1 / 18

COOPERATIVE TECHNIQUES AT THE NETWORK LEVEL

COOPERATIVE TECHNIQUES AT THE NETWORK LEVEL. Anthony Ephremides University of Maryland Colorado State University January 29, 2008. Based on joint work with: R. Liu (UMD), S. Misra (ARL/Cornell U.), A. Sadek (UMD), Y. Sung (Qualcomm), L. Tong (Cornell U.), and L. Yu (UMD). Preview.

saniya
Télécharger la présentation

COOPERATIVE TECHNIQUES AT THE NETWORK LEVEL

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. COOPERATIVE TECHNIQUES AT THE NETWORK LEVEL Anthony Ephremides University of Maryland Colorado State University January 29, 2008 Based on joint work with: R. Liu (UMD), S. Misra (ARL/Cornell U.), A. Sadek (UMD), Y. Sung (Qualcomm), L. Tong (Cornell U.), and L. Yu (UMD)

  2. Preview • Cooperation in a Network can occur at different levels • One example at the MAC layer • One example at the Routing layer • Unless and until we break the barriers among layers, considering cooperation in a broader sense is useful

  3. A little (but important) Background • Much touted question of ‘‘capacity‘‘ • Maximum Throughput Region (packets/slot) (saturated queues)-TR • Maximum Stable Throughput Region (packets/slot) (finite delays)-STR • Capacity Region (bits/s) (reliable communication limits)-C • TR, STR, C : need not coincide Ad Hoc Wireless Network

  4. Example: 2-users in random access collision channel 2 1 1 R TR 2 TR=STR=C Error-free when no collision • “Thorny” model to extend • Interacting Queues-Dominant Systems • No-Cooperation yet! 1 1 STR

  5. More General Network • Similar goal= Find TR, STR, C • Gupta & Kumar (2000)=TR (asymptotically) • Tassiulas and Ephremides (1992) (STR) • Tassiulas & Neely & Geo (2006) (STR) • Point-to-point • Scheduled access (mostly) D Based on “back-pressure” algorithm Equalizes queue loads Delay can be substantial (not “routing-savvy”) 1 1

  6. 1 2 M i 1st Example-Cooperation at the MAC level • Network view of the relay channel • TDMA underlying structure (i): ith-user’s portion (interference-free) • Success if SNR> • Channel sensing is possible • Feedback ACK is perfect M Source Terminals Relay hld hil Destination (d) hid Objective: Exploit the capabilities of the relay

  7. Cooperation Method 1 • Each terminal transmits HOL packet in its assigned slot (if empty, slot is free) • If D receives successfully, it sends ACK (heard by both the relay and the user) • If D does not succeed but R does: at first sensed empty slot R transmits to D the failed packet • If neither D nor R succeed, packet gets retransmitted by the terminal in next frame • Relay does not keep packets after the end of the frame Idle slots are utilized! Stable throughput for the M terminals Remarks: • Relay has always a finite queue (M packets Max) • Individual terminals interact • Successful service of a packet in a frame depends on whether the other terminals are idle or not

  8. Cooperation Method 2 • Each terminal transmits HOL packet in its assigned slot (if empty, slot is free) • If D receives successfully, it sends ACK (heard by both the relay and the user) • If D does not succeed but R does: at first sensed empty slot R transmits to D the failed packet • If neither D nor R succeed, packet gets retransmitted by the relay at next opportunity • Relay keeps all packets it receives correctly Remarks: Again: Idle slots are utilized! Stable throughput for the M terminals and the Relay • Relay has a possibly growing queue • Individual terminals do not interact • They release the unsuccessful packets to the relay • Enhanced version: Relay retransmits only packets of terminals with inferior channels

  9. Relationship to Other Schemes • Plain TDMA (no relay help) • Note that at saturation both schemes reduce to TDMA (no idle slots) • But in 2nd method with losses (what does throughput mean?) • Random Access (no relay help) • Selective Decode-and-Forward (interpreted at Network Level) • Relay forwards if it decodes correctly (in next slot) • Must keep “apples and apples”-hence no saturation • Idle slots not utilized • Gets two chances at twice the rate against a simple chance at the lower rate (per fixed packet) • Others in similar vein

  10. Results for 2-users 2 R1(S1) 2f2d+21(1-f1d)f1lfld 2f2d R2(S1) 1 1f1d 1f1d+21(1-f2d)f2lfld Method 1 at a specific resource sharing vector (1,2) Comparison

  11. Coop-DF: Relay transmits at twice the rate and utilizes one time slots. (Rate and SNR-threshold are related through the Gaussian mutual information formula Coop-DF: Relay transmits at the same rate and utilizes two time slots.

  12. Delay • Notoriously difficult for interacting queues • Symmetric System: 2-users

  13. Questions • Other possible uses of the Relay • Fundamental Relationship of Information-theoretic view of cooperation to Network-level view • More complex networks possibly tractable (max throughput result and methods can be used)

  14. 2nd Example-Cooperation at the Routing Level • Detection of target signal • Objective: maximize PD=prob. of correct detection • Determine routes (not a priori fixed) • Need mapping of PD on link metrics • Sensors cooperate as follows • Every node receives a sufficient statistic from “upstream” • It makes its own measurement as well • Transmits its best sufficient statistic downstream

  15. Tandem Case Fusion center • Markovian signal in space (good physical model) • Detection performance is well approximated by sum of terms along the links (one for each link) • Each link’s terms is monotonically related to its length (assume polynomial attenuation and independent noises) • The longer the link the better (the less correlation the better) • Breakthrough of costs, … except… Target X

  16. Use in “Blind Routing” • First of all who starts? • Chasing farthest nodes • No guaranteed convergence • Poor performance (correlation along path is not monotonic anymore) Target X Fusion center

  17. Possible “Fixes” • Add energy component to metric (moderate the bias toward longer links) • Add exclusion region around visited nodes to enforce directivity • Repeat analysis in 2-dimensions • Remark: • At the heart of the calculation there is an interesting coupling of mutual information and detection performance

  18. Summary • Two examples of “Network-level” cooperation • Stable Throughput Region: Fundamental • Are the differences from the Information Theoretic approach leading to interesting new views of cooperation?

More Related