Rate Control in Wireless Networks
ECE 256
1
Recall 802.11
 RTS/CTS + Large CS Zone
 Alleviates hidden terminals, but trades off spatial reuse
E
CTS
A
B
C
RTS
F
D
2
Recall Role of TDMA
3
Recall Beamforming
Omni Communication
Directional Communication
e
Silenced Node
f
f
a
b
a
c
b
c
d
d
No
Simultaneous Communication
Ok
Simultaneous Communication
4
Also Multi-Channel
 Current networks utilize non-overlapping channels
 Channels 1, 6, and 11
 Partially overlapping channels can also be used
5
Today
Benefits from exploiting channel conditions
– Rate adaptation
– Pack more transmissions in same time
6
What is Data Rate ?
Number of bits that you transmit per unit time
under a fixed energy budget
Too many bits/s:
Each bit has little energy -> Hi BER
Too few bits/s:
Less BER but lower throughput
7
802.11b – Transmission rates
Highest energy
per bit
1 Mbps
2 Mbps
5.5 Mbps
11 Mbps
Time
Lowest energy
per bit
Optimal rate depends on SINR:
i.e., interference and current channel conditions
8
Some Basics
 Friss’ Equation
 Shannon’s Equation
C = B * log2(1 + SINR)
 Bit-energy-to-noise ratio
Eb / N0 = SINR * (B/R)
Leads to BER
Varying
with time
and space
How do we choose the rate of modulation
9
Static Rates
SINR
time
# Estimate a value of SINR
# Then choose a corresponding rate that would transmit packets
correctly (i.e., E b / N 0 > thresh) most of the times
# Failure in some cases of fading  live with it
10
Adaptive Rate-Control
SINR
time
# Observe the current value of SINR
# Believe that current value is indicator of near-future value
# Choose corresponding rate of modulation
# Observe next value
# Control rate if channel conditions have changed
11
Is there a tradeoff ?
Rate = 10
B
A
C
E
D
12
Is there a tradeoff ?
Rate = 10
B
A
C
E
D
Rate = 20
What about length of routes due to smaller range ?
13
Any other tradeoff ?
Will carrier sense range vary with rate
14
Total interference
Rate = 10
B
A
C
E
D
Rate = 20
Carrier sensing estimates energy in the channel.
Does not vary with transmission rate
15
Bigger Picture
 Rate control has variety of implications
 Any single MAC protocol solves part of the puzzle
 Important to understand e2e implications
 Does routing protocols get affected?
 Does TCP get affected?
 …
 Good to make a start at the MAC layer




RBAR
OAR
Opportunistic Rate Control
…
16
A Rate-Adaptive MAC Protocol for
Multi-Hop Wireless Networks
Gavin Holland
HRL Labs
Nitin Vaidya
UIUC
Paramvir Bahl
Microsoft Research
MOBICOM’01 Rome, Italy
© 2001. Gavin Holland
17
Background
 Current WLAN hardware supports multiple data rates
 802.11b – 1 to 11 Mbps
 802.11a – 6 to 54 Mbps
 Data rate determined by the modulation scheme
18
Problem
Modulation schemes have different error characteristics
8 Mbps
BER
1 Mbps
SNR (dB)
But, SINR itself varies
With Space and Time 19
Impact
Large-scale variation with distance (Path loss)
Distance (m)
Mean Throughput (Kbps)
SNR (dB)
Path Loss
8 Mbps
1 Mbps
Distance (m)
20
Impact
Small-scale variation with time (Fading)
SNR (dB)
Rayleigh Fading
2.4 GHz
2 m/s LOS
Time (ms)
21
Question
SNR (dB)
SNR (dB)
Which modulation scheme to choose?
2.4 GHz
2 m/s LOS
Distance (m)
Time (ms)
22
Answer  Rate Adaptation
Mean Throughput (Kbps)
 Dynamically choose the best modulation scheme
for the channel conditions
Desired
Result
Distance (m)
23
Design Issues
 How frequently must rate adaptation occur?
 Signal can vary rapidly depending on:




carrier frequency
node speed
interference
etc.
 For conventional hardware at pedestrian speeds, rate adaptation
is feasible on a per-packet basis
Coherence time of channel higher than transmission time
24
Adaptation  At Which Layer ?
 Cellular networks
 Adaptation at the physical layer
 Impractical for 802.11 in WLANs
Why?
25
Adaptation  At Which Layer ?
 Cellular networks
 Adaptation at the physical layer
Why?
 Impractical for 802.11 in WLANs
RTS/CTS requires that the rate be known in advance
C
Sender
A
D
RTS: 10
CTS: 8
Receiver
8
B
10
 For WLANs, rate adaptation best handled at MAC
26
Who should select the data rate?
A
B
27
Who should select the data rate?
 Collision is at the receiver
 Channel conditions are only known at the receiver
 SS, interference, noise, BER, etc.
A
B
 The receiver is best positioned to select data rate
28
Previous Work
 PRNet
 Periodic broadcasts of link quality tables
 Pursley and Wilkins
 RTS/CTS feedback for power adaptation
 ACK/NACK feedback for rate adaptation
 Lucent WaveLAN “Autorate Fallback” (ARF)
 Uses lost ACKs as link quality indicator
29
Lucent WaveLAN “Autorate Fallback”
(ARF)
A
2 Mbps
Effective Range
1 Mbps
Effective Range
2 Mbps
DATA
B
 Sender decreases rate after
 N consecutive ACKS are lost
 Sender increases rate after
 Y consecutive ACKS are received or
 T secs have elapsed since last attempt
30
SNR (dB)
Performance of ARF
Time (s)
Dropped Packets
Failed to Increase
Rate After Fade
Time (s)
Attempted to Increase
Rate During Fade
–
Slow to adapt to channel conditions
–
Choice of N, Y, T may not be best for all situations
31
RBAR Approach
 Move the rate adaptation mechanism to the receiver
 Better channel quality information = better rate selection
 Utilize the RTS/CTS exchange to:
 Provide the receiver with a signal to sample (RTS)
 Carry feedback (data rate) to the sender (CTS)
32
Receiver-Based Autorate (RBAR)
Protocol
1 Mbps
2 Mbps
C
RTS (2)
A
CTS (1)
DATA (1)
D
2 Mbps
1 Mbps
B
1 Mbps
 RTS carries sender’s estimate of best rate
 CTS carries receiver’s selection of the best rate
 Nodes that hear RTS/CTS calculate reservation
 If rates differ, special subheader in DATA packet
updates nodes that overheard RTS
33
SNR (dB)
Performance of RBAR
Time (s)
RBAR
Time (s)
ARF
Time (s)
34
Question to the class
 There are two types of fading
 Short term fading
 Long term fading
 Under which fading is RBAR better than ARF ?
 Under which fading is RBAR comparable to ARF ?
 Think of some case when RBAR may be worse than ARF
35
Implementation into 802.11
Frame
Control
Duration
DA
SA
FCS
BSSID
Sequence
Control
Body
FCS
Reservation Subheader (RSH)
 Encode data rate and packet length in duration field of frames
 Rate can be changed by receiver
 Length can be used to select rate
 Reservations are calculated using encoded rate and length
 New DATA frame type with Reservation Subheader (RSH)
 Reservation fields protected by additional frame check sequence
 RSH is sent at same rate as RTS/CTS
 New frame is only needed when receiver suggests rate change
WHY
36
Performance Analysis
 Ns-2 with mobile ad hoc networking extensions
 Rayleigh fading
 Scenarios: single-hop, multi-hop
 Protocols: RBAR and ARF
 RBAR
 Channel quality prediction:
• SNR sample of RTS
 Rate selection:
• Threshold-based
 Sender estimated rate:
1E-5
• Static (1 Mbps)
2 Mbps
Threshold
SNR (dB)
8 Mbps
Threshold
37
Performance Results
Single-Hop Network
38
Mean Throughput (Kbps)
Single-Hop Scenario
Distance (m)
A
B
39
Varying Node Speed
UDP Performance
Mean Throughput (Kbps)
RBAR
ARF
CBR source
Packet Size = 1460
Mean Node Speed (m/s)
 RBAR performs best
 Declining improvement with increase in speed
WHY?
 Adaptation schemes over fixed
 RBAR over ARF
 Some higher fixed rates perform worse than lower fixed rates
40
Varying Node Speed
TCP Performance
Mean Throughput (Kbps)
RBAR
ARF
FTP source
Packet size = 1460
Mean Node Speed (m/s)
 RBAR again performs best
 Overall lower throughput and sharper decline than with UDP
 Caused by TCP’s sensitivity to packet loss
 More higher fixed rates perform worse than lower fixed rates
41
No Mobility
ARF
RBAR
Mean Throughput (Kbps)
Mean Throughput (Kbps)
UDP Performance
Distance (m)
CBR source
Packet size = 1460
Distance (m)
 RSH overhead seen at high data rates
WHY?
 Can be reduced using some initial rate estimation algorithm
 Limitations of simple threshold-based rate selection seen
 Generally, still better than ARF
42
No Mobility
UDP Performance
Mean Throughput (Kbps)
RBAR-P
CBR source
Packet size = 1460
Distance (m)
 RBAR-P – RBAR using a simple initial rate estimation algorithm
Why useful ?
 Previous rate used as estimated rate in RTS
 Better high-rate performance
 Other initial rate estimation and rate selection algorithms are a
topic of future work
43
RBAR Summary
 Modulation schemes have different error characteristics
 Significant performance improvement may be achieved by MAClevel adaptive modulation
 Receiver-based schemes may perform best
 Proposed Receiver-Based Auto-Rate (RBAR) protocol
 Implementation into 802.11
 Future work
 RBAR without use of RTS/CTS
 RBAR based on the size of packets
 Routing protocols for networks with variable rate links
44
Questions?
45
OAR: An Opportunistic Auto-Rate
Media Access Protocol for Ad Hoc
Networks
B. Sadeghi, V. Kanodia, A. Sabharwal, E. Knightly
Rice University
Slides adapted from Shawn Smith
46
Motivation
 Consider the situation below
 ARF?
 RBAR?
B
C
A
47
Motivation
 What if A and B are both at
56Mbps, and C is often at
2Mbps?
 Slowest node gets the most
absolute time on channel?
B
Timeshare
A
B
C
C
A
Throughput Fairness vs Temporal Fairness
48
Opportunistic Scheduling
Goal
 Exploit short-time-scale channel quality
variations to increase throughput.
Issue
 Maintaining temporal fairness (time share) of
each node.
Challenge
 Channel info available only upon transmission
49
Opportunistic Auto-Rate (OAR)
 In multihop networks, there is intrinsic diversity
 Exploiting this diversity can offer benefits
 Transmit more when channel quality great
 Else, free the channel quickly
 RBAR does not exploit this diversity
 It optimizes per-link throughput
50
OAR Idea
 Basic Idea
 If bad channel, transmit minimum number of packets
 If good channel, transmit as much as possible
D
A
C
C
A
B
Data Data Data Data
Data Data Data Data
51
Why is OAR any better ?
 802.11 alternates between transmitters A and C
 Why is that bad
D
A
C
C
Data
A
Data
Data
B
Data
Data
Data
Data
Data
Is this diagram correct ?
52
Why is OAR any better ?
 Bad channel reduces SINR  increases transmit time
 Fewer packets can be delivered
D
A
C
C
Data
A
B
Data
Data
Data
Data
Data
53
OAR Protocol Steps
 Transmitter estimates current channel
 Can use estimation algorithms
 Can use RBAR, etc.
 If channel better than base rate (2 Mbps)
 Transmit proportionally more packets
 E.g., if channel can support 11 Mbps, transmit (11/2 ~ 5) pkts
 OAR upholds temporal fairness
 Each node gets same duration to transmit
 Sacrifices throughput fairness  the network gains !!
54
MAC Access Delay Simulation
 Back to back packets in OAR decrease the average
access delay
 Increase variance in time to access channel
Why?
55
OAR Protocol
Channel Condition
Protocol
BAD
MEDIUM
GOOD
Pkts
Rate
Pkts
Rate
Pkts
Rate
802.11
1
2
1
2
1
2
802.11b
1
2
1
5.5
1
11
OAR
1
2
3
5.5
5
11
 Rates in IEEE 802.11b: 2, 5.5, and 11 Mbps
 Number of packets transmitted by OAR ~
Tx Rate
Base Rate
56
Simulations
 Three Simulation experiments
1.
Fully connected networks: all nodes in radio range of each
other
•
Number of Nodes, channel condition, mobility, node location
2. Asymmetric topology
3. Random topologies
 Implemented OAR and RBAR in ns-2 with extension
of Ricean fading model [Punnoose et al ‘00]
57
Fully Connected Setup
 Every node can communicate with everyone
 Each node’s traffic is at a constant rate and
continuously backlogged
 Channel quality is varied dynamically
58
Fully Connected Throughput Results
 OAR has 42% to 56% gain over RBAR
 Increase in gain as number of flows increases
 Note that both RBAR and OAR are significantly better than
standard 802.11 (230% and 398% respectively)
59
Asymmetric Topology Results
OAR maintains time shares of IEEE 802.11
Significant gain over RBAR
60
OAR thoughts
 OAR does not offer benefits when
 OAR may not be suitable for applications like
 With TCP how can OAR get affected ?
61
OAR thoughts
 OAR does not offer benefits when
 Neighboring nodes do not experience diverse channel
conditions
 Coherence time is shorter than N packets
 With TCP can OAR get affected ?
 Back-to-back packets can increase TCP performance
 However, bottleneck bandwidth can get congested quick
 Also, variance of RTT can increase
62
Other Ideas
(Briefly)
63
Exploiting Diversity in Rate Adaptation
 Yet another idea exploits multiple user diversity
 Among many intermediate nodes, who has best channel
 Use that node as forwarding node
 Forwarding node can change with time
• Due to channel fluctuations at different time and space
Channel Conditions
SNR
AP
WLAN
USERS
TIME
64
The Protocol Overview
• MAD using Packet Concatenation (PAC)
DIFS
sender
SIFS
SIFS
SIFS
SF
GRTS
SIFS
DATA 0
DATA 1
DATA 2
CTS 1
user 1
CTS 2
user 2
…
user k
ACK 0~2
CTS k
Since at least one intermediate node is likely to have good channel
condition, transmitter can transmit at a high data rate or concatenate
Multiple packets
• Choosing subset of neighbor-group is important
• Coherence time of channel must be greater than packet chain
• Group needs to really have independent channel gain
•
Correlated channel gains can lead to performance hit.
65
Summary
 Rate control can be useful
 When adapted to channel fluctuations (RBAR)
 When opportunistically selecting transmitters (OAR)
 When utilizing node diversity
 Benefits maximal when
 Channel conditions vary widely in time and space
 Correlation in fluctuation can offset benefits
 OAR and Diversity-based MAC may show negligible gain
 Several more research possibilities with rate control
66
Questions ?
67
What lies ahead ?
 Routing based on rate-control
 Choosing routes that contain high-rate links
 ETX metric proposed from MIT accomodates link character
 Opportunistic routing from MIT again – takes neighbor diversity
into account (best paper Sigcomm 2005)
 Fertile area for a project …
 Dual of rate-control is power control
 One might be better than the other
 Decision often depends on the scenario  open problem
 Directional antennas for DD link for data/ack
 Rate control can be introduced  Not been studied yet
… many many more
68
Infrequent Traffic
UDP Performance
Pareto source
Mean burst interval = 1 sec
Packet size = 1460
Mean Burst Length (ms)
 Similar results for shorter burst intervals
 Similar results for TCP (see tech report)
69
Any other tradeoff ?
Can anyone think of yet another
IMPORTANT tradeoff
Hint: Related to the MAC Layer
70
Total interference
Rate = 10
B
A
C
E
D
Rate = 20
More nodes free to transmit packets (A,B,E)
Interference incident at receiver (D) increases
71
Performance Results
Multi-Hop Network
72
Multi-Hop Network
Mean Throughput (Kbps)
UDP Performance
20 nodes
CBR source
Packet size = 1460
Mean Node Speed (m/s)
 Similar results for TCP
73
Multi-Hop Network
Sensitivity to RSH Loss
B
RSH
C
Flow 2
D
Mean Throughput (Kbps)
Mean Throughput (Kbps)
A
Flow 1
ARF
RSH Loss Probability
 Aggregate throughput is unaffected by RSH loss
 High loss probability results in only slight change in fairness
74
Multi-Hop Network
Sensitivity to RSH Loss
B
RSH
C
Flow 2
D
Mean Throughput (Kbps)
Mean Throughput (Kbps)
A
Flow 1
ARF
RSH Loss Probability
 Similar results
 Slightly more unfairness (vs. ARF) for no loss
 (overall fairness problem due to MAC backoff by node A)
75
802.11b – Transmission rates
 Different modulation
methods for transmitting
data.
Highest energy
per bit
1 Mbps
 Binary/Quadrature Phase
Shift Keying
 Quadrature Amplitude
Modulation
2 Mbps
5.5 Mbps
 Each packs different number
of data in modulation.
 The highest speed has most
dense data and is most
vulnerable to noise.
11 Mbps
Time
Lowest energy
per bit
76