Designing Multi-User MIMO for Energy Efficiency When is Massive MIMO the Answer? Emil Björnson‡* Joint work with: Luca Sanguinetti‡§, Jakob Hoydis†, and Mérouane Debbah‡ ‡Alcatel-Lucent *Dept. Signal Processing, KTH Royal Institute of Technology, Sweden §Dip. Ingegneria dell’Informazione, University of Pisa, Pisa, Italy †Bell 2013-11-07 Chair on Flexible Radio, Supélec, France Laboratories, Alcatel-Lucent, Stuttgart, Germany Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 1 Outline • Presentation is based on E. Björnson, L. Sanguinetti, J. Hoydis, M. Debbah, “Designing Multi-User MIMO for Energy Efficiency: When is Massive MIMO the Answer?,” Submitted IEEE WCNC 2014 Preprint available on arXiv: http://arxiv.org/abs/1310.3843. • Main Question - How should a single-cell downlink system be designed to maximize energy efficiency? - Optimization variables: Number of base station antennas Number of active user equipments Data rate guaranteed per user • Conclusions - Result depends strongly on physical layer precoding scheme - Unconventionally many users and antennas can be optimal! 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 2 Introduction 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 3 What are the Expectations? • Tons of Plenary Talks and Overview Articles - Fulfilling dream of ubiquitous wireless connectivity • Expectation: Many Metrics Should Be Improved in 5G - Higher user data rates Higher area throughput Great scalability in number of connected devices Higher reliability and lower latency Better coverage with more uniform user rates Improved energy efficiency • These are Conflicting Metrics! - Impossible to maximize all metrics simultaneously - Our goal: High energy efficiency (EE) with uniform user rates 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 4 Multi-User MIMO System • Multi-User Multiple-Input Multiple-Output (MIMO) - One base station (BS) with array of π antennas πΎ single-antenna user equipments (UEs) Downlink: Transmission from BS to UEs Share a flat-fading carrier • Multi-Antenna Precoding - Spatially directed signals Signal improved by array gain Adaptive control of interference Serve multiple users in parallel Space-division multiple access (SDMA) 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 5 Multi-User MIMO System (2) • Cell: Area with user location and pathloss distribution • Scheduling: Pick users randomly, with random location Some UE Distribution Clean-Slate Design Select π and πΎ to maximize EE! 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 6 How to Measure Energy Efficiency? • Energy Efficiency in bits/Joule - πΈπΈ = bits channel use Joule Power Consumption channel use Average Sum Throughput • Conventional Academic Approaches - Maximize throughput with fixed power - Minimize transmit power for fixed throughput • New Problem: Balance throughput and power consumption - Crucial: Account for overhead signaling - Crucial: Use reasonable power consumption model 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 7 System Model: Average Sum Throughput 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 8 Time-Division Duplex (TDD) Protocol Assumption: Perfect estimation • Coherence Period: π [channel uses] 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 9 Average Sum Throughput π‘1 • System Model π‘2 - Precoding vector of User π: vπ - Channel vector of User π: hπ ~ πΆπ(π, λπ π) • Random User Selection, - Channel variances λπ : Independent random variables, πλ (π₯) • Achievable Rate of User π: Average over channels and user selection Signal-to-interference+noise ratio (SINR) Cost of estimation 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 10 Impact of Precoding • What Determines User Rates? - Precoding (vector directions and power allocations) - “Optimal” precoding: Extensive computations – Not efficient • Notation - Matrix form: π = [π―1 , … , π―πΎ ], π = [π‘1 , … , π‘πΎ ] - Total radiated power: Ptrans = tr(ππ» π) Maximize signal Minimize interference • Heuristic Closed-Form Precoding - Maximum ratio transmission (MRT): vπ = - Zero-forcing (ZF) precoding: π = π π π» π - Regularized ZF (RZF) precoding: ππ hπ −1 diag(π , … , π ) 1 πΎ π = π(π 2 π Balance signal and interference 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 11 Uniform User Performance • Assumption: Uniform user performance - Same rate at every user: 1 − πΎ π log 2 1 + π π − πΎ - Scaling parameter π ≥ 0 can be optimized Lemma 1 Consider ZF precoding and the user rates above, the average radiated power is Ptrans = E{tr ππ» π) = πΎππ΄λ where π΄λ = E π2 λ depends on UE distribution, propagation, etc. • Consequence: - We use ZF in analysis and other precoding for simulation 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 12 System Model: Power Consumption 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 13 Reasonable Power Consumption Model • What Consumes Power? - Examples will motivate our model • Transmit Power: 1 P η trans - πtrans = Average radiated transmit power -η = Efficiency of power amplifier at BS • Transceiver Chains: π β ππ‘π₯ +ππ π¦π + πΎ β πππ₯ - ππ‘π₯ = Circuit power / BS antenna (converters, mixers, filters) - ππ π¦π = Power of common oscillator at BS - πππ₯ = Circuit power / UE (oscillator, converters, mixer, filters) • Coding and Decoding: πΎ(ππππ + ππππ ) - ππππ = Power for coding at BS / user - ππππ = Power for decoding at each user 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 14 Reasonable Power Consumption Model (2) • Computational Efficiency: πΏ operations per Joule • Uplink Channel Estimation: πΎβπ πΏβπ - Only once per coherence period - π channel components per user, processed separately • Precoding: πΆprecoding πΏβπ - Only once per coherence period - Depends on precoding: πΆprecoding 2πΎπ for MRT = 3πΎ 2 π + 2πΎπ + 2 πΎ 3 for ZF 3 • Architectural Costs: π0 - Control signaling, backhaul infrastructure, load-independent processing, etc. 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 15 Reasonable Power Consumption Model (3) • Summary - General model of power consumption: 1 η πtotal = Ptrans + 3 π πΆ πΎ π,0 π=0 + 2 ππ πΆ πΎ π,1 π=0 for some parameters 0 < η ≤ 1 and πΆπ,π ≥ 0. Energy Efficiency for ZF User rate: 1 − πΎ π log 2 1 + π π − πΎ Radiated power: Ptrans = πΎππ΄λ Average Sum Throughput πΈπΈ = = 1 Power Consumption πΎ πΎ 1 − π log 2 1 + π π − πΎ η πΎππ΄λ + 3 πΆ πΎπ π=0 π,0 + 2 πΆ πΎππ π=0 π,1 Design parameters: π, πΎ, and π 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 16 Optimize System Parameters for Energy Efficiency 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 17 Preliminaries • Our Goal: - Optimize number of antennas π - Optimize the (normalized) transmit power π - Optimize number of active UEs πΎ • Definition - Lambert π function, π(π₯), solves equation π π₯ π π(π₯) = π₯ - The function is increasing and satisfies π 0 = 0 - π π(π₯) behaves almost as linear: 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 18 Optimal Number of BS Antennas • Find π that maximizes EE with ZF precoding: πΎ πΎ 1 − π log 2 1 + π π − πΎ maximize π: π ≥ πΎ 1 η πΎππ΄λ + 3 π π=0 πΆπ,0 πΎ + 2 π π=0 πΆπ,1 πΎ π Theorem 1 (Optimal π) • Observations - Increases sublinearly with power π but linearly at high π - Increases with circuit power coefficients independent of π - Decreases with circuit power coefficients multiplied with π 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 19 Optimal Transmit Power • Find π that maximizes EE with ZF precoding: πΎ πΎ 1 − π log 2 1 + π π − πΎ maximize π≥0 1 η πΎππ΄λ + 3 π π=0 πΆπ,0 πΎ + 2 π π=0 πΆπ,1 πΎ π Theorem 2 (Optimal π) • Observations - Increases power with number of antennas as π/ log π - Opposite to recent claim: Power should decrease as 1/ π - Intuition: Higher circuit power ο Use more transmit power 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 20 Optimal Number of Active UEs • Find πΎ that maximizes EE with ZF precoding: πΎ πΎ 1 − π log 2 1 + πtot π½ − 1 maximize πΎ ≥ 0 1 tot η π π΄λ + where πtot = ππΎ and π½ = π πΎ 3 π π=0 πΆπ,0 πΎ + 2 π+1 π=0 πΆπ,1 π½πΎ are fixed. Theorem 3 (Optimal πΎ) Solution is a root to a quartic polynomial: Closed-form but very large expressions • Observations - Decreases with circuit power coefficients multiplied with π or πΎ - Increases with the static hardware power πΆ0,0 - Increases with the propagation parameter π΄λ 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 21 Numerical Illustrations 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 22 Simulation Scenario • Main Characteristics - Circular cell with radius 250 m - Uniform user distribution with 35 m minimum distance - Uncorrelated Rayleigh fading, typical 3GPP pathloss model 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 23 Optimal System Design: ZF Precoding Optimum π = 165 πΎ = 85 π = 4.6 User rates: as 256-QAM Massive MIMO! Very many antennas, π/πΎ ≈ 2 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 24 Optimal System Design: MRT Optimum π=4 πΎ=1 π = 12.7 User rates: as 64-QAM Single-user transmission! Only exploit precoding gain 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 25 Why This Huge Difference? • Interference is the Limiting Factor - ZF: Suppress interference actively - MRT: Only indirect suppression by making π β« πΎ Only 2x difference in EE 100x difference in throughput • More results: RZF≈ZF, same trends under imperfect CSI 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 26 Energy Efficient to Use More Power? • Recall Theorem 2: Transmit power increases with π - Figure shows EE-maximizing power for different π Essentially linear growth • Intuition: More Circuit Power ο Use More Transmit Power 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 27 Conclusions 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 28 Conclusions • What if a Single-Cell System Designed for High EE? - Need: Reasonable throughput model - Need: Reasonable power consumption model • Contributions - General power consumption model - Closed-form results for ZF: Optimal number of antennas Optimal number of active UEs Optimal transmit power - Observations: More circuit power ο Use more transmit power • Numerical Example - ZF/RZF precoding: Massive MIMO system is optimal - MRT precoding: Single-user transmission is optimal - Small difference in EE, but huge difference in throughput! 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 29 Thank You for Listening! Questions? Main reference: E. Björnson, L. Sanguinetti, J. Hoydis, M. Debbah, “Designing Multi-User MIMO for Energy Efficiency: When is Massive MIMO the Answer?” Submitted IEEE WCNC 2014 Preprint available on arXiv: http://arxiv.org/abs/1310.3843 2013-11-07 Greentouch Open Forum, Designing Multi-User MIMO for Energy Efficiency, E. Björnson (Supélec & KTH) 30