PaperlayerView exact revision on arXiv

Source-linked AI summary

Prospective Multiple Antenna Technologies for Beyond 5G

Jiayi Zhang, Emil Björnson, Michail Matthaiou, Derrick Wing Kwan Ng, Hong Yang, David J. Love

arXiv:1910.00092v3cs.ITeess.SP

TL;DR

The paper addresses difficult beyond-5G requirements where additional bandwidth may not translate directly into higher bit rates. It surveys three prospective multiple-antenna directions, presents recent technical progress, and discusses open problems and deployment visions.

  • Problem

    Beyond-5G requirements are difficult to meet, and using 10 times more mmWave bandwidth may not increase the bit rate.

  • Method

    The paper surveys cell-free massive MIMO, beamspace MIMO, and intelligent reflecting surfaces through system models, performance analyses, signal processing schemes, and deployment visions.

  • Results

    The survey presents recent technical advances for three beyond-5G research directions associated with improving capacity, coverage, and QoS over legacy cellular networks.

  • Takeaways & Limitations

    The paper identifies crucial open problems in each surveyed area and outlines prospective research directions for beyond-5G networks.

Abstract

from arXiv · show

Multiple antenna technologies have attracted large research interest for several decades and have gradually made their way into mainstream communication systems. Two main benefits are adaptive beamforming gains and spatial multiplexing, leading to high data rates per user and per cell, especially when large antenna arrays are used. Now that multiple antenna technology has become a key component of the fifth-generation (5G) networks, it is time for the research community to look for new multiple antenna applications to meet the immensely higher data rate, reliability, and traffic demands in the beyond 5G era. We need radically new approaches to achieve orders-of-magnitude improvements in these metrics and this will be connected to large technical challenges, many of which are yet to be identified. In this survey paper, we present a survey of three new multiple antenna related research directions that might play a key role in beyond 5G networks: Cell-free massive multiple-input multiple-output (MIMO), beamspace massive MIMO, and intelligent reflecting surfaces. More specifically, the fundamental motivation and key characteristics of these new technologies are introduced. Recent technical progress is also presented. Finally, we provide a list of other prospective future research directions.

I. INTRODUCTION

Beyond-5G networks face demands that current cellular approaches, including mmWave-based solutions, may not fully address. The paper surveys cell-free massive MIMO, beamspace massive MIMO, and intelligent reflecting surfaces as prospective multiple-antenna directions.

  • Beyond-5G requirements include exceptionally high bit rates, uniform user performance, ultra-low latency, energy efficiency, and robustness against blocking and jamming.
  • mmWave communications remain vulnerable to signal blockage, and their shorter wavelengths reduce coherence time.The paper states that blockage sensitivity has not been resolved despite substantial research effort.
  • The paper surveys cell-free massive MIMO, beamspace massive MIMO, and intelligent reflecting surfaces, including theory, technical progress, and open research problems.It also discusses applications involving UAV-supported communication and sub-THz bands.
  • Cellular networks provide stronger improvements for cell-center users, while inter-cell interference and handover issues limit cell-edge performance.Cell-edge traffic congestion and mediocre 95%-likely user data rates remain concerns in 5G networks.
  • Cell-free massive MIMO addresses the cellular paradigm by allowing users to be served by suitable access points without cell boundaries.The user-centric approach lets each access point collaborate with different access-point sets for different user equipments.

B. Basics of Cell-Free Massive MIMO

Cell-free massive MIMO uses many distributed APs to jointly serve users, with coherent processing enabled by local CSI and fronthaul-connected CPUs. Analyses and benchmarks report strong gains in user fairness, spectral efficiency, and energy efficiency across varied deployment assumptions.

  • Network architecture: A cell-free massive MIMO network uses many APs that jointly serve UEs, with CPUs connected through fronthaul for centralized encoding and decoding.APs can use local channel estimates, while CPUs are normally assumed to know long-term channel qualities.
  • Performance evidence: Cell-free massive MIMO works across Rician and Rayleigh fading, multi-antenna APs and UEs, correlated fading, and finite-resolution fronthaul scenarios.The surveyed studies conclude that the approach is suitable for a variety of deployment scenarios.
  • Performance evidence: Nearly fivefold improvement in 95%-likely per-user spectral efficiency is reported over small-cell operation with the same APs.Multiple antennas at both APs and UEs can enhance the 95%-likely per-user performance further.
  • Performance evidence: Cell-free operation emphasizes more uniform coverage-area performance rather than higher peak performance, benefiting weak UEs through geographically distributed antennas.Cellular massive MIMO remains preferred for cell-center UEs, while distributed antennas provide macro-diversity.
  • Performance evidence: Nearly ten times higher bit/Joule energy efficiency than cellular massive MIMO has been reported.The surveyed motivation also highlights higher 95%-likely spectral efficiency and improved service for weak users.
  • System model: Users can be served by partially overlapping, user-centric subsets of APs, while TDD uplink pilots provide channel estimates for coherent transmission and reception.Each user k is served by a subset M_k, and APs estimate channels to all UEs from uplink pilots.

2) Downlink Data Transmission:

Downlink transmission distributes each UE’s signal across serving APs using locally selected precoding vectors and AP-specific power allocation. Leakage-aware precoding can improve spectral efficiency over maximum-ratio precoding, but downlink power selection remains difficult.

  • Downlink transmission: Each serving AP maps a UE’s signal to its antennas using a precoding vector selected from locally available CSI.The vector combines a spatial-directivity component with an AP-assigned transmit-power factor.
  • Precoding methods: Maximum-ratio precoding directs transmission toward the intended receiver without accounting for other UEs’ interference.Its expectations can often be computed in closed form for common channel models.
  • Precoding methods: SLNR precoding improves spectral efficiency by balancing intended-signal power against interference leakage to non-intended receivers.It outperforms MR even with single-antenna APs, with almost the same computational complexity.
  • Power allocation: Downlink power allocation is harder than uplink control because each AP must distribute power among served UEs under decentralized channel knowledge.Global optimization becomes infeasible in large networks as its complexity grows with K.
  • Power allocation: Further work is required to handle downlink objectives beyond maximizing the worst UEs’ performance.The surveyed schemes include several power-allocation approaches, but their practical performance can be difficult to evaluate.

D. Scalable Large-Scale Deployment

Scalable cell-free deployment requires distributed association, signal processing, and power control whose complexity and fronthaul demands remain manageable as networks grow. Fixed user-serving limits can preserve spectral efficiency, but several implementation and evaluation challenges remain.

  • Scalability: Global optimization of AP association, signal processing, and power control is fundamentally unscalable as network size increases.The paper identifies scalable network architecture as the main cell-free massive MIMO design challenge.
  • Scalability: A scalability definition requires each AP’s computational complexity and fronthaul capacity requirement to remain independent of K as K →∞.This can be approached by restricting each AP to serve a fixed number of UEs.
  • Scalability: A fixed number of served UEs per AP has been shown to have negligible impact on spectral efficiency.The restriction is motivated by users being distributed near different APs.
  • Open problems: No quantitative comparison of different network infrastructures is available, and effective distributed selection of combining weights remains poorly understood.These gaps accompany unresolved questions about fronthaul, multiple CPUs, and real-world channel scaling.
  • Scalability: Partially centralized precoding and combining can suppress interference between APs and substantially increase spectral efficiency, but scalable implementations remain early-stage.The paper also notes that real-time second-order-cone power-control methods are not fast enough.
  • Power control: Downlink power control must coordinate AP decisions under per-AP constraints, while finite discrete power levels and fairness–sum-efficiency trade-offs remain open issues.Max-min fairness may sacrifice too much sum spectral efficiency to provide absolute fairness.
  • Power control: Full-power uplink transmission may work well in many practical scenarios, contrasting with cellular uplink operation.The evidence marks a departure from cellular thinking about uplink power control.

2) Fronthaul/Backhaul Provisioning:

Cell-free networks impose heavier fronthaul/backhaul demands because many geographically distributed APs cooperate over a broad coverage area. Proposed responses include decentralized processing, quantization, semi-distributed methods, and new infrastructure architectures, while large arrays motivate beamspace processing.

  • Fronthaul/backhaul provisioning: Cell-free networks create a much heavier fronthaul/backhaul burden than traditional cellular systems because many APs are distributed across the coverage area.Optical-fiber provision can be cost-prohibitive except in premium venues or when serial connections are possible.
  • Fronthaul/backhaul provisioning: Wireless fronthaul is viable but must address spectrum availability and the difficulty of reliably delivering ultra-high data rates.A proposed dual-layer architecture uses cellular massive MIMO to haul a cell-free massive MIMO system.
  • Fronthaul/backhaul provisioning: Decentralized processing, heavily quantized fronthaul signals, and distributed quantization are proposed to reduce fronthaul/backhaul requirements.The survey identifies opportunities to exploit fronthaul architecture for improved utilization and semi-distributed methods.
  • Beamspace massive MIMO: Large antenna arrays and higher carrier frequencies increase implementation complexity, motivating methods that exploit channel and hardware spatial structure.Beamspace massive MIMO is presented as the general concept underlying hybrid beamforming and future successors.
  • Beamspace massive MIMO: As arrays grow toward roughly one hundred antennas, beamspace formulations offer a way to rethink linear-precoding processing and implementation.The survey argues that beamspace terminology, notation, and thinking will be critical for 5G and beyond systems.
  • Beamspace massive MIMO: At mmWave frequencies and higher, beamspace processing is described as indispensable for large arrays and non-traditional hardware implementations.Subspace processing uses virtual or effective channels because sounding every array element becomes impractical or impossible.

B. Signal Model

The beamspace signal model decomposes precoding into two stages and can include linear receive processing. This creates a virtual channel whose sparsity may simplify signal processing and support practical implementations.

  • The standard single-user MIMO model represents the received signal using channel matrix H, transmitted signal s, and additive noise n.
  • Beamspace precoding factors the precoder as W = W1W2, separating a first precoder from a lower-dimensional second precoder.
  • The first precoder and receiver are selected first, after which W2 is chosen using the virtual channel Hv = Z^HHW1.
  • When receive processing is not explicit, the model can assume Z = I_Nr, and the resulting input-output model is illustrated in Fig. 4.
  • If Hv is sparse, beamspace processing can simplify W2; in the diagonal extreme, W2 performs virtual subchannel selection.
  • DFT-based precoding and linear reception uniformly sample AoD and AoA spaces, while RF phase-shifter implementations offer practical hardware benefits.

E. Beamspace Using Lens Arrays

Lens arrays focus electromagnetic energy from different directions onto different ports, producing sparse beamspace channels. Selecting dominant beams reduces effective channel dimension and can lower RF-chain, hardware, and power requirements, while channel estimation remains challenging.

  • Lens arrays can provide nearly orthogonal beams without lossy phase shifters and offer substantial hardware and power savings.
  • Lens arrays focus electromagnetic power from different directions onto different ports, transforming the spatial MIMO channel into a sparse beamspace representation.
  • Selecting a small number of dominant beams that carry most electromagnetic energy reduces the effective MIMO channel dimension and associated RF chains.
  • Narrowband channel estimation: Beamspace channel estimation exploits sparsity through dominant-beam selection, but scanning all beams can require pilot symbols proportional to antenna number.
  • Wideband channel estimation: Wideband signaling causes beam squint because angles of arrival and departure become frequency-dependent across massive arrays.
  • Open challenges: Open problems include frequency-dependent channel reconstruction, 3D lens estimation, and channel estimation with coarse ADC quantization.

2) Hardware imperfections:

Lens arrays suffer hardware imperfections including spillover, imperfect switching, and RF-component distortions. These effects reduce focusing accuracy and can undermine theoretically predicted performance, motivating joint circuit and signal-processing design.

  • Lens-array focusing is imperfect because finite antenna sampling leaks desired beam power into neighboring ports.
  • Numerical simulations found that only 0.55 Watt leaves all array ports when 1 Watt enters a Rotman lens beam port.
  • At off-broadside incidence, including φ = 50°, Rotman-lens focusing exhibits stronger spillover and reflections toward opposite ports.
  • Open challenges: Hardware-imperfection characterization requires cooperation between communications and microwave engineers because the communities have often worked in isolation.
  • Practical switches reflect energy back to lens ports and leak energy into neighboring switches because of poor isolation.
  • Non-ideal mmWave RF components create in-band and out-of-band distortions whose aggregate impact can seriously undermine predicted performance.

3) Physical implementation:

Physical implementations span lens-array prototypes and intelligent reflecting surfaces. Lens arrays can deliver high-frequency multibeam operation with reported performance gains, while IRSs manipulate propagation using passive, controllable reflecting elements with low operational power.

  • A 71–76 GHz beam-steerable lens prototype with 64 feed elements delivered 700 Mbit/s over 55 m.
  • A 28 GHz constant-dielectric lens array systematically outperformed a ULA and Rotman lens because of sharper electromagnetic focusing.
  • Intelligent reflecting surfaces: IRSs use passive MIMO antennas to address the cost and energy concerns associated with large numbers of active RF chains.
  • Intelligent reflecting surfaces: An IRS is a reconfigurable metasurface whose sub-wavelength meta-atoms assign controllable phase shifts, jointly steering reflected beams.
  • Intelligent reflecting surfaces: For 8 × 8 mm meta-atoms, reported IRS energy consumption was 125 mW/m2.
  • Intelligent reflecting surfaces: IRSs can be thin and conformable, enabling flexible deployment to mitigate coverage holes or add capacity without replacing conventional massive MIMO.

B. Signal Model of IRS

The IRS-assisted MIMO model considers a multi-antenna transmitter serving a single-antenna user through direct and IRS-reflected paths, with controllable element phases. Joint beamforming and phase optimization can substantially improve SNR, especially when the user is near the IRS, but the resulting problem is non-convex and computationally difficult.

  • System setup: An IRS with N meta-atoms assists communication from an M-antenna transmitter to a single-antenna user by dynamically adjusting reflected-signal phases.The IRS controller controls each meta-atom, while the system includes direct, transmitter-to-IRS, and IRS-to-user channels.
  • Signal model: The received signal combines the direct transmitter-to-user channel with an IRS-reflected path and receiver noise.The channel model defines beamforming, channel matrices, phase shifts, and independent Gaussian noise.
  • Optimization: The achievable rate depends on the composite channel known at the receiver, while transmission design jointly optimizes the beamforming vector and IRS phase-shift matrix.The phase constraints are 0 ≤ θ_n ≤ 2π for every reflection element.
  • Optimization: For M ≥2, coupling between beamforming and constant-modulus IRS phases generally creates a non-convex optimization problem without a systematic globally optimal solution.Brute-force global optimization can have prohibitively high complexity even for moderate-sized systems, motivating suboptimal methods such as SDR and alternating optimization.
  • Numerical illustration: The suboptimal scheme closely approaches the benchmark upper bound and substantially outperforms the no-IRS baseline in simulated SNR.The largest SNR gains occur when the user is close to the IRS because the reflected path is otherwise too weak.

C. Open Research Problems

IRS-assisted MIMO systems introduce open problems in channel estimation, control, hardware modeling, deployment, and multiplexing. The main practical challenge is achieving the promised performance gains with affordable estimation, signaling, computation, and energy requirements.

  • Channel estimation: Estimating the transmitter-to-IRS and IRS-to-user channels is difficult because IRS elements are passive and cannot initiate transmissions.Control signaling and limited controller computation can further delay estimation of G and h_r.
  • Channel estimation: Large IRS apertures can require high-dimensional channel estimation, creating substantial signal-processing and energy burdens.The paper identifies low-cost estimation using parameterizable channel models or sparsity as an open research direction.
  • Channel estimation: Practical IRS channel estimation remains unresolved because existing approaches do not account for their energy costs and signaling overhead.The desired algorithms should jointly reduce energy consumption, signaling overhead, and computational complexity.
  • Control: IRS control requires CSI exchange, real-time beamforming, and synchronization, but practical control protocols and link choices remain unavailable or unclear.The controller must adjust reflection amplitude and phase while communicating with the transmitter.
  • Hardware: Low-cost IRS hardware may have finite phase resolution, mutual coupling, phase noise, and other impairments that require thorough performance analysis.The paper notes that phase shifting may provide only two states.
  • Deployment and multiplexing: A strong transmitter-to-IRS LoS path can produce a low-rank MIMO channel with limited spatial multiplexing, motivating multiple IRSs as controllable scatterers.Optimal IRS positioning and joint control and channel estimation for multiple IRSs remain unsolved.

5) Identifying a “Killer Application”:

The paper argues that IRSs still lack a demonstrated application or metric in which they create a decisive performance shift over existing beamforming and relaying approaches. It therefore situates IRSs within broader beyond-5G requirements and complementary technologies.

  • 5) Identifying a “Killer Application”:: IRS research needs a demonstrated “killer application” or metric where IRS-aided transmission produces a paradigm shift in performance.The paper asks what key benefit IRSs provide beyond conventional beamforming and relaying use cases.
  • Beyond-5G requirements: Future networks may need reconfiguration to support stringent applications including AR, VR, holography, and ultra-reliable coverage.These applications are described as having requirements that a single fixed network configuration may not meet.
  • MIMO-UAV communication: UAVs can provide rapidly deployable aerial base stations or relays when terrestrial infrastructure is unavailable because of disasters, outages, maintenance, or difficult terrain.Their high maneuverability supports temporary hotspots and blockage-compensating links.
  • MIMO-UAV communication: MIMO-UAV systems face size, weight, energy, interference, low-rank LoS-channel, and trajectory-optimization challenges.These constraints limit onboard massive MIMO, spatial multiplexing, and globally optimal joint trajectory and resource-allocation design.
  • MIMO-UAV communication: Multiple cooperative UAVs could form distributed arrays that improve interference mitigation and information transmission across user clusters.Realizing this gain requires user-clustering and UAV-cooperation algorithms, potentially using cell-free massive MIMO methodology.

1) 3D Beamforming:

Beyond-5G systems require adaptive three-dimensional beamforming for mobile UAV deployments and new MIMO designs for increasingly directional, bandwidth-rich sub-THz and mmWave links. Physical constraints, limited coverage, and difficult propagation make array and beamforming design central challenges.

  • 1) 3D Beamforming:: Fixed beam patterns suit stationary two-dimensional user distributions, whereas mobile UAVs serve three-dimensional users and therefore require 3D beamforming.A two-dimensional array mounted beneath a UAV can adapt antenna magnitudes and phases toward different locations.
  • 1) 3D Beamforming:: Limited UAV size, weight, and energy restrict onboard antennas and beamwidth, causing lower SNR and limited coverage.A small field of view can force a UAV to move repeatedly or fly higher, increasing delay or energy consumption.
  • 1) 3D Beamforming:: Multiple UAVs can divide a service area into clusters and form cooperative distributed MIMO arrays through high-rate air-to-air links.Cell-free massive MIMO methods may support a lean and scalable architecture, but cooperation algorithms remain necessary.
  • 1) 3D Beamforming:: Compact MIMO arrays are necessary at mmWave frequencies to preserve antenna aperture as individual antenna size shrinks, but their implementation remains challenging.The main mmWave drawbacks include more complicated hardware design and unfavorable propagation conditions.
  • 1) 3D Beamforming:: Sub-THz bands offer at least 50 GHz from 90–200 GHz and 100 GHz from 220–320 GHz, but propagation is more directional and range is more limited than in 5G.With 100 GHz bandwidth, data rates are projected to reach 1 Tbit/s at cell center and 100 Gbit/s at cell edge.

1) Hybrid Beamforming:

The survey presents hybrid beamforming and related beyond-5G antenna architectures as ways to balance hardware complexity, beamforming flexibility, coverage, and propagation constraints. It highlights cell-free deployments, beamspace processing, intelligent reflecting surfaces, and UAVs as prospective components of heterogeneous networks.

  • Hybrid Beamforming: Hybrid beamforming compromises between hardware complexity and beamforming flexibility, simplifying initial transceiver design.It combines digital and analog beamforming, potentially using phase shifters.
  • Hybrid Beamforming: Smaller numbers of RF chains can preserve maximum beamforming gain, but make beamforming design more cumbersome.This architecture is motivated by the difficulty of placing a dedicated RF chain behind every antenna element.
  • Hybrid Beamforming: Sub-THz systems face beam-squinting, uncertain propagation models, hardware impairments, and increasingly influential phase noise and nonlinearities.These challenges complicate channel estimation and signal processing over wide bandwidths.
  • Hybrid Beamforming: The beamspace approach is presented as a suitable methodology for designing signal processing in sub-THz bands.The paper positions it as a response to channel and hardware constraints that must first be characterized.
  • Hybrid Beamforming: Cell-free massive MIMO and many intelligent reflecting surfaces can improve propagation conditions and support spatial multiplexing in beyond-5G deployments.Distributed deployments are also motivated by low-rank channels that limit spatial diversity and multiplexing.
  • Hybrid Beamforming: The survey combines conventional massive MIMO with cell-free deployments, IRS support, beamspace processing, sub-THz bands, and UAVs in prospective heterogeneous networks.It also discusses open research problems and the need for practical, scalable, energy-efficient technologies.
Loading 1910.00092v3…