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 makes detection more challenging as antenna dimensions grow, while optimal ML/MAP detection already has exponentially increasing complexity. This survey synthesizes MIMO detection fundamentals and five decades of methods, distinguishes two LS-MIMO regimes, and extracts strategies and lessons for complexity-scalable detection.

  • Problem

    LS-MIMO detection requires scalable approaches because optimal ML/MAP detection complexity increases exponentially with problem size.

    Source: §IX · p. 32

  • Method

    The paper surveys MIMO detection fundamentals and 50 years of methods, then organizes LS-MIMO detection strategies by two system types.

    Source: §X · p. 36

  • Results

    The survey extracts insights and lessons from existing MIMO detection methods for designing complexity-scalable algorithms applicable to LS-MIMO.

    Source: §X · p. 36

  • Takeaways & Limitations

    LS-MIMO detection should be considered across distinct system types, with strategies informed by the accumulated history of MIMO detection.

    Source: §X · p. 36

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 … C. The Contributions of This Paper

The paper motivates LS-MIMO as a response to rapidly growing mobile traffic and spectrum scarcity, then frames MIMO detection as a central but computationally difficult problem. It contributes a unified review emphasizing complexity-scalable detectors and addressing underexplored general LS-MIMO uplinks.

GLOSSARY

  • The glossary defines the paper’s recurring abbreviations, including LS-MIMO, CCI, ML, MAP, PDA, SDPR, and ZF.It also lists terms spanning communication systems, detection algorithms, performance metrics, and implementation technologies.Source: GLOSSARY · pp. 1–2

A. Why are Large-Scale MIMOs Important?

  • 81% growth in global mobile data traffic in 2013 and a predicted nearly 11-fold increase from 2013 to 2018 motivate LS-MIMO for high spectral and energy efficiency,.The forecast reaches 15.9 EB per month by 2018, while spectrum scarcity further constrains wireless-system design.Source: A. Why are Large-Scale MIMOs Important? · p. 2
  • LS-MIMO, millimetre-wave, and small-cell technologies form complementary 5G enablers, providing high dimensionality in spatial, frequency, and network-cell domains, respectively.Millimetre-wave systems operate at about 30–300 GHz and may require small-cell heterogeneous networks because of shorter wavelengths and smaller coverage areas.Source: A. Why are Large-Scale MIMOs Important? · pp. 2–3, Figure 3 · p. 3

B. Why is MIMO Detection Important and Challenging?

  • MIMO detection must recover concurrently transmitted symbols despite noise and interference, with joint detection improving performance over separate detection at higher computational cost.Co-channel interference is a fundamental limitation in MIMO systems, making joint detection central to realizing MIMO benefits.Source: B. Why is MIMO Detection Important and… · p. 3
  • Optimum MIMO detection is NP-hard, so ML- and MAP-based algorithms have complexity that grows exponentially with the number of decision variables –.Power limits in CMOS integrated circuits mean even modest-complexity detectors can be unsuitable for battery-powered devices, motivating low-complexity high-performance alternatives.Source: B. Why is MIMO Detection Important and… · pp. 3–4

C. The Contributions of This Paper

  • The paper reviews representative MIMO detection methods from the past fifty years in a unified mathematical model, emphasizing complexity-scalable algorithms for LS-MIMO.Covered families include linear, interference-cancellation, tree-search, lattice-reduction, probabilistic-data-association, and semidefinite-programming-relaxation detectors.Source: C. The Contributions of This Paper · p. 4
  • The paper addresses gaps in prior reviews by covering advanced detectors and focusing on LS-MIMO, whereas earlier work emphasized CDMA, conventional small-/medium-scale MIMO, or common suboptimum detectors –.The review explicitly includes tree-search, lattice-reduction, PDA, and SDPR methods alongside classic linear and interference-cancellation approaches.Source: C. The Contributions of This Paper · p. 4
  • The paper targets the limited attention given to uplinks of general LS-MIMO systems, beyond research focused mainly on precoding and beamforming for asymmetric downlinks.Its stated goal is to fill this gap in the open literature, with the paper’s organization summarized in Fig. 5.Source: C. The Contributions of This Paper · p. 4

II. THE NATURE OF CO-CHANNEL INTERFERENCE

Co-channel interference arises when transmissions overlap in signal features or occupy non-orthogonal channel subspaces across frequency, time, or space. Recovery requires distinguishability in at least one domain, while practical multiplexing schemes manage unavoidable overlap through deliberate orthogonalization or attenuation.

Definition and physical domains

  • CCI is interference from multiple transmissions on mutually non-orthogonal channels, and recovery becomes impossible when they are highly non-orthogonal in frequency, time, and space.The desired signal must be distinguishable from interference in at least one of these domains.Source: §II · p. 4, Figure 6 · p. 5

Practical signal separation

  • Although radio-signal overlap is theoretically inevitable across frequency, time, and space, practical systems deliberately maximize distinguishability in one domain while permitting overlap in the others.Filters and sufficient spatial separation can reduce practical interference, despite idealized signals extending across domains.Source: §II · pp. 4–5

Multiple-access schemes

  • FDM/FDMA, TDM/TDMA, and SDM/SDMA mitigate CCI by orthogonalizing access in frequency, time, and space, respectively, while CDM/CDMA spreading-code orthogonality is a special case of time-domain orthogonality.Practical code systems often use quasi-orthogonal sequences because perfectly orthogonal sequences are limited, producing substantial interference.Source: §II · pp. 5–6

Detection implications

  • Space-division and code-division systems generally pose more difficult detection problems than frequency- and time-division systems because physical constraints, random channels, and degraded code orthogonality create substantial interference.Guard intervals more readily preserve frequency- and time-domain resolution, whereas spatial resolution is undermined by transmitter/receiver size and propagation conditions.Source: §II · p. 6

III. CONCEPT AND GENERALITY OF MIMO DETECTION

MIMO detection is the joint recovery of multiple symbols when input subchannels are nonorthogonal and interfere, providing a generic framework spanning ISI equalization, point-to-point MIMO, and multiuser systems. Its difficulty and applicable processing depend strongly on whether transmitters and receivers are collocated, with distributed interference and X channels remaining especially challenging.

Concept and generality

  • MIMO detection jointly detects multiple information-bearing symbols over multi-input multi-output channels precisely when nonorthogonal subchannels create interference.The framework is fundamental to high-throughput communications and supports varied physical interpretations of inputs and outputs.Source: §III · p. 6
  • The same mathematical problem covers ISI-channel equalization, point-to-point multi-antenna MIMO, multiuser SDMA, and uplink CDMA detection.In these cases, inputs and outputs represent interfering symbol sequences, collocated antenna signals, distributed-user transmissions, or matched-filter outputs, respectively.Source: §III · pp. 6–7, Figure 8 · p. 7, Figure 9 · p. 7

Collocation and cooperation

  • Collocated transmitters or receivers enable cooperative joint encoding or decoding, making joint MIMO transmission and detection feasible, whereas multiple-access uplinks are typically asynchronous because users cannot jointly encode.The point-to-point MIMO example has collocated transmit and receive antennas, while CDMA and SDMA uplinks generally lack user-side joint encoding.Source: §III · p. 7

Downlink multiuser MIMO

  • Downlink multiuser MIMO shifts most sophisticated processing to BS transmit precoding, leaving detection at each user less challenging and outside this paper’s scope.The BS can exploit collocated inputs for cooperative joint encoding; transmit preprocessing is not elaborated further.Source: §III · p. 7

Distributed interference channels

  • Distributed-transmitter and distributed-receiver interference and X channels are substantially harder than single-user MIMO, and their capacity and signal-processing solutions remain largely open.The X channel generalizes multiple-access, broadcast, and interference channels; cooperation such as joint multicell processing and interference alignment is used to make these settings more tractable, but those topics are outside scope.Source: §III · pp. 7–8

IV. FORMAL DEFINITION OF THE MIMO DETECTION PROBLEM · V. MIMO SYSTEM MODEL FOR LINEAR MEMORYLESS CHANNELS

The paper defines MIMO detection generically as estimating finite-alphabet inputs from noisy outputs of a non-orthogonal linear multi-input, multi-output system, while focusing primarily on coherent detection. For linear memoryless channels, the canonical model is a flat-fading VBLAST-style system in which one received signal vector suffices for detection.

IV. FORMAL DEFINITION OF THE MIMO DETECTION PROBLEM

  • MIMO detection estimates a finite-alphabet input vector from the received vector and channel matrix in a noisy linear system whose channel columns are non-orthogonal.The alphabet may contain real or complex symbols, and the additive noise need not be Gaussian.Source: §IV · p. 8
  • The generic formulation covers any multiple-input, multiple-output system with additive random noise, but the paper restricts MIMO detection to systems with non-orthogonal channel columns.The real- and complex-valued cases are represented by F = R or F = C, while the alphabet and input/output dimensions are normally treated as known constants.Source: §IV · p. 8
  • The paper focuses on coherent MIMO detection because space-time-coded models often permit separate symbol-by-symbol or pairwise decoding, whereas the single-time-slot model is the usual setting for MIMO detection.The multi-time-slot space-time model generalizes the single-slot model, but typically does not present the targeted detection problem for orthogonal or quasi-orthogonal codes.Source: §IV · pp. 8–9
  • Coherent detection uses the instantaneous channel matrix, whereas noncoherent detection relies on statistical channel knowledge or implicit estimation and generally incurs complexity and power-efficiency tradeoffs.Noncoherent schemes commonly use differential encoding and block sequence detection, although differential space-time modulation can admit linear-complexity symbolwise or pairwise decoding.Source: §IV · pp. 8–9

V. MIMO SYSTEM MODEL FOR LINEAR MEMORYLESS CHANNELS

  • The linear memoryless model applies when current outputs are independent of previous inputs, including narrowband synchronous VBLAST-style SDM-MIMO over flat-fading channels.It also applies separately to each subcarrier of frequency-selective multiple-antenna OFDM systems.Source: §V · p. 9
  • For this model, Nt and Nr denote transmit and receive antennas, with hj,i representing the complex impulse response from transmit antenna i to receive antenna j.The antenna counts correspond to the generic dimensions as Nt = NI and Nr = NO.Source: §V · p. 9
  • One-shot detection based on a single received vector is sufficient for linear memoryless MIMO channels, with transmitted-symbol energy typically normalized to E(si) = 1 or E(s) = 1.This normalization supports fair comparison with single-input single-output systems.Source: §V · p. 9

VI. MIMO SYSTEM MODEL FOR DISPERSIVE CHANNELS EXHIBITING MEMORY · VII. COMPLEX-VALUED VERSUS REAL-VALUED MIMO SYSTEM MODEL

Dispersive MIMO channels require sequence detection over multiple received vectors, with zero-padding producing a banded Toeplitz system model. Although real- and complex-valued formulations can often be converted and yield equivalent performance, complex-valued models generally offer broader constellation flexibility and lower implementation complexity, while real-valued models can benefit ordering-dependent detection.

VI. MIMO SYSTEM MODEL FOR DISPERSIVE CHANNELS EXHIBITING MEMORY

  • Dispersive MIMO links are modeled as finite-impulse-response channels with memory length L, making one-shot detection nonoptimal and requiring sequence detection across multiple received vectors.The channel may be complex-valued, and each path contributes an antenna-pair channel gain.Source: §VI · p. 9
  • Zero-padding with P = L −1 samples removes interblock interference and forms a frame of K = N + P sampling intervals, while alleviating noise enhancement or error propagation.The information-bearing portion occupies N intervals, followed by P consecutive zero vectors.Source: §VI · pp. 9–10
  • The frame model stacks K Nr-dimensional received vectors and N Nt-dimensional transmitted vectors, yielding a channel matrix of size (KNr × NNt) with banded Toeplitz structure.Each block H_l is an Nr × Nt matrix containing the gains for all transmit–receive antenna pairs on path l.Source: §VI · p. 10

VII. COMPLEX-VALUED VERSUS REAL-VALUED MIMO SYSTEM MODEL

  • The generic MIMO model can be formulated over either R or C, with complex-valued models typically preferred for QAM and PSK because they are more concise.The two formulations are mutually convertible under the stated uncorrelated real- and imaginary-part assumption.Source: §VII · p. 10
  • Real-valued decomposition is limited for PSK, star QAM –, and near-Gaussian constellations,, while complex arithmetic can require lower silicon complexity than doubled real-valued storage.A complex multiplier can be implemented with either 4 real multipliers and 2 real adders or 3 real multiplications and 5 real additions.Source: §VII · p. 10
  • Although complex- and real-valued detectors usually achieve equivalent performance, real-valued VBLAST can outperform its complex counterpart when extra detection-ordering freedom matters.The potential gain applies more generally to detection algorithms whose performance depends on symbol ordering, at the cost of redundant channel-matrix storage.Source: §VII · p. 11
  • For improper or noncircular complex signals, pseudo-covariance and advanced complex-valued processing – are needed because conventional proper or circularity assumptions may be unjustified.Such processing can achieve better performance in some applications, –.Source: §VII · p. 11

VIII. HISTORY AND STATE-OF-THE-ART OF MIMO DETECTION · A. Optimum MIMO Detector · 1) Matched Filter (MF) versus Optimum MIMO Detector:

MIMO detection evolved from early equalization of crosstalk and ISI into diverse optimum and suboptimum detection frameworks spanning single-, multiuser, and large-scale antenna systems. The matched-filter comparison established that joint detection can substantially outperform conventional single-user filtering when interference carries exploitable information.

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

  • MIMO detection originated in 1967 with Shnidman’s optimum linear receiver, which jointly addressed intersymbol interference and waveform crosstalk in bandwidth-limited pulse modulation.This work extended the classic Nyquist equalization problem to multiple simultaneously transmitted waveforms.Source: §VIII · p. 11
  • The field progressed through four periods: early single-user FDM/TDM crosstalk suppression, CDM/CDMA multiuser detection, joint detection in small-/medium-scale MIMO, and detection in large-scale MIMO systems.Representative detectors are categorized along dimensions including optimum versus suboptimum, linear versus nonlinear, sequential versus one-shot, and hard versus soft decision.Source: §VIII · pp. 11–12
  • MIMO detection repeatedly adapted ISI-equalization methods, including ML sequence estimation, ZF, MMSE, decision-feedback, adaptive, and blind equalization –.The shared structure between ISI equalization and generic MIMO detection explains the transfer of these techniques.Source: §VIII · p. 12

A. Optimum MIMO Detector

  • The first optimum MIMO detector appeared in 1976, when van Etten derived an ML sequence-estimation receiver for simultaneous ISI and interchannel interference, with asymptotic optimality under specified conditions.The result was later recognized as research interest in CDMA and multiple-antenna systems increased.Source: A. Optimum MIMO Detector · p. 12

1) Matched Filter (MF) versus Optimum MIMO Detector:

  • Verdú’s 1983 optimal multiuser detector disproved the belief that a conventional single-user matched filter was essentially optimal, revealing a substantial performance gap in general,,,.Poor and Verdú also developed nonlinear single-user detectors after identifying the non-Gaussian nature of multiuser interference.Source: 1) Matched Filter (MF) versus Optimum… · p. 12
  • Even Gaussian multiuser interference does not make the single-user matched filter optimal, because its desired-user output is not sufficient in the presence of nonorthogonal interference.Joint detectors can exploit the interfering users’ matched-filter outputs, which contain additional information about the desired symbol.Source: 1) Matched Filter (MF) versus Optimum… · p. 12
  • The meaning of “optimum” depends on the goodness criterion and assumptions, so theoretical optima primarily serve as benchmarks or bounds and may mislead when their assumptions do not match realistic conditions.Changing the criterion or assumptions can change which detector is optimal.Source: 2) Optimum Decision Criteria · p. 12

2) Optimum Decision Criteria:

The section defines MAP as the minimum-error-probability criterion for MIMO detection and shows that uniform priors reduce it to ML, equivalent to finite-set least-squares or minimum Euclidean distance. It also distinguishes criteria by application and objective, noting that VER-optimal MAP/ML/MED need not minimize BER or SER, while linear detectors optimize SNR, SIR, SINR, MMSE, or BER under their respective criteria.

Optimum Decision Criteria

  • MAP detection minimizes vector-error rate (VER) under observed signals and hypotheses, using posterior probabilities and likelihood-based inference.Communication error probability may also be measured by BER, SER, PER, FER, BLER, or VER.Source: 2) Optimum Decision Criteria · p. 13
  • With identical prior probabilities, MAP becomes ML, which can be reformulated as finite-set constrained least-squares or minimum Euclidean distance detection.The equivalence follows because the observation density is independent of the transmitted signal vector; ML is commonly used in FEC-uncoded systems, whereas MAP supports iterative detection and decoding with decoder-provided priors.Source: 2) Optimum Decision Criteria · pp. 13–14
  • MAP, ML, and MED minimize VER but do not guarantee minimum BER or SER, which are especially important in many FEC-coded applications.The distinction is relevant because BER and SER are separate performance metrics from vector-level error probability.Source: 2) Optimum Decision Criteria · p. 14
  • Among linear detectors, MF maximizes received SNR, ZF maximizes received SIR, and MMSE minimizes mean-squared error while also maximizing received SINR.The MMSE and SINR optimality statements are restricted to linear detectors.Source: 2) Optimum Decision Criteria · p. 14
  • The linear MBER detector achieves the lowest BER among linear detectors.Its detailed treatment is identified as part of the paper’s later detector-design discussion.Source: 2) Optimum Decision Criteria · p. 14

3) Computational Complexity: … 1) MF Detector:

The section contrasts exponentially complex optimum MIMO detection with lower-cost suboptimum approaches, then introduces linear detectors and the MF detector’s low-complexity, interference-limited behavior. It also identifies computational-complexity metrics relevant to practical implementations and conditions where MF can approach ML performance.

3) Computational Complexity:

  • Brute-force solution of the optimum detector searches |A|^NI symbol vectors, producing exponentially increasing complexity as the number of users or decision variables grows.For BPSK with NI = 2, the search contains M^NI = 4 possible transmitted vectors.Source: 3) Computational Complexity · p. 14
  • Optimum MUD for CDMA has per-bit complexity O(2^K) asynchronously and O(2^K/K) synchronously, making detection intractable for many active users.The problem was characterized as NP-hard and NP-complete, so known optimal algorithms have exponential complexity,.Source: 3) Computational Complexity · p. 14
  • Practical detector assessment should include average, worst-case, and complexity-distribution measures, along with hardware complexity such as silicon area and NAND2-gate count.Tree-search algorithms can exhibit scenario-dependent complexity, motivating multiple computational metrics.Source: 3) Computational Complexity · pp. 14–15

4) Milestone Contributions:

  • The optimum detector’s performance and complexity gap relative to conventional SUMF motivated suboptimum detectors, including linear, interference-cancellation, tree-search, PDA, SDPR, and LR approaches.These representative detector classes are organized in Fig. 15 and developed in subsequent sections.Source: 4) Milestone Contributions · p. 15

B. Linear MIMO Detectors

  • Linear MIMO detectors apply a linear transformation to y and offer appealingly low complexity, but generally incur substantial performance loss relative to ML detection.Their filtering matrix is designed according to different criteria.Source: B. Linear MIMO Detectors · p. 15

1) MF Detector:

  • The MF detector has the lowest computational complexity among MIMO detectors and is the optimal linear filter for maximizing output SNR under additive stochastic noise.Its approach is based on a linear transformation of the received signal.Source: 1) MF Detector · p. 15
  • Although MF detection follows a single-user philosophy and performs poorly in CCI-limited systems, it can approach optimal ML performance in certain LS-MIMO contexts,.This motivates joint-detection MIMO detectors while preserving MF’s relevance for selected large-scale-array regimes.Source: 1) MF Detector · p. 15

2) Linear ZF Detector: · 3) Linear MMSE Detector: · 4) Other Linear Detectors:

The section presents linear ZF and MMSE detectors, emphasizing their tradeoffs in interference suppression, noise enhancement, and low-SNR performance. It also surveys alternative criteria addressing high-SNR errors, bias, non-Gaussian noise, and parameter uncertainty.

2) Linear ZF Detector:

  • ZF detection uses the channel inverse or Moore–Penrose pseudoinverse to solve the linear system and eliminate multi-input interference, but it augments noise power.For full-rank square H, H† = H^-1; for Nr > Nt, H† = (H^H H)^-1 H^H.Source: 2) Linear ZF Detector · p. 15
  • The ZF detector was developed for multiplexing and CDMA, achieving ML-equivalent decisions with unknown user energies and near-far resistance at lower complexity,,,,,.It was later studied in SDM-based VBLAST systems –.Source: 2) Linear ZF Detector · pp. 15–16

3) Linear MMSE Detector:

  • MMSE detection chooses the linear transformation minimizing mean-square error and jointly balances multiuser-interference suppression against noise enhancement.The detector therefore outperforms ZF at low SNRs.Source: 3) Linear MMSE Detector · p. 16
  • The MMSE detector originated in multiplexed PAM systems and was extended to multiple-channel, CDMA, and multi-antenna systems, with performance analyzed through output SINR and related metrics,,,, –, –.These analyses include error probability, outage probability, diversity-multiplexing tradeoff, and asymptotic SINR distributions.Source: 3) Linear MMSE Detector · pp. 16–17

4) Other Linear Detectors:

  • The MAME detector targets vanishing-noise bit-error performance and requires known fixed user energies, unlike ZF,.Asymptotic multiuser efficiency measures high-SNR performance loss caused by interference.Source: 4) Other Linear Detectors · p. 17
  • WLS removes the transmitted-symbol bias of ZF and MMSE, while MBER can outperform MMSE under high signature cross-correlation or non-Gaussian noise,,.MBER was first considered for multiplexing with ISI and ICI.Source: 4) Other Linear Detectors · p. 17
  • Many linear detectors assume known phase, power, and delay, requiring parameter estimation and receiver updates; linear detection can also be formulated as equalization because their mathematical models are similar,,.MIMO interference comes from simultaneous transmissions, whereas equalization interference comes from sequentially transmitted symbols.Source: 4) Other Linear Detectors · p. 17

C. Interference Cancellation Aided MIMO Detectors · D. Tree-Search Based MIMO Detectors

Interference-cancellation detectors improve on linear detection through nonlinear cancellation but trade complexity and error propagation against performance. Tree-search detectors offer flexible complexity–ML-performance tradeoffs, although conventional sphere decoding can remain exponentially complex while fixed-complexity variants provide near-ML alternatives.

C. Interference Cancellation Aided MIMO Detectors

  • Interference-cancellation detectors generally outperform linear detectors, but higher complexity and decision-error propagation prevent reliable attainment of optimum ML performance.They include SIC, PIC, MIC, and DFD variants, with benefits especially pronounced without channel coding and when interfering users are stronger than the desired user.Source: C. Interference Cancellation Aided… · pp. 17–18
  • SIC detects and subtracts symbols sequentially, making detection ordering critical and rendering the method effective for unequal-power streams but sensitive to error propagation.Its iterations and reordering increase linearly with the number of transmitted symbols, increasing complexity and processing delay in high-dimensional systems; accurate power control can mitigate degradation.Source: C. Interference Cancellation Aided… · pp. 18–19
  • PIC detects all symbols simultaneously and iteratively cancels interference, reducing processing delay and inter-stream error propagation relative to SIC while favoring similar-power signals.PIC has inferior near-far resistance to SIC, which performs better for different-power streams.Source: C. Interference Cancellation Aided… · p. 19
  • MIC cascades multiple detection stages for successive interference cancellation, whereas DFD combines SIC with optimized feedforward and feedback filters and detection ordering.MIC can begin with SUMF, ZF/MMSE, SIC, or another suboptimal detector; DFD was developed for synchronous and asynchronous CDMA systems,,,,.Source: C. Interference Cancellation Aided… · p. 19

D. Tree-Search Based MIMO Detectors

  • Tree-search detectors became prominent because sphere decoding and related methods flexibly trade computational complexity against near-ML or ML performance.Their development was propelled by the SD algorithm’s application to multi-antenna detection and by results on the closest lattice-point problem,.Source: D. Tree-Search Based MIMO Detectors · pp. 19–20
  • Tree-search methods span depth-first, breadth-first, and best-first variants, extending earlier CDMA and trellis-decoding approaches,, –, –.The SE refinement was reported more efficient than VB sphere decoding, while improved sphere-decoding algorithms reduced complexity relative to VB-SD and SE-SD.Source: D. Tree-Search Based MIMO Detectors · pp. 20–21
  • At fixed SNR, sphere-decoding expected complexity can remain exponential, O(M^(βN_t)), making conventional tree-search detection inefficient at low SNR or with many inputs.This contradicts claims that expected complexity is often roughly cubic across broad SNR and antenna ranges [58],, –, because β ∈ (0,1] depends on SNR,.Source: D. Tree-Search Based MIMO Detectors · p. 21
  • FCSD achieves near-ML performance with complexity O(M^√N_t), independent of SNR, offering a fixed-complexity alternative suited to efficient hardware implementation.It was proposed to avoid the varying-complexity characteristic of conventional tree-search detectors,.Source: D. Tree-Search Based MIMO Detectors · p. 21

E. Lattice-Reduction Aided Detectors · F. Probabilistic Data Association Based Detectors

Lattice-reduction preprocessing improves MIMO channel quality by finding near-orthogonal bases, while probabilistic data association offers low-complexity detection through iterative Gaussian approximation. Across both detector classes, the paper emphasizes practical complexity–performance tradeoffs and extensions for diverse MIMO settings.

E. Lattice-Reduction Aided Detectors

  • LR preprocessing transforms the channel basis toward orthogonality, improving the quality of MIMO channel matrices while avoiding the prohibitive cost of optimal basis search.The transformed basis uses a unimodular matrix whose real and imaginary entries are integers and whose determinant is ±1 or ±j.Source: E. Lattice-Reduction Aided Detectors · p. 21
  • The LLL algorithm is the predominant practical LR method because it finds a basis within a factor of optimality in polynomial time, despite unbounded worst-case complexity.Real- and complex-valued LLL variants target the corresponding MIMO system models.Source: E. Lattice-Reduction Aided Detectors · p. 21
  • LR can be combined with linear and nonlinear ZF or MMSE detectors, yielding substantial performance gains with little added complexity and enabling full receive diversity in MIMO decoding.LR has also been applied to precoded systems, soft-decision detection, VLSI implementations, and large-scale MIMO linear detectors.Source: E. Lattice-Reduction Aided Detectors · pp. 21, 23

F. Probabilistic Data Association Based Detectors

  • PDA detection originated in radar tracking and combines candidate associations into statistically likely updates, while JPDA extends this framework to joint decisions for multiple targets.The approach has also been used in sonar, sensor networks, navigation, and computer vision.Source: F. Probabilistic Data Association… · pp. 23–24
  • PDA provides a reduced-complexity alternative to optimal soft-decision MAP detection, achieving near-optimal performance in early CDMA applications and extending to asynchronous CDMA and QAM-based SDM-MIMO.RPDA was formulated for M-QAM using a real-valued model, while A-CPDA approximated complex Gaussian statistics using matched mean and covariance.Source: F. Probabilistic Data Association… · pp. 24–25
  • CPDA outperformed RPDA and A-CPDA by using pseudo-covariance to fully characterize the complex-valued Gaussian distribution.PDA performance depends strongly on the accuracy of the iterative Gaussian approximation to multimodal interference-plus-noise.Source: F. Probabilistic Data Association… · p. 25
  • Subsequent PDA developments improved approximation and receiver design through joint multi-symbol detection, direct bit-based QAM detection, Bayesian iterative receivers, and distributed soft combining.For FEC-coded MIMO, conventional PDA outputs were identified as normalized symbol likelihoods rather than true APPs, motivating approximate and exact Bayesian receivers.Source: F. Probabilistic Data Association… · p. 25

MiNI · G. Semidefinite Programming Relaxation Based Detectors

The merged sections review PDA-based and semidefinite-programming-relaxation MIMO detectors, emphasizing their operating principles, performance conditions, complexity, and modulation-dependent limitations. SDPR detectors offer polynomial-time worst-case complexity and strong BPSK/QPSK performance, but high-order QAM remains challenging.

MiNI

  • PDA detectors are inherently soft-input soft-output, supporting integration with convolutional, turbo,, and LDPC codes,.Source: MiNI · p. 25
  • PDA performance improves as transmit antennas or users increase when the channel is neither overloaded nor rank-deficient, although its approximation-and-iteration behavior remains insufficiently understood.Source: MiNI · p. 26
  • PDA-based detectors approximate interference-plus-noise distributions with Gaussian mixtures, illustrated using a four-modal 4PAM-like example for a (2 × 2)-element VBLAST system.Source: MiNI · p. 26

G. Semidefinite Programming Relaxation Based Detectors

H. Detection in Rank-Deficient and Overloaded MIMO Systems · I. Impact of Soft-Decision and Transmit Preprocessing on MIMO Detection · J. Guidelines on Choosing the Right MIMO Detectors

The sections address detection under rank deficiency or overload, receiver designs enhanced by soft decisions and transmit preprocessing, and criteria for selecting detectors according to performance, complexity, and operating conditions.

H. Detection in Rank-Deficient and Overloaded MIMO Systems

  • Rank-deficient channels arise from antenna correlation or keyhole effects, reducing spatial degrees of freedom and MIMO capacity, even when channel entries are uncorrelated.Keyhole effects can produce rank(H) = 1 despite uncorrelated channel entries –.Source: H. Detection in Rank-Deficient and… · p. 28
  • In rank-deficient or overloaded systems, standard ZF/MMSE, SIC, PDA, and sphere detectors often perform unacceptably because they solve under-determined noisy systems.Scenario-dependent alternatives include pseudo-inverse, group, generalized sphere-decoding, modified PDA or SDPR, metaheuristic, and search-based detectors,,,,,,,,, –.Source: H. Detection in Rank-Deficient and… · p. 29

I. Impact of Soft-Decision and Transmit Preprocessing on MIMO Detection

  • FEC, space-time coding, and precoding or beamforming shift recovery effort toward transmission, while orthogonal or quasi-orthogonal STBCs and ZF precoding simplify detection by reducing interference,, –.The resulting encoder or precoder can improve performance or reduce decoder and detector complexity.Source: §I · p. 29
  • Soft-input soft-output detectors support turbo-processing IDD receivers, with MAP, EM, MMSE, sphere-decoding, PDA, SDPR, and LS-MIMO iterative designs represented in prior work,, –.EXIT charts analyze iterative convergence and assist near-capacity wireless-system design,.Source: §I · p. 29

J. Guidelines on Choosing the Right MIMO Detectors

  • Detector choice has no universal optimum because goodness criteria and performance-complexity profiles depend on application assumptions; MAP/ML offers optimal VER but exponential O(M^Nt) complexity.MAP/ML is insensitive to channel imperfections and rank deficiency or overloading, with best error-probability performance across SNR, but exact implementation is practical only for modest dimensions.Source: J. Guidelines on Choosing the Right… · pp. 29, 31
  • Linear detectors minimize complexity, while interference-cancellation detectors improve error probability; MF, ZF, and MMSE suit noise-dominated, interference-dominated, and balanced conditions respectively, with rank-deficient limitations.Their typical complexity ranges are O(N_t^2)–O(N_t^3) for linear detectors and O(N_t^3)–O(N_t^4) for interference cancellation.Source: J. Guidelines on Choosing the Right… · p. 31
  • Tree-search detectors trade complexity for near-ML performance but have nondeterministic, potentially exponential complexity, low-SNR weakness, and poor suitability for large-scale systems with many inputs.LR preprocessing offers polynomial complexity and improves effective-channel quality without changing baseline detector tradeoffs, whereas SDPR/PDA and metaheuristics remain scenario-dependent options.Source: J. Guidelines on Choosing the Right… · p. 31

IX. DETECTION IN LS-MIMO SYSTEMS

LS-MIMO systems exploit large antenna arrays to provide high spatial diversity, reliability, and throughput without increasing bandwidth, while introducing deployment and detection challenges that require adapting conventional MIMO insights. Their detection problems remain intrinsically large-scale, but random-matrix-theoretic properties can enable distinct approaches.

LS-MIMO benefits

  • MIMO reliability can improve exponentially with N_t and/or N_r, while achievable rate scales linearly with min{N_t, N_r} log2(1 + SNR).These gains provide higher spatial degrees of freedom without increasing occupied bandwidth.Source: §IX · p. 32

Deployment scenarios

  • LS-MIMO can use hundreds of BS antennas to serve dozens of multi-antenna users, but point-to-point deployments require spatial separation and high SNR for multiplexing gains.High-frequency implementations may have limited coverage, making indoor and small-cell scenarios particularly suitable.Source: §IX · p. 32

Antenna configurations

  • Linear, rectangular, cylindrical, and spherical arrays trade physical area and complexity, while distributed arrays can enhance indoor coverage or outdoor cooperation.Linear arrays radiate in two dimensions; the other compact configurations support three-dimensional directions.Source: §IX · p. 32, Figure 26 · p. 32

Detection motivation

  • Optimal ML/MAP detection already has exponentially increasing complexity with problem size, making LS-MIMO detection intrinsically large-scale rather than wholly novel.Its practical importance is amplified because earlier research largely focused on small-scale MIMO when large arrays were considered impractical.Source: §IX · p. 32
  • Random matrix theory gives LS-MIMO distinct advantages that may circumvent signal-processing difficulties, so small-scale detection insights must be carefully adapted.The section therefore categorizes LS-MIMO detection according to application scenarios.Source: §IX · pp. 32–33

A. Detection in Single-Cell/Noncooperative Multi-Cell LS-MIMO Systems · B. Detection in Cooperative Multi-cell Multiuser LS-MIMO Systems · C. Applicability of Existing MIMO Detection Algorithms to LS-MIMO

The paper distinguishes LS-MIMO detection by antenna configuration and cooperation: Type-II systems can approach near-optimal detection with simple linear methods, whereas Type-I systems require complexity-scalable algorithms. Channel hardening motivates these simplifications, but pilot contamination and cooperation backhaul remain important limitations.

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

  • As Type-I dimensions grow together, channel singular values become deterministic and diagonal dominance produces channel hardening, enabling approximate inversion and low-complexity detection.The approximation can support ZF-, MMSE-, and PDA-aided detectors through series expansions [401], while preserving good large-scale-MIMO performance.Source: A. Detection in… · p. 33
  • In Type-II LS-MIMO, high receive diversity, asymptotic user orthogonality, and favorable conditioning enable even MF detection to achieve near-optimum performance,.Increasing BS antennas also mitigates fading and uncorrelated noise, helping compensate for low SNR or poor channel estimates,,,.Source: A. Detection in… · pp. 33–34
  • Noncooperative multi-cell LS-MIMO remains interference-limited because pilot contamination from reused adjacent-cell pilots becomes the major performance constraint,.BS cooperation and multicell joint processing are therefore proposed to enhance achievable performance.Source: A. Detection in… · p. 34, C. Applicability of Existing MIMO… · p. 34

B. Detection in Cooperative Multi-cell Multiuser LS-MIMO Systems

  • Cooperative multi-cell detection centralizes distributed BSs into a physical or virtual super-BS, providing an upper bound through interference cancellation or data fusion.Cooperation can use high-capacity optical-fiber or microwave links, but substantially increases backhaul traffic,,.Source: B. Detection in Cooperative Multi-cell… · p. 34

C. Applicability of Existing MIMO Detection Algorithms to LS-MIMO

  • Most existing detectors can serve Type-II LS-MIMO, where low-complexity linear detection may be near-optimal and sophisticated sphere decoding can become unnecessary.This applicability depends on identifying the LS-MIMO type before selecting a detector.Source: C. Applicability of Existing MIMO… · p. 34
  • Tree-search detectors such as SD become less feasible for Type-I systems because their worst-case complexity grows exponentially with the number of transmit antennas.SD may remain usable in Type-II systems, whose antenna configuration is more favorable for low-complexity detection.Source: C. Applicability of Existing MIMO… · p. 34
  • PDA offers Type-I systems an attractive performance-complexity tradeoff through polynomial-time stochastic interference modeling, while convex SDPR detectors may also apply [130]–,.Both approaches avoid the exponential worst-case scaling associated with tree-search detection.Source: C. Applicability of Existing MIMO… · p. 34

D. Recent Advances in LS-MIMO Detection · E. Applications of MIMO Detection Techniques in Other Areas · X. SUMMARY AND CONCLUSIONS

Recent LS-MIMO detection advances span asymptotically analyzed receivers, local-search and message-passing methods, soft-output algorithms, and hardware implementations, while MIMO detection also extends to cognitive radio, radar, and optical communications. The survey concludes that detector choice must follow LS-MIMO type, with several low-complexity families offering promising performance–complexity tradeoffs.

D. Recent Advances in LS-MIMO Detection

  • Asymptotic analyses established that PDA detectors approach optimal performance, while BI-GDFE and MMSE-SIC can approach the single-user MFB at sufficiently high SNR.The results rely on Gaussian MUI approximations and random matrix theory, including deterministic and asymptotically Gaussian SINR characterizations.Source: D. Recent Advances in LS-MIMO Detection · pp. 34–35
  • Local-search detectors extended LS-MIMO detection to systems with up to 600 transmit and receive antennas, with LAS attaining near-SISO AWGN performance at O(N_tN_r) average per-bit complexity.RTS improves on LAS through an efficient local-minima exit strategy, while BP, R-MCMC, randomized search, Gibbs-sampling, and LR-based detectors were also studied.Source: D. Recent Advances in LS-MIMO Detection · p. 35
  • Low-complexity SUMIS and approximate-message-passing detectors provide soft-input soft-output LS-MIMO detection, while non-binary LDPC systems can achieve near-capacity performance with MMSE detection.These developments broaden iterative and coded LS-MIMO detection beyond hard-output local-search methods.Source: D. Recent Advances in LS-MIMO Detection · p. 35
  • The first ASIC LS-MIMO detector using truncated Neumann expansion achieves 3.8 Gb/s with 128 BS antennas and 8 users in LTE-A.This provides a concrete hardware demonstration of scalable LS-MIMO detection.Source: D. Recent Advances in LS-MIMO Detection · p. 35

E. Applications of MIMO Detection Techniques in Other Areas

  • MIMO detection techniques extend beyond wireless communications to MIMO cognitive radio, MIMO radar, and mode-division-multiplexing multimode-fiber optical communications.Cognitive radio combines MIMO with software-defined radio, software-defined networking, and envisaged NFV-based 5G architectures.Source: E. Applications of MIMO Detection… · p. 36

X. SUMMARY AND CONCLUSIONS

  • The survey identifies LS-MIMO detection as a paradigm shift and emphasizes selecting detector strategies according to whether the system is Type-I or Type-II.Sphere-decoding-based detectors may be infeasible for Type-I systems, whereas low-complexity linear detectors may be near-optimal for Type-II systems.Source: §X · p. 36
  • Local-neighborhood metaheuristics, Bayesian message passing, and convex optimization offer promising performance–complexity tradeoffs for LS-MIMO detector design.The conclusion draws these lessons from the survey’s historical review and analysis of LS-MIMO detection strategies.Source: §X · p. 36

Related papers

Loading 1507.05138v1…