PaperlayerView exact revision on arXiv

Source-linked AI summary

Fifty Years of MIMO Detection: The Road to Large-Scale MIMOs

Shaoshi Yang, Lajos Hanzo

arXiv:1507.05138v1cs.IT

TL;DR

Large-scale MIMO introduces very large antenna arrays and new detection challenges, especially because optimum MIMO detection becomes computationally difficult at large problem sizes. The paper surveys MIMO detection fundamentals and five decades of methods, organizes LS-MIMO detection strategies by system type, and reviews their applicability to LS-MIMO. Its conclusion is a set of historical insights and detection strategies for different LS-MIMO settings, including recent advances.

  • Problem

    LS-MIMO creates large-dimensional detection problems, while optimum MIMO detection has complexity that increases exponentially with problem size.

  • Method

    The paper presents a unified review of MIMO detection fundamentals and representative methods developed during 1965–2015, then examines complexity-scalable algorithms and strategies for two LS-MIMO types.

  • Results

    The survey extracts insights from fifty years of MIMO detection research and discusses distinct detection strategies, algorithm applicability, and recent advances for LS-MIMO systems.

  • Takeaways & Limitations

    Existing MIMO detection research provides relevant lessons for designing complexity-scalable algorithms potentially applicable to LS-MIMO systems.

Abstract

from arXiv · show

The emerging massive/large-scale MIMO (LS-MIMO) systems relying on very large antenna arrays have become a hot topic of wireless communications. Compared to the LTE based 4G mobile communication system that allows for up to 8 antenna elements at the base station (BS), the LS-MIMO system entails an unprecedented number of antennas, say 100 or more, at the BS. The huge leap in the number of BS antennas opens the door to a new research field in communication theory, propagation and electronics, where random matrix theory begins to play a dominant role. In this paper, we provide a recital on the historic heritages and novel challenges facing LS-MIMOs from a detection perspective. Firstly, we highlight the fundamentals of MIMO detection, including the nature of co-channel interference, the generality of the MIMO detection problem, the received signal models of both linear memoryless MIMO channels and dispersive MIMO channels exhibiting memory, as well as the complex-valued versus real-valued MIMO system models. Then, an extensive review of the representative MIMO detection methods conceived during the past 50 years (1965-2015) is presented, and relevant insights as well as lessons are inferred for designing complexity-scalable MIMO detection algorithms that are potentially applicable to LS-MIMO systems. Furthermore, we divide the LS-MIMO systems into two types, and elaborate on the distinct detection strategies suitable for each of them. The type-I LS-MIMO corresponds to the case where the number of active users is much smaller than the number of BS antennas, which is currently the mainstream definition of LS-MIMO. The type-II LS-MIMO corresponds to the case where the number of active users is comparable to the number of BS antennas. Finally, we discuss the applicability of existing MIMO detection algorithms in LS-MIMO systems, and review some of the recent advances in LS-MIMO detection.

GLOSSARY

The paper motivates LS-MIMO through rapidly growing mobile traffic and spectrum constraints, then frames MIMO detection as joint recovery under interference. It reviews detection methods from the past fifty years to identify complexity-scalable approaches for LS-MIMO.

  • A. Why are Large-Scale MIMOs Important?: Global mobile data traffic was forecast to increase nearly 11-fold from 2013 to 2018, reaching 15.9 EB per month by 2018.The forecast corresponds to a 61% compound annual growth rate over 2013–2018.
  • A. Why are Large-Scale MIMOs Important?: LS-MIMO, millimetre-wave communications, and small-cell HetNets are presented as closely related technologies for 5G wireless communications.LS-MIMO provides high dimensionality in the spatial domain, whereas millimetre-wave systems provide it in the frequency domain.
  • B. Why is MIMO Detection Important and Challenging?: MIMO detection jointly recovers multiple transmitted symbols despite random noise and co-channel interference.Joint detection accounts for the characteristics of the other symbols rather than detecting each symbol independently.
  • B. Why is MIMO Detection Important and Challenging?: The optimum MIMO detection problem is NP-hard, with optimal algorithms whose complexity grows exponentially with the number of decision variables.This computational difficulty motivates complexity-scalable detection methods.
  • C. The Contributions of This Paper: The paper unifies and reviews representative linear, interference-cancellation, tree-search, lattice-reduction, PDA, and SDPR detection methods from the previous fifty years.Its focus is on methods potentially applicable to LS-MIMO systems.

II. THE NATURE OF CO-CHANNEL INTERFERENCE

Co-channel interference arises when multiple transmissions overlap in signal features or occupy non-orthogonal channels. The paper relates interference management to distinguishability across frequency, time, and space, while noting that practical filtering can reduce leakage.

  • Nature of Co-Channel Interference: Co-channel interference is defined as interference from multiple transmissions on mutually non-orthogonal channels.Equivalently, the interfering and desired signals span subspaces with a non-empty intersection.
  • Nature of Co-Channel Interference: CCI originates from signal-feature overlap among multiple transmissions in frequency, time, or space.This overlap is common in spectrum-efficient CDM/CDMA and SDM/SDMA systems.
  • Signal Distinguishability: Multiple signals must be distinguishable in at least one of the frequency, time, or space domains.Practical designs make signals as distinguishable as possible in one domain while allowing overlap in the others.
  • Signal Distinguishability: Frequency-, time-, and space-division multiple-access techniques seek to avoid CCI by orthogonalizing channel access in the corresponding domain.Space-division systems generally retain stronger non-orthogonality than frequency- or time-division systems.
  • Nature of Co-Channel Interference: CCI is mainly considered in SDM/SDMA and CDM/CDMA systems, where transmissions often overlap simultaneously or partially over the same frequency.It may also be called ISI, ICI, IAI, MUI, MAI, or MSI depending on the application.

III. CONCEPT AND GENERALITY OF MIMO DETECTION

MIMO detection jointly detects symbols transmitted through multiple-input, multiple-output channels when subchannels are non-orthogonal. The framework covers point-to-point, multiple-access, interference, and X-channel settings, with collocation determining whether cooperative processing is feasible.

  • Concept of MIMO Detection: MIMO detection is joint detection of several information-bearing symbols transmitted through a multiple-input, multiple-output channel.The problem arises when the input subchannels are non-orthogonal and therefore interfere at the outputs.
  • Point-to-Point and Multiple-Access MIMO: The same generic detection framework applies to point-to-point MIMO and multiple-access MIMO channels.Examples include collocated antennas in point-to-point systems and distributed mobile stations transmitting to a multi-antenna base station.
  • Collocation and Processing: Whether transmitters and receivers are collocated determines whether cooperative joint encoding or decoding can be performed.Collocated antennas enable joint processing, making joint MIMO transmission and detection feasible.
  • Multiuser MIMO Broadcast Channels: In multiuser MIMO downlinks, transmit preprocessing at the base station can make detection at each user less challenging.The base station can jointly encode signals because its transmit inputs are collocated.
  • Interference and X Channels: Geographically distributed transmitters and receivers form interference or X channels with distinct message-to-receiver relationships.Interference-alignment problems for these channels are outside the paper’s detailed scope.

IV. FORMAL DEFINITION OF THE MIMO DETECTION PROBLEM

MIMO detection estimates transmitted input vectors from noisy, generally non-orthogonal multi-input/multi-output systems. The paper formalizes coherent and noncoherent settings and relates the generic model to memoryless and dispersive channels.

  • Formal scope: MIMO detection usually denotes symbol detection in narrow-band spatial-division-multiplexing multiple-antenna systems, although related problems predate the term.The term became widespread with multiple-antenna techniques in the mid-1990s.
  • Formal scope: The generic problem uses an NI-input linear system with non-orthogonal channel columns and additive random noise, whose distribution need not be Gaussian.Input vectors are drawn from a finite constellation set.
  • Formal scope: The received model y = Hs + n uses y for the received vector, H for the channel matrix, and n for additive noise over either R or C.The considered MIMO detection problem specifically requires a channel matrix with non-orthogonal columns.
  • Formal scope: Space-time coding yields a more general matrix model across multiple time slots, while the vector model is recovered when only one time slot is considered.The matrix model is mainly used for space-time-coding-aided systems.
  • Detection settings: Detection estimates s using y and H; exact instantaneous H supports coherent detection, whereas statistical channel knowledge supports noncoherent detection.Noncoherent schemes typically use differential encoding and block-based sequence detection.
  • Channel models: For dispersive ISI channels, the received signal depends on multiple transmitted symbols through channel memory, producing a sequence-detection problem.The model is y = Hs + w, with H determined by the ISI channel memory.
  • Channel models: The same generic model applies in time or frequency domains to both memoryless and dispersive channels; a flat-fading VBLAST-style system is a memoryless example.In that application, Nt and Nr denote transmit and receive antenna counts, and h_j,i is the impulse response between antenna pairs.

VI. MIMO SYSTEM MODEL FOR DISPERSIVE CHANNELS EXHIBITING MEMORY

Dispersive MIMO channels require sequence detection because multipath couples symbols across time. The paper models this with zero-padded frames, a banded Toeplitz channel matrix, and discusses equivalent real-valued formulations and their limits.

  • Dispersive channel model: A frequency-selective VBLAST system can be represented by linear finite-impulse-response dispersive channels with possibly complex-valued impulse responses.The channel memory length L is the maximum number of multipath components in a link.
  • Dispersive channel model: Because channel memory makes one-shot detection non-optimal, detection must use multiple received vectors for sequence detection.The required sequence spans multiple Nr-element received signal vectors.
  • Frame construction: The received frame concatenates K noisy Nr-element vectors, while transmitted symbols and noise are similarly stacked across sampling instants.The construction produces vectors y, s, and n for matrix-form detection.
  • Matrix representation: The resulting channel matrix has dimension KNr × NNt and a banded Toeplitz structure formed from the Nr × Nt path-gain matrices H_l.Each H_l contains gains between every transmit and receive antenna pair for path l.
  • Related systems: The model also connects linear memoryless and memory-bearing MIMO formulations to synchronous and asynchronous CDMA systems, respectively.Asynchronous CDMA can additionally be represented in the z domain.
  • Complex versus real models: Complex and real-valued MIMO models are often mutually convertible, and commonly deliver equivalent achievable performance under stated assumptions.A pairwise real-valued formulation can reduce complexity relative to the conventional real-valued model.
  • Complex versus real models: The real-valued decomposition is limited to real-valued or rectangular-QAM constellations and can require a channel representation twice as large.Complex arithmetic may also be efficient for hardware implementation, while improper signals require pseudo-covariance-based processing.

VIII. HISTORY AND STATE-OF-THE-ART OF MIMO

MIMO detection developed from early equalization and multiuser-detection work into joint detection for multiple-antenna systems and large-scale MIMOs. The history shows that interference structure and receiver assumptions determine detector optimality.

  • Historical development: MIMO detection traces its embryonic concept to the 1960s, including Shnidman’s 1967 equalization treatment of simultaneously transmitted waveforms.The early system used multiple amplitude-scaled waveforms over one physical channel.
  • Methodological heritage: A four-period historical account places early equalization, multiuser detection, joint MIMO detection, and large-scale MIMO detection in one lineage.This frames modern MIMO detection as an extension of earlier interference-management problems.
  • Historical development: The field passed through CDM/CDMA multiuser detection, joint detection for small- and medium-scale MIMO, and detection for large-scale multiple-antenna systems.These periods span the 1980s through the large-scale MIMO era.
  • Methodological heritage: Techniques developed for ISI equalization were often extended to MIMO detection because the underlying equalization and generic MIMO problems are similar.Adapted methods include maximum-likelihood sequence estimation using the Viterbi algorithm.
  • Methodological heritage: In 1976, van Etten derived an ML sequence-estimation receiver addressing both intersymbol and interchannel interference.Under certain conditions, its performance asymptotically approached the optimum receiver for an idealized interference-free system.
  • Interference and optimality: The conventional belief that multiuser interference was white Gaussian made single-user matched-filter detection appear essentially optimal.This belief persisted until the early 1980s.
  • Interference and optimality: Verdú’s 1983 optimal multiuser detector disproved that conventional view and revealed a substantial performance gap between SUMF and optimal MUD.The result followed recognition that multiuser interference is generally non-Gaussian.
  • Interference and optimality: Even Gaussian interference does not make SUMF optimal, because the desired-user matched-filter output is not sufficient when multiplexed signals are non-orthogonal.Interfering users’ matched-filter outputs contain information useful for joint detection.

2) Optimum Decision Criteria:

Optimum MIMO detection depends on the chosen error criterion and available prior information. MAP, ML, and MED formulations target vector-error minimization, whereas other criteria address different performance objectives.

  • MF versus optimum detection: An infinite-population limit can make multiuser interference asymptotically Gaussian when interferer amplitudes decrease while total interference power remains fixed.This is presented as a specific situation where the central limit theorem applies rigorously.
  • Detector tradeoffs: The detector overview presents tree-search methods as generally suboptimal but able to trade achievable performance against computational complexity.In some scenarios, tree-search detectors can attain optimum ML performance.
  • Optimum criteria: Under Bayesian inference, MAP minimizes error probability using observed signals, hypotheses, and prior probabilities.The MAP criterion uses a posteriori probabilities of candidate transmitted vectors.
  • Optimum criteria: When all candidate vectors have equal prior probability, MAP simplifies because the prior and evidence terms do not affect the maximizing candidate.The resulting simplification leads to the ML criterion.
  • Optimum criteria: MAP is commonly used in iterative detection and decoding for coded systems, while ML is commonly used when forward-error correction is absent.The distinction follows from whether symbol priors are supplied by decoder information.
  • Optimum criteria: ML detection can be reformulated as finite-set constrained least-squares optimization, equivalently the minimum Euclidean distance criterion.The estimate is selected from the finite constellation-constrained candidate set.
  • Optimum criteria: The matched filter is optimal for received-SNR maximization under additive stochastic noise, but this objective differs from minimum detection error.The paper distinguishes several detector-design criteria.
  • Optimum criteria: MAP, ML, and MED detectors minimize vector-error rate but do not guarantee minimum bit-error rate or symbol-error rate.BER and SER are especially important in many forward-error-corrected applications.

3) Computational Complexity:

Optimal MIMO detection can require exponentially increasing complexity, motivating suboptimum detectors that trade performance against computational cost. Complexity must be assessed across average, worst-case, distributional, and hardware dimensions.

  • Computational growth: Brute-force search examines |A|^NI symbol vectors, producing exponentially increasing computational complexity.For BPSK with M=2 and N_I=2, four transmitted symbol-vector realizations must be considered.
  • Computational growth: Optimal MIMO detection is linked to NP-hard closest lattice-point search, while polynomial-time optimality would resolve open NP-complete problems.The closest lattice-point formulation was shown NP-hard, and polynomial-time solutions for NP-complete problems remain unresolved.
  • Complexity metrics: MIMO-detector complexity should be examined through average, worst-case, and distributional measures, alongside silicon area and NAND2-gate requirements.The latter metrics capture hardware implementation complexity.
  • Complexity-performance trade-off: Suboptimum detector families seek good performance at lower computational cost than the optimum detector.Representative classes include linear, interference-cancellation, tree-search, PDA, and SDPR-based detectors.
  • Complexity-performance trade-off: Linear detectors apply a transformation to y and offer low complexity but can lose considerable performance relative to ML detection.Their transformation matrix is designed according to different criteria.

2) Linear ZF Detector:

The zero-forcing detector uses a pseudoinverse-based linear transformation to eliminate multi-input interference. This comes at the cost of increased noise power, while MMSE design balances interference suppression and noise enhancement more effectively at low SNR.

  • 2) Linear ZF Detector: For full-column-rank H with N_r>N_t, ZF uses the left-multiplying pseudoinverse H†=(H^H H)^−1H^H.When H is square and invertible, H† coincides with H^−1.
  • 2) Linear ZF Detector: ZF completely eliminates inter-stream interference but augments noise power, yielding d=s+H†n.This expresses the central ZF trade-off in the detected signal.
  • 2) Linear ZF Detector: The ZF detector was introduced in 1975 for multiple-channel multiplexing with ISI and ICI and later studied extensively for CDMA.Its CDMA interpretation included the linear decorrelating multiuser detector.
  • 3) Linear MMSE Detector: MMSE designs T to minimize mean-square error between transmitted data and transformed channel output.The MMSE criterion jointly accounts for multiuser interference and noise.
  • 3) Linear MMSE Detector: MMSE achieves a better MUI–noise-enhancement balance and outperforms ZF at low SNRs.The formulation assumes noise power σ^2 per real dimension and E(s)=1.

4) Other Linear Detectors:

Beyond ZF and MMSE, linear detectors use alternative criteria, while interference-cancellation detectors improve performance through nonlinear processing. Their trade-offs depend on signal-power disparities, error propagation, and dimensionality.

  • 4) Other Linear Detectors: MAME-based linear detection targets minimum bit-error probability as noise approaches zero by optimizing asymptotic multiuser efficiency.AME characterizes high-SNR performance loss caused by interference from other active users.
  • 4) Other Linear Detectors: WLS detection addresses the bias of ZF and MMSE by providing unbiased symbol estimates.Its practical use requires estimating system parameters and regularly adapting the receiver structure.
  • 4) Other Linear Detectors: MBER detection can outperform MMSE when signature cross-correlation is high or background noise is non-Gaussian.Linear detection can also be designed from the mathematically similar equalization perspective.
  • Interference cancellation: SIC detects symbols sequentially, subtracting reconstructed interference and prioritizing the strongest interferer.It benefits near-far systems but can propagate decision errors when received signal strengths are insufficiently separated.
  • Interference cancellation: SIC complexity and processing delay grow with the number of transmitted symbols because detection reordering is required at each iteration.Its performance is strongest when simultaneously transmitted symbols have substantially different received strengths.
  • Interference cancellation: PIC detects all symbols simultaneously and iteratively subtracts regenerated interference, reducing delay and inter-stream error propagation relative to SIC.PIC suits similar-power signals, whereas SIC performs better for different-power streams.
  • Tree-search detectors: Tree-search detectors offer a performance–complexity trade-off, but their expected complexity can remain exponential at fixed SNR.The reported scaling is O(M^(βN_t)), with β∈(0,1] depending on SNR.

E. Lattice-Reduction Aided Detectors

Lattice-reduction aided detectors transform the MIMO channel basis toward orthogonality before detection, improving linear and nonlinear detectors with limited added complexity. Their practical algorithms trade optimality for computational feasibility, with LLL variants offering polynomial average-complexity bounds but potentially unbounded worst-case complexity.

  • E. Lattice-Reduction Aided Detectors: Lattice-reduction aided detectors reinterpret MIMO detection through lattice geometry and seek a channel basis closer to orthogonality.The received vector is a noisy observation of a lattice point generated by the channel matrix.
  • E. Lattice-Reduction Aided Detectors: Optimal lattice bases can be computationally prohibitive, so practical LR algorithms instead search for improved bases using tractable approximations.Gaussian, Minkowski, and KZ reductions can find optimal bases but are unsuitable for communications systems because of their complexity.
  • E. Lattice-Reduction Aided Detectors: LLL algorithms have polynomial average-complexity bounds, but their worst-case computational complexity can be infinite.The cited bounds depend on the system dimensions and logarithmic factors, while the worst-case scenario is reported to have zero practical probability.
  • E. Lattice-Reduction Aided Detectors: LR can be combined with ZF, ZF-SIC, MMSE, and MMSE-SIC detectors to obtain substantial performance gains with little additional computational complexity.Real- and complex-valued LLL variants support corresponding MIMO system models.
  • E. Lattice-Reduction Aided Detectors: The paper situates LR among major MIMO detector families and reviews its development alongside tree-search, PDA, and related approaches.The detector families are presented as suboptimal alternatives motivated by the performance-complexity gap of optimum detection.

MiNI

PDA-based detectors approximate multimodal interference-plus-noise distributions with iteratively updated Gaussian models and produce soft information for detection and decoding. They offer low or polynomially scaling complexity and can perform well under suitable, non-overloaded channel conditions, while high-order QAM remains a challenge for SDPR-based methods.

  • MiNI: PDA outputs are inherently soft-input soft-output, making them suitable for integration with convolutional, turbo, and LDPC codes.This supports iterative detection and decoding architectures.
  • MiNI: PDA performance can improve as the number of transmit antennas or users increases, provided the channel is neither overloaded nor rank-deficient.The paper also notes that the approximation-and-iteration nature of PDA remains less well understood than mature detector families.
  • MiNI: PDA detectors use iterative Gaussian approximation to model multimodal interference-plus-noise distributions while estimating soft symbol information.Their performance depends strongly on the accuracy of this approximation.
  • MiNI: SDPR detectors relax optimum MIMO detection into semidefinite programming and offer polynomial-time worst-case complexity with high performance in certain circumstances.The complexity increases polynomially with the number of users or transmit antennas.
  • MiNI: High-order modulation is a key limitation for SDPR detectors, whose performance is less promising for high-order QAM than for BPSK or QPSK.A bit-based high-order-QAM SDPR variant improves on an earlier method, but further improvement is still needed while retaining low complexity.

H. Detection in Rank-Deficient and Overloaded MIMO Systems

Rank-deficient and overloaded MIMO channels reduce available spatial degrees of freedom and challenge standard detectors. The paper compares scenario-dependent remedies and emphasizes that detector choice depends on performance criteria, system assumptions, and complexity behavior.

  • H. Detection in Rank-Deficient and Overloaded MIMO Systems: Rank deficiency arises from correlated antenna channels or keyhole effects and reduces spatial degrees of freedom and MIMO capacity.A keyhole channel can have rank one even when its entries are uncorrelated.
  • H. Detection in Rank-Deficient and Overloaded MIMO Systems: Overloaded systems lack the usual favorable dimensional relationship, creating detection difficulties alongside rank-deficient channels.The paper describes overloaded systems through a fat channel matrix that lacks full column rank.
  • H. Detection in Rank-Deficient and Overloaded MIMO Systems: Proposed remedies include pseudo-inverse linear detection, group detection, generalized sphere decoding, modified PDA and SDPR detectors, and metaheuristics.These approaches are presented as scenario-dependent strategies rather than a single universal solution.
  • H. Detection in Rank-Deficient and Overloaded MIMO Systems: Detector optimality depends on the chosen performance criterion and application assumptions, so no algorithm is universally best.Performance may include error probability or robustness, while complexity may include computation or hardware cost.
  • H. Detection in Rank-Deficient and Overloaded MIMO Systems: Tree-search detectors have exponentially increasing complexity and may be unsuitable for low-SNR or large-scale systems despite possible fixed-complexity variants.Their enumeration strategies are described as poorly suited to systems with many inputs.

IX. DETECTION IN LS-MIMO SYSTEMS

LS-MIMO systems use many antennas to increase spatial degrees of freedom, reliability, and achievable rate, but their detection problem requires complexity-scalable algorithms. The section introduces deployment configurations and frames large-scale detection as distinct from conventional MIMO through its system dimensions and application scenarios.

  • IX. DETECTION IN LS-MIMO SYSTEMS: The section focuses on detection for systems with dozens or hundreds of antennas, where LS-MIMO can achieve high spectral efficiency and diversity order.These benefits motivate complexity-scalable detector designs.
  • IX. DETECTION IN LS-MIMO SYSTEMS: Large antenna arrays can provide diversity that increases reliability and achievable rates that scale with min(Nt, Nr).The paper states that more antennas increase spatial degrees of freedom without increasing occupied bandwidth.
  • IX. DETECTION IN LS-MIMO SYSTEMS: LS-MIMO systems may use point-to-point backhaul or other deployments requiring different antenna-array configurations and physical layouts.Linear, rectangular, cylindrical, and spherical arrays differ in coverage direction, compactness, and physical area.
  • IX. DETECTION IN LS-MIMO SYSTEMS: Although MIMO detection is intrinsically large-scale because optimum ML/MAP complexity grows exponentially with problem size, LS-MIMO environments require careful adaptation of existing insights.The section categorizes the detection problem according to the application scenario.

A. Detection in Single-Cell/Noncooperative Multi-Cell LS-MIMO Systems

Single-cell and noncooperative multi-cell LS-MIMO systems split into Type-I and Type-II configurations with different detection implications. Type-II systems benefit from channel hardening and favorable conditioning, whereas pilot contamination limits noncooperative multi-cell performance and motivates BS cooperation.

  • System types: Type-I systems scale transmit and receive dimensions together, while Type-II systems use many collocated BS antennas with substantially fewer active transmit antennas.Type-II systems correspond to underloaded uplinks and exhibit a highly unbalanced antenna configuration.
  • Type-I systems: For Type-I systems, singular-value distributions converge to deterministic limits described by the Marčenko–Pastur law.As channel dimensions grow, singular values become less sensitive to the distributions of the i.i.d. channel entries.
  • Type-I systems: Channel hardening makes large random channel matrices increasingly deterministic and improves diagonal dominance in H^H H.This enables series approximations for matrix inversion and can support low-complexity detection.
  • Type-II systems: In Type-II systems, high receive diversity, diminishing MUI and noise impact, asymptotic user orthogonality, and favorable conditioning allow MF detection to approach optimum performance.Increasing BS antennas can also compensate for low SNR or poor channel estimates in the discussed downlink setting.
  • Multi-cell cooperation: Pilot contamination is the major limiting factor in noncooperative multi-cell systems, so centralized or distributed BS cooperation can enable multi-cell joint processing.Pilot contamination arises from reusing pilot sequences in adjacent cells.
  • Detection strategies: Tree-search detectors with exponential worst-case complexity are less feasible for Type-I systems, while polynomial-time SDPR methods may remain applicable.Many existing detectors can still be used in Type-II systems, where low-complexity linear detectors may achieve near-optimum performance.

D. Recent Advances in LS-MIMO Detection

Recent LS-MIMO detection advances build on decades of MIMO and large-system research, emphasizing low-complexity algorithms and scalable implementations. The survey concludes that several algorithm families offer promising performance–complexity tradeoffs, while MIMO detection techniques also extend to radar and optical-fiber communications.

  • Recent advances: Large-scale MIMO detection research draws on asymptotically optimal MMSE multiuser detectors and BI-GDFE methods developed for large-scale CDMA and random MIMO channels.BI-GDFE approaches the single-user matched-filter bound when received SNR is sufficiently high.
  • Recent advances: Large-scale systems have been addressed with local-search metaheuristics, including LAS extensions demonstrated for Type-I systems with up to 600 transmit and receive antennas.Other surveyed approaches include randomized search, Markov chain Monte Carlo, mixed Gibbs sampling, multiple restart, and lattice-reduction methods.
  • Recent advances: Soft-input soft-output detectors based on SUMIS and approximate message passing provide relatively low-complexity LS-MIMO detection alternatives.Approximate message passing has also been applied to FEC-coded large-scale MIMO-OFDM systems.
  • Hardware implementation: A first ASIC design using truncated Neumann series expansion achieved 3.8 Gb/s in an LTE-A LS-MIMO system with 128 BS antennas and 8 users.The reported implementation demonstrates a hardware-oriented route to high-throughput detection.
  • Applications: MIMO detection methods also support MIMO radar and mode-division-multiplexed multimode-fiber optical communication systems.The applications extend beyond wireless communications into radar design and high-speed optical communications.
  • Conclusions: The survey identifies local-neighborhood search, Bayesian message passing, and convex optimization as promising ways to balance LS-MIMO detection performance and complexity.These conclusions are drawn from the surveyed detection heritage and recent LS-MIMO advances.
Loading 1507.05138v1…