1. 1 Multi-User Diversity in Single-Radio OFDMA Ad Hoc Networks Based on Gibbs Sampling Marzieh Veyseh J.J. Garcia-Luna-Aceves Hamid R. Sadjadpour

2. Motivation Concurrency Frequency Code Space 2 fc1 Channel 1 fc2 fc3fc4fc5 Channel 1 Channel 3 fc Channel 1 Code 1 Code 3 Code 1 Code 2 Code 3 fc

3. Motivation Adaptivity Time Frequency, through different modulations… Diversity To use fading in our advantage and to improve rate… MIMO OFDAM-Multiuser diversity 3 t1t1 t2t2 t3t3 t4t4 t5t5 t3t3 t1t1

4. Proposed for infrastructure-based OFDMA networks We think that utilizing OFDMA in ad hoc networks gives us the ability to achieve our goals: 1.Concurrency: multiple nodes use different portions of BW 2.Diversity: multi-user diversity 3.Adaptivity: subchannels can have different sizes OFDMA 4 fc Subchannel i Subchannel 1 Subchannel 2 Subchannel 3 Subchannel 4

5. OFDMA Synchronization 5 C-Tx In an ad hoc multi-transmitter scenario to avoid loss of orthogonality at a common neighbor to multiple transmitters, a quasi-synchronous network is required. Quasi-synchronous : transmitter start sending data at the same time Multi-transmission synchronization achieved via control message exchanges

6. Previous work Not much work on protocols for OFDMA for ad hoc networks CTRMA: In our previous work, we assigned non-overlapping channels unique within the two hop neighborhood based on some priority order. [OFDMA Based Multiparty Medium Access Control in Wireless Ad Hoc Networks, ICC, 09]) CBD: We added diversity when distributing the assigned subchannels among neighbors [Cross-Layer Channel Allocation Protocol for OFDMA Ad Hoc Networks, Globecom 10] 6

7. Gibbs Sampling Goal: Maximize concurrency – Design a MAC that assigns subchannels to each directional link on- demand and based on the present interference, and node’s needs Maximize diversity – Subchannel selection should be done based on minimizing fading – Gibbs sampling previously used to distribute channels among multiple 802.11 Aps : min direct interference 7 GSA C-Tx_1 C-Tx_2 C-Rx aa bb

8. Gibbs Sampling 8 Graph, and state X…want to find

9. GSA Gibbs Subchannel Assignment (GSA) We propose to use Gibbsian method by defining a new energy model and a MAC protocol that works to select subchannels (Schannel) for each communicating link. 9 C-Tx (a) C-Tx (c) C-Rx b v u Subchannel k

10. MDMA Freq Control FreeTx Schannel_1 FreeTx Schannel_2 C-Tx a b d RTM-T time c

11. Freq Control FreeTx Schannel_1 FreeTx Schannel_2 C-Tx a b d RTM-T Pilot time Pilot c MDMA

12. Freq Control FreeTx Schannel_1 FreeTx Schannel_2 C-Tx a b d RTM-T a b c d CTR Pilot time Pilot c v SINR(k)>Thr e MDMA

13. Freq Control FreeTx Schannel_1 FreeTx Schannel_2 C-Tx a b d RTM-T a b c d CTR Data: C-Tx->a Pilot time Broadcast Data: C-Tx->b Pilot c MDMA

14. C-Rx a b c d RTM-R Freq time Control FreeRx Schannel_1 FreeRx Schannel_2 MDMA

15. C-Rx a b c d RTM-R Freq time a b c d CTT a b c d CTT Control FreeRx Schannel_1 FreeRx Schannel_2 MDMA

16. C-Rx a b c d RTM-R Freq time STT a b c d CTT a b c d CTT Control FreeRx Schannel_1 FreeRx Schannel_2 MDMA

17. C-Rx a b c d RTM-R Freq time STT a b c d CTT Data: a->C-Rx a b c d CTT Data: b->C-Rx Control FreeRx Schannel_1 FreeRx Schannel_2 MDMA

18. Range = 18 vv uu aa vv Analysis Average =4

19. 19 Analysisvv uu aa vv 1.3

20. Analysis Result 20 Analytically GSA is outperforming ideal scheduling 2.6 times

21. 21 Simulations C-Tx Assumptions Subcarriers2048 Subchannel32 (wimax) Randomly selected neighbors(K) 2 to 5 Channel, Rayleigh, Portable 20-taps, defined in EN-300-744 SNR10 SEEDS10 random b a c u Matlab Simulations

22. 22 Simulations C Assumptions Subcarriers2048 Subchannel32 (wimax) Randomly selected neighbors(K) 2 to 5 Channel, Rayleigh, Portable 20-taps, defined in EN-300-744 SNR10 SEEDS10 random A A AB B B Matlab Simulations

23. 23 MDMA Comparison McHOrthogonal channel assignment, adaptive rate selection, CH=3 CBDOrthogonal channel assignment, adaptive subchannel assignment, GSAAdaptive subchannel assignment GSA increases concurrency comparing to CBD 63% 2.3 times Matlab Simulations

24. Model Developed in QUALNET Number of nodes 50 stationary SNR10 Packet length 512 B topology10 random Average number of node’s neighbors 10 Bit rate11Mbit/sec Percentage of the communicating neighbors 24 MDMA 1.6 times Qualnet Simulations

25. Previous cross-layer MAC fail to effectively utilize OFDMA We improved our previous works by Utilizing Gibbs method Increasing two-hop neighbor concurrency Attaining diversity Our analysis and simulations proved that MDMA’s performance with the same BW is 1.6 times better than CBD 2.2 times better than CTRMA 2.4 times better than traditional multi-channel networks Future work – Mobility – 802.11n OFDMA 25 Conclusion

26. Questions