Deep Learning-Based OFDM Receiver Using Fully Convolutional Neural Networks: Implementation, Evaluation, and Analysis

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Ali, Ghader
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901699238559744
author Ali, Ghader
author_facet Ali, Ghader
contents <p>Deep learning has emerged as a promising paradigm for physical layer processing in wireless communication systems. In this paper, we present a comprehensive implementation and evaluation of DeepRx, a fully convolutional neural network (CNN) that replaces the entire receiver processing chain in a 5G-compliant orthogonal frequency-division multiplexing (OFDM) system. The proposed architecture jointly performs channel estimation, equalization, and soft demapping by processing the frequency-domain received signal directly into bit-level log-likelihood ratios (LLRs). We construct a three-component input tensor comprising the received signal, known pilot symbols, and raw channel estimates, enabling the network to leverage both pilot and unknown data symbols for improved channel tracking. Through extensive simulations using 3GPP-defined tapped delay line (TDL) channel models with varying signal-to-noise ratios (0-24 dB) and Doppler shifts (10-500 Hz), we demonstrate that DeepRx achieves up to 1.5x lower bit error rate (BER) compared to a traditional least squares/linear minimum mean square error (LS/LMMSE) receiver. Per-bit analysis reveals that the performance gains are predominantly attributed to improved amplitude estimation of the least significant bits, suggesting that the network implicitly learns a blind equalization strategy. The complete source code is available at: https://github.com/G-ALI007/DeepRx-OFDM-PyTorch</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19024220
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Deep Learning-Based OFDM Receiver Using Fully Convolutional Neural Networks: Implementation, Evaluation, and Analysis
Ali, Ghader
Deep Learning
OFDM
5G NR
Channel Estimation
Convolutional Neural Networks
Equalization
Signal Processing
PUSCH
Bit Error Rate
Wireless Communications
<p>Deep learning has emerged as a promising paradigm for physical layer processing in wireless communication systems. In this paper, we present a comprehensive implementation and evaluation of DeepRx, a fully convolutional neural network (CNN) that replaces the entire receiver processing chain in a 5G-compliant orthogonal frequency-division multiplexing (OFDM) system. The proposed architecture jointly performs channel estimation, equalization, and soft demapping by processing the frequency-domain received signal directly into bit-level log-likelihood ratios (LLRs). We construct a three-component input tensor comprising the received signal, known pilot symbols, and raw channel estimates, enabling the network to leverage both pilot and unknown data symbols for improved channel tracking. Through extensive simulations using 3GPP-defined tapped delay line (TDL) channel models with varying signal-to-noise ratios (0-24 dB) and Doppler shifts (10-500 Hz), we demonstrate that DeepRx achieves up to 1.5x lower bit error rate (BER) compared to a traditional least squares/linear minimum mean square error (LS/LMMSE) receiver. Per-bit analysis reveals that the performance gains are predominantly attributed to improved amplitude estimation of the least significant bits, suggesting that the network implicitly learns a blind equalization strategy. The complete source code is available at: https://github.com/G-ALI007/DeepRx-OFDM-PyTorch</p>
title Deep Learning-Based OFDM Receiver Using Fully Convolutional Neural Networks: Implementation, Evaluation, and Analysis
topic Deep Learning
OFDM
5G NR
Channel Estimation
Convolutional Neural Networks
Equalization
Signal Processing
PUSCH
Bit Error Rate
Wireless Communications
url https://doi.org/10.5281/zenodo.19024220