1 / 31

MURI Review Meeting, UCSD - September 14, 2006

MURI Review Meeting, UCSD - September 14, 2006. PHY-aware MAC protocol design for MIMO ad hoc networks. Michele Zorzi. University of California at San Diego, and University of Padova, Italy Contributors: Paolo Casari, Francesco Rossetto, Marco Levorato and Stefano Tomasin.

nusa
Télécharger la présentation

MURI Review Meeting, UCSD - September 14, 2006

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. MURI Review Meeting, UCSD - September 14, 2006 PHY-aware MAC protocol design for MIMO ad hoc networks Michele Zorzi University of California at San Diego, and University of Padova, Italy Contributors: Paolo Casari, Francesco Rossetto, Marco Levorato and Stefano Tomasin US Army Research Office Multi-University Research Initiative (MURI) Grant # W911NF-04-1-0224

  2. Outline • Issues in ad hoc networks with multiple antennas • A new solution for the beamforming gain asymmetry • A cross-layer solution for MIMO ad hoc networks • Backoff and coordination of transmissions • Conclusions

  3. Multiple antennas in ad hoc networks • Recently introduced as a means for • Improving parallelism (greater spatial reuse) • Achieving higher bit rates • Bridging longer distances • Directional communications • Exploit higher transmit/receive antenna gain andnull signal propagation toward some directions • MIMO links • Many (not necessarily independent) PDUs are superimposedin space (need for channel tracking and adaptive weighing) • Greater channel capacity • Higher complexity at the receiver

  4. Random access and MIMO ad hoc nets • We focus on random access techniques • Best known is IEEE 802.11 • Main ingredients/features of 802.11 • CSMA and collision avoidance via RTS/CTS • Clear view of network activity • Backoff and retransmissions • These aspects need to be revisited if MIMO is used • Collisions need to be controlled rather than avoided • Signaling is trickier if directionality is exploited • Difficult to know who is active/available • How is backoff applied (source vs. destination)? • We deal with some of these issues and try to get insight on developing effective random access schemes

  5. Beamforming gain asymmetry [1],[2] • Parallelism may be significantly improved using multiple antennas • Control packets need to be sent omnidirectionally • Otherwise, lack of control information may impair correct data transmissions between parties • Beamforming may extend the range of transmissions, but most of the neighborhood will not detect the signal • Extending signaling range through directional transmissions only covers a fraction of space • Solution by Tassiulas et al.: use sequential directional transmissions • Solution proposed here: use proper Space-Time Codesto achieve omnidirectionality and range extension

  6. Circular RTS approach • Sweeping the horizon increases latency, causing a potential throughput reduction • Needs a LOS environment for beamforming to be effective • Tassiulas assumes the azimuth can be perfectly sliced • As many slices as antennas • Actually, this is not true • Considering the 3dB beamwidths of endfire and broadside beams, the number of needed beams is • 5 beams for 4 antennas • 10 beams for 8 antennas Broadside 3dB beamwidth

  7. Proposed signaling solution • Extend range maintaining omnidirectionality through STC • Design a solution that works in a general Ricean env. • Note that transmit CSI is not available (one different channel per user) • Solution • Use RX maximal ratio combining for array gain (needs training) • Use STBC at the transmitter to get coding gain instead of beamforming gain • Codes should be full-rate (for delay) and full-diversity (for outage) • Fading improves STBC performance through greater diversity • Generalized ABBA codes used(2x2  Alamouti, 4x4, 8x8) Space Time

  8. BER results with 4 antennas • 4x4 STBC approach compared to ideal Circular RTS (perfect slices) • Total Rayleigh vs. total LOS environment • STBC outperforms CRTS in Rayleigh fading at higher SNR and vice versa, but in either case the difference is at most 2 dB • If the BER loss is tolerable, the delay abatement can be fully exploited

  9. Network results • In a simulated network, the delay has been reduced by 20% to 50% • STBC always provides the required level of performance at lower delay • STBC works in every channel condition • With 8 antennas, only a half-rate 8x8 STBC has comparable BER wrt CRTS • Introduces a delay of 2… • …which is still lower than the 10-fold delay of CRTS in this scenario

  10. Gain asymmetry: conclusions • An STC-based scheme can achieve increased-range communications omnidirectionally without delays • Can be used in signaling phase of common protocols • The delay decrease positively impacts the overall networking performance • Publications: [1] F. Rossetto, M. Zorzi; "A Space-Time based Approach to Solving the Gain Asymmetry in MIMO ad hoc Networks"; Proc. of IEEE VTC Spring, Melbourne, Australia, May 7-10, 2006. [2] F. Rossetto, M. Zorzi; "On gain asymmetry and broadcast efficiency in MIMO ad hoc networks"; Proc. of IEEE ICC, Istanbul, Turkey, June 11-15, 2006.

  11. Decision Feedback Multiuser Detection(V-BLAST) in ad hoc networks [3],[4] • Great advantage can be derived from the use of BLAST in MIMO ad hoc networks • Incoming signals have highly different attenuations (better case for multiuser detection) • More packets can be simultaneously sent (spatial multiplexing, SM) • Upon reception of multiple signals: • Select the one with highest SNR • Obtain ZF weights by (pseudo)-inverting the correlation matrix • Detect the signal • Cancel the detected contribution • Restart with the 2° highest SNR • …and so on Canceled Under detection SNR ZF nulling

  12. Network design – Frame structure • The major drawback of BLAST is that receivers may easily become overwhelmed if too many signals • How to limit transmissions w/o hampering throughput? • Choice: organize transmissions in frames • RTS slots used to declare transmit needs (heard by all receivers) • Receivers use the CTS to limit own traffic and avoid overload • Incoming traffic estimates can be derived from the RTSs • DATA packets are sent according to info in CTSs • In the ACK phase, the receivers confirm any correct detection • NO NEED for symbol synchronization (it’s a complexity issue…)

  13. Network design – CTS policy • In high traffic scenario, nodes cannot be allowed to send everything they request • Need for a policy to limit the number of TX granted • Reception and interference cancellation capabilities are limited by the maximum number of trackable channels • Choice: Follow Traffic (FT) policy • From RTSs, understand which packets are wanted and which are interferers • Always grant at least one wanted request • Take other requests in order of decreasing power, and grant them if they are wanted • Otherwise, account for the fact that they are interferers to cancel • Terminate when the maximum number of trackable channels is reached

  14. Example of FT operations • Three wanted requests (4+4+2 packets in total) • Five interfering requests (14 pkt) • The red one isgranted first (greatest RX power) • Other are either granted (if wanted) or estimated and canceled (if interfering) • Some of them (lightest yellow) are neglected Transmission requests SNR Expected interference Mapped into channel estimation resources Result: a tradeoff betweenthroughput and protection from interference From RTSs

  15. Other details on CTS policies • Balance between the needs for • High reuse and superposition of transmissions • Protection from stronger (highest power) interference • Acceptable receiver load (wanted + unwanted estimated PDUs) • Acceptable unestimated interference • Other examples of CTS policies • Partially Follow Traffic (PFT): according to the previous figure • If possible, the receiver grants all transmission requests first • If there are some estimation resources left, as many interferers as possible are accommodated, in order of decreasing RX power • Expected to be unable to sufficiently suppress interference • Do Not Follow Traffic (NFT): according to the previous figure • As many wanted transmissions as possible are granted • No interfering PDU is estimated and canceled • Expected to perform poorly, since it does not exploit MIMO benefits

  16. Network simulation and parameters

  17. PHY level approximation [5],[6] • A fully detailed bit-level simulation of a MIMO ad hoc network is overly complex • Need to reproduce the behavior of a multiuser detector • Need to transmit a large number of packets to achieve confidence • Solution: evaluate BER through a Gaussian approximation of the power of residual errors after cancellation • One-shot calculation of the SINR of the current detection Extracted signal power Space-filtered noise power UnestimatedinterferingPDUs’ power Error signal power(due to imperfectcancellations) Power of still uncanceled PDUs

  18. Throughput performance • Throughput = # of correctly received wanted PDUs per frame • NFT shows the poorest performance • PFT sustains more traffic but soon shows congestion (insufficient interference protection) • FT does not experience deadlock because of its good balance between throughput and interference cancellation • Analysis and fully detailed simulation are in very good agreement

  19. PDU reception success performance • % of correctly received PDUs • Performance agrees with the throughput results • The poor PFT and NFT behavior is mainly explained with low success in detecting incoming signals • Message: the tradeoff between throughput and interference protection is crucial in a MIMO ad hoc network

  20. Cross-layer MAC/PHY: conclusions • Potential of MIMO PHY can be exploited via careful PHY-aware MAC design • Interesting tradeoffs to be explored • Important to have effective simulation tools • Publications: [3] P. Casari, M. Levorato, M. Zorzi; "On the Implications of Layered Space-Time Multiuser Detection on the Design of MAC Protocols for Ad Hoc Networks"; Proc. of IEEE PIMRC 2005. [4] P. Casari, M. Levorato, M. Zorzi; "Some Issues Concerning MAC Design in Ad Hoc Networks with MIMO Communications"; WPMC Symposium, Aalborg, Denmark, September 2005. [5] M. Levorato, S. Tomasin, P. Casari, M. Zorzi; "Analysis of Spatial Multiplexing for Cross-Layer Design of MIMO Ad Hoc Networks"; IEEE VTC Spring, Melbourne, Australia, May 7-10, 2006. [6] M. Levorato, S. Tomasin, P. Casari, M. Zorzi; "An Approximate Approach for Layered Space-Time Multiuser Detection Performance and its Application to MIMO Ad Hoc Networks"; Proc. of IEEE ICC, Istanbul, Turkey, June 11-15, 2006. [7] M. Zorzi, J. Zeidler, A. Anderson, A. L. Swindlehurst, M. Jensen, S. Krishnamurthy, B. Rao, J. Proakis; "Cross-Layer Issues in MAC Protocol Design for MIMO Ad Hoc Networks"; IEEE Wireless Communications Magazine (special issue on smart antennas), August, 2006.

  21. On channel access persistency and coordination among nodes [8,9] • The previous scenarios comprise frame-structured transmissions, a CTS policy, and a random exponential backoff in case a CTS is not received • The rationale is to limit persistency in transmission attempts until local overload has passed • Persistency reduction techniques, such as backoff • Reduce local traffic, avoiding congestion • Allow FT and similar policies to work on a smaller number of requests, thus being more efficient • Backoff policies have a great impact on network performance and deserve a deeper understanding • More effective concept: try to avoid TX to receivers who can’t hear • New challenge: how to know receivers’ availability?

  22. Exponential backoff schemes • In order to understand which backoff configuration achieves the best performance, we compare • Destination-wise backoff (DEST–LOCK) which blocks transmission toward a single destination each time • Node-wise backoff (NODE–LOCK), which blocks any transmission • In either case, the silence time lasts for a number of frames randomly chosen in the interval [1,Bmax], with Window growth parameter Number of consecutive failures

  23. NODE–LOCK throughput • Node-Lock prevents all transmissions upon unheard response to RTS • Result: channel access is more limited • Tuning the window growth parameter W changes the saturation throughput value • W = 1 is the best choice here (balances silence time and need to transmit) Increasing W

  24. DEST–LOCK throughput • Dest-Lock only blocks transmissions toward unavailable receivers • Thus on average, nodes are left more free to transmit • Unlike Node-Lock, here we need to increase W, so that dest-wise silences are longer • Best performance is obtained with W = 16 (best transmission/silence balance) • Node-Lock outperforms the best Dest-Lock Increasing W

  25. Beyond backoff: DSMA [8] • Distributed Scheduling for MIMO Ad hoc networks • Better results are expected if nodes could know exactly who is free to transmit to (impossible in random access) • Receivers: piggy-back in the ACK packet a special reservation message directed to wanted destinations • Transmitters and idle nodes: receive this ACK and, if reserved, refrain from transmissions in the following frame • This way, there is a higher probability that a wanted party is not engaged in something else, thus more links activated • Better parallelism and spatial reuse • More unlikely that no CTS response because of a busy receiver

  26. DSMA: further details • Idle nodes’ probability of transmitting is constrained, so that they are somehow forced to listen to reservation even in high traffic situations: this prob is called ptx • Receivers do not always reserve a node, but do this with a tunable probability, called preserve • The maximum number of different nodes that can be reserved is dmax. • Thus, upon reservation any node scans its queue until it finds at most dmax different nodes to reserve • We tested this protocol and compared it to the previous schemes based on random access with backoff

  27. DSMA throughput (ptx = 0.1) • The reservation scheme used in DSMA achieves a good coordination • In turn, this means better throughput… • …with almost no further overhead (reservations are included inside ACKs) • Drawback: the parameters (e.g., preserve) have different optimal values depending on traffic Increasing pr

  28. Backoff and DSMA: conclusions • Different levels of aggressiveness: nodelock vs destlock • Interesting tradeoffs can be studied • Distributed scheduling shows promise for improvements • Actual behavior depends on various parameters • These are preliminary results, more under way • Publications: [8] M. Levorato, P. Casari, M. Zorzi, "On the Performance of Access Strategies forMIMO Ad Hoc Networks", accepted to IEEE GLOBECOM, San Francisco, CA, USA, Nov 27 - Dec 1, 2006. [9] P. Casari, M. Levorato, M. Zorzi; "DSMA: an Access Method for MIMO Ad Hoc Networks Based on Distributed Scheduling"; Proc. of ACM IWCMC, Vancouver, Canada, July 3-6, 2006.

  29. Conclusions • Multiple antennas in ad hoc networks are a very promising enabling technology for next generation wireless systems • The new challenges posed call for protocol design strategies that explicitly account for the special PHY • Cross-layer design is very helpful – information exchange among layers is essential • A mere stacking of an old MAC on a (though higher-capacity) MIMO PHY cannot do the job • Our study focused on the effects and consequences, pros and cons of such interactions, toward a deeper understanding of the use of MIMO in ad hoc networks

  30. Current and future work • Include multihop operation in the study • Evaluate the performance of such schemes • Identify possible design tradeoffs, and MAC/routing strategies • Include waveform design in the study • We are evaluating LASTMUD with CDMA – other possibilities • Study the effect of power control • Study the effect of imperfect channel estimates • Look for improvements of DSMA • Study dependencies on traffic and other parameters • Compare random access and scheduled schemes

  31. Published papers Effective use of Space-Time and Directional Communications [1] F. Rossetto, M. Zorzi; "A Space-Time based Approach to Solving the Gain Asymmetry in MIMO ad hoc Networks"; Proc. of IEEE VTC Spring, Melbourne, Australia, May 7-10, 2006. [2] F. Rossetto, M. Zorzi; "On gain asymmetry and broadcast efficiency in MIMO ad hoc networks"; Proc. of IEEE ICC, Istanbul, Turkey, June 11-15, 2006. Decision-Feedback Multiuser Detection in ad hoc nets: PHY evaluation and MAC design [3] P. Casari, M. Levorato, M. Zorzi; "On the Implications of Layered Space-Time Multiuser Detection on the Design of MAC Protocols for Ad Hoc Networks"; Proc. of IEEE PIMRC 2005. [4] P. Casari, M. Levorato, M. Zorzi; "Some Issues Concerning MAC Design in Ad Hoc Networks with MIMO Communications"; WPMC Symposium, Aalborg, Denmark, September 2005. [5] M. Levorato, S. Tomasin, P. Casari, M. Zorzi; "Analysis of Spatial Multiplexing for Cross-Layer Design of MIMO Ad Hoc Networks"; IEEE VTC Spring, Melbourne, Australia, May 7-10, 2006. [6] M. Levorato, S. Tomasin, P. Casari, M. Zorzi; "An Approximate Approach for Layered Space-Time Multiuser Detection Performance and its Application to MIMO Ad Hoc Networks"; Proc. of IEEE ICC, Istanbul, Turkey, June 11-15, 2006. [7] M. Zorzi, J. Zeidler, A. Anderson, A. L. Swindlehurst, M. Jensen, S. Krishnamurthy, B. Rao, J. Proakis; "Cross-Layer Issues in MAC Protocol Design for MIMO Ad Hoc Networks"; IEEE Wireless Communications Magazine (special issue on smart antennas), August, 2006. [8] M. Levorato, P. Casari, M. Zorzi, "On the Performance of Access Strategies forMIMO Ad Hoc Networks", accepted to IEEE GLOBECOM, San Francisco, CA, USA, Nov 27 - Dec 1, 2006. [9] P. Casari, M. Levorato, M. Zorzi; "DSMA: an Access Method for MIMO Ad Hoc Networks Based on Distributed Scheduling"; Proc. of ACM IWCMC, Vancouver, Canada, July 3-6, 2006.

More Related