Title: Physical Carrier Sensing and Spatial Reuse in Multirate and Multihop Wireless Ad Hoc Networks
1Physical Carrier Sensing and Spatial Reuse in
Multirate and Multihop Wireless Ad Hoc Networks
- Hongqiang Zhai and Yuguang Fang
- Dept of Electrical Computer Engineering
- University of Florida
- Presented by Tae Hyun Kim
2Contents
- Problem Statement
- Analysis for optimum CS distance
- Interference models
-
- End-to-end throughput
- Simulation
- Conclusion
- Some comments
3Problem Statement
- To find optimum CS (Carrier Sense) distance that
maximizes throughput
- Considered factors
- MAC overhead (frame headers and IFSs)
- Bidirectional handshaking
- Multirate different TX ranges, RX sensitivities
and required SINR - Multihop forwarding, hidden/exposed nodes, and
random topology
4Analysis for Optimum CS distance
- Notations
- Two worst case interference models
- Multiple CS distances for multiple rates?
- Exposed/hidden node problems
- Bidirectional handshaking intensifies
interference - End-2-end throughput
5Notations
Payload size
Frame header size
Fixed-lengthoverhead
6Two Interference Models
- Worst case of 6 interferers
7Two Interference Models (cont.)
- Non-overlapping area of each TX
- of concurrent transmissions
- Aggregate throughput
- Thus, given SINR, we can find optimum CS distance
while fixing transmission distance dt
8Two Interference Models (cont.)
- Optimum CS distance is,
- SINR plays major role other than protocol
overhead
10-10
10-7
9Two Interference Models (cont.)
- 6 interferers scenario may be too conservative
- Worst case of only ONE strong interferer
- Compute dc using same method
- Optimum CS area reduces to 2589 of 6
interferers - As shorter CS distance may greatly increase
spatial reuse, we may be allowed to decrease dc
10Multiple CS distances for Multiple Rates?
- Optimum X in interference model varies much,
given SINR requirement - Fortunately,RX sensitivities vary much, too
- Recall
11Multiple CS distances for Multiple Rates?
- Optimum CS distances and thresholds by 6
interferers model - Single CS distance is sufficient to maximize
throughput
12Multiple CS distances for Multiple Rates?
- We do have more reasons for single CS distance
- High complexity to adapt multiple CS distances
for multiple rates - Mobility, distance, channel fading, etc.
- Multiple CS distances may introduce additional
collisions
13Exposed/Hidden Nodes
- Exposed node problem
- Nodes that are unnecessarily shut up
- Lets define interference range
- (X-1)dt dc-dt from receiver
- Exposed-area ratio
- E.g.) By using 6 interferers model,54 Mbps?
d0.24, 0.56 when X10, 5, ?gt3
Exposed area
14Exposed/Hidden Nodes (cont.)
- Shorter dc could
- Alleviate exposed nodes problem
- Achieve higher spatial reuse
- Have potentially larger hidden nodes
- Hidden node problem
- As TX is not sensed by C
- C may interfere TX from A to B
- Increase collisions
- Large CS distance can reduce hidden nodes
15Exposed/Hidden Nodes (cont.)
- Summary
- Tradeoff between degrees of exposed nodes and
hidden nodes
16Bidirectional Handshaking
- Bidirectional handshaking incurs
- Packet collision by immediate ACK
- Receiver blocking (permanent link failure)
- Packet collision by immediate ACK
- After successfully receiving DATA, ACK is
transmitted without CS - RTS may mitigate this as following CTS is sent
when channel is idle
17Bidirectional Handshaking (cont.)
- Receiver blocking
- Before transmitting either CTS or DATA, CS is
performed - If there is nearby on-going transmission,
receiver never replies to RTS - MAC decides that link has been broken
A
B
C
D
DATA
RTS
A receiver does not replyas channel is busy
18Bidirectional Handshaking (cont.)
- Receivers of previous interferers become new
interferers closer to yellow receiver - Modified 6 interferers model
- Intuitively, larger dc required to prevent
interferers from transmitting
19Bidirectional Handshaking (cont.)
- Compute optimum CS distance, again
- For SINR gt - 3dB,
- Thus,
20Bidirectional Handshaking (cont.)
- This solution,
- Sacrifices spatial reuse
- Increases potential exposed nodes
- Incurs MAC contention
- But, this also reduces
- Potential hidden terminals
- Packet collision by immediate ACKs
- Receiver blocking
21Optimum CS distance
- Summary of previous observations
- Tradeoff between larger and smaller dc
- For protocol stability, larger dc might be better
- Optimum CS distance is determined by optimum X
- Simulation study will find µ
22Multihop flow consideration
- End-2-end throughput
- Conditions for maximized spatial reuse along the
path - Distance between TXs be less than dc
- Not corrupting each others packet
- N of hops between nearest concurrent TXs
- 1/N spatial reuse ratio of a multihop flow
- Then, throughput upperbound is
N3
A
B
C
D
E
F
Hop distance
23Multihop flow consideration (cont.)
- Upperbound for one multihop flow throughput
- Observations
- Higher rate does not necessarily generate higher
throughput IF MAC overhead is taken into account
24Multihop flow consideration (cont.)
- Consider interference from nearby concurrent
transmissions in a regular chain topology - Achievable maximum E2E throughput
- This is proportional to BDiP (
) - May not be maximum in general topology (?!)
dc'
dc'-dt
A
B
C
D
E
F
25Simulation
- Modified Ns-2 cumulative interference
- 150 nodes in 1000m x 1000 m area
- To obtain one hop optimum CS distance
- Observations
- Maximum throughput when 60ltCSthlt70 dBm
- For some high rates, CSth lt RXse starving flows
exist - Max throughput can be sustained with some
starving flows
RX sensitivities for different rates
26Simulation (cont.)
- Optimum CS distance for multihop flows
- Observations
- CSth for single hop does not work well
- CSth 91 dBm (smaller CS distance)
- Single CS distance could be optimal
- Higher rates do not necessarily generate higher
throughput
Optimum CSth
CSth lt RXse
randomly selected 20 TCP connections with
500600 E2E distance
27Conclusion
- This paper analyzes impact of CS distance to
throughput from various perspectives - Found optimum CS distance
- Single CS distance is sufficient
- dc may be less than dc due to conservativeness
of 6 interferers model - dc dc dt due to bidirectional handshaking
- dc µ(dc / dt 1)
- dt dh to get maximum E2E throughput
- µ can be found by simulation according to network
setup
28Comments
- Based on simplified interference models and
extreme cases - Through analysis no relationships between any
factors are drawn only some intuitions - Ignorance on random access MAC overhead much
larger than frame header and IFSs overhead - Packet with higher rate has more overhead
proportion, thus penalizing higher rates - Hard to compute BDiP
- Does 801.11 do CS before sending CTS and DATA??
- It does not. Nevertheless, receiver blocking can
happen due to virtual CS - If rate is adapted, then single CSth may not be
a good strategy
29THANK YOU!ANY QUESTIONS?
30Backup slides
31Multihop flow consideration (cont.)
- Worst case model for condition (2)
- Equivalent to bidirectional handshaking model
- We have,
- Bound for one multihop flow throughput
32Multihop flow consideration (cont.)
Ns-2 default SINR10 dB, ?4 ?Spatial reuse
ratio 1/3