Deep-Learning-Aided Joint Detection and Channel Estimation for Massive MIMO-OFDM Systems

Celio A Souza Junior, Jaime L. Jacob, Taufik Abrão

DOI: 10.14209/sbrt.2025.1571157310

Evento: XLIII Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2025)

Keywords:

Abstract

Accurate channel state information (CSI) and ef- ficient multi-user detection are critical for massive MIMO- OFDM. We present a frame-level deep-neural-network (DNN) receiver that jointly estimates the channel and detects symbols. Monolithic and spectrum-partitioned architectures are trained with Monte-Carlo data under EPA-7 frequency-selective fading channel model, AWGN, and QAM. Compared with least-squares (LS) and time-domain MMSE (TD-MMSE) followed by zero- forcing, the partitioned DNN achieves up to 10 dB bit-error-rate (BER) gain at 10−3 in a 64 × 64 uplink antenna while keeping latency comparable. The model is robust to cyclic-prefix (CP) suppression and to a 4× pilot reduction, indicating suitability for 5G/6G deployments. Keywords- Massive MIMO, OFDM, Channel estimation, Deep learning, 5G, 6G.

Download