EqDeepRx: Learning a Scalable MIMO Receiver

TL;DR

EqDeepRx tackles scalability and explainability challenges in ML-based MIMO receivers by blending conventional signal processing with compact neural modules. The core idea is to keep linear receivers in the loop, using parallel LMMSE and RZF equalizers and a per-MIMO-layer DetectorNN, so complexity grows near-linearly with the number of layers. DenoiseNN enhances channel estimation in the pilot domain, while DemapperNN translates soft symbols to LLRs for LDPC decoding, yielding robust performance across 5G/6G OFDM scenarios. End-to-end simulations across CDL-C/CDL-D/UMa channels show ~2–4 dB SINR gains at 10% BLER and 15–25% gains in spectral efficiency over a strong baseline, with flexibility to support varying MIMO configurations without retraining.EqDeepRx delivers robust BLER and throughput gains over conventional receivers while preserving low computational complexity by (i) denoising channel estimates with a lightweight DenoiseNN, (ii) running two complementary equalizers in parallel to avoid learning matrix inverses, and (iii) performing per-MIMO-layer data-aided detection with shared weights. The architecture maintains MIMO-layer count invariance and exhibits resilience under mobility and inter-cell interference, making it suitable for 5G/6G deployments. Ablation studies confirm the necessity of the LMMSE+RZF pair and the DenoiseNN, and show per-layer detectors outperform monolithic designs for arbitrary layer counts. Overall, EqDeepRx offers practical, scalable gains with a favorable complexity-performance trade-off and clear pathways for further optimization.keywords_backbone

Abstract

While machine learning (ML)-based receiver algorithms have received a great deal of attention in the recent literature, they often suffer from poor scaling with increasing spatial multiplexing order and lack of explainability and generalization. This paper presents EqDeepRx, a practical deep-learning-aided multiple-input multiple-output (MIMO) receiver, which is built by augmenting linear receiver processing with carefully engineered ML blocks. At the core of the receiver model is a shared-weight DetectorNN that operates independently on each spatial stream or layer, enabling near-linear complexity scaling with respect to multiplexing order. To ensure better explainability and generalization, EqDeepRx retains conventional channel estimation and augments it with a lightweight DenoiseNN that learns frequency-domain smoothing. To reduce the dimensionality of the DetectorNN inputs, the receiver utilizes two linear equalizers in parallel: a linear minimum mean-square error (LMMSE) equalizer with interference-plus-noise covariance estimation and a regularized zero-forcing (RZF) equalizer. The parallel equalized streams are jointly consumed by the DetectorNN, after which a compact DemapperNN produces bit log-likelihood ratios for channel decoding. 5G/6G-compliant end-to-end simulations across multiple channel scenarios, pilot patterns, and inter-cell interference conditions show improved error rate and spectral efficiency over a conventional baseline, while maintaining low-complexity inference and support for different MIMO configurations without retraining.

Figures (15)

• Figure 1: Overall block diagram of the considered system. In this article, we consider different alternatives for the receiver algorithm part, including the conventional baseline as well as the proposed EqDeepRx ML receiver.
• Figure 2: EqDeepRx high-level architecture. MIMO layers are handled in parallel separately after the EQs, allowing for efficient inference and layer count generalization. Blocks with gray background contain trainable parameters.
• Figure 3: Architecture of the DetectorNN and DemapperNN. This part of the model handles each MIMO layer separately in parallel.
• Figure 4: The subsampling residual block for DetectorNN. DenoiseNN has similar structure, but both convolutions have Nx1 kernels for frequency-only operation.
• Figure 5: Simulated (a) uncoded BER and (b) BLER with CDL-C, under speed of 10--15 m/s. Data rate matched MCSs.