Practical Many-Antenna Base Argos Stations Clayton W. Shepard Hang Yu, Narendra Anand, Li Erran Li, Thomas Marzetta, Richard Yang, Lin Zhong Motivation • Spectrum is scarce • Hardware is cheap 2 MU-MIMO Theory • More antennas = more capacity – Multiplexing – Power – Orthogonality 3 4 Why not? • Nothing scales with the number of antennas – CSI Acquisition – Computation – Data Transportation 5 Background: Beamforming = Destructive Interference Constructive Interference ? = 6 Background: Channel Estimation Measured Due PathtoEffects environment channels (Walls)and are not terminal The Align CSI uplink, the is phases then send Uplink? aat pilot the from receiver atantenn the the to mobility reciprocal due tocalculated differences has to AFor pilot is estimation sent from each BSoccur terminal ensure terminal constructive and then sent calculate interference to CSI theatBS BS quickly in the Tx and and periodically Rxback hardware! Tx R x + Tx BS R x = Tx + R x Tx R x 7 Background: Multi-User Beamforming 8 Background: Multi-User Beamforming 9 Background: Null Steering 10 Background: Null Steering 11 Background: Multi-User Beamforming 12 Background: Scaling Up 13 Background: Scaling Up 14 Background: Scaling Up 15 Background: Scaling Up 16 Background: Linear Precoding • Calculate beamweights – Every antenna has a beamweight for each terminal • Multiply symbols by weight, then add together: s W s 17 Recap 1) Acquire CSI 2) Calculate Weights 3) Apply Linear Precoding 18 Scalability Challenges 1) Acquire CSI – M+K pilots, then M•K feedback 2) Calculate Weights – O(M•K2), non-parallelizable, centralized data 3) Apply Linear Precoding – O(M•K), then O(M) data transport 19 Argos’ Solutions 1) Acquire CSIO(M•K) → O(K) – New reciprocal calibration method 2) Calculate WeightsO(M•K2) → O(K)* – Novel distributed beamforming method O(M•K) → O(K) * 3) Apply Linear Precoding – Carefully designed scalable architecture 20 Solutions • Reciprocal Calibration • Distributed Beamforming • Scalable Architecture 21 Channel Reciprocity • Pilot transmission source? – Basestation – Terminal • Base station pilot transmission – Requires feedback – M pilots (M ≥ K) • Terminal pilot transmission – No feedback – K pilots 22 Channel Reciprocity TxA RxA TxB Tx TxAA+C+Rx +RxC-Tx C-Tx C- CC-Rx RxA A Can Tx/Rx we Chain do this differences without terminal require involvement? calibration TxB+RxC-TxCRxB TxC RxC Channel Estimation RxB Transmission 23 = Key Idea Any constant phase shift results in same beampattern! = 24 Channel Reciprocity TxA TxATx +Rx A-Rx C-Tx A CRxA RxA TxB TxC TxBTx +Rx B-Rx C-Tx B CRxB RxC Channel Estimation RxB Transmission 25 Internal Reciprocal Calibration TxA • Find phase difference between A and B TxA+C+RxB TxB+C+RxA RxA TxA+RxB - TxBRxA TxB RxB • Tx from A: • Phase offset = TxA-RxA add phase difference • Tx from B: + (TxA+RxB - TxBRxBA)) = • (TxB-Rx TxA26 RxA Solutions • Reciprocal Calibration • Distributed Beamforming • Scalable Architecture 27 Problems with Existing Methods • Central data dependency • Transport latency causes capacity loss • Can not scale – Becomes exorbitantly expensive then infeasible 28 Conjugate Beamforming • Requires global power scaling by constant: • Where, e.g.: • This creates a central data dependency 29 Conjugate Beamforming Power Bad Channel Good Channel BS Okay Channel 30 Conjugate Beamforming Power Low Power High Power BS Normal Power 31 Distributed Conjugate Beamforming • Scale power at each antenna: • Maximizes utilization of every radio – More appropriate for real-world deployments • Quickly approaches optimal as K increases – Channels are independent and uncorrelated 32 Distributed Conjugate Beamforming High Power High Power BS High Power 33 Solutions • Reciprocal Calibration • Distributed Beamforming • Scalable Architecture 34 Architectural Design Goals • Scalable Data – Support thousands of BS antennas ? – Cost scales linearly with # of antennas … • Cost-effective 35 Linear Precoding … … K … K … K … K M 36 Scalable Linear Precoding … … K … K … K … K M 37 Scalable Linear Precoding … … K … K … K K … Common Databus! M 38 MUBF Linear Precoding: Uplink K K … K … … … … K M 39 Scalable Linear Precoding Constant Bandwidth! K K … K … … … … K M 40 Partition Ramifications • CSI and weights are computed and applied locally at each BS radio – No overhead for additional BS radios • No central data dependency – No latency from data transport – Constant data rate common bus (no switching!) • Unlimited scalability! 41 How do we design it? • Daisy-chain (series) … – Unreliable – Large end to end latency • Flat structure … – Unscalable – Expensive, with large fixed cost • Token-ring | Interconnected … – Not amenable to linear precoding – Variable Latency – Routing overhead 42 Solution: Argos Architecture Data Backhaul Central Controller Argos Hub Module Argos Hub Module … Module Argos Hub Module Module … Module … Radio Radio … Radio 43 Argos Implementation WARP Module Ethernet Central Controller (PC with MATLAB) Central Controller Argos Hub Daughter FPGA WARP Module Cards WARP Module Daughter FPGA Radio Power PC Daughter Cards FPGA Cards 1 Radio Radio Argos Hub 1 Radio 2 1 Peripherals Hardware Radio FPGA Fabric Argos Ethernet Radio Argos and Other I/O Model Radio 2 FPGA Fabric Interconnect Module Interconnect 3 Peripherals Hardware 2 Radio Peripherals and Other I/O Hardware Model Radio Sync Pulse Module and Other I/O Model 3 Radio Clock Board 4 3 Radio Clock Radio Clock Board 4 Module Clock Board 4 Distribution … Power PC PowerFPGA PC Fabric 44 16 45 Central Controller WARP Module s Argos Interconnects Sync Distribution Argos Hub Clock Distribution Ethernet Switch46 System Performance 47 Linear Gains as # BS Ant. Increases Capacity vs. M, with K = 15 48 Linear Gains as # of Users Increases Capacity vs. K, with M = 64 49 Scaling # of Users with 16 BS Ant. Capacity vs. K, with M = 16 50 Zero-forcing is not always better! Capacity vs. K, with M = 16 | Low Power 51 Calibration is stable for hours! 52 Conclusion • First many-antenna beamforming platform – Real-world demonstration of manyfold capacity increase • Devised novel techniques and architecture – Unlimited Scalability http://argos.rice.edu http://recg.rice.e du 53