1. Improving Performance of Higher Layer Protocols with MIMO based MAC
Guided By Prof. Anirudha Sahoo
Presented By Ankit Jindal
SYENRG MTP-3 Workshop
24th June 2009

2. Outline Introduction Problem Definition MAC layer enhancements
Introduction to MIMO
MIMO at different layers
Introduction
Problem Definition
MAC layer enhancements
TCP layer enhancements
Simulation
Routing layer enhancements
Conclusion & Future work
Work at TCP layer
Work at Routing layer
Basic concept
Application
Basic concept
Modifications for MIMO based MAC
Setup
Results
Basic concept
Objective
Different approaches

3. Introducing MIMO . Channel Matrix H s1 s2 sM s r1 r2 rM r
User data stream .
Channel
Matrix H s1 s2 sM s r1 r2 rM r
Transmitted vector
Received vector
h11
h12
h11 h …….. hM1
h12 h …….. hM2
h1M h2M …….. hMM
…….. . N
hij is a Complex gain between the ith transmit and jth receive antenna
Where H =

4. MIMO at different layers
TCP
Routing Layer
Physical Layer
MAC Layer
Category A
Category A
No Modifications
Spatial Multiplexing
Category B
Diversity
Parallel Communications
Parallel Communications
Rate Links
Rate/Range/Reliable link
Physical Layer Schemes
Range/Reliable Links
Modified MAC for MIMO

5. Problem Definition
Design the TCP parameter with MIMO based MAC to improve the performance.
2. Propose the modifications to the traditional ad-hoc routing layer protocols with MIMO based MAC.

6. MIMO based MAC: Basic Concept
User data stream .
Channel
Matrix H s1 s2 sM s r1 r2 rM r
Transmitted vector
Received vector
h11
h12
h11 h …….. hM1
h12 h …….. hM2
h1M h2M …….. hMM
…….. . N
hij is a Complex gain between the ith transmit and jth receive antenna
Where H =

7. Basic Concept Signal received at antenna i Beamformer Output
Complex Gain

8. Basic Concept Importance of Complex gain
Complex Gain = 0 implies Signal Nullified
Otherwise, Signal Received
Proper Selection of Weights enable the Nodes to Receive Only Signal of Interest

9. Application on Multi Hop Network
Beamformer
Output
Node C can nullify signal of B by having Gain between B and C = 0

10. Weight Selection Algorithm
Transmitter Weights Selection
If No CTS overheard then use default weight vector
Else Solve the following set of equations
For all R belongs to set K
= 0
Where WR is weight vector of node R and K is set of all nodes whose CTS have been overheard
Receiver Weights Selection
If No RTS overheard other then intended one, then select weight such that Gain = 1
Else Solve the following set of equations
For all T belongs to set K
Gain = 1 and = 0
Where WT is weight vector of node T and K is set of all other nodes whose RTS have been overheard

11. Problem Definition: Recall
TCP
No Modifications
Design TCP Parameter
Modified MAC for MIMO

12. TCP Layer Enhancements
CWL = Path’s BDP
Path’s BDP <= BWmin * Round Trip Delay
Forward Path Delay <= n * (S/BWmin)
Reverse Path Delay <= m * (S/BWmin)
Path’s BDP <= BWmin * (n * S/BWmin + m * S/BWmin)
Path’s BDP <= (n + m) * S
CWL = k * (n + m) * S
Where 0 <= k <= 1

13. TCP Layer Enhancements
CWL = k * (n + m) * S
Where 0 <= k < = 1
CWL can be set to constant multiple of round trip hop count
Value of k depends upon the number of packets that the path from source to destination can accommodate
Implies k depends upon the MAC protocol used

14. TCP Layer Enhancements
IEEE MAC
MIMO based MAC
Maximum h/2 packet can be in transit
Maximum h/3 packet can be in transit
Implies, k <= 1/2 for MIMO based MAC
Implies, k <= 1/3 for IEEE MAC

15. TCP Layer Enhancements
Lower Bound on k
Limit the number of DATA packets in transit to half of the maximum in order to accommodate ACK
Implies, k >= 1/6 for IEEE MAC
k >= 1/4 for MIMO based MAC

16. TCP Layer Enhancements
CWL = k * (n + m) * S
Where k is a constant such that
1/4 <= k <= 1/2 For MIMO based MAC
1/6 <= k <=1/3 For IEEE MAC
Appropriate value of k can be found out by finding optimal CWL for different length chain topology

17. Simulation: Setup 100,000 Packets Single TCP flow
Buffer Size = 25 packets
Shadowing model used
Poisson Arrival
Offered load 3 Mbps
Two Antennae per node
Each Channel Matrix entry = Gaussian Distribution with mean 0 and variance 0.5
Transmission range is assumed to be same as Interference range

18. Throughput v/s Congestion Window Limit
Simulation: Results
Throughput v/s Congestion Window Limit
There exists a CWL beyond which there is no significant improvement in throughput

19. Simulation: Results Average RTD v/s RTHC
There exists a CWL beyond which there is no significant improvement in throughput but increase in average round trip packet delay.

20. Setting k = 1/2.8 approx the Optimal CWL found out by simulations
Simulation: Results
Optimal CWL v/s RTHC
Setting k = 1/2.8 approx the Optimal CWL found out by simulations

21. Setting k = 1/2.8 approx the Optimal CWL found out by simulations
Simulation: Results
Throughput v/s RTHC
Setting k = 1/2.8 approx the Optimal CWL found out by simulations

22. CWL depends upon the underlying MAC protocol used
Simulation: Results
Throughput v/s RTHC
CWL depends upon the underlying MAC protocol used

23. Problem Definition: Recall
Routing Layer
Category A
Modify Routing Protocol
Parallel Communications
Rate/Range/Reliable link
E.g. MIR routing protocol

24. Routing Layer Enhancements: Basic Concept
Number of communications in the collision domain depends upon the available degree of freedom
Maximum N communications can happen in same collision domain, where N is number of antennae
Objective: To select path that allows more number of parallel communications.

25. Example Topology: Two Antennae per node
Objective 1 S1 D1 1 2 S2 D2 3
Example Topology: Two Antennae per node

26. Different Approaches
Suppose each node knows its number of active neighbors. S1 D1
Node append its active neighbor count in the route request packet C
Destination on receiving multiple route request messages has to rank the path B S2 D2
We have different approaches in mind for destination to rank the path A

27. Different Approaches Proposal 1 Sum of ANC along the path
Hop count
< Number of Antennas 1 1 1 2 3
Use hop count to decide the path if above equation is not satisfied for any path
Problem: A path is rejected because of congestion at only few intermediate nodes

28. Different Approaches Proposal 2
Count nodes satisfying, ANC < Number of Antennae
Select path having max value of C/h 1 1 1 2 2
Problem: A path having C/h = 1 of higher hop count say 20 nodes, is accepted compare to path having C/h = 0 of low hop count say 5 is rejected

29. where 0 <= alpha <= 1
Different Approaches
Proposal 3
Allow destination to entertain route request for only some time in order to reject large hop count path
Or, give some weight age to h and C
Select the path which has maximum value of
alpha * C + (1 – alpha) * h,
where 0 <= alpha <= 1
Favorable value of alpha can be found out by performing experiments for various topologies.

30. Active Neighbor Count Calculation
Different Approaches
Active Neighbor Count Calculation A B B
Routing Table
ANC & Nodes

31. Conclusion and Future Work
Traditional MAC protocols are not suitable for MIMO based systems
MIMO based MAC allows more number of parallel communications
For such MIMO based MAC, TCP CWL should be appropriately modified
Simulation data shows that TCP CWL can be empirically set to 1/2.8 of round trip hop count to improve performance
For such MIMO based MAC, routing protocol can also be designed to improve the performance.
Knowing the active neighbor count, appropriate routing protocol can be designed for such MIMO based MAC

32. Conclusion and Future Work
Propose the appropriate modifications to the AODV routing protocol
Detailed study of such routing protocol is required as part of future work
Performance evaluation of various approaches can be done to design the efficient routing protocol for MIMO based MAC

33. QUESTIONS??? Thank You !!! sahoo@cse.iitb.ac.in

34. Routing Layer Enhancements: Backup Slides
Y axis
Equality 1
(0,0)
X axis
Equality 0
Equality 0

35. Routing Layer Enhancements: Backup Slides
With N antennae, we want to have N + 1 communications in the same collision domain.
Then, we have N + 1 equations each of N variables, out of which N equation are of type equality 0 and one equation of equality 1
With N + 1 equations in N dimension, there can be only one point of intersection
Origin is the point of intersection of N equations of type equality 0, which does not satisfy last equation of type equality 1

36. Spatial Multiplexing: Backup Slides
Each receiver antenna receives the
superposition of all of the transmitted data streams.
Receiver can retrieve the transmitted
streams, as each stream have different spatial signature due to multi path.
Transmission of independent streams of
data through each antenna provides linear increase in capacity
C = min(M;N)log2(1 + P)

37. Diversity: Backup Slides
Because of multi path, each stream is independent & hence probability of each stream facing poor channel is very less
Thus, Diversity helps in reducing BER or increase in SNR at the output of combiner
Reduce BER on the link.
For the required BER on the link, increase in SNR implies increased communication range
For the fixed BER and SNR, the transmit power consumption can be minimized.

38. Beta is path loss exponent & alpha is shadowing or standard deviation
Shadowing Model
The free space model and the two-ray model predict the received power as a deterministic function of distance.
In reality, the received power at certain distance is a random variable due to multi path propagation effects, which is also known as fading effects.
Shadowing Model
Beta is path loss exponent & alpha is shadowing or standard deviation

39. Backup Slides: Beam Forming
Controls the phase and relative amplitude of the transmitted signal at each transmitter
Create a pattern of constructive and destructive interference in the wave front.
Received streams are combined in such a way that the expected pattern of radiation is preferentially observed.
Antenna should be separated by at least ½ of transmit signal wavelength

40. Backup Slides: Space Time Code
Improve the reliability of data transmission in wireless communications using multiple transmit antennas.
Transmit multiple, redundant copies of a data stream to the receiver
Probability of each stream facing poor channel is very less
Antenna should be separated by at least 4 to 10 times the wavelength to keep the signal through each multi-path independent