See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/4375798
Intelligent approach to video transmission over 2.4 GHz wireless technology
Conference Paper in IEEE International Conference on Fuzzy Systems · July 2008
DOI: 10.1109/FUZZY.2008.4630372 · Source: IEEE Xplore
CITATIONS
READS
2
25,310
2 authors:
Prof. Hassan Kazemian
Soamsiri Chantaraskul
London Metropolitan University
The Sirindhorn International Thai-German Graduate School of Engineering (TGGS), …
80 PUBLICATIONS 635 CITATIONS
37 PUBLICATIONS 159 CITATIONS
SEE PROFILE
All content following this page was uploaded by Prof. Hassan Kazemian on 04 August 2014.
The user has requested enhancement of the downloaded file.
SEE PROFILE
Intelligent approach to video transmission over 2.4 GHz
wireless technology
H. B. Kazemian Senior Member, IEEE and S. Chantaraskul
Abstract—Video transmission over dynamic channels like 2.4
GHz wireless technology is unpredictable, because of
interferences caused by other wireless devices in the same ISM
(Industrial, Scientific and Medicine) frequency band and/or
general channel noises. Transmitting Moving Picture Expert
Group (MPEG) video stream over wireless technology also
presents a challenge, as MPEG demands large bandwidth.
Considering these issues, an intelligent transmission is
introduced in this research so that the controller may adjust
itself to the current state of the wireless channel in order to
sustain MPEG video quality during the communication process.
A neural-fuzzy controller and a rule-based fuzzy controller are
implemented in the design to monitor input – output of a traffic
shaping buffer and offer suitable parameters for the MPEG
encoder for video transmission over the network. Matlab
computer simulation results show that the proposed method
reduces data loss and improves image quality as compared with
traditional MPEG video transmission over 2.4 GHz wireless
technology.
I. INTRODUCTION
M
PEG video compression uses the similarities and/or
redundancies between consecutive frames or images in
order to reduce the size of video stream. In Variable Bit Rate
(VBR) coding, the complex and intricate scenes usually
contain more data bits, which demands a larger bandwidth
for transmission [1]. This kind of video coding offers
constant video quality by statically setting the quantization
scale for I- (Intra), P- (Predicted), and B- (Bi-directional)
frames while the bit rate is allowed to vary. This paper
utilizes an intelligent control system to guarantee that the
VBR coding from the host satisfies the traffic situation of the
wireless channel during the transmission period.
Many researchers have applied soft computing techniques
such as neural networks and fuzzy logic to wireless
technology. For example, Kumar and Venkataram proposed
a Hopfield neural-network-based multicast routing algorithm
for constructing a reliable multicast tree that connects the
participants of a multicast group for video-on-demand data
transmission. Computer simulation results show that the
proposed scheme facilitates a multicast routing algorithm for
high-speed mobile networks [2]. Zhang and Zhu used fuzzy
logic technique to QoS routing allowing efficient use of
network resources for transmission and distribution of audio
and video. Computer simulation results show that the fuzzy
routing algorithm makes use of network resources more
effectively [3]. Mar and Kao used Cascade Fuzzy Logic
Control (CFLC)-based Movable Boundary Wireless Multiple
Access in the UMTS (MBWIMA/UMTS) protocol for voice
and video data rate control. The simulation results
demonstrate that the CFLC-based MBWIMA/UMTS
protocol using the first-duplicated Space Division Multiple
Access (SDMA) scheduling can greatly improve both the
voice-video dropping probability and data packet delay,
compared with the MBWIMA/UMTS and GPRS/UMTS
protocols using the first-duplicated SDMA scheduling [4].
Razavi et al have recently researched into fuzzy logic control
of automatic repeat request (ARQ). Bluetooth’s default
(ARQ) scheme is not suitable for video distribution, resulting
in missed display and decoded deadlines. Razavi et al have
therefore used fuzzy ARQ with active discard of expired
packets from the sent buffer. In summary simulation results
demonstrate that in poor channel conditions, fuzzy control of
ARQ provides at least 4 dB improvement in video quality
[5].
Furthermore, there has been a substantial amount of
research work in the application of hybrid artificial
intelligence techniques, such as fuzzy logic and neural
networks to wireless technology. For instance, Su et al uses a
fuzzy neuron network for General Packet Radio Services
congestion control. The proposed scheme is capable of
adjusting input and output, increase robustness, stability, and
working speed of the network. As shown by experimental
results, the algorithm is capable of significantly preventing
the congestion for video transmission, which proves that the
method is more practical and effective than traditional
methods for congestion control in wireless networks [6].
Kazemian and Meng applied fuzzy logic to MPEG video
transmission over Bluetooth. The computer simulation
results demonstrate that the use of the fuzzy controllers
reduces excessive data loss at the Host Controller Interface
as compared with an open loop VBR/CBR video
transmission in Bluetooth [7]. This paper takes the research
further by introducing neural-fuzzy to control the departure
rate and speeding up the quantization level using fuzzy logic,
which results into an intelligent methodology suitable for
real-time transmission.
II. MOVING PICTURE EXPERT GROUP (MPEG)
H. B. Kazemian (e-mail: h.kazemian@londonmet.ac.uk) and S.
Chantaraskul, London Metropolitan University, Computing Dept., 100
Minories, London EC3N 1JY, UK.
In digital video a major setback of packet loss is that it
may lead to degradation in video quality as MPEG
246
c
978-1-4244-1819-0/08/$25.002008
IEEE
compression uses temporal dependencies and motion
compensation. Therefore, handling and recovering from
packet loss has been widely researched upon in video
transmission. In this work, we propose such system that
recognizes the channel situation and provides the
transmission speed in order to keep in line with bandwidth
availability, hence providing less packet loss and delay
during the transmission process. There are three main
components, temporal model, spatial model and entropy
encoder, briefly described below [8].
Ropen(k+1)
RBF Rate Control
Rd(f last)
^
Ra(k+2)
X(k)
Qscale
V(k)
RBFC
Ra (k+1)
MEPG
Encoder
X(f)
Traffic
shaper
Rd(f)
rd(f)
X(f)
Y(f)
Ra (k+1)
Y(f)
ractual
NFC
Token
bucket
ractual
HC INTERFACE
Wireless Hardware
BLUETOOTH MODULE
Air
Fig. 1. Intelligent video transmission in wireless technology.
The P and B pictures first pass through the temporal
model. In temporal model, the reference frame is used to
achieve the motion vector, known as predictive-coded
picture. This process is called motion estimation, where each
small block of P or B picture is searched against blocks
within the reference frame to get the motion vector. Spatial
compression is applied to a single frame of video. The
degree of spatial compression affects the overall video
quality, and this is the part where our design focuses in order
to manipulate the video transmission speed. In general, the
concern is to balance between the picture quality and
compression degree. However, video transmission speed is a
crucial factor in choosing suitable configuration for spatial
compression as well as the picture quality and compression
degree.
Discrete Cosine Transform (DCT) and Quantization are
the main functions in spatial model. DCT is a method of
compressing an N×N block of data into a weighted sum of
spatial frequencies. The following equation represents a 2-D
DCT for an N×N block, which is employed here.
S pq
N 1 N 1
ª S (2 x 1) p º
ª S (2 y 1)q º
D pD q ¦¦ Txy cos «
»¼ cos «¬
»¼
2N
2N
¬
x 0 y 0
(1)
where p, q = horizontal and vertical frequency indices =
0,1,2,…,N-1
Dp and Dq are given by:
2008 IEEE International Conference on Fuzzy Systems (FUZZ 2008)
247
D
p
­
°
°
®
°
°¯
1
N
for
p
2
N
for
p z 0
0
(2)
Inverse DCT can be calculated by using Txy where x, y =
0,1,2,…,N-1. Quantization takes the DCT coefficients and
maps them into a quantized signal with a reduced range of
values. A scalar quantizer is used here, where each
coefficient is mapped to obtain an individual quantized
value. In an entropy encoder process, probabilities of the
symbols, which represent elements of video sequence are
assigned. These probabilities are then used in order to
produce a compressed bit sequence, which is ready for the
transmission.
III. INTELLIGENT VIDEO TRANSMISSION
Fig. 1 outlines the overall diagram of a rule-based fuzzy
(RBF) Controller and a neural-fuzzy (NF) Controller for
MPEG VBR video transmission over 2.4 GHz wireless
technology. As the diagram shows, in this research a trafficshaping buffer is introduced to manipulate and co-ordinate
the VBR encoding video prior entering the wireless channel
[9]. The shaper buffer’s role is to smooth the video output
traffic and to partially eliminate the burstiness of the video
stream entering the network. The inputs to the RBF
controller are the fuzzified mean value X (k ) and the
fuzzified standard deviation V (k ) of the queue length from
the traffic-shaping buffer. The output from the RBF
controller is the fuzzified desired arrival rate RÖ a (k 2) . The
departure rate of the last frame Rd ( flast ) and the estimated
data-rate Ropen (k 1) for an open-loop algorithm are used to
calculate the desired arrival rate RÖ a (k 2) . The inputs to the
NF controller are the fuzzified queue length X ( f ) from the
traffic-shaping buffer and the fuzzified available tokens from
the generic cell rate algorithm [10-11] commonly known as
token bucket Y ( f ) . It should be noted that token bucket or
generic cell rate algorithm is usually associated with data
transmission over most wired and wireless standards.
However, token bucket rate policing method is largely
inadequate to sufficiently regulate transmission of video data
sources over wireless standards. The output from the NF
controller is the fuzzified departure data rate rd ( f ) . f is
denoted for a frame. The range of the departure rate Rd ( f )
is between the arrival rate Ra (k 1) and the actual
architecture [12] and uses fuzzy sets (i.e. fuzzy membership
functions) as its weights at the input and output layers. The
nodes of the hidden layer embody the fuzzy IF-THEN rules.
The conventional back-propagation procedure [13] for multilayer neural networks is utilized to train the fuzzy
membership functions. The parameters associated with the
membership functions change through the learning process
such that the network interprets the desired input/output map
of the controller as accurately as possible. Finally, the
parameters attained from the training procedure are fed back
into the fuzzy system to facilitate the best control
performance. The NF controller is based on Sugeno or
Takagi-Sugeno-Kang method [14]. Therefore, the proposed
adaptive scheme is a Sugeno RBF1 controller whose output
membership functions are of a first-order. Sugeno method
works well with optimization and adaptive techniques.
This paper introduces a novel rule-based fuzzy set to
automate the rate control in the MPEG encoder. In Fig. 1,
the RBF Rate Control is placed before the MPEG encoder.
The desired RÖ a (k 2) is an input to the RBF Rate Control.
In this approach, Qscale for I picture is attained first, followed
by Qscale for P picture and then B picture. The process starts
by using initial Qscale and find the difference between the
offered Ra and the desired RÖ a (k 2) . The result will
determine the next Qscale to be tested. For example, if the
offered Ra is higher than the desired RÖ a (k 2) , Qscale could
be decreased. Hence the new Qscale to be tested is the
minimum Qscale. The same process continues to find Qscale for
P picture then B picture. Eventually, a set of Qscale for I, P
and B is offered for the new GOP to be used in the MPEG
encoder.
The rules of this RBF Rate Control are reasonably straight
forward yet effective. The input of the RBF Rate Control
consists of 11 membership functions shown in Fig. 2 as level
1 to level 11. These levels represent 11 range of values
between 0 to 1. The input arrived at the RBF Rate Control is
then mapped into this membership function and passed
through the rule-based fuzzy set to offer the output. The
major advantage of this novel approach is the speed [15],
which is one of the most important aspects in video
transmission.
TABLE 1. MATRIX RELOADED – COMPARISON OF SIMULATION RESULTS
FROM VBR AND INTELLIGENT SYSTEM
Average GOP size
(kbit)
Standard deviation
of GOP size (kbit)
Open-loop system
Intelligent system
416.82
454.61
123.06
29.519
transmission (token) rate ractual.
An adaptive technique is incorporated into the control
scheme through the use of a neural-fuzzy controller in Fig. 1.
First, a neural network represents a rule-based fuzzy
controller. This network comprises of a three-layered
248
2008 IEEE International Conference on Fuzzy Systems (FUZZ 2008)
3000
Ra (kbit/s)
level1 level2 level3 level4 level5 level6 level7 level8 level9 level10 level11
1
1000
0
0.6
0
100
200
300
400
500
600
700
800
900
1000
700
800
900
1000
700
800
900
1000
Frame Number
3000
0.4
ractual (kbit/s)
Degree of membership
0.8
2000
0.2
2000
1000
0
0
0
100
200
300
400
500
600
Frame Number
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
normalised Rdif
Fig. 2. Membership functions of RBF Rate Control input.
IV. COMPUTER SIMULATION RESULTS
The MPEG decoder developed here needs to be able to
cope with missing information. The designed system
described in section III shows two buffers, the traffic shaper
and the token bucket. At some points in the simulation, when
one or both of these buffers are full and data stream is still
provided from the MPEG encoder, the excessive data will
have to be dropped. In this research, only the B picture can
be dropped as the loss of B picture will not affect other
frames in the GOP. As a result, the MPEG decoder needs to
be modified in order to cope with the loss of B picture. This
is done by replacing the missing information with the default
values to complete the stream for each component. The
components are then ready to be decoded, described in
section II.
Video clip “Matrix Reloaded” is used in this research to
simulate the proposed intelligent system [16]. A Bluetooth
channel is used to represent 2.4 GHz wireless frequency for
the adaptive video streaming technique. Simulations are
carried out for both open-loop encoding system or VBR
system and the proposed intelligent system in order to
compare the results. Table 1 gives summarized values for
both plots showing that the standard deviation of the
intelligent system is much lower than the standard deviation
obtained for the open-loop encoding system, which results in
reduction in burstiness and data loss. Moreover in Table 1,
the GOP size is averaged for both systems and there are
more GOP sizes for the intelligent system than the open-loop
system demonstrating more data for transmission resulting in
better picture quality.
Ra and ractual (kbit/s)
0
3000
2000
1000
0
0
100
200
300
400
500
600
Frame Number
Fig. 3. Matrix Reloaded – Open-loop encoding system results for arrival
rate (Ra) and actual transmission rate (ractual).
Fig. 3 and Fig. 4 summarise the results obtained from the
conventional open-loop system and the proposed intelligent
system respectively. It can be seen that in terms of
transmission rate, the intelligent technique can manipulate
the MPEG encoder as well as regulate the bit rate in order to
keep the output bit rate Rd in line with the actual
transmission speed ractual much better than the open-loop
system. The burstiness of video data in Fig. 4 is much better
controlled for the departure rate Rd than the arrival rate Ra
resulting in better image quality. The overall data drop is
much more for the open-loop system in Fig. 3 than the
intelligent system in Fig. 4.
Picture frames from both systems were captured for the
comparison purposes. It was observed that the frames from
the open-loop system display data loss, noise and distortion.
However, the original image could roughly be seen and this
is because of the advanced technique in spatial compression
offered by MPEG scheme. The motion vector is a very
important part in reconstructing the image for the B picture,
which was allowed to be dropped in this work. With full set
or part of motion vector, the image can be reconstructed
from the reference frame(s). Furthermore, the open-loop
system experienced severe loss in many frames in fast
moving events, while the images captured from the
intelligent system showed no data loss. In this scenario, no
data survives for the particular pictures in the open-loop
system, and without motion vectors, the images cannot be
reconstructed.
2008 IEEE International Conference on Fuzzy Systems (FUZZ 2008)
249
Ra (kbit/s)
3000
REFERENCES
2000
[1]
1000
0
0
100
200
300
400
500
600
700
800
900
1000
[2]
Frame Number
Rd (kbit/s)
3000
[3]
2000
1000
0
0
100
200
300
400
500
600
700
800
900
1000
[4]
Frame Number
ractual (kbit/s)
[5]
3000
2000
1000
0
0
[6]
100
200
300
400
500
600
700
800
900
1000
Ra, Rd, and ractual (kbit/s)
Frame Number
[7]
3000
2000
[8]
1000
0
0
100
200
300
400
500
600
700
800
900
1000
[9]
Frame Number
[10]
Fig. 4. Matrix Reloaded – Intelligent system results for traffic shaper
arrival rate (Ra) and departure rate (Rd), and actual transmission rate
(ractual).
[11]
[12]
[13]
V. CONCLUSION
Two intelligent schemes, a rule-based fuzzy controller and
a neural-fuzzy controller are developed to oversee the arrival
rate to a traffic shaping buffer and the departure rate from
the buffer respectively. This work aims to reduce MPEG
video data loss and standard deviation, while increase the
overall data distribution and the size of GOP of data
transmission, which results in improving picture quality.
Furthermore, the research work uses a novel fuzzy approach
to quantization level control, which speeds up the
transmission rate. In conclusion, the intelligent system is
based on simple algorithms, which accommodate for the data
loss at the receiving end and could be used for real-time
implementations in delay-intolerant MPEG video services in
wireless devices.
[14]
[15]
[16]
Sen, P. Maglaris, B. Rikli, N.-E. Anastassiou, D., “Models for
packet switching of variable-bit-rate video sources,” IEEE Journal on
Selected Areas in Communications, vol. 7, issue 5, pp. 865-869, June
1989.
B.P.V. Kumar and P. Venkataram, “Reliable multicast routing in
mobile networks: a neural-network approach,” IEE ProceedingsCommunications, vol. 150, no. 5, pp. 377-384, Oct 2003.
R.T. Zhang and X.M. Zhu, “Fuzzy routing in QoS networks,” Fuzzy
Systems and Knowledge Discovery, PT 2, Proceedings lecture notes
in Artificial Intelligence, 3614, Springer-Verlag Berlin, pp. 880-890,
Part 2, 2005.
J. Mar and C.Y. Kao, “Intelligent MBWIMA/UMTS protocol using
SDMA scheduling in UTRA TDD mode,” Wireless Personal
Communications, vol. 36 no. 2, pp. 143-162, JAN 2006.
R. Razavi, M. Fleury, and M. Ghanbari, “Fuzzy Logic Control of
Adaptive ARQ for Video Distribution over a Bluetooth Wireless
Lin,k” Advances in Multimedia, Hindawi Publishing Corporation,
vol. 2007, Article ID 45798, July 2007.
J.L. Su, Y. Chen, and Z. Ouyang, “GPRS congestion control
algorithm based on fuzzy kernel neural networks,” ISDA 2006: Sixth
International Conference on Intelligent Systems Design and
Application, vol. 1, pp. 954-959, Oct 16th-18th, 2006.
H.B. Kazemian and L. Meng, “A fuzzy approach to video
transmission over Bluetooth ACL links,” 1st IEEE Consumer
Communications and Networking Conf. (CCNC 2004), Las Vegas,
NV, USA, ISBN 0-7803-8146-7, 5th –8th Jan. 2004.
J. Watkinson, MPEG Handbook, Publisher: Focal Press; 1st edition,
ISBN-13: 978-0240516561, September 2001.
Kazemian H. and Meng L., “A fuzzy control scheme for video
transmission in Bluetooth wireless,” INS – Information Sciences,
Elsevier Journals, vol. 176, issue 9, pp 1266-1289, May 2006.
W.K. Wong, H.Y. Tang, and V.C.M. Leung, “Token bank fair
queuing: a new scheduling algorithm for wireless multimedia
services,” Int. Journal of Communication Systems, vol. 17, no. 6, pp.
591-614, Aug 2004.
J.R. Gallardo, P. Medina, and W.H. Zhuang, “QoS mechanisms for
the MAC protocol of IEEE 802.11 WLANs,” Wireless Networks, vol.
13, no. 3, pp. 335-349, June 2007.
S. Haykin, Neural Networks – a comprehensive foundation, PrenticeHall, Inc., 2nd Ed. ISBN 0-13-908385-5, 1999.
I-C. Chang, C-H. Wang, and L-H. Chang, “An area-based vertical
motion estimation on heterogeneous wireless networks, Wireless
Algorithms, Systems, and Applications,” proceedings lecture notes in
Computer Science, 4138, Publisher: Springer-Verlag, Berlin, pp.
187-198, 2006.
L-X Wang, A Course In Fuzzy Systems and Control, Prentice-Hall,
Inc., Upper Saddle River, NJ, USA, ISBN:0-13-540882-2, 1996.
H.L. Wang, S. Kwong, C.W. Kok, “Efficient predictive model of zero
quantized DCT coefficients for fast video encoding,” Image and
Vision Computing, Elsevier Science BV, vol. 25 no. 6, pp. 922-933,
1st June 2007.
http://www.movie-list.com/nowplaying.shtml.
ACKNOWLEDGMENT
This work was funded by the Emerald Grant UK. The work
was carried out at London Metropolitan University, UK.
250
View publication stats
2008 IEEE International Conference on Fuzzy Systems (FUZZ 2008)