell payload.
6.7
Conclusions:
To carry the broadband communication services to remote areas over the satelliteintegrated network model, performance of ATM layer was evaluated. The simulation
results to quantify the satellite BER performance required to support the ATM QoS
parameters in terms of CLR and CER have been presented. These results show that
the level of performance required by ATM layer can be delivered by implementing
concatenated coding scheme.
130
Chapter 6: Performance of ATM over satellite links
Implementation of different FEC schemes at different level of protocol hierarchy can
be found in the scientific literature to improve either the QoS of ATM or error rate of
TCP [2] [3] [4] [13] [14]. In this study, ATM and TCP packets are sent through the
same concatenated coding scheme implemented in the physical layer that is efficient
in optimising the ATM QoS as well as TCP segment error ratio necessary to deliver
services in broadband communication. Simulation result of TCP segment error ratio
performance shows that at BER level of 10 8 , TCP segment error ratio is 10 6 ,
achieved at Eb/No of 2.95 dB. At this segment error ratio, retransmission due to packet
loss will be reduced significantly which improves throughput efficiency.
Besides delivering ATM QoS, concatenated coding scheme ensures a more reliable
channel for TCP to perform optimally over the satellite link. However, the inherently
large propagation delay of satellite links still poses an adverse effect on the essentially
feedback-triggered TCP protocol. Enhancement of the TCP protocol to mitigate the
degrading effects of a large propagation delay is the focus of the next chapter.
131
Chapter 6: Performance of ATM over satellite links
References:
1. T. Kaltenschnee, S. Ramseier, “ Impact of Burst Errors on ATM over SatelliteAnalysis
and Experimental
Results”,
Digital
Satellite Communications,
May,1995, pp. 236-243.
2. E. G. Cuevas, B. Doshi and S. Dravida, “Performance Models for ATM
Applications over 45 Mb/s Satellite Facilities”, Digital Satellite Communications,
May, 1995, pp. 228-235.
3. E. G. Cuevas, “The Development of Performance and Availability Standards for
Satellite ATM Networks”, IEEE Communications Magazine, July, 1999, pp. 7479.
4. J. Lunsford, S. Narayanaswamy, D. Chitre and M. Neibert, “ Link Enhancement
for ATM over Satellite Links”, ICDSC, vol. 1, 1995, pages 129-136.
5. IEE Colloquium on “ATM over Satellite”, 1996.
6. W. Stallings, Data and Computer Communication, Pearson Prentice Hall, New
Jersey, 2007.
7. B Sklar, Digital Communications: Fundamentals and Applications, Prentice
Hall, 2001.
8. L. H. Charles lee, Convolution Coding- Fundamentals and Applications, Artech
House, 1997.
9. S. Haykin, Digital Communications, John Wiley and Sons, 1988.
10. ITU-T, “B-ISDN ATM Layer Cell Transfer Performance”, ITU-T Rec. I.356,
2000.
11. Z. Sun, Satellite Networking Principles and Protocols, John Wiley and Sons,
2005.
132
Chapter 6: Performance of ATM over satellite links
12. Y. Chotikapong and Z. Sun, “Evaluation of Application Performance for TCP/IP
via satellite Links.” IEE Aerospace Group Seminar: Satellite Services and the
Internet, London, 2000.
13. N. Celandroni, “Comparison of FEC Types with Regard to the Efficiency of TCP
Connections over AWGN Satellite Channels”, IEEE Transactions on Wireless
Communications, Vol. 5, No. 6, 2006, pp. 1-11.
14. W. Hamouda and P. McLane, “An integrated FEC coding scheme for ATM
transmission over regenerative satellite networks”, International Journal of
Satellite Communications and Networking, Vol. 23, 2005, pp. 33-46.
133
Chapter 7: TCP/IP performance over satellite links
7 CHAPTER 7
TCP/IP Performance Evaluation over Satellite Links
134
Chapter 7: TCP/IP performance over satellite links
7.1
Introduction:
The Internet Protocol suite employs two transport control protocols: The connection
oriented Transmission Control Protocol (TCP) and the connectionless User Datagram
Protocol (UDP). TCP is reliable as it ensures that transmitted data is delivered and
that packets are received in the correct order whereas UDP does not ensure the data
delivery and is unreliable. A significant amount of today’s internet traffic including
web pages namely world wide web (www) or hypertext transfer protocol (http), file
transfer such as file transfer protocol (ftp), email based on Simple Mail Transfer
Protocol (SMTP) and remote access traffic e.g. Telnet, is carried by the TCP.
TCP is an end-to-end protocol originally developed for terrestrial Local Area
Network (LAN) or Wide Area Network (WAN) etc. that are practically error-free
(BER of 10-8 ~10-10) with negligible propagation delay. A satellite channel manifests
large propagation delay and higher channel errors. The flow control mechanism of
TCP based on the slow start and congestion control algorithms, is unable to utilise
available bandwidth efficiently in networks having large delay nay the large delaybandwidth-product such as satellite link. Moreover, any packet loss in the network is
seen by TCP as congestion indication and consequently it cuts back its window. For
error prone satellite links such behaviour leads to a significant deterioration of TCP
performance. When TCP performs poorly, channel utilization is low.
Different link level, end-to-end [1] and proxy based [2] solutions have been proposed
and studied for reliable and efficient performance of TCP in satellite environment.
However, it is vital to improve the link quality for optimum TCP performance, and
this study focused on two-fold performance enhancing techniques. Firstly,
135
Chapter 7: TCP/IP performance over satellite links
concatenated coding scheme has been implemented in chapter 6 to improve the BER
as low as 10-8 that ensures very low packet error ratio. Secondly, the throughput
performance of TCP has been optimised by tuning two TCP mechanisms namely
window scaling and path MTU (Maximum Transmission Unit) discovery over a
single satellite link in isolation. The design, implementation and evaluation of this
second strategy are the focus of the rest of this chapter.
Consideration of a single link may seem a limitation as there exists other competing
TCP connections in the wide area internet which dominate the satellite connection’s
performance. But due to its inherent features the satellite part might receive a different
treatment and attention with respect to the cabled parts of the networks;
methodologies such as TCP splitting and TCP spoofing bypass the concept of end-toend service by either dividing the TCP connection into segments or introducing
intermediate gateways, with the aim of isolating the satellite link [1].
7.2
TCP overview:
TCP protocol is a congestion control mechanism for smooth data flow between the
send and receive side. Congestion implies that more traffic resides on the network
than what it is designed for. Congestion manifests in terms of

Lost packets (buffer overflow at routers)

Long delays (due to queuing at routers)
Congestion control of TCP connections can be done in two ways:
136
Chapter 7: TCP/IP performance over satellite links

Source based- A transmit window maintained at the sender calculated from
last byte sent and last byte received, which is continuously compared with the
available receiver window throughout the data flow.

Gateway based- By controlling the congestion itself i.e. overflowing queues at
gateways. One method for gateways to notify the source of congestion is to
drop packets, done automatically when queue is full.
The congestion control mechanism of TCP is composed of four algorithms: slow start,
congestion avoidance, fast retransmit and fast recovery. Based on these flow control
mechanisms many variants of TCP exist. The TCP flavours or types regarded as
conformant to the Internet Engineering Task Force (IETF) standard are as follows:
7.2.1
TCP vanilla or Tahoe:
It implements the slow start and congestion avoidance mechanism introduced by
Jacobson in 1988 [3]. TCP switches between two states imposed by receiver such as
receiver maximum window (RCVWND) and slow start threshold (ssthresh), usually
half of RCVWND. Another variable called congestion window (CWND) is
maintained at the sender, which cannot grow beyond receiver’s advertised window
thereby allowing TCP to increase the data transmission at a rate such that the
intermediate router is not overwhelmed. Slow start begins by sending one segment
and waiting for an acknowledgement (ACK). For each acknowledgement the sender
receives, it injects two segments into the network, leading to an exponential increase
in the amount of data being sent. Slow start ends when ssthresh is reached and
congestion avoidance takes over where data transmission rate is slower than ssthresh
state. During congestion avoidance phase, for each acknowledged segment, the
137
Chapter 7: TCP/IP performance over satellite links
congestion window is increased by 1/CWND. This adds one extra segment at the end
of each RTT and hence window grows linearly with every RTT.
Congestion avoidance is used to probe the network for available bandwidth by
sending one additional segment for each Round Trip Time (RTT) up to the receivers
advertised window. Once RCVWND is hit, TCP window grows no further and
continue sending data at this constant rate. In Tahoe, when the sending TCP detects
segment loss (indicating congestion) through three duplicate acknowledgements, it
retransmits the lost segment and drops back into the slow start until the packet
sending rate is half the rate at which the loss was detected and then begins the
congestion avoidance phase.
7.2.2
TCP Reno:
With original slow start/congestion avoidance, Reno also incorporates fast retransmit
and fast recovery [4]. Fast Retransmit reduces the time it takes a TCP sender to detect
a single dropped segment. Rather than waiting for the retransmit time-out (RTO), the
TCP sender can retransmit a segment if it receives three duplicate acknowledgements
for the segment sent immediately before the loss occurrence.
After fast retransmit sends what appears to be the missing segment, congestion
avoidance but not slow start is performed. This is the fast recovery algorithm. It is an
improvement that allows high throughput under moderate congestion, especially for
large windows. The reason for not performing slow start in this case is that the receipt
of the duplicate acknowledgements tells TCP more than just a packet has been lost.
Since the receiver can only generate the duplicate acknowledgement when another
138
Chapter 7: TCP/IP performance over satellite links
segment is received, that segment has left the network and is in the receiver buffer.
That is, there is still data flowing between the two ends and TCP does not want to
reduce the flow abruptly by going into the slow start phase.
Although the joint uses of fast retransmit and fast recovery makes possible for a TCP
Reno connection to achieve high throughput during moderate congestion, it is not
effective in dealing with multiple packet losses in a single window. Each invocation
of these algorithms can retransmit a single segment. At every loss, the congestion
window is reduced to half of its size. Therefore, when multiple losses occur, the
window size is reduced significantly. The return to its original size occurs only after a
considerable delay.
If three or more packets are lost in a row, the TCP sender is forced to wait for a RTO
to find out that a loss occurred, keeping the connection idle for a relatively long
period. Moreover, after an RTO, the TCP sender is forced to enter in a costly slow
start phase.
7.2.3
TCP New Reno:
The New-Reno TCP includes a small change to the Reno algorithm at the sender that
eliminates Reno’s wait for a retransmit timer when multiple packets are lost from a
window [5].
139
Chapter 7: TCP/IP performance over satellite links
7.2.4
SACK:
As in Reno, the SACK TCP implementation enters fast recovery when the data sender
receives three duplicate acknowledgements. The sender retransmits a packet and cuts
the congestion window in half. During fast recovery, SACK maintains a variable
called pipe that represents the estimated number of packets outstanding in the path
(This differs from the mechanisms in the Reno implementation). The sender only
sends new or retransmitted data when the estimated number of packets in the path is
less than the congestion window [5].
7.3
TCP throughput:
Throughput is defined as the total number of bytes delivered to the destination
application (excluding retransmission and losses) divided by the total connection
time. Throughput depends on line or channel speed, TCP window size and round trip
delay [6].
The window size is the number of outstanding data units allowed for a connection.
More specifically, TCP window size is the amount of data that a TCP receiver allows
a TCP sender to send before having to wait for an acknowledgement. It is limited by
the numbering space allotted to the sequence number of a data unit. TCP/IP protocols
are usually lumped together and are embedded in the software for operating systems
and browsers such as Windows and Netscape. Windows 2000 window size is 64
Kbytes and Linux default window size is 32 Kbytes.
140
Chapter 7: TCP/IP performance over satellite links
If the largest window allowed is 64 Kbytes, the maximum theoretical throughput
achievable via satellite is
=
window _ size
Round _ trip _ time( RTT )
Within the defined window size, data is sent with a certain rate. For instance, if the
line rate is 1.53 Mbps, it takes 0.3268 s to output the 64 Kbytes of data. This is the
transmission delay. The Round Trip Time (RTT) or delay is simply given by the sum
of the transmission delay and the propagation delay on the satellite link. Assuming
transmission of 64 Kbytes packet at 1.53 Mbps (reverse transmission time for
acknowledgements is ignored since acknowledgements are only 32 bits) and a fixed
two way propagation delay of 0.58 s we get an RTT of 0.9068 s. A TCP sender with
64 Kbytes of maximum window will be limited to transfer at 64 Kbytes /0.9068 s or
approximately at 71 Kbytes/sec.
TCP performance depends not upon the transfer rate itself, but rather upon the product
of the link bandwidth (link bit rate or channel capacity) and the round trip delay time.
In networking jargon bit rate, Rb is usually mentioned as bandwidth, B, when in fact
Rb  B. The proportionality constant, k depends on the modulation scheme used. For
instance, if Binary Phase Shift Keying (BPSK) is used then k = 1 and Rb = B. The
“bandwidth-delay product” (BDP) measures the amount of data that would “fill the
pipe”- it is the minimum buffer space required at sender and receiver to obtain
optimum throughput on the TCP connection over the path. BDP is the amount of
unacknowledged data that TCP must handle in order to keep the pipeline full. Buffer
141
Chapter 7: TCP/IP performance over satellite links
size required in the network for high TCP performance is proportional to the BDP of
the network [7].
7.4
Enhancing TCP over satellite channels:
In keeping with the objective that, where possible, the TCP should not be modified to
accommodate the satellite links, Request for Comments 2488 (RFC-2488)
recommends the following [8], which are compliant with IETF standards-
7.4.1
Path MTU discovery:
Path MTU discovery is a method that allows the sender to find the largest packet and
hence, largest TCP segment size that can be sent without fragmentation. As TCP
sender increases its window by segments, larger packets allow the congestion window
to increment faster in terms of number of bytes carried. However, large packets will
be fragmented if they encounter a network with small MTU. This proposed solution
lets TCP detect the largest packet that can cross the Internet without incurring the cost
of fragmentation and reassembly. Although there are delay as well as complexity
involved in implementing Path MTU discovery, overall, it improves the performance
of TCP over satellite links.
7.4.2
TCP window scaling:
TCP layer uses a 16-bit word to notify the send TCP of the size of the receive
window. Allowing 1 byte for certain overheads, the biggest window size that can be
declared for the receive window is 216 – 1 = 65535 bytes. As part of the changes to
142
Chapter 7: TCP/IP performance over satellite links
add timestamps to the sequence numbers, IETF enhanced TCP to negotiate a window
scaling option. The option multiplies the value in the window field by a constant. The
effect is that the window can only be adjusted in units of the multiplier. So, if the
multiplier is 4, an increase of 1 in the advertised window means the receiver is
opening the window by 4 bytes. The maximum multiplier permitted is 214. This means
the maximum window size is 230 or 1 Gbytes [9]. Applications must negotiate the
large sending and receiving buffer sizes to trigger the use of window scaling during
connection setup.
The two mechanisms Protection Against Wrapped Sequence (PAWS) and Round Trip
Time Measurement (RTTM), are extensions that should be used with large windows.
By including a timestamp in the TCP header, the RTT can be measured. TCP provides
32 bits for specifying a sequence number for the segment or frame. With large
windows the corresponding space of 231 sequence numbers can be exhausted quickly
and reused again. As a result numbering of old sequences might overlap with new, a
condition known as wrap-around. The PAWS is an algorithm that also makes use of
the timestamp. With such timestamp technique, the ambiguity caused by wrap-around
can be eliminated.
It needs to be mentioned that for a satellite link shared among many flows, large
windows may not be necessary. For example, two long-lived TCP connections each
using a window of 64 Kbytes, can fully utilise a T1 GEO satellite channel as
explained below:
BDP = 1.53 Mpbs × 0.58 s ~ 113 Kbytes
143
Chapter 7: TCP/IP performance over satellite links
which is the effective window needed to fill the pipe. The required number of
connections to fill the pipe can be obtained by dividing the effective window by single
connection window size as
113
~ 2
64
7.4.3
Selective Acknowledgements:
When multiple losses occur in a window of data, current TCP’s cumulative
acknowledgement carrying only the sequence number of the next pack expected by
the receiver is unable to recover from more than one lost segment per RTT. Selective
Acknowledgements (SACK) make it possible for TCP to acknowledge data lost or
received out of order. Sender is able to identify the packets correctly received and
detect the gap in the receiver buffer and able to retransmit multiple lost packets in the
same RTT resulting in faster recovery if there are enough acknowledgements
returning to the sender.
7.4.4
Forward error correction:
Higher error rates in satellite networks impact TCP performance in two ways. Firstly,
errors corrupt the datagram, which will have to be retransmitted. Secondly, the lost
packet due to transmission errors is seen by TCP as due to congestion invoking
cutting down the window i.e. data rate. It is required to either reduce the error rate to a
level acceptable to TCP or adopt a mechanism that will allow TCP to differentiate
between congestion and corruption error and thus TCP will not reduce transmission
rate in case of corruption.
144
Chapter 7: TCP/IP performance over satellite links
Besides the above mitigations there are several other techniques not yet recommended
by IETF such as-
7.4.5
Split TCP connections:
To shield the high latency network from the rest, network can be split into smaller
loops in a way that it is transparent to applications, although this solution breaks the
end-to-end semantics associated with the TCP protocol. The main obstacle to the
deployment of split-connection protocol gateways is their interaction with a security
infrastructure. In particular, any IP security protocols that encrypt the payload of an IP
packet render a split connection gateway useless. TCP may be split in the following
ways [10] [2]-

TCP spoofing- In this approach, the gateway on the network side of the
connection prematurely acknowledges data destined for the satellite host, to
speed up the sender’s data transmission. It then suppresses the true
acknowledgement stream from the host and takes responsibility for resending
any missing data. As long as the traffic is primarily unidirectional, TCP
datagrams are passed through the gateway without alteration. In the reverse
direction, the same strategy is followed. No changes are needed at the satellite
client.

TCP splitting- Instead of spoofing, the connection may be fully split at the
gateway on the network side and a second TCP connection may be used from
the satellite gateway to the satellite host. Logically, there is not much
difference between this approach and spoofing, except that the gateway may
try to run TCP options that are not supported by the terrestrial server. Modern
145
Chapter 7: TCP/IP performance over satellite links
firewall implementations often perform a type of TCP splitting such as
sequence number remapping for security reasons.

Web caching- If satellite based web users connect to a web cache within the
satellite network, the cache is effectively splitting any TCP connection for
requests that result in a cache miss. Therefore, web caching not only can
reduce the latency for users in fetching data from the web, it has the benefit of
splitting the transport connection for cache misses.
7.4.6
Multiple data connections (XFTP):
Several TCP connections can be opened to send a single file. One advantage of this
technique is that this does not require any modifications to standard TCP. Using many
connections simultaneously increases the maximum and the initial windows resulting
in faster growth of throughput. XFTP uses an adaptive algorithm to find the
appropriate number of connections so to not overload the network [11].
7.5
MATLAB simulation environment for TCP protocol:
The MATLAB simulation environment consists of TCP send (server) and TCP
receive (client) sides. The TCP algorithms are implemented from the sender’s
perspective. The packets are generated as a sequence of integer numbers whereas in
actual TCP implementation packets are maintained in bytes. The state variables
ssthresh, RCVWND and CWND are all maintained in packets.
The ACK triggered window evolution of TCP can be seen in Figure 7.1 as in
Jacobson [3] assuming ssthresh of 4 packets (RCVWND of 8 packets). A 6 Kbytes
146
Chapter 7: TCP/IP performance over satellite links
file transfer is considered for illustration. If TCP segment size is of 512 bytes then 12
segments are required to be transferred. The link rate is 1.53 Mbps.
Figure 7.1: The chronology of slow start and congestion avoidance algorithm of
TCP
Initially a packet is sent with slow start mode. If it is not in error then the packet is
declared as received and an ACK is generated. With every ACK next two packets are
sent by send TCP, provided the flight size (last packet sent – last packet received) is
less than ssthresh. When ACK for packet 4 is received at the beginning of fourth
147
Chapter 7: TCP/IP performance over satellite links
RTT, flight size is 3 and TCP is allowed to push only one packet with each ACK
received i.e. congestion avoidance phase takes over. In this mode of operation,
CWND is updated as a fraction of current CWND and at the end of this RTT one
extra packet can be pushed resulting in a linear growth of the window. With packet
12, 6 Kbytes file transfer is completed. If there were more packets to be sent then this
mode of operation would have continued till TCP hit the RCVWND (8 packets), after
which TCP windows grow no further. This is the steady-state condition, where TCP
transfers data at a constant rate i.e. with every ACK one packet is sent.
Figure 7.2: Illustration of RTT calculation
Figure 7.2 shows the corresponding RTT for 6 Kbytes file transfer. The first RTT
starts with sending packet 1 and ends after receiving an ACK for the packet sent.
Upon receiving this ACK, next packets (packet 2 and 3) are sent and next RTT starts
and ends when ACKs for packet 2 and 3 arrive. Similarly, fourth RTT ends upon
148
Chapter 7: TCP/IP performance over satellite links
receiving the ACK for the last packet sent (packet 12). As 6 Kbytes file contain 12
segments, no packets were sent with the ACKs received in this last RTT and the
connection terminates. The elapsed time to transfer a number of packets in any RTT is
calculated as the sum of fixed propagation delay (PD) and transmission time for the
packets sent in the corresponding RTT.
The total connection time at the end of fourth RTT is 2.35 sec. Throughput is obtained
by dividing the number of bytes transferred by the connection time, which is 2.55
Kbytes/sec. Window evolution in the event of packet loss is not described further for
simplicity. It needs to be mentioned that different attributes related to the actual TCP
implementation such as impact of segment size on IP datagram, receiver buffer usage
threshold, maximum ACK delay, timer granularity are not taken into consideration in
MATLAB implementation of TCP protocol. These attributes influence the TCP
performance under different network conditions such as packet loss, delay variation,
queue management in router, etc.
To observe the performance of MATLAB implementation of TCP, simulation was
performed with different parameters such as file size (10 Mbytes), window size (240
Kbytes), propagation delay (0.58 s), segment size (1000 Bytes) and link rate (1.3
Mbps) [10].
149
Chapter 7: TCP/IP performance over satellite links
Figure 7.3: Comparison of results from MATLAB with experiment
In Figure 7.3 our simulation results are compared with the experimental findings of
Henderson and Katz [10], and shows broad agreement. The experimental results are
sample means from several independent transfers of 10 Mbyte files where congestion
loss occurred due to insufficient router buffer. In the MATLAB simulation there was
no congestion loss and error events were introduced by randomly selecting a packet to
be in error which would be the scenario in practice for a high quality link of BER
lower than 10-8. The throughput performance shows good agreement since TCP
responds in a similar way to loss events, whether from congestion or transmission
corruption.
150
Chapter 7: TCP/IP performance over satellite links
7.6
Variation of TCP throughput with propagation delay:
The algorithms of TCP Reno, the most widely implemented TCP variants, were
implemented in MATLAB to analyse the performance in networks having different
propagation delay with current TCP implementation i.e. 64 Kbytes maximum window
size. File of 10 Mbytes were transferred having Local Area Network (LAN) standard
TCP segment size of 512 bytes. The link in the networks has no bit errors or
congestion loss or variations in propagation delay, which, while not representative of
all networks, exemplifies the common case. TCP and IP header bytes were not
excluded in the throughput calculation for better comparison with the link capacity.
Figure 7.4: TCP throughput at different propagation delay
Figure 7.4 shows TCP throughput obtained from MATLAB simulation. As can be
seen from the Figure 7.4 at 0.58 s propagation delay, achievable throughput is 0.5068
151
Chapter 7: TCP/IP performance over satellite links
Mbits/sec (~65 Kbytes/sec) with 64 Kbytes window, which is only 33% of the
channel capacity. At 0.1 s propagation delay, the achievable throughput is 1.135
Mbits/sec, which is 74% of the channel capacity. So the link utilization in the satellite
link with large propagation delay (~500-600 ms) is very low compared to the
terrestrial link where propagation delay is in the region of 100 ms and below.
7.7
TCP (Reno) throughput performance over satellite link:
To investigate the impact of propagation delay on TCP protocol, simulation was
performed for transferring different file sizes from 200 Kbytes to 1 Gbytes, in the
Figure 7.5: Throughput for different file sizes without loss (Maximum TCP
window 64 Kbytes)
152
Chapter 7: TCP/IP performance over satellite links
satellite link transmitting at T1 rate i.e. 1.53 Mbps with 0.58 s propagation delay. For
relatively smaller file sizes such as 200 Kbytes, the overall throughput is low as
evident from Figure 7.5 due to the amount of time spent in the slow start phase. On
the other hand, larger file sizes can achieve throughput up to the TCP limit as the slow
start penalty is amortized over a longer duration of file transfer. Figures 7.6 and 7.7
show the window evolution of file transfer of 200 Kbytes and 10 Mbytes respectively.
File transfer of 200 Kbytes was completed within the slow start and congestion
avoidance phase resulting in overall low throughput. On the other hand, 10 Mbytes
file transfer spent more time at the constant rate of ~71 Kbytes/sec thereby obtaining
higher throughput. Transfer of different ftp file sizes reveals that both the TCP
window size and the TCP slow start and congestion control algorithms contribute to
the observed limits in throughput.
Figure 7.6: Window dynamics for file size of 200 Kbytes (Maximum TCP
window 64 Kbytes)
153
Chapter 7: TCP/IP performance over satellite links
Figure 7.7: Window dynamics for file size of 10 Mbytes (Maximum TCP window
64 Kbytes)
Even if the network has a high capacity and the router has large buffering capability,
the throughput performance of TCP is limited by the maximum allowed window size
i.e. 64 Kbytes. This follows since a TCP source cannot send more than the window
size of data in an RTT. Window scaling i.e. large TCP window as specified in RFC
1323 [12], can allow TCP to send data at higher rates. Window sizes from 64 Kbytes
to 1 Mbytes were considered to transfer file of size 10 Mbytes which is large enough
to take advantages of large windows. It was assumed that router buffer was always
large enough to cache the data when TCP window grows large so that no segment was
lost due to congestion. As can be seen from Figure 7.8, the throughput generally
increases
154
Chapter 7: TCP/IP performance over satellite links
Figure 7.8: Throughput for a 10 Mbytes file transfer for different TCP window
sizes (without loss)
with the TCP window size. The throughput is 145.67 Kbytes/sec for 1 Mbytes TCP
window, which is 76% of the channel capacity. Thus, for a longer duration TCP
connection, which is transferring data at T1 rate over satellite links, achievable
throughput is satisfactory with large TCP window.
It is obvious that smaller file transfer such as http file (www web pages) [1] suffer
from slow start and congestion avoidance attributes of the protocol. Currently, there
are no standardized solutions to overcome the performance problems associated with
small file size, although several techniques have been studied and proposed in the
literature. TCP can be allowed to use an initial congestion window of four segments
maximum (4380 bytes) rather than one segment [11]. Many web pages are usually
155
Chapter 7: TCP/IP performance over satellite links
less than this size, so transfer of http files would then complete in one RTT rather than
two or three. This technique is referred to as “4K slow start” or 4KSS.
There are also investigations of the potential for caching congestion information from
a recently used connection in order to start the new connection from a larger initial
window size [13] [14]. It was shown that if initial window starts from a higher value,
more data can be sent within first few RTTs resulting in better throughput for shorter
transfers [1].
Figure 7.9: Throughput for a 10 Mbytes file transfer for different TCP window
sizes (with loss and without loss)
In the designed satellite-integrated network a low error rate is achieved (i.e. packet
error event is ~ 10-6 at 99.99% of time) by implementing concatenated coding scheme
[chapter 6] which presents TCP with a more reliable link having rare packet errors. To
156
Chapter 7: TCP/IP performance over satellite links
observe throughput efficiency in the presence of transmission errors, loss was
introduced into the simulation by randomly injecting single loss events into a 10
Mbytes file transfer corresponding to a packet error ratio (PER) of 4.88×10-5 . The
throughput generally decreases for all the window sizes, in particular, with 1 Mbytes
window throughput reduced to 64.8% of the channel capacity (Figure 7.9).
Throughput is decreased by no more than ~11 -12 % for all the large window sizes in
the presence of loss. It is obvious that TCP Reno’s fast retransmit and fast recovery
algorithm is efficient in recovering from single packet loss from a window of data
especially for large file transfer.
Figure 7.10: Throughput for a 10 Mbytes file transfer for different TCP window
sizes with large segment size (no loss)
As described in section 7.4, large segment size can be considered to yield faster
growth of congestion window. The TCP maximum segment size is 9180 bytes
157
Chapter 7: TCP/IP performance over satellite links
recommended for TCP connections over long latency link [15]. Figure 7.10 shows the
throughput performance of large segments with large windows for a 10 Mbytes file
transfer. The performance increases proportionately with increasing window size
except that at 1 Mbytes window size the throughput is comparatively lower. This is
due to the fact that there were not enough segments in a 10 Mbytes file to take full
advantage of the use of a large window size along with a large segment size.
Figure 7.11: Throughput for a 100 Mbytes file transfer for different TCP
window sizes with large segment size (no loss)
To observe the simultaneous effect of large window and large segment size, 100
Mbytes file was transferred. According to the simulation results in Figure 7.11, at 1
Mbytes window size with 9180 bytes segment size, the throughput is 169.83
Kbytes/sec, which is 89% of the channel capacity. When a large file is transferred the
throughput is improved even with a standard segment size of 512 bytes since a large
158
Chapter 7: TCP/IP performance over satellite links
proportion of the file transfer time is spent at constant rate rather than in the initial
slow start and congestion avoidance phase with large window sizes.
Figure 7.12: Comparison of percentages of throughput with different window
sizes over terrestrial and satellite links
159
Chapter 7: TCP/IP performance over satellite links
Figure 7.13: Comparison of percentages of throughput with different window
sizes and file sizes over satellite links
The comparisons of throughput obtained with different combinations of window size,
segment size and file size can be seen in the bar graphs (Figure 7.12 and 7.13). For
very large file transfer of 1 Gbytes, the throughput is 90% with large window and
large segment size as in Figure 7.14. This result is very useful for many network
applications such as distance learning, which use massive file transfer.
Figure 7.14 shows the throughput comparisons over satellite link for large file transfer
without loss and with loss. As can be seen one packet error in 1 Gbytes file transfer
has no impact on the overall throughput performance as TCP catches up with
maximum transfer rate after the initial loss over long duration of file transfer. These
results confirm that TCP performance is as efficient as terrestrial link when operating
with large windows and large segment size over satellite link when large ftp file
160
Chapter 7: TCP/IP performance over satellite links
transfer is concerned. Moreover, TCP with large segment size of 9180 bytes could be
efficient for shorter file transfer as file transfer would be completed within 1 or 2
RTTs.
Figure 7.14: Comparison of percentages of throughput over satellite links with
loss and without loss
In our network model, bottleneck link of 8 Mbps can support many users at the same
time (CQN model). In other words, 8 Mbps is available to any users to transfer a file
at any time, hence the link utilisation is low. TCP performance is improved by
applying end-to-end enhancement mechanisms over the network.
161
Chapter 7: TCP/IP performance over satellite links
7.8
For
Conclusions:
efficient
performance
of
satellite-integrated
network
for
broadband
communication, all the TCP enhancement techniques currently standardized by IETF
were implemented such as Forward error correction, window scaling and Path MTU
discovery except Selective Acknowledgements. The results presented in this chapter
show that TCP Reno with large window and large segment size can allow an
application to achieve high throughput over GEO satellite links for large file transfers.
The performance has not been degraded significantly in the presence of a single
packet loss in the file transfer. However, if path MTU discovery is not implemented
with large windows, short-lived file transfers such as web pages still suffer from
protocol attribute such as slow start threshold and congestion avoidance phase,
showing poor performance. Techniques could be adopted as mentioned in section 7.4
for short file transfer over satellite links.
It was assumed that there was only a single packet error in the file transfer. As
concatenated coding scheme in the network improves BER to values as low as 10-8 to
10-10 at very low Eb/No, this assumption is not unrealistic. TCP Reno with large
windows and large segment size works well if the satellite link is more reliable and
there is no or moderate congestion. If network congestion and more channel errors
occur, SACK could be implemented to improve performance. However,
implementation of SACK is currently not within the scope of the thesis.
Link level solutions by implementing FEC and automatic repeat request (ARQ)
mitigate the problem of corruption loss so that TCP performs better. But ARQ is not
very efficient in large delay environment and FEC implemented at link level does not
162
Chapter 7: TCP/IP performance over satellite links
guarantee reliable service at TCP layer. Moreover, none of these schemes address
TCP protocol attributes. In this context, end-to-end solutions with robust coding
scheme in the physical layer are more attractive for TCP performance optimisation.
To achieve a much simpler protocol design, an end-to-end approach was adopted for
TCP so that link characteristics can be ignored. In more complex network
heterogeneity, such an end-to-end approach suffers severely. Proxy-based solutions
rhyme with this approach by isolating hosts from details of link characteristics thereby
allowing TCP to perform optimally. These categories of solutions such as end-to-end,
link level or proxy-based can be used together or on its own in a network by trading
off their different performance objectives and additional system complexities [16].
This chapter concludes the parameter optimisation for a reliable satellite-integrated
network to carry broadband traffic. Finally, system level analysis and simulation of
the designed network model will be found in the next chapter.
163
Chapter 7: TCP/IP performance over satellite links
References:
1. M. Marchese, “TCP modifications over satellite channels: study and performance
Evaluation”, International Journal of Satellite Communications, Vol. 19, 2001,
pp. 93-110.
2. M. Luglio, M. Y. Sanadidi, M. Gerla and J. Stepanek,"On-Board Satellite ‘Split
TCP’ Proxy”, IEEE Journal on Selected Areas in Communications, Vol. 22,
No.2, 2004, pp. 362-370.
3. V. Jacobson, “Congestion Avoidance and Control”, Proceedings of ACM
SIGCOMM’ 88 Conference, 1988.
4. M. W. Murhammer, O. Atakan, S. Bretz, L. R. Pugh, K. Suzuki and D. H. Wood,
TCP/IP Tutorial and Technical Overview, Prentice Hall, 1998.
5. K. Fall and S. Floyd, “Simulation-based Comparisons of Tahoe, Reno and SACK
TCP”, ACM Computer Communications Review, Vol. 26, No. 3, 1996.
6. W. R. Stevens, TCP/IP Illustrated, Vol 1, Addison Wesley, 1994.
7. R. Goyal, R. Jain, M. Goyal, S. Fahmy, B. Vandalore, S. Kota, N. Butts and T.
VonDeak, “Buffer Management and Rate Guarantees for TCP over Satellite-ATM
Networks”, International Journal of Satellite Communications, Vol. 19, 2001, pp.
111-119.
8. M. Allman, S. Floyd and C. Partridge, “Incresing TCP’s Initial Window”, Internet
RFC 2414, 1998.
9. C. Partridge and T. Shepard, “TCP Performance over Satellite Links”, IEEE
Network, Vol. 11, No. 5, 1997, 44-49.
10. T. R. Henderson and R. H. Katz, “Transport Protocols for Internet-Compatible
Satellite Networks”, IEEE Journal on Selected Areas in Communications,
Vol. 17, No 2, 1999, pp. 326-344.
164
Chapter 7: TCP/IP performance over satellite links
11. M. Allman, D. Glover, J. Griner, K. Scott, J. Touch and D. Tran, “Ongoing TCP
Research Related to Satellites”, RFC 2760, 1998.
12. V. Jacobson, R. Braden and D. Borman, “TCP Extensions for High
Performance”, RFC 1323, May 1992.
13. V. Padmanabhan and R. Katz, “TCP Fast Start: A Technique for Speeding Up
Web Transfers”, Proceedings of IEEE Globecom ’98 Internet Mini-Conference,
1998.
14. J. Touch, “TCP Control Block Interdependence”, Internet RFC 2140, 1997.
15. R. Goyal, R. Jain, M. Goyal, S. Fahmy, B. Vandalore and T. VonDeak,
“Traffic Management in ATM Networks over Satellite Links”, NASA/STI
Program, 1999.
16. V. G. Bharadwaj, J. S. Baras and N. P. Butts, “An Architecture for Internet
Service via Broadband Satellite Networks”, International Journal of Satellite
Communications, Vol. 19, 2001, pp. 29-50.
165
Chapter 8: Link budget analysis and simulation
8 CHAPTER 8
Link Budget Analysis and Simulation
166
Chapter 8: Link budget analysis and simulation
8.1 Introduction:
The overall performance of satellite-integrated networks could be assessed by the
quality of satellite links. The physical parameters of the transmission in the network
such as data rate (link bandwidth) and Eb/No are determined by the traffic model
(CQN) and the coding/modulation scheme implemented in chapters 5 and 6. These
two parameters set-up the minimum threshold C/No required to transmit the signal
with desired quality in the network. The satellite link is 8 Mbps and the Eb /No is 2.95
dB for both the uplink and downlink. In the link budget process appropriate selection
of other factors such as transmitter power, transmitter gain, receiver gain and the
calculated parameters such as free space loss, other losses, system noise temperature,
etc. are required for practical implementation.
Link level simulation was performed to optimise the BER level required for
multimedia services at a certain Eb /No. The results then can be used to determine the
overall system availability through system level simulation. The initial set-up for the
system level simulation is link budget analysis. As actual propagation data are not
available for the region, propagation measurement at Sparsholt, UK has been
incorporated to evaluate the performance of the communication system over a
dynamic channel condition. A new approach of seasonal downlink power control to
combat rain attenuation is presented. Implications of site diversity and adaptive fade
mitigation techniques on system availability are also discussed.
167
Chapter 8: Link budget analysis and simulation
8.2 Link budget set up:
The selected station (Dhaka) location is 23.72 0 N and 90.41 0 E with elevation angle of
46.1 0 to a GEO satellite at 60 0 E. The uplink frequency is 19.7 GHz and the downlink
frequency is 18.7 GHz. The system employs Quaternary Phase Shift Keying (QPSK)
modulation and concatenated coding scheme and the access scheme is Time Division
Multiple Access (TDMA) for uplink and Time Division Multiplexing (TDM) for
downlink.
Table 8.1: Earth stations parameters
Transmit e/s
Latitude
Longitude
Elevation angle
23.72
90.41
46.1
Receive e/s
Latitude
Longitude
Elevation
angle
Range
23.72
90.41
46.1
Units
Degrees
Degrees
Degrees
Range
37334
37334
km
Service
Required bit
rate(each e/s)
Modulation
Coding scheme
Required BER
E b /N o
Code rate
Voice,data,video
8
Mbps
QPSK
RS+conv(1/2)
10 8
2.95
dB
Antenna
diameter
Access system
1.8
No of e/s
Burst rate
Threshold C/N
3
24.89
2.37
0.44
Antenna
diameter
1.8
m
TDMA/TDM
1
Mbps
dB
The equivalent isotropically radiated power (EIRP) and G/T for the satellite are 57
dBW and 8 dB/K for the zonal beam and 75 dBW and 19.91 dB/K for the spot beam.
168
Chapter 8: Link budget analysis and simulation
Tables 8.1 and 8.2 summarise the link budget parameters for earth station and
satellite.
Table 8.2: Satellite parameters
Satellite(GEO)
Longitude
Uplink
frequency
Downlink
frequency
EIRP(saturated)
G/T
Bandwidth
8.3
60
19.7
Units
Degrees
GHz
18.7
GHz
57(zonal beam)
8 (zonal beam)
54
75 (spot beam)
19.91 (spot beam)
dBW
dB/K
MHz
Link budget analysis:
Table 8.3 shows the most significant link budget parameters. The link budget analysis
results in 16.9 dB of margin under clear air condition.
In satellite communication network, atmospheric attenuation especially rain
attenuation degrades the link quality of high frequency band (Ku, Ka) significantly.
At 0.01% of time the predicted rain attenuation is 34 dB at 18.7 GHz frequency band
for the selected station and the link budget fails (Table 8.4), which suggests that the
year round operation of high-availability links will not be feasible. At 0.1% of time,
the rain attenuation is 16 dB and the link margin is around 1 dB to keep the link
operational for low availability services.
169
Chapter 8: Link budget analysis and simulation
Table 8.3: Most significant link budget parameters (clear air) for zonal beam
Uplink:
Uplink frequency
e/s Tx power
e/s Antenna gain
Path Loss
Miscellaneous Losses
Sat transponder G/T
19.7 GHz
20 dBW
49.52 dB
209.77 dB
0.5 dB
8 dB/K
Boltzmann’s constant
Noise bandwidth
-228.6 dB/Hz/K
74.54 dBHz
Uplink C/N
21.30 dB
Downlink:
Downlink frequency
Sat EIRP(saturated)
EIRP@req data rate
Path Loss
Miscellaneous Losses
e/s Antenna gain
18.7 GHz
57 dBW
54.99 dBW
209.32 dB
0.5 dB
49.07 dB
System temperature
Boltzmann’s constant
Noise bandwidth
21.76 dBK
-228.6 dB/Hz/K
74.54 dBHz
Downlink C/N
23.53 dB
Overall C/N
Threshold C/N
Link margin
19.26 dB
2.37 dB
16.9 dB
However, rain attenuation over Bangladesh has very strong seasonal dependence.
Most of the rainfall occurs during monsoon months, whereas rain attenuation is 11 dB
during non-monsoon period at 0.01% of time. So, a high availability communication
link can be provided during non-monsoon period in this region and others of similar
climate.
170
Chapter 8: Link budget analysis and simulation
Table 8.4: Most significant link budget parameters (under rain) for zonal beam
Rain only in downlink
Downlink frequency
Availability
Rain attenuation
∆N
Downlink C/N under rain
Link margin under rain
18.7 GHz
99.99 %
34 dB
4.02 dB
-14.49 dB
-16.86 dB
Availability
Rain attenuation
∆N
Downlink C/N under rain
Link margin under rain
99.9 %
16 dB
3.96 dB
3.57 dB
1.13
Generally, for efficient utilisation of satellite communication resources, a simple
seasonal link budget approach could be considered by setting up the transmitter power
of earth station at two different levels, one for rainy months and a much lower setting
for dry winter months. However, increasing transmitter power from the earth station
does not improve the link quality any further as the overall link is downlink limited.
To combat the severe weather condition during monsoon month, an innovative
downlink power control technique is implemented. In this approach the satellite link
is switched to a high power rating transponder (i.e. transmit power in the range of 240
W or more) with spot beam having higher EIRP of 75 dBW. By considering rain
attenuation at 0.01% of time the link margin is around 1 dB with spot beam
configuration as in Table 8.5. So, high availability communication services can be
provided during rainy months by switching the link to a spot beam from the zonal
beam configuration for dry winter months.
171
Chapter 8: Link budget analysis and simulation
Table 8.5: Most significant link budget parameters (under rain) for spot beam
Uplink:
Uplink frequency
e/s Tx power
e/s Antenna gain
Path Loss
Miscellaneous Losses
Sat transponder G/T
19.7 GHz
20 dBW
49.52 dB
209.77 dB
0.5 dB
19.91 dB/K
Boltzmann’s constant
Noise bandwidth
-228.6 dB/Hz/K
74.54 dBHz
Uplink C/N
33.21 dB
Downlink:
Downlink frequency
Sat EIRP(saturated)
EIRP@req data rate
Path Loss
Miscellaneous Losses
e/s Antenna gain
18.7 GHz
75 dBW
69.99 dBW
209.32 dB
0.5 dB
49.07 dB
System temperature
Boltzmann’s constant
Noise bandwidth
21.76 dBK
-228.6 dB/Hz/K
74.54 dBHz
Downlink C/N
41.53 dB
Overall C/N
Threshold C/N
Link margin
32.62 dB
2.52 dB
30.23 dB
Rain only in downlink
Downlink frequency
Availability
Rain attenuation
∆N
18.7 GHz
99.99 %
34 dB
4.02 dB
Overall C/N under rain
Link margin under rain
3.5 dB
1.13 dB
As the selected operating frequency bands for the Fixed Satellite Service (FSS) are
also allocated for terrestrial Fixed Service (FS), radio-navigation etc. [1] over the
172
Chapter 8: Link budget analysis and simulation
region, it is desirable to estimate the maximum power flux density produced at the
surface of the earth by emissions from a satellite to avoid harmful interference. For all
conditions and methods of modulation, the power flux density threshold is [2]
-115
dB(W/m2 )
for
θ ≤ 5o
-115 + 0.5(θ-5)
dB(W/m2 )
for
5o < θ ≤ 25o
-105
dB(W/m2 )
for
25o < θ ≤ 90o
where θ is the angle of arrival of the radio-frequency wave (degrees above the
horizontal).
The power flux density with spot beam configuration (EIRP = 75 dBW) is calculated
as
EIRP - Bn – 10×log10(4πR2) = -161.98 dBW/m2
where Bn is data bandwidth and R is slant range. The resulting power flux density is
well below the threshold level. Therefore, other services will not be affected by the
satellite spot beam transmission.
173
Chapter 8: Link budget analysis and simulation
8.4
Time series simulation:
Accurate estimates of the point-to-point link conditions are needed to characterise the
influence of rain attenuation on the quality of communication link. In the statistical
link budget process, the time percentage occurrence of outages corresponding to a
given attenuation level is calculated to evaluate the system performance. To simulate
link performance over dynamic channel conditions, time series of attenuation
Figure 8.1: Comparison of Attenuation level for Sparsholt and Dhaka
measurements is required, which is not available over the region. A pre-processed
attenuation time series on 26th Deccember, 2003 at 20 GHz measured at Sparsholt is
considered although Sparsholt is of different climatic conditions than Dhaka.
However, by comparing the predicted attenuation level for Sparsholt and Dhaka
174
Chapter 8: Link budget analysis and simulation
(Figure 8.1) at 18.7 GHz frequency band and 46.1 0 path elevation a factor is
estimated, which is applied to generate the attenuation time series of Dhaka.
Figure 8.2: Attenuation time series for Sparsholt and Dhaka
Figure 8.2 shows the comparison of attenuation time series of Sparsholt and Dhaka.
Although this approach of time series generation does not provide sample-by-sample
accuracy, it nevertheless gives a good statistical representation of the rain attenuation
dynamics in Dhaka.
The resulting time series of attenuation is then applied to the simulation algorithm
which consists in deriving time varying E b /N o ratios over the satellite channel
comprising the 19.7 GHz uplink, a bent pipe transponder with zonal and spot beam
configuration and the 18.7 GHz downlink. Figures 8.3 and 8.4 show the instantaneous
175
Chapter 8: Link budget analysis and simulation
behaviour of the link in terms of E b /N o variation for a rain event occurring on the
downlink with zonal beam and spot beam mode of operation respectively. As can be
seen from the Figure 8.3 and 8.4, the overall E b /N o is always above the minimum
required threshold (2.95 dB) and the link remains available during the event except at
a particular instant when rain fade goes down to ~60 dB.
Figure 8.3: Instantaneous link behaviour with rain event for Dhaka (Zonal
beam)
Attenuation characteristics of Bangladesh suggest that at 0.005% of time the
attenuation is in the range of 40 dB and deep fade in the range of 40 to 60 dB are not
unexpected during monsoon period. Studies show that convective rain cells extend
over more limited regions in tropical regions, a typical cell size ranging from 2-5 km
and often lasting for only 10- 20 minutes [3] [4]. High availability services would be
disrupted for several seconds to minutes during a rain event which can not be
176
Chapter 8: Link budget analysis and simulation
improved with higher satellite power as is suggested by time series simulation where
fade depth of around 10 to 20 dB can be compensated for. Different Fade Mitigation
Techniques (FMT) such as power control, data rate reduction could be considered to
improve the link availability under severe weather conditions. However, these
techniques were not implemented in the current study due to the lack of attenuation
measurements over the region.
Figure 8.4: Instantaneous link behaviour with rain event for Dhaka (Spot beam)
Besides switching to spot beam configuration to compensate for high rain attenuation,
implementation of site diversity i.e. setting up two earth stations at different locations
could be considered. Analysis of rain cell size distribution presented in chapter 4
shows that the diameters of intense rain cells are in the range of 10 km. Site diversity
works because the cell structure of heavy rain means that the probability of such rain
occurring simultaneously at two sites separated by, say, 10 km is very small compared
177
Chapter 8: Link budget analysis and simulation
to the probability of the same rainfall rate occurring at one of the sites. The
improvement in diversity performance can be characterised in two ways:

The diversity improvement factor is defined as the ratio of the percentage of
time that a given attenuation is exceeded at a single site to the percentage of
time that the same attenuation is exceeded at both sites simultaneously.

Diversity gain is the difference (in dB) between the rain attenuation exceeded
for a given percentage of the time at a single site and the attenuation exceeded
for the same percentage of the time taking account of diversity.
Large improvements in the performance of a system may thus be obtained by
providing two separate earth stations and using the transmission path which has the
lowest instantaneous attenuation.
Selection of techniques to compensate rain fade depends on the complexity of the
scheme and the cost incurred as well as on the required availability of different
services. Site diversity is considered to be very efficient in combating rain fade at the
expense of a second site. Besides extra earth station cost, dedicated terrestrial
interconnection,
switching
complexity,
required
data
buffering
and
site
synchronisation add significant system overhead. In general, site diversity is best
suitable for network control centres and major gateways and not applicable to lowcost earth stations unless applied with public switched networks to interconnect the
terminals. In the current telecommunication network design using higher satellite
power is considered rather than implementing site diversity technique to provide a
large rain margin to provide services at 99.99% availability.
178
Chapter 8: Link budget analysis and simulation
8.5
Conclusions:
A reliable communication network to route multimedia data services requires accurate
estimates of the impact of propagation impairments on system performance. In this
chapter a realistic communication system is conceived through the link budget
analysis. According to the link budget analysis, high availability services (at 99.99 to
99.999% of time) can be provided during dry months as rainfall over the region is
highly seasonal. To operate communication services at 99.99% availability during
rainy months requires switching the link over the spot beam configuration.
The generated attenuation time series is incorporated to observe the link behaviour
over dynamic channel conditions during a rain event. This simulation suggests that
fade depth in the range of 10 ~ 20 dB can be mitigated by using higher satellite
power. However, switching the link over the spot beam mode can not improve the
link when rain fades in the range of 30 ~ 60 dB occur, which are not unexpected over
the region during heavy rain events. So, to operate with very high availability services
at 99.999 % of time different FMTs need to be considered to restore the link quality
and thus to improve the availability.
179
Chapter 8: Link budget analysis and simulation
References:
1. BTRC, Bangladesh National Frequency Allocation Plan (NFAP), Vol.3.1, July,
2005.
2. ITU-R, “Maximum permissible values of power flux density at the surface of
earth produced by satellites in the fixed-satellite service using the same frequency
bands above 1 GHz as line-of-sight radio relay systems”, Rec. ITU-R SF.358-5,
1995.
3. H. E. Green, “Propagation impairments on Ka-Band SATCOM Links in
Tropical And Equatorial regions”, IEEE Antennas and Propagation Magazine,
Vol. 46, No. 2, 2004, pp.31-45.
4. Q. W. Pan and G. H. Bryant, “Results of 12 GHz Propagation Measurements in
Lae (PNG), Electronics Letter, 28, 1992, pp. 2022-2024.
180
Chapter 9: Conclusions and further work
9 CHAPTER 9
Conclusions and Further Work
181
Chapter 9: Conclusions and further work
9.1
Conclusions:
The presented investigation of a reliable satellite-integrated network for broadband
service provision yielded the following interesting findings:

Seasonal behaviour of rainfall was an important factor in improving the link
availability over the network.

Intense rain cell has extension in the range of 10 km.

CQN model is more suitable in dimensioning the link bandwidth of
multimedia traffic.

Concatenated coding scheme is efficient in improving the link quality at a
level required for ATM and TCP performance.

TCP performance is quite satisfactory with window scaling and Path MTU
discovery over satellite link.

High availability satellite services can be realised in tropical climates such as
Bangladesh through a strategy of seasonal downlink power control whereby
there is a switch from a global/zonal beam to spot beams during the monsoon
months. This delivers an extra 20 dB power flux density on the ground to cope
with the higher level of rain attenuation during this season.
Besides these specific solutions for multimedia communication in the network many
other conclusions may be drawn from all the stages of the work. An effort has been
made to organise the conclusions in a format relevant to the chapter wise progression
of the research.
182
Chapter 9: Conclusions and further work
Long-term rainfall data were analysed to devise a more reliable attenuation prediction
model over the region. Along with the attenuation model, seasonal variation of
rainfall proved to be important elements in improving the availability of the network.
It was also found that annual rainfall was normally distributed with 2055 mm of mean
annual rainfall over Dhaka. Spectral analysis of rainfall time series revealed that
rainfall pattern oscillates between high to low in every three years period.
Radar and rain gauge data over Sparsholt, UK were analysed to derive the rain cell
size distribution. Statistics of rain cell size is an important factor in setting up the
diversity stations for site diversity implementation to overcome the severe rain fade
over a location. The result was not integrated in the design procedure directly as the
method of switching to spot beam configuration was proved to be more suitable
solution for the network to perform optimally in the latter part of the research.
However, implication of the cell size distribution is discussed as a more general case
in rain fade compensation technique.
Once propagation criteria over the region were solved, design of the network started
with appropriate calculation of the link capacity required to route the traffic over the
satellite link to connect a remote location with the existing networks in city areas. To
calculate the link capacity CQN model was applied, which gives per-flow or perconnection bandwidth for multimedia communication. According to the model, 50
users could be supported in the network with 8 Mbps link capacity. Unlike the
conventional traffic models CQN model takes account of the self-similar
characteristics of multimedia traffic and hence was more appropriate to calculate the
link capacity.
183
Chapter 9: Conclusions and further work
Current telecommunication infrastructure over the region was analysed before
justifying the suitability of a satellite-integrated network. ITU telecommunication
indicators showed that cellular communication was improving faster although internet
usage was very low and limited to city areas. Moreover, broadband link is frequently
disrupted due to meteorological conditions over the region. Deployment of a satelliteintegrated network was an appropriate solution to extend broadband communication
all over the country.
Concatenated coding scheme, a standard coding scheme for deep space
communication was implemented to optimise link quality. It was shown through the
simulation that an interleaving depth of 5 in the block interleaver is sufficient to
improve BER at required level to deliver ATM QoS parameters as well as TCP packet
error ratio. The interleaving delay in this scheme is negligible compared to the delay
associated with ATM cell header interleaving or header bit interleaving which are
usually adopted for ATM performance optimisation over satellite links. The extra
required bandwidth for concatenated coding scheme is only 13% compared to the
bandwidth for standard coding scheme i.e. convolution coding over satellite links.
The ITU-T performance objectives of 7.5×10 8 for CLR and 1.4×10 6 for CER over
satellite link were met at required Eb/No of 2.95 dB and 2.88 dB respectively. It was
shown that CLR varies proportionally with BER, the proportionality constant being
15.65. The SER of TCP was also related with BER through a proportionality constant
of 124.89. The implemented coding scheme ensures successful TCP transfer by
presenting a more reliable satellite channel, where BER of 10-8 or lower were
achieved with nominal Eb /No.
184
Chapter 9: Conclusions and further work
TCP performance enhancing mechanisms such as window scaling and Path MTU
discovery were implemented for optimum delivery of broadband services over the
designed network system. It can be concluded from the simulation that TCP
performance over satellite link is as efficient as terrestrial link with these
enhancement mechanisms.
The performance of the whole network was evaluated through link budget analysis. It
was shown that seasonal variability of rainfall proved to be an advantage to operate
the network with high availability services. During winter there is no or little rainfall
and high availability services can be operated without any disruption. Switching the
link over the spot beam configuration was considered to be a more suitable option to
combat deep rain fade during monsoon months. However, very high availability
services at 99.999% of time still would be disrupted with spot beam operation as
higher satellite power able to improve the link from fade depth of 10 ~ 20 dB. During
the occurrence of large rain fade as large as 30 ~ 40 dB often lasting for only a few
minutes, the higher satellite spot beam EIRP was not sufficient to maintain the link. In
this case, implementations of different FMTs, most notably site diversity, were
suggested to improve the link quality and to provide the high availability services.
An important design criterion for a newly constructed network is to ensure that the
interference levels must remain within the acceptable limits set by ITU to avoid
harmful signal distortion. Interference level due to spot beam transmission was
evaluated to comply with ITU level so that other existing communication links
sharing the same frequency bands would not be degraded.
185
Chapter 9: Conclusions and further work
Finally, it can be concluded that the designed network with optimised parameters at
different layers of protocol stacks is a feasible solution to efficiently deliver high
availability broadband services to facilitate communication to large population
especially for users separated by long distances including rural areas and remote
islands. Although it remains to be seen whether the network is economically viable,
the solution is a significant contribution towards the current trend of satellite
broadband networks.
9.2
Future prospects:
The investigation suggests the following interesting directions for further work:

More rain cell could be tracked to obtain more reliable statistics of rain cell
translation velocity.

Bursty traffic could be generated to analyse the impact of queuing delay in the
network model.

More functionality of actual TCP behaviour could be extended in MATLAB
implementation of TCP protocol.

SACK could be implemented in MATLAB to observe the impact of window
scaling and Path MTU discovery in the event of successive packet error in a
window.

Synthetic rain field for Dhaka could be generated to obtain attenuation time
series for application in link simulations.
186
Appendix
Appendix-A
Kinematics of the horizontal wind field
y
y
V
(x,y)
v
(x0,y0)
x
x
u
Figure A.1: Horizontal wind field
Let us consider the coordinate system for motion vector V as in Figure A.1. To
estimate the wind field at an arbitrary point x, y from the wind at a nearby point x 0 ,
y 0 , we can consider the Taylor’s expansions in terms of x and y assuming that the
measurement circle is horizontal-
u  u0 
u
u
x
y
x
y
v
v
v  v0  x 
y
x
y
(A.1)
considering (x 0 , y 0 ) at origin and neglecting higher order terms. We can write
v
1 v
1 v
x
x
x
x
2 x
2 x
So that,
187
Appendix
u  u0 
v  v0 
1 u
1 u
1 u
1 u
x
x
y
y
2 x
2 x
2 y
2 y
(A.2)
1 v
1 v
1 v
1 v
x
x
y
y
2 x
2 x
2 y
2 y
We can also write,
0
1  v v  1  v v 
y     x  
2  x x  2  y y 
(A.3)
0
1  u u  1  u u 
y     x  
2  x x  2  y y 
By combining equations (A.2) and A.3), the equation for u and v can be written as,
1  u v  1  v u 
1  u v  1  v u 
u  u0     x     y     x     y
2  x y  2  x y 
2  x y 
2  x y 
(A.4)
1  u v 
1  v u  1  u v 
1  v u 
v  v0     y     x     y     x
2  x y 
2  x y  2  x y 
2  x y 
Equation (A.4) gives the horizontal wind field as in Figure A.1. Any linear wind field
and a non-linear wind field (approximately) can be characterised by the five terms as
in equation (A.4) such as:
Translation (first term in equation (A.4)): u 0 and v 0
188
Appendix
Divergence (second term in equation (A.4)):
u v

x y
∂u/∂x, is the rate at which u rotates clockwise, which is the horizontal gradients of the
horizontal wind component and ∂v/∂x, is the rate at which v rotates anti-clockwise.
When the sum of these two terms is positive, the area is increasing, which is
divergence and when the sum is negative, the area is decreasing resulting
convergence.
Physically,
divergence
“dilutes”
rotation
and
convergence
“concentrates” rotation.
Relative vorticity (third term in equation (A.4)):
v u

x y
Physically, vorticity is the curl of the vector wind, defined positive for counterclockwise rotation as  ×V. As the velocity components are measured relative to a
frame of reference fixed in the earth, this quantity is known as the relative vorticity.
Stretching deformation (fourth term in equation (A.4)): u  v
x y
When the horizontal gradient of the horizontal wind component is more than the
vertical gradient of the vertical wind component, the area is stretched along the
horizontal direction.
Shearing deformation (fifth term in equation (A.4)):
u v

y x
189
Appendix
This effect is caused by the rate at which u and v rotates anticlockwise. The above
states of wind field can be visualised in Figure A.2.
y
y
x
x
Convergence: Change in area, no change
in orientation, shape, location
Translation: Change in
location, no change in area,
orientation,
y shape
y
x
x
Vorticity: Change in orientation, no change
in area, shape, location
Stretching deformation: Change in shape,
no change in area, orientation, location
y
x
Shearing deformation: Change in shape, no change
in area, orientation, location
Figure A.2: Different states of wind flow
190
Appendix
Appendix-B
Wind field from Radar observation:
Figure B.1: Radar geometry.
Let us consider a Cartesian coordinate system, Oxyz, whose origin O is the location of
the radar. The spherical coordinates r, α and β of the measurement point are
associated with the Cartesian system, where r is the range, α is the elevation angle and
β is the azimuth angle starting from the north. The components of the velocity of
scatterers (Doppler velocity) along Ox, Oy and Oz are denoted by u, v and w
respectively. The radial velocity can be written as
Vr  u cos  cos   v cos  sin   w sin 
(B.1)
191
Appendix
Vr  [u 0 
u
u
v
v
x
y ] cos  cos   [v0  x  y ] cos  sin   w sin 
x
y
x
y
u
u
x cos  cos  
y cos  cos  
x
y
v
v
v0 cos  sin   x cos  sin   y cos  sin   w sin 
x
y
(B.2)
Vr  u 0 cos  cos  
Replacing
(B.3)
x = rcosαcosβ and y = rcosαsinβ
u
u
r cos  cos   cos  cos   r cos  sin   cos  cos  
x
y
v
v
v0 cos  sin   r cos  cos   cos  sin   r cos  sin   cos  sin   w sin 
x
y
Vr  u 0 cos  cos  
u
u
r cos 2  cos 2   r cos 2  cos  sin  
x
y
v
v
v0 cos  sin   r cos 2  cos  sin   r cos 2  sin 2   w sin 
x
y
Vr  u 0 cos  cos  
(B.4)
Using identity,
1
sin 2
2
1 1
sin 2    cos 2
2 2
1 1
cos 2    cos 2
2 2
sin  cos  
We obtain the equation as,
192
Appendix
1
u v 1
u v
Vr  u 0 cos  cos   v0 cos  sin   r cos 2  (  )  r cos 2  (  ) cos 2 
2
x y 2
x y
1
v u
r cos 2  (  ) sin 2  w sin 
2
x y
Vr is a periodic function with base period 2π that can be written in the form of a
Fourier series expansion:
Vr 

1
a 0   a n cos n  bn sin n 
2
n 1
(B.5)
Equation (B.5) has the form of such an expansion limited to the first three
components. The Fourier coefficients are given by
a0 
 u v 
1
r cos 2      w sin 
2
 x y 
a1  v0 cos 
(B.6)
(B.7)
b1  u0 cos 
(B.8)
 u v 
1
a 2   r cos 2    
2
 x y 
(B.9)
b2 
 u v 
1
r cos 2    
2
 y x 
(B.10)
The kinematic properties of the local field are derived directly from the Fourier
coefficients. The first component gives horizontal divergence as
193
Appendix
u v
2


(a0  w sin  )
x y r cos 2 
(B.11)
The horizontal wind field can be obtained from second component as
uo = b1/cosα and vo = a1/cosα
(B.12)
The third component contains the stretching deformation as
2a 2
u v


x y
r cos 2 
(B.13)
And shearing deformation as
2b2
u v


y x r cos 2 
(B.14)
194
Appendix
Appendix-C
List of significant MATLAB file
File/Function
Description
npwr_dha.m
Perform spectral analysis of a time-series
(annual rainfall)
probtest.m
Implementation of an algorithm to find
optimum bin size for the distribution (normal,
exponential etc.) to best fit into the
population frequency curve
RiceHolm.m
Obtain rain rate distribution from annual
rainfall (Rice-Holmberg model)
atten_dha_rf.m
Improved rain attenuation model for Dhaka
crane_risk_estimation.m
Implementation of crane ad hoc model to
estimate associated risk is prediction
filt_reflectivity_correlation.m
Obtain translation velocity from successive
radar scans by correlation method
id_filter.m
Calculate the Fourier transform of a signal
(such as Doppler velocity) and perform ideal
filtering
doppler_fft.m
Find the Doppler speed from a PPI scan
rain_event.m
Implement an algorithm to find the rain event
from rain gauge data
calculate_cell_diameter.m
Compute the rain cell diameter from radar
scans
traffic.m
Calculate capacity by traffic model (Erlang B
and Erlang C)
ATM_RS.m
Simulation of ATM cell loss ratio and cell
error ratio with concatenated coding scheme
TCP_IP.m
Simulation of TCP segment error ratio with
concatenated coding scheme
TCP.m
Implementation of TCP protocol (Tahoe)
TCPRENO.m
Implementation of TCP protocol (Reno)
195
Appendix
link_budget_TDMA.m
Link budget analysis
link_budget_EbNo.m
Simulation algorithm to derive time varying
Eb/No over the satellite channel
196