A Transformer Inspired AI-based MIMO receiver

TL;DR

AttDet reframes MIMO detection as a Transformer-style sequence prediction task by treating transmit layers as tokens and learning inter-stream interference through channel-derived queries and keys. Values initialized from a matched-filter output are iteratively refined across attention layers, bridging model-based insight and data-driven optimization. Link-level simulations on 3GPP 5G channels show AttDet can closely match near-optimal detectors (e.g., ML/K-best) for SU- and MU-MIMO with high-order QAM, while preserving a polynomial, scalable complexity. The approach offers interpretable attention patterns tied to channel correlation and generalizes across antenna configurations and modulation orders, suggesting practical applicability in real receivers.

Abstract

We present AttDet, a Transformer-inspired MIMO (Multiple Input Multiple Output) detection method that treats each transmit layer as a token and learns inter-stream interference via a lightweight self-attention mechanism. Queries and keys are derived directly from the estimated channel matrix, so attention scores quantify channel correlation. Values are initialized by matched-filter outputs and iteratively refined. The AttDet design combines model-based interpretability with data-driven flexibility. We demonstrate through link-level simulations under realistic 5G channel models and high-order, mixed QAM modulation and coding schemes, that AttDet can approach near-optimal BER/BLER (Bit Error Rate/Block Error Rate) performance while maintaining predictable, polynomial complexity.

Figures (8)

• Figure 1: AttDet: overall neural architecture
• Figure 2: Breakdown of the Attention block
• Figure 3: BLER for SU-MIMO, $N_r=8$, $N_l=2$, 16QAM
• Figure 4: BLER for SU-MIMO, $N_r=32$, $N_l=2$, 16QAM
• Figure 5: BLER for SU-MIMO, $N_r=8$, $N_l=2$, 64QAM