Ch. 10 — MIMO Communications

Chapter 10 of Wireless Communications (2nd ed. Draft) — Andrea Goldsmith

Abstract

This chapter establishes the theoretical foundations and practical architectures for Multiple-Input Multiple-Output (MIMO) wireless communication systems. It analyzes channel capacity under varying assumptions of Channel State Information (CSI) availability, specifically contrasting scenarios with and without Transmitter/Receiver CSI. The core technical contribution is the formulation of the fundamental trade-off between diversity gain and multiplexing gain, alongside the presentation of space-time coding and detection algorithms necessary to realize MIMO benefits in fading environments.

Key Concepts

• Outage Capacity without Transmitter CSI: In fading channels where the transmitter lacks CSI, the transmission rate must be fixed, resulting in an outage probability if the channel capacity falls below the rate. The capacity with outage is determined by the tail of the probability distribution of the instantaneous capacity, which can be improved by disabling certain transmit antennas to reduce spectral spreading when outage probability is high.
• MIMO Beamforming: This strategy utilizes array and diversity gain by transmitting a single symbol weighted by complex factors across all transmit antennas, creating a rank-one input covariance matrix. Optimal performance requires knowledge of the channel matrix $\mathbf{H}$ at both ends to align transmit and receive weight vectors with the principal left and right singular vectors, maximizing the received SNR to ${\sigma }_{\max }^2\rho$.
• Diversity–Multiplexing Trade-off (DMT): A fundamental design constraint defining the relationship between data rate (multiplexing gain $r$) and reliability (diversity gain $d$). For block fading channels with high SNR and sufficient block length, the optimal trade-off curve is given by $d_{opt}(r)=(M_t-r)(M_r-r)$, illustrating that maximizing rate comes at the expense of error performance.
• Maximum Likelihood (ML) Detection: The optimal receiver algorithm minimizes the probability of symbol detection error by searching over all possible transmitted symbol vectors $\mathbf{x}$ to minimize the Euclidean distance $\Vert \mathbf{y}-\mathbf{Hx}{\Vert }^2$. While performance-optimal, it suffers from prohibitive complexity exponential in the number of transmit antennas, necessitating suboptimal approximations for practical systems.
• Linear Receivers (ZF and MMSE): To reduce complexity, Zero-Forcing (ZF) and Linear Minimum Mean-Square Error (L-MMSE) receivers spatially decorrelate channels via matrix inversion. ZF eliminates Inter-Symbol Interference (ISI) but amplifies noise when the channel matrix is ill-conditioned, whereas L-MMSE regularizes the inversion to better balance interference cancellation and noise enhancement.
• Sphere Decoding (SD): An algorithm that approximates ML performance with reduced complexity by restricting the search space to lattice points within a hypersphere of radius $r$. By leveraging the upper triangular nature of the QR-decomposition of $\mathbf{H}$, SD prunes candidate solutions via depth-first search, trading search radius for complexity performance.
• Space-Time Code Design Criteria: Reliable communication over MIMO fading channels requires codes that satisfy the rank and determinant criteria to maximize diversity and coding gain. The rank criterion dictates that the difference matrix $\Delta$ between any two codewords must have full rank $M_t$, while the determinant criterion requires maximizing the minimum geometric mean of the nonzero eigenvalues of $\Delta$.
• BLAST Architectures: Bell Labs Layered Space-Time algorithms, such as V-BLAST and D-BLAST, implement spatial multiplexing using layered encoding. V-BLAST transmits independent streams from each antenna using successively cancel interference, while D-BLAST rotates stream symbols across antennas to achieve full diversity with linear decoding complexity, albeit with some efficiency loss.
• Smart Antennas: Distinct from MIMO multiplexing, smart antennas utilize directional gain via sectorization or phased arrays to suppress interference and increase range. This approach introduces a trade-off between increasing data rates through multiplexing, robustness through diversity, and interference reduction through directionality, often requiring complex tracking of signal angular locations.
• Frequency-Selective MIMO Channels: When bandwidth exceeds the channel coherence bandwidth, MIMO systems experience Intersymbol Interference (ISI) across spatial and temporal dimensions. Effective mitigation combines spatial processing with time-domain equalization or Orthogonal Frequency Division Multiplexing (OFDM) to convert the channel into parallel narrowband MIMO channels.

Key Equations and Algorithms

• Beamforming Received SNR: $\gamma ={\sigma }_{\max }^2\rho$. This expression calculates the Signal-to-Noise Ratio achieved when optimal precoding vector $\mathbf{v}$ and shaping vector $\mathbf{u}$ align with the maximum singular value ${\sigma }_{\max }$ of the channel matrix $\mathbf{H}$, effectively reducing the MIMO system to a SISO channel.
• Beamforming Capacity: $C=B{\log }_2(1+{\sigma }_{\max }^2\rho )$. This equation defines the spectral efficiency of a beamforming scheme under perfect CSI, showing that capacity is bounded by the strongest spatial mode rather than the sum of all parallel modes available in the singular value decomposition.
• DMT Multiplexing Definition: $R(\text{SNR})\dot{=}{\text{SNR}}^r$. This asymptotic definition characterizes the transmission rate in the high-SNR regime, where $r$ represents the multiplexing gain, indicating how many independent data streams can be supported relative to the SNR logarithm.
• DMT Diversity Definition: $P_e(\text{SNR})\dot{=}{\text{SNR}}^{-d}$. This expression defines the diversity gain $d$ in the high-SNR regime, describing the slope of the probability of error curve as a function of SNR on a log-log scale.
• Optimal DMT Limit: $d_{opt}(r)=(M_t-r)(M_r-r)$. This specific result for block fading channels with $T\ge M_t+M_r-1$ quantifies the exact achievable diversity gain for a given multiplexing gain $r$, establishing the boundary of MIMO performance trade-offs.
• ML Detection Objective: $\hat{\mathbf{x}}=\arg {\min }_{\mathbf{x}\in {\mathcal{X}}^{M_t}}\Vert \mathbf{y}-\mathbf{Hx}{\Vert }^2$. This minimization problem defines the maximum likelihood estimator, which searches the discrete constellation space $\mathcal{X}$ to find the vector $\mathbf{x}$ closest to the received vector $\mathbf{y}$ in Euclidean distance.
• ZF Receiver Solution: ${\mathbf{H}}^{†}=({\mathbf{H}}^H\mathbf{H}{)}^{-1}{\mathbf{H}}^H$. This pseudo-inverse formula represents the equalization matrix used to cancel channel effects when $M_t\le M_r$, but it is prone to noise amplification if the condition number of $\mathbf{H}$ is large.
• L-MMSE Receiver Solution: $\mathbf{A}=({\mathbf{H}}^H\mathbf{H}+\frac{M_t}{\rho }\mathbf{I}{)}^{-1}{\mathbf{H}}^H$. This equation provides the optimal linear estimator minimizing mean square error by incorporating the noise variance ${\sigma }^2$ (represented by $M_t/\rho$) into the regularization term to prevent singular matrix inversion.
• Sphere Decoding Constraint: $\Vert \mathbf{y}-\mathbf{Hx}{\Vert }^2\le r^2$. This inequality defines the search criterion for the Sphere Decoder, where $r$ is a radius parameter that controls the trade-off between computational complexity and proximity to the ML solution; $r\to \infty$ recovers ML.
• Pairwise Error Probability Bound: $P(\mathbf{X}\to \hat{\mathbf{X}})\le {\prod }_{k=1}^{R_{\Delta }}({\lambda }_k(\Delta )\gamma {)}^{-M_r}$. This upper bound on the probability of mistaking codeword $\mathbf{X}$ for $\hat{\mathbf{X}}$ depends on the rank $R_{\Delta }$ and eigenvalues ${\lambda }_k(\Delta )$ of the difference matrix, driving the rank and determinant design criteria.

Key Claims and Findings

• Capacity Limits without CSI: In i.i.d. ZMSW block fading channels without transmitter or receiver CSI, increasing transmit antennas beyond the block duration does not increase capacity, preventing linear growth in data rate despite additional spatial dimensions.
• Beamforming Gain Limits: For frequency-selective fading models without CSI, capacity at very high SNRs grows double logarithmically rather than linearly, meaning there is no multiplexing gain associated with multiple antennas when the channel is unknown.
• Diversity-Multiplexing Constraint: It is impossible to achieve full diversity and full multiplexing gains simultaneously for finite blocklengths; a trade-off is mandatory where increasing the rate $r$ linearly reduces the achievable diversity order $d$.
• Linear vs. Nonlinear Detection: Linear receivers (ZF and MMSE) offer significantly lower complexity than Maximum Likelihood detection but introduce BER performance degradation due to incomplete interference cancellation or noise enhancement.
• Space-Time Code Efficiency: Space-Time Trellis Codes (STTCs) achieve both full diversity and coding gain but incur exponential decoding complexity, whereas Space-Time Block Codes (STBCs) like Alamouti provide full diversity with linear receiver complexity but lack coding gain.
• BLAST Diversity Orders: V-BLAST architectures achieve at most a diversity order of $M_r$ because each codeword is transmitted from a single antenna, whereas D-BLAST achieves full $M_t{M}_r$ diversity by spreading codewords across all transmit antennas via stream rotation.
• Smart Antenna Trade-offs: The use of directional antennas increases the signaling range and suppresses interference but requires angular location tracking which is impediment for mobile systems, favoring sectorized antennas in cellular base stations over dynamic phasing.

Terminology

• CSIT/CSIR: Acronyms for Channel State Information at the Transmitter and Receiver, respectively, which determine the achievable capacity and coding strategies; the chapter distinguishes results based on whether this information is available (e.g., no CSIT/CSIR scenarios).
• Beamforming: A technique where the input covariance matrix has unit rank, transmitting a single complex-weighted symbol across multiple antennas to achieve array and diversity gain through coherent combining at the receiver.
• Multiplexing Gain: A parameter $r$ representing the asymptotic data rate growth relative to the logarithm of the SNR, quantifying the number of independent data streams supported by the spatial channel dimensions.
• Diversity Gain: A parameter $d$ representing the asymptotic decay rate of the error probability relative to the SNR on a log-log scale, quantifying the system robustness against fading.
• Outage Probability: The probability that the instantaneous channel capacity falls below a fixed transmission rate $R$, requiring the system to define performance metrics like capacity with outage when CSI is unavailable at the transmitter.
• Space-Time Block Code (STBC): A class of space-time codes designed to achieve full diversity order using orthogonal code matrix structures that allow for linear complexity decoding at the receiver.
• Space-Time Trellis Code (STTC): A code constructed using a trellis diagram to map sequences of bits to space-time symbols, optimized using the rank and determinant criteria to achieve superior coding gain over STBCs.
• V-BLAST: Vertical BLAST, an architecture using parallel encoding where independent data streams are transmitted from separate antennas, typically employing successive interference cancellation for detection.
• D-BLAST: Diagonal BLAST, an architecture using diagonal encoding with stream rotation to transmit codeword symbols across different antennas over time, aiming to achieve full diversity with linear decoding complexity.
• Condition Number: A measure of the relative magnitudes of the singular values of the channel matrix $\mathbf{H}$, defined as the ratio of the largest to smallest singular value, indicating the susceptibility to noise amplification in ZF receivers.