Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SpikingRx: From Neural to Spiking Receiver

Ankit Gupta, Onur Dizdar, Yun Chen, Stephen Wang

TL;DR

This work proposes a deep convolutional SNN with spike-element-wise ResNet layers which takes a whole OFDM grid compliant with 5G specifications and provides soft outputs for decoded bits that can be used as log-likelihood ratios.

Abstract

In this work, we propose an energy efficient neuromorphic receiver to replace multiple signal-processing blocks at the receiver by a Spiking Neural Network (SNN) based module, called SpikingRx. We propose a deep convolutional SNN with spike-element-wise ResNet layers which takes a whole OFDM grid compliant with 5G specifications and provides soft outputs for decoded bits that can be used as log-likelihood ratios. We propose to employ the surrogate gradient descent method for training the SpikingRx and focus on its generalizability and robustness to quantization. Moreover, the interpretability of the proposed SpikingRx is studied by a comprehensive ablation study. Our extensive numerical simulations show that SpikingRx is capable of achieving significant block error rate performance gain compared to conventional 5G receivers and similar performance compared to its traditional NN-based counterparts with approximately 9x less energy consumption.

SpikingRx: From Neural to Spiking Receiver

TL;DR

This work proposes a deep convolutional SNN with spike-element-wise ResNet layers which takes a whole OFDM grid compliant with 5G specifications and provides soft outputs for decoded bits that can be used as log-likelihood ratios.

Abstract

In this work, we propose an energy efficient neuromorphic receiver to replace multiple signal-processing blocks at the receiver by a Spiking Neural Network (SNN) based module, called SpikingRx. We propose a deep convolutional SNN with spike-element-wise ResNet layers which takes a whole OFDM grid compliant with 5G specifications and provides soft outputs for decoded bits that can be used as log-likelihood ratios. We propose to employ the surrogate gradient descent method for training the SpikingRx and focus on its generalizability and robustness to quantization. Moreover, the interpretability of the proposed SpikingRx is studied by a comprehensive ablation study. Our extensive numerical simulations show that SpikingRx is capable of achieving significant block error rate performance gain compared to conventional 5G receivers and similar performance compared to its traditional NN-based counterparts with approximately 9x less energy consumption.
Paper Structure (25 sections, 19 equations, 15 figures, 4 tables)

This paper contains 25 sections, 19 equations, 15 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Contributions
  4. System Model
  5. PDSCH Signal Transmission-Reception
  6. 5G-NR Receiver Processing
  7. Proposed SNN-based Receiver (SpikingRx)
  8. Spiking Neuron
  9. SpikingRx Design
  10. Training Methodology
  11. Generation of Training and Testing Data
  12. Robustness of SpikingRx
  13. Ablation Study
  14. Training Activation Probability
  15. Neuron Activation and Communications Performance
  16. ...and 10 more sections

Figures (15)

  • Figure 1: Block diagram of SISO downlink PDSCH transmission. The receiver consists of conventional signal-processing blocks.
  • Figure 2: Representation of the LIF neuron.
  • Figure 3: Illustration of SpikingRx. (a) OFDM resource grid for a TTI, (b) LIF neuron with input and output spikes, (c) Spike activities in LIF neuron, (d) Gradient values concerning membrane potential, and (e) SpikingRx architecture with concatenated received signal and pilot information as inputs, 7 ResNet blocks, and mean soft-probabilistic as output.
  • Figure 4: Block diagram of SISO downlink PDSCH transmission. The receiver consists of SpikingRx.
  • Figure 5: Block diagrams of ResNet and SEW Resnet.
  • ...and 10 more figures