Matched-Filter Precoding for MIMO

• Matched-filter precoding is a linear transmit strategy leveraging channel conjugation to maximize received power in multi-antenna systems.
• It underpins massive MIMO and network MIMO deployments due to its scalability, energy efficiency, and robustness to imperfect CSI.
• Its implementation requires only channel conjugation and scalar normalization, ensuring low computational complexity in large arrays.

Matched-filter (MF) precoding is a foundational linear transmit strategy for multi-antenna communication systems, notable for its extremely low computational complexity and analytic tractability. Sometimes referred to as conjugate beamforming or maximum-ratio transmission in the literature, matched-filter precoding leverages channel conjugation to maximize received signal power in the intended direction but does not directly suppress multi-user interference. It is widely adopted as a baseline in massive MIMO, network MIMO, and multi-user deployments for both its scalability and energy efficiency, especially in scenarios with imperfect Channel State Information at the Transmitter (CSIT) or hardware limitations (Mehana et al., 2012, Zhao et al., 2024, Lee et al., 2012, Shekar et al., 6 Oct 2025, Zhao et al., 2024).

1. Mathematical Definition and Structure

In point-to-point $M \times N$ MIMO channels with flat fading, the transmit MF precoder is given by:

$W_{MF} = \beta H^H$

where $H \in \mathbb{C}^{M \times N}$ is the channel matrix, $\beta = [\operatorname{tr}(HH^H)]^{-1/2}$ normalizes the output power, and $H^H$ is the Hermitian (conjugate transpose) of the channel (Mehana et al., 2012). For a multi-user MIMO (MU-MIMO) system with $L$ antennas and $K$ single-antenna users under Rayleigh fading,

$w_k = h_k^* / \|h_k\|$

where $h_k\in\mathbb{C}^L$ is user $k$'s channel, and $W = [w_1, ..., w_K]$. In the context of network MIMO and holographic MIMO, MF precoding generalizes naturally to include excitation and mutual coupling matrices, where the final transmit vector involves the conjugate of the end-to-end channel including all hardware and propagation effects (Shekar et al., 6 Oct 2025, Lee et al., 2012).

2. Performance Analysis: SINR and Rate Statistics

The instantaneous Signal-to-Interference-plus-Noise Ratio (SINR) for user $k$ under matched-filter precoding in the canonical MU-MIMO model is:

$$\mathrm{SINR}_k = \frac{P_t/K \,\|h_k\|^2}{\sigma_k^2 + (P_t/K) \sum_{i\ne k} \frac{|\langle h_k, h_i \rangle|^2}{\|h_i\|^2}}$$

where $P_t$ is the total transmit power, $\sigma_k^2$ is the noise variance, and the interference sum results from the non-orthogonality of users' channels in the transmit domain (Zhao et al., 2024, Lee et al., 2012).

The distribution of the SINR in Rayleigh fading is fully characterized by:

• $$Y = K \sigma_k^2 / [P_t \|h_k\|^2] \sim \mathrm{Inv}\text{-}\mathrm{Gamma}(L, a)$$
• Each normalized interference term $$X_i = |u_k^T h_i^*|^2/\|h_i\|^2\sim \mathrm{Beta}(1, L-1)$$.

The SINR can be exactly analyzed by considering the sum $Z = Y + \sum_{i\ne k} X_i$; the cumulative distribution function (CDF) is accessible via characteristic-function inversion. At high SNR, the SINR converges in distribution to the reciprocal of a Beta-sum, reflecting the interference-limited nature of MF in this regime (Zhao et al., 2024). In the massive MIMO limit ($L\to\infty, K$ fixed), the empirical interference sum becomes Gamma distributed, providing deterministic-equivalent results for rate (Zhao et al., 2024).

The ergodic rate, $R = \mathbb{E}[\log_2(1+\mathrm{SINR})]$, has asymptotic and second-order closed-form approximations that are accurate uniformly over SNR, reflecting fundamental limits for practical system design (Zhao et al., 2024). In practical network massive MIMO, sum-rate bounds under matrix and vector normalizations can be tightly approximated by Jensen bounds:

$$R_{MF} \approx \log_2\left(1 + \frac{P(M+1)}{P(K-1)+K}\right)$$

for $M$ antennas and $K$ users (Lee et al., 2012).

3. Diversity, Multiplexing, and Spectral Efficiency Limits

The Diversity-Multiplexing Tradeoff (DMT) for MF precoding in MIMO is flat: for any positive multiplexing gain $r>0$, the diversity is identically zero,

$d_{MF}(r) = 0 \quad \forall r > 0$

indicating the presence of an irreducible error floor at any nonzero data rate. At fixed spectral efficiency $R$ (corresponding to multiplexing gain $r=0$), a threshold emerges:

$R_{th} = N \log \left(\frac{N}{N-1}\right)$

such that full diversity $M N$ is achieved for $R < R_{th}$, but diversity collapses to zero for $R > R_{th}$. Thus, MF is only diversity-optimal at sufficiently low rates; at high rates, multi-stream interference is dominant (Mehana et al., 2012).

In massive MIMO, the ergodic rate converges almost surely to

$\ln\left(1 + \frac{c P_t}{P_t + \sigma^2}\right)$

where $c = L/K$ is the antennas-to-users ratio, showing MF precoding achieves nearly optimal scaling as arrays grow large (Zhao et al., 2024).

4. Normalization, Implementation, and Complexity

Implementations of MF precoding require only channel conjugation and scalar normalization—no matrix inversion is necessary. In "cloud base station" or distributed MIMO architectures, the correct normalization is critical:

• Matrix normalization (normalize the full precoding matrix Frobenius norm): always outperforms vector normalization for MF.
• Vector normalization (normalize each user-precoding vector individually): can be optimal for ZF, but is suboptimal for MF (Lee et al., 2012).

For matched-filter precoding, the recommendation is to always use matrix normalization, as it achieves the best possible sum-rate bound for any system parameters:

$$R_{\text{MF}} \approx \log_2 \left(1 + \frac{P(M+1)}{P(K-1)+K} \right)$$

(Lee et al., 2012).

A key mode-switching rule is given by the user "crossing point":

$K_{\rm cross} \approx \frac{P(M+1)}{1+P}$

If $K \geq K_{\rm cross}$, MF should be preferred; if $K < K_{\rm cross}$, ZF becomes optimal. This rule is especially relevant at low SNR or at the cell boundary (Lee et al., 2012).

Computational complexity is $\mathcal{O}(KL)$ per symbol vector, and no matrix inversion or NP-hard optimization is involved, in stark contrast to ZF and MMSE/RZF (Mehana et al., 2012, Zhao et al., 2024).

5. Robustness, Channel State Information, and Practical Regimes

MF precoding is highly robust to imperfect CSIT. Analysis in holographic and massive MIMO regimes demonstrates that MF precoding gracefully degrades as channel estimation quality decreases, whereas more interference-cancelling approaches like ZF or beamforming relying on precise nulling may collapse (Shekar et al., 6 Oct 2025, Zhao et al., 2024). In partial or no-CSI cases, approximate analytic expressions for the average throughput are available:

$\bar{R}_k \approx \ln(1 + \bar{\Sigma})$

with $\bar{\Sigma}$ determined by channel statistics and mutual coupling for holographic arrays, and analytic closed-forms exist for both partial and no-CSI regimes (Shekar et al., 6 Oct 2025).

In the noise-limited regime (low SNR), MF closely matches the optimal beamformer; at higher SNR, MF performance saturates due to unmitigated interference, underscoring its utility for low and moderate SNR, or in hardware-constrained contexts, or where CSI is costly (Shekar et al., 6 Oct 2025).

6. Extensions: RSMA, Array Architectures, and Design Guidelines

Recent work extends MF precoding to advanced access protocols such as Rate-Splitting Multiple Access (RSMA), where a single MF precoder suffices for both common and private streams, eliminating the need for separate MRT or MVDR designs for broadcasting. Notably, in the large-system regime, MF-precoded RSMA achieves ergodic rates identical to conventional MRT+MF RSMA but with considerably reduced complexity and power dissipation (Zhao et al., 2024).

In multi-user holographic MIMO and network massive MIMO, MF precoding remains analytically tractable in the presence of mutual coupling and ultra-dense antenna layouts. Operational guidelines include maintaining inter-element spacing near or above half-wavelength to prevent coupling-induced throughput degradation, and selecting MF over BF/optimization-based beamforming when SNR is moderate/low or array size is large (Shekar et al., 6 Oct 2025).

Regime	MF Precoding Advantage	Limitation
Low/moderate SNR	Full diversity, array gain	Outperformed by BF at high SNR
High user count ($K\sim M$)	High throughput with matrix normalization	ZF fails due to rank deficiency
Imperfect/Statistical CSI	Robustness, analytic modeling	May not exploit all available CSI
Massive MIMO ($L\gg K$)	Deterministic-equivalent rates, low complexity	NA

7. Comparative Perspective and Mode Selection

MF precoding is best understood as a low-complexity, energy-efficient baseline for multi-antenna systems. Unlike ZF or RZF, it is interference-limited at high rates/SNR, but it outperforms ZF at low SNR, especially in user-dense or poorly conditioned channels (Mehana et al., 2012, Shekar et al., 6 Oct 2025). Optimized regularization (Wiener/MMSE) achieves nonzero diversity at all rates and avoids the error floor, offering a middle ground at increased complexity.

In deployment, MF is preferred when:

• The system operates at low-to-moderate rates/SNR,
• Hardware or computational constraints rule out matrix inversion,
• Large arrays provide sufficient beamforming/array gain,
• CSI is partial or imperfect,
• Energy and analytic tractability are priorities.

These considerations are central in current research for scalable, low-cost, and robust wireless infrastructure (Mehana et al., 2012, Zhao et al., 2024, Shekar et al., 6 Oct 2025).