PaperlayerView exact revision on arXiv

Source-linked AI summary

Ultra-Dense Cell-Free Massive MIMO for 6G: Technical Overview and Open Questions

Hien Quoc Ngo, Giovanni Interdonato, Erik G. Larsson, Giuseppe Caire, Jeffrey G. Andrews

arXiv:2401.03898v3cs.ITeess.SP

TL;DR

Ultra-dense CF-MMIMO is being considered for 6G connectivity, but practical deployment remains constrained by architecture, coordination, synchronization, access, and channel-acquisition challenges. This article surveys the technology and proposes research directions by examining practical infrastructures, processing methods, capacity bounds, resource allocation, and related open questions.

  • Problem

    Ultra-dense CF-MMIMO must support future ubiquitous connectivity while unresolved practical issues remain in architecture, coordination, synchronization, access, and channel acquisition.

  • Method

    The article provides a comprehensive survey of ultra-dense CF-MMIMO research and examines practical implementations, processing, capacity bounding, massive access, resource allocation, synchronization, calibration, and channel acquisition.

  • Results

    The survey identifies key ingredients and suitable approaches for ultra-dense CF-MMIMO, including O-RAN-based implementation, channel-subspace projection, and alternative bounds when channel hardening is unavailable.

  • Takeaways & Limitations

    The article provides a roadmap for future CF-MMIMO research and development under ultra-dense, realistic deployment constraints.

  • Takeaways & Limitations

    Current O-RAN standards lack a well-defined inter-O-DU coordination framework needed to exchange data and signaling for CF-MMIMO cooperation.

Abstract

from arXiv · show

Ultra-dense cell-free massive multiple-input multiple-output (CF-MMIMO) has emerged as a promising technology expected to meet the future ubiquitous connectivity requirements and ever-growing data traffic demands in 6G. This article provides a contemporary overview of ultra-dense CF-MMIMO networks, and addresses important unresolved questions on their future deployment. We first present a comprehensive survey of state-of-the-art research on CF-MMIMO and ultra-dense networks. Then, we discuss the key challenges of CF-MMIMO under ultra-dense scenarios such as low-complexity architecture and processing, low-complexity/scalable resource allocation, fronthaul limitation, massive access, synchronization, and channel acquisition. Finally, we answer key open questions, considering different design comparisons and discussing suitable methods dealing with the key challenges of ultra-dense CF-MMIMO. The discussion aims to provide a valuable roadmap for interesting future research directions in this area, facilitating the development of CF-MMIMO MIMO for 6G.

I. INTRODUCTION

6G is motivated by ubiquitous connectivity and growing traffic demands, while ultra-dense CF-MMIMO is presented as a promising approach whose practical deployment remains unresolved. The paper surveys the technology and examines implementation challenges and open questions.

  • 6G motivation: 6G is expected around 2030 and targets ubiquitous connectivity alongside enhanced capabilities beyond eMBB, URLLC, and mMTC.These capabilities include AI-related and multi-sensory-interaction use cases.
  • CF-MMIMO concept: Ultra-dense CF-MMIMO uses many distributed APs that coherently serve users, with total AP antennas exceeding jointly scheduled users.It combines ultra-dense deployment with AP cooperation and massive-MIMO principles.
  • Research scope: The paper addresses the gap between established CF-MMIMO theory and challenging practical implementation in increasingly dense networks.Its approach combines a comprehensive survey with discussion of computational-complexity and performance tradeoffs.
  • Scope: The paper focuses on sub-6G bands because they are expected to remain primary 6G traffic carriers and offer more favorable propagation than higher bands.This is stated as the paper’s frequency-scope assumption.
  • Paper organization: The article is organized around CF-MMIMO and ultra-dense-network overviews, challenges, a user-centric mathematical model, open questions, and conclusions.The paper also introduces notation and abbreviations for its technical development.

M GHG

Multiple-antenna systems gain spatial multiplexing from channel structure, but propagation rank and pilot overhead constrain achievable gains. The section traces this progression from point-to-point MIMO through MU-MIMO and massive MIMO.

  • Point-to-point MIMO: At high SNR, point-to-point MIMO achievable rate scales with min(M, K), the multiplexing gain, when the channel is full rank.This scaling does not generally hold in low-SNR or low-rank propagation conditions.
  • Point-to-point MIMO: Channel rank is limited by min(M, K, L), determining the number of independent data streams and rate scaling with log2(SNR).Poorly separated path angles can reduce the number of dominant streams below the matrix rank.
  • Point-to-point MIMO: In a rank-one line-of-sight channel, increasing K improves user SNR but not spatial degrees of freedom or multiplexing gain.Only one spatial degree of freedom is available in this case.
  • Multiuser MIMO: MU-MIMO can achieve full spatial degrees of freedom when users’ angles of arrival are distinct, although high rates require well-separated dominant singular values.Linear ZF or MMSE processing can perform well despite the availability of nonlinear capacity-achieving methods.
  • Multiuser MIMO: MU-MIMO with many base-station antennas faces downlink pilot dimensions at least equal to antenna count, limiting simultaneous dense-user service.This pilot overhead constrained early MU-MIMO proposals to relatively small antenna arrays.

C. Massive MIMO

Massive MIMO uses many base-station antennas for spatial multiplexing and gains, but cellular boundaries and inter-cell interference limit connectivity. Network MIMO and CF-MMIMO increase cooperation, with CF-MMIMO eliminating fixed cells through scalable user-centric AP selection, while introducing processing, fronthaul, and standardization challenges.

  • Massive MIMO: Massive MIMO equips base stations with many antennas, exceeding users’ antennas and enabling spatial multiplexing on the same time-frequency resource.Its operation typically relies on TDD and uplink-downlink channel reciprocity.
  • Massive MIMO: High array and multiplexing gains, favorable propagation, and channel hardening allow simple processing to provide strong energy and spectral efficiency.Examples include linear precoding, combining, and channel-statistics-based resource allocation.
  • Cellular limitations: Cellular massive MIMO retains boundary effects: users near cell edges can experience poor SINR because of inter-cell interference and high path loss from their serving base station.This limitation motivates cooperation across base stations.
  • Network MIMO and cooperation: Network MIMO pools antennas from nearby cooperating base stations, but fixed clusters preserve cells and boundaries, while user-centric clusters require flexible dynamic fronthaul.Network MIMO also entails complicated signal co-processing, substantial fronthaul or backhaul overhead, and deployment costs.
  • Cell-free massive MIMO: CF-MMIMO removes fixed cells by distributing many APs across an area and serving users through cooperation, either canonically through all APs or scalably through user-centric AP subsets.The canonical all-AP configuration is an idealized model that lacks scalability; user-centric clusters change with user locations and loads.
  • Cell-free massive MIMO: CF-MMIMO can provide uniformly high-quality connectivity with simple maximum-ratio processing, outperforming co-located massive MIMO in location uniformity and supporting favorable propagation for multi-stream LoS transmission.Its practical operation assumes directly measured CSI, TDD reciprocity, and repeated training, payload, and downlink activities within coherence intervals.

III. ULTRA-DENSE NETWORKS: STATE OF THE ART AND KEY CHALLENGES

Ultra-dense networks place multiple APs over each location and become interference-limited, creating both opportunities and practical barriers. CF-MMIMO can use AP cooperation and directional beamforming to reduce interference, while cost, permitting, and deployment constraints remain central challenges.

  • Ultra-dense network regime: Ultra-dense operation means each location is covered by multiple APs and, without cooperation, the network is strongly interference-limited rather than noise-limited.Equivalently, as the AP count grows, SINR approaches SIR and noise becomes negligible relative to cumulative interference.
  • Ultra-dense network regime: CF-MMIMO reduces cumulative interference through directional beamforming and AP cooperation, addressing densification collapse in which SINR and sum throughput can saturate and then decline.The passage identifies interference reduction as a key role for CF-MMIMO in dense deployments.
  • Deployment status: Most cellular deployments remain coverage-limited rather than ultra-dense, despite potential gains from cell splitting and improved coverage.The text presents high densification as difficult to achieve in practice.
  • Practical barriers: Cost is the main barrier to greater densification, with small-cell site acquisition and deployment often not substantially cheaper per site than macrocell deployment.Major costs include site rental, permitting and legal paperwork, and backhaul or power provisioning.
  • Practical barriers: Legal and permitting constraints can remain difficult to overcome irrespective of cost, including regulations limiting aggregate electromagnetic radiation.The paper identifies regulatory burdens as a continuing hurdle for dense cellular deployment toward 6G.
  • 6G relevance: Densification is increasingly relevant for 6G because higher density supports higher carrier frequencies, joint communication and sensing, lower radiated power, and edge-cloud aggregation.These potential benefits are presented alongside the deployment barriers rather than as evidence that densification is already widespread.

C. Ultra-Dense Network Meets Cell-Free Massive MIMO: Key Challenges

Ultra-dense CF-MMIMO combines user-centric AP cooperation with massive MIMO, but practical deployment requires scalable architectures and channel acquisition procedures. The system model defines AP–user associations, TDD phases, pilot transmission, and local or CPU-based MMSE estimation.

  • System model: User-centric CF-MMIMO assigns each user a preferred AP set, with corresponding user sets identifying which users each AP serves.The model contains M APs and K single-antenna users; each AP has N antennas.
  • Channel model: The channel between AP m and user k combines large-scale fading β_mk with an N×1 small-scale fading vector h_mk.The small-scale coefficients are normalized, and practical channels may contain both LoS and non-LoS components.
  • TDD operation: Each TDD coherence interval contains UL channel estimation, UL payload transmission, and DL payload transmission, with τ_c = τ_p + τ_u + τ_d.The pilot phase uses τ_p symbols, followed by τ_u UL and τ_d DL payload symbols.
  • Channel acquisition: During UL training, all users transmit pilots to every AP, and each AP receives a noisy pilot matrix for channel estimation.The pilot sequence has unit norm, while the normalized pilot SNR and noise matrix characterize reception.
  • Channel acquisition: Channel estimates may be computed locally at APs or centrally at the CPU, with centralized estimation requiring pilot forwarding over fronthaul links.The AP applies linear MMSE estimation, and the resulting channel estimate is uncorrelated with its estimation error.

B. Uplink Payload Data Transmission

UL payload transmission sends user data to serving APs, which detect symbols using channel-based combining. Combining can be local or centralized, depending on whether APs use local channel estimates or a CPU uses global estimates.

  • Uplink transmission: During UL payload transmission, each user sends a data symbol with an associated transmit-power coefficient to the APs.The received signal at each AP is formed from the users’ transmitted symbols, channels, and noise.
  • Uplink detection: To detect user k, signals from APs serving that user are combined using channel estimates acquired during training.The combining vector is designed according to the selected processing architecture.
  • Processing architectures: Distributed processing computes the combining vector locally at each AP, whereas centralized processing computes it at the CPU from global channel estimates.The distributed case uses local AP-to-user channel estimates; the centralized case uses estimates from all APs to all users.
  • Downlink processing: In the DL, each AP precodes symbols only for its served users using channel estimates from UL training.The precoding vector depends on the processing level, and the power-control coefficient satisfies the AP power constraint.
  • Downlink reception: The received DL signal at each user consists of the intended precoded transmission, interference, and additive noise.The receiver model represents the noise as complex Gaussian with unit variance.

V. KEY OPEN QUESTIONS

Ultra-dense CF-MMIMO faces architectural, coordination, and fronthaul challenges as user-centric clusters change with network conditions. O-RAN provides a promising implementation framework, but inter-O-DU cooperation remains insufficiently specified.

  • Architecture scalability: User-centric CF-MMIMO is difficult to scale because changing user loads and locations require frequent cluster control and substantial signaling.Network-centric clustering still requires large connection bandwidth from APs to CPUs as network size grows.
  • O-RAN alignment: O-RAN splits PHY processing between O-RUs and O-DUs, resembling the division between APs and CPUs in user-centric CF-MMIMO.O-RAN also adds RIC and orchestration functions for broader network control.
  • Synchronization: O-RAN supports clock and time synchronization for multiple radio units, including over-the-air reciprocity calibration validated with commercial O-RUs.This synchronization capability supports the practical mapping of CF-MMIMO functions onto O-RAN components.
  • O-RAN alignment: O-RAN can map CPUs to O-DUs and APs to O-RUs, with Near-RT RIC coordination providing centralized network performance control.The architecture therefore offers a possible practical realization of user-centric CF-MMIMO.
  • Functional split: Category B O-RUs can perform precoding, whereas Category A O-RUs cannot.This distinction affects where downlink processing can be implemented in the O-RAN deployment.
  • Open coordination questions: CF-MMIMO requires inter-O-DU cooperation for user data and combining or precoding signaling, but current O-RAN lacks a well-defined coordination interface.Flexible fronthaul and dynamically migrating cluster processors are proposed as promising scalability measures, while their implementation remains complex.

B. What User-Centric Processing Should Ultra-Dense Cell-

User-centric processing distributes CF-MMIMO computation between APs and CPUs in two principal ways. Local processing reduces CPU complexity and fronthaul demands while retaining cluster-level UL combining and corresponding DL precoding operations.

  • Centralized processing: Centralized processing connects cluster APs to a CPU that designs UL combining and DL precoding, typically using MMSE processing.It exchanges pilots, received data, data symbols, and precoding vectors over fronthaul links, creating substantial CPU complexity in UDNs.
  • Local processing: Local processing computes channel estimates, UL combining, and DL precoding at each AP, sending only locally detected UL scalars to CPUs.For DL transmission, only data symbols need to be sent from CPUs to APs.
  • Local processing: Local processing is considered suitable for ultra-dense CF-MMIMO because it reduces CPU computation, tolerates AP additions or failures, and performs well with multi-antenna APs.These benefits motivate focusing subsequent processing analysis on the local approach.
  • Local uplink detection: In local UL processing, each serving AP forms a detected version of user k’s symbol using a local detection vector computed from its channel estimates.Maximum-ratio combining uses the estimated channel, while local ZF, local MMSE, and TMMSE offer higher performance with greater complexity.
  • Cluster-level combining: Each AP sends its detected symbol to the user’s cluster processor, which combines the |A_k| signals using coefficients that manage interference and noise.Poor interference-and-noise-dominated local detections receive smaller combining coefficients.
  • Processing implementation: The UL scheme is called local detection with cluster-level combining, and the same processing structure can be reused for DL precoding without extra computation.The APs can send weighted detected signals for CPU combination or over-the-air aggregation, depending on implementation.

2) Downlink Payload Data Transmission:

Downlink precoding can reuse uplink processing through approximate UL-DL duality, avoiding extra computation when both directions share the same scheme. Local processing therefore balances complexity, resilience, and performance, but requires flexible fronthaul routing.

  • 2) Downlink Payload Data Transmission:: Approximate UL-DL duality lets APs derive downlink precoding vectors and power-control coefficients from uplink combining vectors and cluster-level coefficients.The same processing structure can provide the downlink quantities directly.
  • 2) Downlink Payload Data Transmission:: Using the same uplink and downlink processing scheme creates no extra computation for downlink precoding.The downlink precoding vectors and power-control coefficients are obtained from the uplink processing quantities.
  • 2) Downlink Payload Data Transmission:: Local processing is presented as an ideal choice for ultra-dense CF-MMIMO because it balances computational complexity, system resilience, and system performance.Its remaining implementation challenge is flexible fronthaul capable of dynamically routing traffic among changing clusters.

C. How to Dynamically Route Fronthaul Traffic and Quantize Signals?

Ultra-dense CF-MMIMO requires joint fronthaul routing, cluster-processor placement, computation allocation, and signal quantization. The design treats uplink and downlink traffic differently: strong uplink observations are selectively quantized, while downlink information bits are preferred over modulated samples.

  • C. How to Dynamically Route Fronthaul Traffic and Quantize Signals?: Scalable operation requires flexible fronthaul connecting APs, CPUs, and routing-capable nodes because user-centric clusters depend on user locations.Each CPU hosts a bounded number of software-defined cluster processors implementing user-specific PHY functions.
  • C. How to Dynamically Route Fronthaul Traffic and Quantize Signals?: Fronthaul design jointly optimizes UL/DL routing, cluster-processor placement, computation capacity, and bits per signal dimension.Remote processor placement can increase multihop traffic, while nearby placement may conflict with CPU capacity constraints.
  • C. How to Dynamically Route Fronthaul Traffic and Quantize Signals?: A mixed-integer linear program efficiently addresses joint routing and computation-resource optimization for fairly large networks.The uplink routing component is a multiple-unicast problem carrying quantized projected received signals.
  • C. How to Dynamically Route Fronthaul Traffic and Quantize Signals?: Uplink quantization adapts the number of bits to observation strength, discarding observations when their variance is at most the distortion target D.Only strong observations are quantized and forwarded for cluster-level combining.
  • C. How to Dynamically Route Fronthaul Traffic and Quantize Signals?: Downlink fronthaul should forward user information bits rather than precoded complex-valued streams, making the O-RAN 7.2 split inefficient for this role.Encoding, modulation, and precoding can instead be performed at the APs; downlink routing is a multiple-multicast problem.

D. When Does the Hardening Bounding Technique Work?

The hardening bound is reliable when propagation provides channel hardening, but double-scattering channels can create a large gap from full-side-information rates. Training overhead, coherence conditions, and traffic dynamics determine which alternative is appropriate.

  • D. How Much Prior Information is Really Available and How: In CF-MMIMO, subset-based service and unequal AP large-scale fading can weaken channel hardening and cause hardening-based rates to underestimate practical rates.The gap is usually small in centralized setups but can become substantial with MR processing or single-antenna APs.
  • D. How Much Prior Information is Really Available and How: Under double-scattering channels, full-side-information rates are much higher than hardening rates, especially in dense networks.The example uses double-scattering fading, large-scale-fading AP selection, random pilots, and centralized or local MMSE precoding.
  • D. How Much Prior Information is Really Available and How: Accounting for downlink channel-estimation overhead can make the hardening technique outperform full-side-information throughput in dense networks.Full-side-information operation requires effective-channel knowledge and an additional K-symbol downlink training duration.
  • D. How Much Prior Information is Really Available and How: The hardening bound works fairly well under hardening propagation environments, even with small clusters or many users.Its analytical simplicity makes it widely used, but it is not an approximation of ergodic capacity.
  • D. How Much Prior Information is Really Available and How: When channels do not harden, alternatives include beamforming training, non-coherent bounds for τd ≫ K, blind estimation for τd ≫ 1 and K ≫ 1, or enhanced normalization.The non-coherent bound avoids explicit instantaneous-gain estimation but requires a coherence interval much larger than the user count.
  • D. How Much Prior Information is Really Available and How: The hardening bound may be unsuitable for bursty traffic because codewords must span many fading realizations while active users change across transmission intervals and frequency bands.This limitation arises from the ergodic-rate interpretation of the bound.

E. How Much Prior Information is Really Available and How

CF-MMIMO channel acquisition depends on uplink pilots, reciprocity, and channel statistics. Spatial subspace projection can substantially improve estimation when co-pilot users occupy well-separated directional subspaces, but covariance information must also be estimated.

  • E. How Much Prior Information is Really Available and How: Each AP estimates associated-user channels from uplink pilots and uplink/downlink reciprocity, although limited pilot dimensions create pilot contamination.The channel estimate is obtained by projecting received pilot symbols onto the corresponding pilot sequence.
  • E. How Much Prior Information is Really Available and How: Channel-subspace projection significantly improves uplink channel estimation by projecting contaminated estimates onto dominant covariance eigenvector subspaces.When co-pilot users are well separated in highly directional channel subspaces, the method nearly eliminates pilot contamination.
  • E. How Much Prior Information is Really Available and How: The method is most effective when propagation uses a few angular clusters and users have nearly orthogonal channel subspaces in the angular domain.These conditions separate co-pilot users through spatial correlation and directional preference.
  • E. How Much Prior Information is Really Available and How: Channel covariance matrices or dominant subspaces must be estimated for every AP-user pair, but channel geometry varies on a much longer timescale than small-scale fading.This supports treating the channel processes as locally wide-sense stationary for second-order statistics.

F. Aspects of Synchronization and Calibration

Distributed CF-MMIMO requires time, frequency, and phase alignment, while practical coordination and control must balance performance against synchronization, signaling, computation, and scalability constraints.

  • Synchronization requirements: Distributed APs require time alignment, a common frequency reference, and joint reciprocity calibration for coherent transmission and reception.Time alignment need not exceed the OFDM cyclic-prefix duration, whereas joint reciprocity-based downlink beamforming also requires phase alignment.
  • Synchronization architectures: Phase synchronization can use distributed RF signals, baseband data with a low-frequency reference, or other interconnection architectures.Time synchronization is comparatively easy, while phase alignment is more difficult for distributed APs.
  • Calibration trade-offs: Over-the-air reciprocity calibration removes costly synchronization cables but adds signaling overhead and disrupts the TDD flow.Bidirectional AP measurements can perform calibration, with beamforming helping when APs have multiple antennas.
  • Power-control coordination: Centralized power control can maximize network-wide utility but is generally neither scalable nor low-complexity because it requires CSI sharing over the fronthaul.Distributed control avoids some coordination requirements but presents different optimization trade-offs.
  • Power-control objectives: Full uplink transmit power is quasioptimal for sum spectral efficiency with MMSE combining, whereas fairness motivates fractional power control.In the downlink, centralized fractional power control trades off sum spectral efficiency and fairness, while distributed optimization can improve both objectives.
  • Scalable implementation: Scalable power control should be heuristic and fully distributed, or sub-optimal and centralized with only partial CSI sharing.Learning-based methods may reduce online computation, but their practical scalability is constrained by training-data generation, fronthaul, and changing propagation conditions.
  • Massive access: Massive access is difficult because sporadic, tiny uplink packets make conventional signaling inefficient and require new random-access and control-plane techniques.Compressed sensing and non-orthogonal pilots are promising, but channel acquisition consumes additional resources and channel models may be inaccurate.

I. How to Implement Dynamic Scheduling?

Dynamic scheduling in user-centric CF-MMIMO remains underexplored when many users compete for a smaller set of simultaneously active transmissions. Candidate architectures combine fairness-aware scheduling with distributed or sequential processing to preserve scalability and fronthaul efficiency.

  • I. How to Implement Dynamic Scheduling?: A useful scheduling rule selects active users to trade off aggregate spectral efficiency, instantaneous per-user rate, and fair long-term throughput.A cited rule of thumb sets active users at roughly half the total system antennas.
  • I. How to Implement Dynamic Scheduling?: Dynamic scheduling is not widely investigated when total users greatly exceed the number active on each time-frequency resource block.The scheduler must balance total spectral efficiency, scheduled-user rates, and long-term fairness.
  • I. How to Implement Dynamic Scheduling?: The network-utility formulation uses a concave α-fairness utility and Lyapunov drift-plus-penalty priority adaptation.Each scheduling interval solves a constrained weighted sum-rate maximization problem.
  • I. How to Implement Dynamic Scheduling?: CF-MMIMO deployment must jointly address costly widespread infrastructure, accurate AP synchronization, coherent coordination, and scalable resource allocation.These requirements make practical deployment more demanding than the idealized cooperative model.
  • I. How to Implement Dynamic Scheduling?: Radio Stripes integrate antenna elements, baseband processing, and fronthaul into serially connected APUs sharing power, synchronization, and data transfer.Sequential processing moves signal operations beside the antennas and forwards cumulative results along the stripe.
  • I. How to Implement Dynamic Scheduling?: Sequential serial-fronthaul processing achieves optimal uplink and nearly optimal downlink spectral efficiency while reducing fronthaul requirements relative to centralized implementations.The fronthaul advantage is substantial when CF-MMIMO has more APs than active users.
  • I. How to Implement Dynamic Scheduling?: Radio Stripes offer flexible, low-power, resilient deployments, but PoE-based implementations with up to 10 Gbps are suited to sub-6 GHz frequencies.Routing can tolerate node failures, while distributed components ease heat dissipation and non-invasive installation.

K. Other Design Aspects of Ultra-Dense CF-MMIMO

Ultra-dense CF-MMIMO design must manage distributed interference, CSI-sharing overhead, hardware impairments, and deployment complexity. The paper surveys practical architectures and identifies remaining questions about scalable processing and realistic hardware operation.

  • 1) Low-complexity processing:: Partial zero-forcing and local partial MMSE reduce complexity by suppressing only the most significant interference contributions without additional fronthaul overhead.Their distributed design trades interference cancellation against desired-signal enhancement and does not suppress inter-AP interference.
  • 1) Low-complexity processing:: CSI sharing can dominate fronthaul overhead because distributed APs need tightly timed channel information even when data sharing is confined to AP subsets.Cooperation among only a few APs or transmitter-specific CSI can limit this burden.
  • Other design aspects: Inexpensive hardware reduces deployment cost but can introduce phase noise, carrier-frequency offsets, inter-carrier interference, and nonlinear distortion.Nonlinearities are especially important because their distortion is generally signal-dependent and can be correlated across antennas.
  • Other design aspects: Downlink hardware impairments are harder to mitigate than uplink impairments, and strict out-of-band emission requirements can remain challenging.Downlink precoders can partly adapt to nonlinearities but do not remove the practical constraint.
  • Other design aspects: Approximating hardware distortion as uncorrelated across antennas is valid only under special rapidly varying fading and simultaneous multi-user beamforming conditions.Static line-of-sight channels are an important counterexample to that approximation.
  • Other design aspects: CF-MMIMO combines broad connectivity goals with practical requirements for widespread infrastructure, synchronization, coordination, and resource allocation.The survey frames these requirements as central design challenges for beyond-5G and 6G systems.
  • Other design aspects: The survey identifies open questions involving densification level, mobility, backscatter communication, multi-antenna users, reconfigurable surfaces, near-field transmission, and AI.These questions extend beyond the addressed implementation challenges and motivate further research.
Loading 2401.03898v3…