Understanding and Mitigating the Impact of
RF Interference on 802.11 Networks
Ramki Gummadi (MIT), David Wetherall (UW)
Ben Greenstein (IRS), Srinivasan Seshan (CMU)
Presented by Lei Yang in CS595H, W08
1
The impact of interference?
 Shannon Channel Capacity
 Capacity = Bandwidth*log(1+Signal/Noise)
signal
sender
noise
receiver
 Very useful in communication theory
 Give the upper bound
 Give directions for approaching the upper bound
 E.g. Telephone line moderm
 Very difficult to extend to wireless networks
 Too many factors and optimization goals
2
Extend to wireless networks?
 Current status of network information theory
 Succesfully extend to broadcast and multi access channel
 But the simplest cases are still unknown
 The simplest relay channel
 The simplest interference channel
 Real networks are much more sophisticated
 Factors are difficult to model






3
Real interference are not stationary, white and additive
MAC scheduling
Delay
User cooperation/relay/multi-hop
Dynamic traffic
Hardware implementation factors
Interference
channel
What can we do?
 Information theoretical view
 CT: bit-meters/second
 Scaling law results O(n), O(√n)…
 Estimate performance in real networks
 Modeling based on simplifications
 Markov process for MAC behavior
 Poisson arrival of traffic
 Capture hardware impacts…
 Simulations
 Experiments
 Design better schemes to improve performance
 Physical layer: modulation, coding
 MAC/Network layer: scheduling, routing
 Joint optimization: network coding, distributed source coding
4
Another interference paper
Last week’s paper
This paper
Theoretical approach
Experimental approach
 In wireless networks,
Blog(1+S/N) doesn’t
consider CSMA MAC and
traffic demand
 Develop a model to
estimate pairwise
throughput and packet loss
ratio
 Propose a new scheduling
method?
5
 Even for a single 802.11 link,
show interference generates
worse impacts by
experiments
 Extend existing SINR model
to capture the impacts
 Show that simple solutions
don’t work, propose a
channel-hopping method to
reduce the impacts
Motivations
 Growing interference in unlicensed bands
 Anecdotal evidence of problems, but how severe?
 Characterize how 802.11 operates under interference in
practice
Other 802.11
6
What do we expect?
linearly with interference
 There to be lots of options for
802.11 devices to tolerate
interference
 Bit-rate adaptation
 Power control
 FEC
 Packet size variation
 Spread-spectrum processing
 TX/RX diversity
7
Throughput (linear)
 Throughput to decrease
Interferer power
(log-scale)
 Effects of interference more
severe in practice
 Caused by hardware limitations
of commodity cards, which
theory doesn’t model
Throughput (linear)
What we see
Interferer power
(log-scale)
8
Overview of this paper
 Characterize the impact of interference
 Extend the SINR model
 Simple interference mitigation methods don’t help much
 Apply channel hopping to tolerating interference
9
Experimental setup
Access
Point
UDP flow
802.11 Interferer
802.11
Client
10
802.11 receiver path
PHY
To RF Amplifiers
Amplifier
control
RF
AnalogSignal ADC
signal
Timing
Recovery
6-bit
samples
PHY
MAC
MAC
AGC
Barker
Correlator
Demodulator
Descrambler
Data
(includes
beacons)
Preamble Detector/
Header CRC-16 Checker
Receiver
SYNC
SFD
CRC Payload
PHY header
Extend SINR model (in paper) to capture these vulnerabilities
11
Interested in worst-case natural or adversarial interference
Timing recovery interference
 Interferer sends continuous SYNC pattern
 Interferes with packet acquisition (PHY reception errors)
1000
100
1200
Weak interferer
Moderate
interferer
1000
800
Throughput
Log-scale
600
10
400
Latency
1
200
0.1
0
−∞ -20 -12
12
-2
0
8
12
15
Interferer Power (dBm)
20
Latency
(microseconds)
Throughput (kbps)
10000
Dynamic range selection
1000
100
700
Throughput
10
500
300
Latency
1
100
0.1
-100
−∞ -20 -12
13
-2
0
8
12
15
Interferer Power (dBm)
20
Latency
(microseconds)
Throughput (kbps)
 Interferer sends on-off random patterns (5ms/1ms)
 AGC selects a low-gain amplifier that has high processing noise
(packet
10000CRC errors)
900
Header processing interference
 Interferer sends continuous 16-bit Start Frame Delimiters
 Affects PHY header processing (header CRC errors)
Unsynchronized interferer
1000
800
100
600
Throughput
10
400
Latency
1
200
0.1
0
−∞
14
1000
-20
-12
0
8
12
15
Interferer Power (dBm)
20
Latency
(microseconds)
Throughput (kbps)
10000
Extending the SINR Model
 Original Model
 Extended Model
 Accounting for processing gain
 Accounting for AGC behavior
 Acouting for non-linearity in receiver sensitivity
 Not very useful
15
Interference mitigation options
 Lower the bit rate
 Decrease the packet size
 Choose a different modulation scheme
 Leverage multipath (802.11n)
 Move to a clear channel
16
Impact of 802.11 parameters
 Rate adaptation, packet sizes, FEC, and varying CCA
parameters do not help
Throughput (kbps)
10000
With and
Changing CCA
without FEC
mode
Rate adaptation
1000
100
10
11Mbps, PBCC
1Mbps
11Mbps
CCA Mode 1,
11Mbps
2Mbps
5.5Mbps,
5.5Mbps
Changing
PBCC
100-byte packets,
11Mbps
1
packet size
0.1
−∞
17
-20
-12
-2
8
12
Interferer Power (dBm)
15
20
Impact of 802.11g/n
Throughput (kbps)
100000
High throughputs 1000
without
interference
Significant
drops with
802.11n
Throughput
weak interferer800
10000
1000
100
802.11g
Throughput
10
802.11g Latency
600
400
200
1
802.11n Latency
0.1
0
−∞
18
-20
-12
0
8
12
15
Interferer Power (dBm)
20
Latency
(microseconds)
 No significant performance improvement
Impact of frequency separation
• But, even small frequency separation (i.e., adjacent
802.11 channel) helps
– Channel hopping to mitigate interference?
Throughput (kbps)
10000
15MHz separation
1000
10MHz separation
Same channel
(poor performance)
100
10
1
0.1
−∞
19
-20
-12
0
8
12
Interferer Power (dBm)
15
20
Rapid channel hopping
 Use existing hardware
 Design dictated by radio PHY and MAC properties
(synchronization, scanning, and switching latencies)
 Design must accommodate adversarial and natural
interference  channel hopping
 Test with an oracle-based adversary
 Design overview
 Packet loss during switching + adversary’s search speed 
10ms dwell period
 Next hop is determined using a secure hash chain
 Triggered only when heavy packet loss is detected
20
Evaluation of channel hopping
• Good TCP & UDP performance, low loss rate
Throughput (kbps)
10000
CH, UDP traffic
1000
CH, TCP traffic
Weak interference,
Moderate interference,
17% degradation 1Mbps throughput
100
No CH, UDP traffic
10
1
No CH, TCP traffic
0.1
0
5
10
15
Interferer Power (dBm)
21
20
Evaluation of channel hopping
1800
450
1600
400
Throughput (kbps)
1400
350
1200
300
1000
250
800
200
600
150
400
200
22
Three orthogonal
802.11 interferers
0
Interferers
Linear scale
100
50
0
Latency (microseconds)
 Acceptable throughput even with multiple interferers
Conclusions
 Lot of previous work on RF interference
 We show 802.11 NICs have additional PHY and MAC fragilities
 Interference causes substantial degradation in commodity
NICs
 Even weak and narrow-band interferers are surprisingly
effective
 Changing 802.11 parameters does not mitigate interference,
but rapid channel hopping can
23
Pros & Cons
 Pros
 Clear structure
 Useful results
 Clearly separation of hardware limitations
 Cons
 Extended model is not useful
 Channel Hopping solution is not novel
24