Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Low-Complexity OFDM Deep Neural Receivers

Ankit Gupta, Onur Dizdar, Yun Chen, Fehmi Emre Kadan, Ata Sattarzadeh, Stephen Wang

TL;DR

The paper addresses the high computational cost and slow training convergence of OFDM NeuralRx receivers. It introduces a low-complexity ResNet-SS design that uses channel split/shuffle, GELU activation, and small 3×3 kernels with unit dilation to cut NFLOPs and MAC, while preserving or improving decoding accuracy. The proposed architecture, combining ResNet-SS with ResNet-T blocks, achieves faster training convergence and up to 1 dB decoding gains, and reduces energy consumption compared to prior NeuralRx designs and baselines. These contributions offer practical improvements for online retraining and deployment of neural receivers in OFDM-based wireless systems across realistic channel conditions.

Abstract

Deep neural receivers (NeuralRxs) for Orthogonal Frequency Division Multiplexing (OFDM) signals are proposed for enhanced decoding performance compared to their signal-processing based counterparts. However, the existing architectures ignore the required number of epochs for training convergence and floating-point operations (FLOPs), which increase significantly with improving performance. To tackle these challenges, we propose a new residual network (ResNet) block design for OFDM NeuralRx. Specifically, we leverage small kernel sizes and dilation rates to lower the number of FLOPs (NFLOPs) and uniform channel sizes to reduce the memory access cost (MAC). The ResNet block is designed with novel channel split and shuffle blocks, element-wise additions are removed, with Gaussian error linear unit (GELU) activations. Extensive simulations show that our proposed NeuralRx reduces NFLOPs and improves training convergence while improving the decoding accuracy.

Low-Complexity OFDM Deep Neural Receivers

TL;DR

The paper addresses the high computational cost and slow training convergence of OFDM NeuralRx receivers. It introduces a low-complexity ResNet-SS design that uses channel split/shuffle, GELU activation, and small 3×3 kernels with unit dilation to cut NFLOPs and MAC, while preserving or improving decoding accuracy. The proposed architecture, combining ResNet-SS with ResNet-T blocks, achieves faster training convergence and up to 1 dB decoding gains, and reduces energy consumption compared to prior NeuralRx designs and baselines. These contributions offer practical improvements for online retraining and deployment of neural receivers in OFDM-based wireless systems across realistic channel conditions.

Abstract

Deep neural receivers (NeuralRxs) for Orthogonal Frequency Division Multiplexing (OFDM) signals are proposed for enhanced decoding performance compared to their signal-processing based counterparts. However, the existing architectures ignore the required number of epochs for training convergence and floating-point operations (FLOPs), which increase significantly with improving performance. To tackle these challenges, we propose a new residual network (ResNet) block design for OFDM NeuralRx. Specifically, we leverage small kernel sizes and dilation rates to lower the number of FLOPs (NFLOPs) and uniform channel sizes to reduce the memory access cost (MAC). The ResNet block is designed with novel channel split and shuffle blocks, element-wise additions are removed, with Gaussian error linear unit (GELU) activations. Extensive simulations show that our proposed NeuralRx reduces NFLOPs and improves training convergence while improving the decoding accuracy.

Paper Structure

This paper contains 18 sections, 6 equations, 5 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. Proposed OFDM NeuralRx Design
  4. Preliminaries on NeuralRx Design
  5. Proposed Efficient ResNet Blocks
  6. Kernel Size and Dilation Rate
  7. Filter Size
  8. ResNet-SS Block Design
  9. Proposed NeuralRx Architecture and Training
  10. Proposed NeuralRx Architecture
  11. NeuralRx Training and Evaluation
  12. Generalizability
  13. Performance Evaluation
  14. Baseline Methods
  15. Training Convergence
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Block diagram of 5G-NR transceiver and NeuralRx.
  • Figure 2: Architecture of NeuralRx with ResNet blocks.
  • Figure 3: Training convergence for Y-NeuralRx (parameters in Table \ref{['tab:arch_resnet_t']}) and proposed NeuralRx (parameters in Table \ref{['tab:arch_resnet_ss']}).
  • Figure 4: Decoding performance of traditional NeuralRx (parameters in Table \ref{['tab:arch_resnet_t']}) and proposed NeuralRx (parameters in Table \ref{['tab:arch_resnet_ss']}).
  • Figure 5: Impact of different components in ResNet-SS.