Increasing cellular capacity using cooperative networks

1. Increasing cellular capacityusing cooperative networks Shivendra S. Panwar Joint work with Elza Erkip, Pei Liu, Sundeep Rangan, Yao Wang Polytechnic Institute of NYU

2. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

3. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

4. Cellular Networks are becoming heterogeneous • Macrocell based network architecture isexpensive and cannot keep up with user demand (Cisco’s 66X traffic increase prediction) • Heterogeneous networks enableflexible and low-cost deployments andprovide a uniform broadband experience • The network becomes a mix of macro, pico, femto base stations and operator deployed relay stations • The dense deployment greatly improves network capacity, and provides richer user experience and in-building coverage • Reduces operating cost, such as backbone cost, site acquisition cost, and utility cost for operators

5. Emerging trends and our research • The future network architecture is heterogeneous, with macro-, pico- and femto-cells, along with WiFi and (some) ad hoc nodes • A large part of the 66x increase predicted by Cisco will be drained by increased deployment of WiFi, femto/picocells for stationary or slow moving users • Femtocells, in particular, are the carrier’s Trojan Horses! • Macrocell bandwidth is precious and should be used only when there is no alternative (like satellite networks are today) • Cooperative networking can be used in such emerging environments by using user end devices, femtocells, WiFi access points, picocells, and macrocell infrastructure as the devices that constitute the cooperating nodes

6. Cooperation and Heterogeneity • Cooperation performs much better if the number of relays is large • In a macrocell based deployment, the number of operator deployed relay stations will be limited • In traditional networks, the performance gain for cooperation is limited unless user (MS) cooperation is enabled • But user cooperation gives rise to the following problems: battery consumption, synchronization, securityand incentive • The proliferation of pico/femto base stations will provide enough relays (“femtorelays”) • They do not have the battery consumption problem • They are easier to synchronize:stationary, backbone connection and better radio design • They are more secure because they are part of the operator’s network

7. Motivation for Cooperation • Wireless channel by nature is a broadcast one. • The broadcast channel can be fully exploited for broadcast traffic. • But it is considered more as a foe than a friend, when it comes to unicast. • Cooperative communications allow the overheard information be treated as useful signal, instead of interference. • Relays process this overheard information and forward to destination. • Network performance improved because edge nodes transmit at higher rate thus improving spectral efficiency. • Candidate relays?Mobile user, macro/pico-cell BS, fixed relays, femtocell BS, etc. • What are the incentives? Throughput, power, interference. • A cross-layer design encompassing physical, MAC, network and application layers is required to address this problem.

8. Relaying in commercial systems • Cooperative / multihop communications have been adopted in the next generation wireless systems. • IEEE 802.11sEnables multihop and relays at MAC layer, does not provide for joint PHY-layer combining. • IEEE 802.16jExpands previous single-hop 802.16 standards to include multihop capability. Integrated into IEEE 802.16m draft. • 3GPP LTECooperative multipoint is supported with joint transmissions and receptions to enable cost-effective throughput enhancement and coverage extension.

9. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

10. Robust Cooperative MIMO Design • Limitations of previous cooperative methods: • Single relay: low spatial diversity gain • Multiple relays: consume more bandwidth resource when several relays sequentially forward signal • Any alternative? • Distributed Space-Time Coding (DSTC) • How does DSTC work? • Recruit multiple relays to form a virtual MIMO • Each relay emulates an indexed antenna • Each relay transmits encoded signal corresponding to its antenna index • Pros: Spatial diversity gains • Cons: • Tight synchronization required • Relays need to be indexed, leading to considerable signaling cost • Global channel state information needed • Good DSTC might not exist for an arbitrary number of relays • Unselected relays cannot forward, sacrificing diversity gain

11. Robust Cooperative MIMO • Randomized cooperation strategies provide powerful PHY layer coding techniques that • alleviate the previous problems and allow robust and realistic cooperative transmission with multiple relays. • randomize distributed space-time coding (R-DSTC) for diversity. • randomized distributed spatial multiplexing (R-DSM) for spatial multiplexing. • Highlights of randomized cooperation: • Relays are not chosen a-priori to mimic particular antennas • Multiple relays can be recruited on-the-fly • Relays are used opportunistically according to instantaneous fading levels • Signaling overheads and channel feedback greatly reduced • Performance comparable to centralized MIMO is attained

12. R-DSTC: A New Solution • Randomized Distributed Space-Time Coding (R-DSTC) • How does R-DSTC work in PHY? • Two-hop network: source station, relays, destination station. • Relays re-encode the first-hop signals and forward over the second hop • Unlike DSTC, R-DSTC relay does NOT transmit the signal from a specific indexed antenna • Instead, each relay transmits a weighted linear combination of all streams of an underlying STC codeword of size L x K. • As long as the number of relays N>L-1, a diversity order of L is achieved.

13. R-DSTC Advantages

14. R-DSTC Performance (WiFi) • Underlying orthogonal STBC codeword size: 2, 3, 4. • PHY layer rates: 6, 9, 12, 18, 24, 36, 48, 54 • BPSK, QPSK, 16-QAM, 64-QAM; Convolutional code 1/2, 2/3, 3/4 • 20 MHz bandwidth • Contention window: 15 -1023 • Transmit power: 100mW

15. CoopMAX: A Cooperative Relaying Protocol in Mobile WiMAX Network • CoopMAX enables robust cooperation in a mobile environment with low signaling overheads. • It is robust to mobility and imperfect knowledge of channel state. • Simulation shows 1.8x throughput gain for a single cell with mobility, and 2x throughput gain for multicell deployment. Single cell deployment Multicell deployment

16. R-DSM for spatial multiplexing • Mismatch in the number of antennas on BS and MS • Assuming each mobile station has only one antenna and the base station has L antennas • Randomized Distributed Spatial Multiplexing (R-DSM) is based BLAST scheme • The channel capacity between the relays and the destinations scales linearly with min(N,L), where N is the number of relays • How does R-DSM work in PHY? • Two-hop network: SISO transmission from source to relays first, followed by relays transmitting together to the destination using R-DSM. • Each relay independently generates a random coefficient and then transmits a weighted sum of the signals for each antenna in BLAST scheme

17. Performance • Our results demonstrate that R-DSM scheme delivers MIMO system performance • Average data rate for the second hop (relays-destination link) scales with the number of relays • For direct transmissions, the peak data rate is supported at a short range • R-DSM can increase the number of stations that can transmit near the peak data rate

18. Cooperative Video Multicast • Performance of conventional video multicast schemes in an access network is limited • Source transmits atthe lowesttransmission rate • Receivers withgood channel quality unnecessarily suffer

19. Cooperative Video Multicast with R-DSTC • Source station transmitsa packet • Nodes who receivethe packets become relays which re-encodethe first-hop signals and forward over the second hop • Each relay transmits aweighted linear combination of all streams ofan underlying STCwith a dimension of

20. Results: Single Layer Schemes

21. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

22. Cooperative MIMO for Heterogeneous Networks • For high mobility MSs or MSs that are covered by any femtocell, cooperative MIMO • enables fully opportunistic use of all available surrounding radios. • increases network capacity and helps to reduce coverage holes.

23. Cooperative handoff for Pico/Femtocells • Handoffs happen much more frequentlyfor MSs in a heterogeneous network • Smaller BS coverage area • Loosely planned or unplanned deployment • Higher signaling overheads and more dropped calls • Cooperative handoffs in Heterogeneous Networks • Separate signaling and data paths • Macrocell BS orchestrates handoff and allocates radio resources for data transmissions • User data goes through surroundingpico/femtocellBSseither through their backhaul or by cooperative relaying

24. Cooperative Handoff for delay tolerant applications • Macrocell BS tracks the locations of the MS and makes handoff predictions based on which pico/femtocell BSs the MS is moving to. • In the downlink • Macrocell BS pre-fetches user data packets to acluster of pico/femtocell BSs via their backhauls • Macrocell BS allocates frequency/time slots for thedownlink data transmission • Pico/femtocell BSs cooperatively transmit to the MSusing R-DSTC • In the uplink • Macrocell BS broadcasts the allocated frequency/time slotsfor the MSs • A pico/femtocell BS that successfully decodes an uplink user packet forwards it to the Macrocell BS via its backhaul

25. Cooperative Interference Coordination • Pico/femtocell BS deployments are unplanned with vastly different power levels compared to macrocell BS deployments • The interference patterns are significantly different • Current cellular systems treat interference as noise, which is not effective for high interference levels • Dynamic orthogonalization orHan-Kobayashi is needed Han-Kobayashi Orthogonalization Treat interference as noise

26. Changing Interference Conditions Macro cell - planned Macro - unplanned Loss from randomness (~2dB) Very bad links (restricted assoc) Short-range model Very good links (SNR>10 dB)

27. Iterative message passing algorithm Widely used in coding, non-Gaussian estimation, machine learning Pass “beliefs” along edges of graphs representing estimates of the marginal distribution Natural distributed implementation for wireless. Similar methods used in many approximate BP algorithms for CDMA multiuser detection & non-Gaussian estimation: Caire, Boutros (’02), Guo-Wang (‘06), Tanaka-Okada (‘05), Neirroti-Saad (‘05), Kabashima (‘05), Donoho, Maleki, Montanari (‘09), Bayati-Montanari (‘10), Rangan (’10) Belief Propagation Solution

28. BP Multi-Round Protocol Interference Desired link Interference TX1 TX2 RX2 RX1 TX vector x1(0) TX vector x2(0) Round 0 Sensitivity D2(0) Interference z1(0) and sensitivity D1(0) TX vector x2(1) TX vector x1(1) Round 1 Sensitivity D2(1) Interference z1(1) and sensitivity D1(1) Data scheduled along TX vector x1 Data transmission

29. Interference Coordination with Relays • Still an open problem • What are the optimal strategies for transmitters, relays and receivers to maximize spectrum efficiency? • What is the best strategy for relays - Forwarding signal or forwarding interference? • Preliminary information theoretical results show both signal relaying and/or interference forwarding could be optimal under certain regimes (Elza Erkip) • Missing Components: • Practical coding and signal processing schemes for cooperative interference coordination • MAC design that handles the signaling between different entities participating in the cooperative interference coordination

30. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

31. The Second-Last Mile Problem • Explosively growing traffic demand • More than 5 billion cell phones by 2010 • Increasing number of data intensive applications • 3G/4G standards are pushing up the macrocell data rates (~100 Mbps) • Poor cellular infrastructure • Most of the BS backhauls use four to six T1/E1 lines (~8 Mbps) • Adding BSs or updating data lines is expensive(more than $10,000 per line and $50,000 per site annually) • Macrocell backhaul has become the bottleneck!

32. Solution: FemtoHaul • System Architecture for FemtoHaul • FemtoHaul is a novel solution to the macrocell backhaul problem. • In FemtoHaul, the femtocell backhaul is used to carry non-femto user traffic by forwarding through a relay. • Detailed Design • Channel allocation mechanism based on OFDMA WiMAX; • Policy for base stations to schedule user transmissions.

33. FemtoHaul Performance Evaluation • Backhaul Supply Rate Comparison • Average Download Rate in Stationary Scenario • Simulations demonstrate that our solution can significantly reducethe macrocell backhaul traffic while still guaranteeinga high rate to the subscribers

34. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

35. Cooperative Networking Testbeds • Goal: Build a large scale experimental, deployable and scalable cooperative network (Erkip, Korakis, Panwar, Liu, Wang, Bertoni) • Funding from NSF (MRI, CRI), WICAT, NYU-Poly • We have taken two approaches • PHY layer: Software Defined Radio (SDR) platform • MAC layer: Open Source Driver Platform on Linux

36. Implementing Cooperative PHY • Cooperative protocols require changes inthe PHY layer • Commercial devices do not give access to PHY • Use Wireless Access Research Platform (WARP), a SDR by Rice University • We have a basic three node system operating, consisting of one source, one relay and one receiver • Cooperative coding using convolutional codes and soft decision decoding implemented • We also have basic R-DSTC implemented

37. WARP System

39. Implementing Cooperative MAC for IEEE 802.11 • Use open source drivers and commercial WiFi cards • Advantages • Backward compatible with 802.11 • Can be used in large testbeds such as ORBIT • Disadvantages: • No access to PHY(but still gains from Cooperative MAC)

41. Outline • Motivation for Cooperation • Robust Cooperative MIMO Design • Randomized Space Time Coding • Randomized Spatial Multiplexing • Cooperation in Heterogeneous Network • Cooperative Handover • Cooperative Interference Coordination • Combating Macrocell Backhaul Bandwidth Shortage • Implementation Efforts • Conclusions

42. Conclusions • Cooperation is a perfect match for the emerging heterogeneity in wireless communications • Robust cooperative schemes (R-DSTC, R-DSM) require little overhead and well suited even for MSs with high mobility • Heterogeneous networks provide many capable relays for cooperation • Cooperative handoff • Cooperative interference coordination • FemtoHaul: Offload traffic from constrained macrocell backhaul to abundant femtocell backhaul

43. Thank You! Our Cooperative Research website: http://coop.poly.edu

45. Backup

46. Synchronization Issues • Nodes cooperating without central control will encounter the practical problem of synchronizing their access to the channel. • Distributed relays have no access to a global clock. • Relays need to be synchronized both in time and frequency. • Synchronization accuracy affects physical layer performance of cooperative MIMO system. • How to achieve synchronization? • 4G systems (LTE and WiMAX) synchronize the transmissions from UE both in time and frequency via close-loop control. • In a wireless LAN, relays can be synchronized by letting relays lock to a common reference signal. For example, the source can continuously transmit a reference carrier. • R-DSTC performs well under residual synchronization errors1. 1. M. Sharp, A. Scaglione and B. Sirkeci-Mergen, “Randomized cooperation in asynchronous dispersive links”, IEEE Transactions on Communications, vol. 57, no. 1, pp. 64-68, January 2009.

47. Incentives for cooperation • Cooperative relaying improves network capacity and reduces delay. • In a wireless LAN, throughput for each individual node can be improved. • In a cellular network, the BS can provide incentive for relays by allocating more time/frequency resources to relays. • Battery consumption • Average Joule/Bit performance is improved. • Energy consumption for nodes acting as relays (CoopMAC) is also reduced in wireless LANs2. • By employing several relays, the energy consumption for each individual relay is just 1/L of the case of employing one relay. • It is possible that a node’s battery drains faster because it acts as a relay for multiple sources, possibly as a result of its position. • Not an issue for dedicated fixed relays, or femtocells acting as relays. 2. S. Narayanan and S. Panwar, “To Forward or Not to Forward - That is the Question”, Wireless Personal Communications, Special issue on cooperation in wireless networks, Vol.43, No.1, pp. 65-87, 2007