Communication over Bidirectional Links A. Khoshnevis, D. Dash, C Steger, A. Sabharwal TAP/WARP retreat May 11, 2006

Wireless Networks Higher throughput TAP: 400 Mbps WiMax/Mesh 4G

Network of Unknowns Queue Topology Interference Channel Battery

Medium Access Example If S 1 knows q 2 and S 2 knows q 1 –No need for handshaking –TDMA scheduling –No collision As load increases –Probability of queue empty reduces –Network utility increases Having the “knowledge” about Queue states, increases the utilization 1 2 q2q2 q1q1 S1S1 S2S2 D

How to learn about unknowns There is gain in knowing unknown parameters The information can be gathered –Directly Feedback Training Dedicated link, information sharing –Indirectly Overhearing Passive sensing H + X W Y Q(H) S1S1 S2S2 D

Need for Bidirectional links Indirect –Limited –Highly depends on the topology and availability Direct –Amount of information can be controlled An explicit sharing of information requires flow of information in both directions among all communicating nodes, hence Communication over Bidirectional Links

Cost-Benefit of learning the unknowns Catch –We don’t care about the unknown Only care about sending data –Time varying in nature Periodic measurements Spend resources for non-data If considering the true cost of knowing the unknown, is there still any gain left?

Our research Unknown Channel –Chris, Farbod, Ashu, Behnaam Allerton’05, ISIT’06, JSAC’06 Resource allocation algorithm Uncertainty of noise –Farbod, Dash, Ashu CTW’06, Asilomar’06 Coding scheme Randomness of source –Upcoming NSF proposal Access mechanism S1S1 D h S1S1 S2S2 D

Multiple Access Channel: MAC The system is modeled by Information theory answers: What is the maximum rate (R 1,R 2 ) at which X 1 and X 2 can transmit with arbitrary small probability of error X1X1 X2X2 Y

Standard solution method Finding an achievable upper bound –Achievability proof –Converse proof Typical solution to MAC R1R1 R2R2

MAC with Bidirectional links Time is slotted –Forward channel: multiple access –Reverse channel: feedback from receiver Superposition coding Decoded Tx Rx Decodable From Feedback Un-decodable New Information Un-decoded

Our model j,l I’,k’

Contribution and results Considering resources in feedback –Time –Power (P f ) Coding scheme to compress the feedback information P f / e P

Interpretation of result In second timeslot –Both user help to resolve uncertainty Co-operation induced by feedback

Cooperative link Anticipate the exponential feedback power is resolved Under investigation –Rate region –Coding strategies X1X1 X2X2 Y

What if… Receiver has information for senders Superimpose feedback information with its own information

Achievable rate region R3R3 A B A:  = 0 –Only Broadcast B:  = 1 –Only MAC

Channel state vs. data feedback So far, receiver sends back unresolved information In fading environment using channel state –Power / rate control increases the throughput Feedback can be used to send back channel state information h1h1 h2h2 h1h1 h2h2

Randomness of source X1X1 X4X4 X3X3 X2X2 Challenges: –K is random –Under delay constraint –Access mechanism is required Each node needs to know the number of active users

Recap Ongoing work: Gaining information about the unknowns increases the throughput Obtaining information is best when it is explicit and direct –Requires resources (power and time) to be allocated to unknowns –Requires bidirectional communication link Capacity of MAC increases with “realistic” feedback –Power in the feedback link is large Up coming: Cooperative link Channel state vs. data feedback Randomness of the source