Source-linked AI summary
Capacity Limits and Multiplexing Gains of MIMO Channels with Transceiver Impairments
Emil Björnson, Per Zetterberg, Mats Bengtsson, Björn Ottersten
TL;DR
Ideal MIMO channels have a high-SNR capacity slope equal to min(Nt,Nr), but this letter asks whether that scaling survives physical transceiver distortions. Modeling those impairments analytically, it proves finite high-SNR capacity for any channel distribution, zero classic multiplexing gain, and at least the ideal-transceiver relative MIMO gain.
Problem
The paper examines whether ideal MIMO high-SNR multiplexing gains persist when physical transceiver impairments affect systems of any size.
Method
The letter analyzes a generalized MIMO channel with transmitter distortion modeled from physical transceiver impairments and derives its capacity analytically.
Results
The capacity has a finite high-SNR limit for any channel distribution, so the classic multiplexing gain is zero.
Takeaways & Limitations
Despite zero classic multiplexing gain, physical MIMO systems retain capacity growth roughly linear in min(Nt,Nr), and their relative gain is at least as large as with ideal transceivers.
Takeaways & Limitations
The point-to-point capacity limit is an upper bound when extra constraints such as distributed power and limited coordination apply, including network MIMO.
Abstract
from arXiv · showhide
The capacity of ideal MIMO channels has a high-SNR slope that equals the minimum of the number of transmit and receive antennas. This letter analyzes if this result holds when there are distortions from physical transceiver impairments. We prove analytically that such physical MIMO channels have a finite upper capacity limit, for any channel distribution and SNR. The high-SNR slope thus collapses to zero. This appears discouraging, but we prove the encouraging result that the relative capacity gain of employing MIMO is at least as large as with ideal transceivers.
I. INTRODUCTION
Ideal MIMO capacity scales with the minimum number of transmit and receive antennas, but physical transceiver impairments fundamentally affect spectral efficiency and produce a finite high-SNR capacity limit. Despite zero classic multiplexing gain, the relative capacity gain from MIMO is at least as large as with ideal transceivers.
- MIMO capacity in the ideal high-SNR regime scales as M log2(SNR)+O(1), where M=min(Nt,Nr) is the multiplexing gain.
- Finite channel coherence time can limit channel acquisition and coordination, creating a spectral-efficiency ceiling regardless of power or antenna count.
- Physical RF transceivers exhibit impairments including amplifier nonlinearities, IQ imbalance, phase noise, quantization noise, and frequency or sampling offsets.
- The letter shows that transceiver impairments impose a finite high-SNR capacity limit for any channel distribution, collapsing the multiplexing gain to zero.
- The relative capacity increase of MIMO over single-antenna channels is at least as large with physical transceivers as with ideal transceivers.
II. GENERALIZED CHANNEL MODEL
The generalized model extends the classical flat-fading MIMO channel by adding transmitter distortion caused by physical impairments. Distortion is modeled as uncorrelated Gaussian noise whose variance depends on transmit power and can include multi-carrier leakage.
- The channel has Nt transmit antennas, Nr receive antennas, random full-rank H, normalized expected gain, and circular-symmetric Gaussian thermal noise.
- Ideal hardware models only multiplicative channel transformation and additive thermal noise, whereas physical hardware leaves residual distortion after compensation.
- Transmitter distortion ηt represents the mismatch between the intended and actually radiated signals and is modeled as uncorrelated Gaussian noise.
- Under tr(Q)=1 with Q=E{xxH}, distortion variance increases with each antenna's signal power and neglects antenna cross-correlation.
- The parameter α∈[0,1] interpolates between one and many subcarriers, while κ>0 sets impairment level; the model is approximately linear in the assumed dynamic range.
III. ANALYSIS OF CHANNEL CAPACITY
The analysis derives capacity expressions and shows that physical MIMO channels have finite high-SNR capacity bounds, characterized by antenna counts and impairment level. The bounds are tight under isotropic channel distributions and deterministic full-rank channels with suitable covariance adaptation.
- Capacity expression: The capacity is optimized by Gaussian signaling x ∼ CN(0, Q) over feasible covariance matrices Q ⪰ 0.For fixed channel realization and SNR, the impaired channel is equivalent to a classical MIMO channel with an impairment-dependent noise covariance.
- Finite high-SNR limit: The asymptotic capacity limit is finite for any channel distribution and depends only on the antenna numbers and impairment level κ.This contrasts with the unbounded high-SNR capacity of ideal transceivers.
- Capacity bounds: The lower and upper capacity bounds coincide when Nt ≤ Nr, whereas only the upper bound grows with Nt when Nt > Nr.The bounds are tight when the high-SNR covariance is isotropic over appropriate transmit subspaces.
- Capacity bounds: For right-rotationally invariant channel distributions, isotropic covariance Q = 1/Nt I achieves capacity and makes the lower bound asymptotically tight.This includes uncorrelated transmit-side Rayleigh fading and other isotropically directed channel distributions.
- Capacity bounds: For deterministic full-rank channels, covariance adaptation along the nonzero eigenvectors achieves the upper bound asymptotically for any Nt.The capacity-achieving covariance uses the eigendecomposition of H^H H and waterfilling over its nonzero modes.
M UMUH
The analysis establishes that isotropic signaling remains asymptotically achievable in the worst case when the transmitter lacks channel-distribution knowledge. It also connects impairment-induced saturation with selective transmission and covariance design in deterministic channels.
- M UMUH: For deterministic channels with Nt ≥ Nr, increasing Nt improves the capacity limit because selective transmission uses the Nr nonzero channel dimensions while distortion remains isotropic across Nt dimensions.The same optimal waterfilling allocation applies as with ideal transceivers.
- M UMUH: When the channel distribution is unknown at the transmitter, Q = 1/Nt I maximizes the worst-case mutual information.This guarantees asymptotic achievability of the lower capacity bound without distribution knowledge.
A. Numerical Illustrations
Numerical examples show that ideal and impaired transceivers agree at low and medium SNR but diverge at high SNR, where impaired capacity saturates. The saturation value depends on impairment level and channel structure, while a medium-SNR DoF regime can still appear.
- A. Numerical Illustrations: The asymptotic difference between synthetic uncorrelated and measured correlated channels vanishes, so impairment level κ determines the capacity limit in the Nt = Nr = 4 example.The comparison uses κ ∈ {0.05, 0.1}.
- A. Numerical Illustrations: When Nt increases with Nr = 4, deterministic-channel capacity limits increase, whereas random-channel limits remain unchanged and convergence becomes faster.The deterministic cases also have larger limits for α = 1 than for α = 0 because α makes distortion more isotropic.
- A. Numerical Illustrations: A medium-SNR range can exhibit roughly the ideal M-slope before the high-SNR saturation regime, demonstrating a DoF regime for physical MIMO channels of any size.This behavior arises from transceiver impairments rather than only from large-network coherence-time limitations.
IV. GAIN OF MULTIPLEXING
Physical transceiver impairments make high-SNR capacity saturate, so the classical multiplexing-gain definition collapses even though multiple antennas can still provide substantial relative capacity gains over SISO.
- Asymptotic behavior: Physical MIMO channels have a finite upper capacity bound, unlike ideal channels whose capacity grows as M log2(SNR) + O(1).The resulting high-SNR multiplexing slope is therefore zero under the classical asymptotic definition.
- Definition: The finite-SNR multiplexing gain M(SNR) is defined as the ratio of MIMO to SISO capacity at a given SNR.This ratio measures the capacity improvement from spatial multiplexing.
- Bounds: Theorem 2 bounds finite-SNR multiplexing gain using channel norms, with upper bounds for full-rank deterministic channels and lower bounds for right-rotationally invariant distributions.The upper bounds are achieved with α = 1, while the lower bounds apply to right-rotationally invariant channel distributions.
- Proof strategy: At low SNR, the finite-SNR multiplexing gain is characterized using Taylor approximation and isotropic signaling, while high-SNR behavior follows from the capacity-limit theorem.The per-realization-optimal covariance uses the dominant eigenvector of H H^H for the upper bound.
- Practical interpretation: Physical systems can achieve M(SNR) > M at high SNR, whereas ideal transceivers achieve only M, despite the physical channel’s finite capacity limit.Thus, the classical high-SNR slope does not fully capture the practical capacity benefit of multiple antennas.
A. Numerical Illustrations
Simulations confirm the theoretical finite-SNR multiplexing-gain limits for Rayleigh and deterministic channels, revealing similar gains for ideal and impaired transceivers but faster impaired-channel convergence at high SNR.
- Numerical setup: The simulations use uncorrelated Rayleigh fading and deterministic channels with Nt ∈ {4, 8, 12}, Nr = 4, κ = 0.05, and α = 1.These settings correspond to the numerical evaluations of finite-SNR multiplexing gain.
- Results: The simulations confirm Theorem 2’s limits and show remarkably similar finite-SNR multiplexing gains for physical and ideal transceivers.The similarity follows because the asymptotic limits are nearly the same across impairment levels.
- High-SNR behavior: At high SNR, impaired channels converge faster to their limits, while deterministic channels achieve an asymptotic gain higher than M when Nt > Nr.The latter effect occurs in the deterministic-channel case rather than generally across all channel distributions.
V. CONCLUDING REMARKS
Physical MIMO capacity saturates at high SNR because transceiver-distortion power scales with signal power, collapsing the classic multiplexing gain to zero. Nevertheless, capacity grows roughly linearly with M = min(Nt, Nr) over the SNR range, while additional constraints can lower finite-SNR multiplexing gain.
- High-SNR capacity saturates because transceiver-distortion power is proportional to signal power, so the classic multiplexing gain becomes zero.
- Capacity nevertheless grows roughly linearly with M = min(Nt, Nr) across the SNR range, enabling substantial gains from MIMO and spatial multiplexing.
- The point-to-point capacity limit upper-bounds systems with extra constraints such as network MIMO, whose capacity also saturates at high SNR.
- Additional constraints decrease finite-SNR multiplexing gain, while impairments limit the asymptotic accuracy of channel-acquisition schemes.
- For Nt ≤ Nr, the upper and lower capacity bounds coincide, making Q = 1 asymptotically optimal.