Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Channel Autoencoders for Wideband Communications leveraging Walsh-Hadamard interleaving

Cel Thys, Rodney Martinez Alonso, Sofie Pollin

TL;DR

The paper tackles energy-efficient, wideband communication under hardware constraints by integrating Walsh-Hadamard domain interleaved converters with end-to-end autoencoder-based coded modulation. It develops a comprehensive WH-domain transceiver model, derives a hardware-aware power/energy framework, and trains WH-domain autoencoders for short block lengths to jointly optimize modulation, coding, and HW effects. The key contributions include a detailed power model, hyperparameter optimization demonstrating near-polar performance with substantially improved energy efficiency (average ~29% over TI-AE and up to 4.8× over CNN-AE), and a Pareto analysis guiding architecture selection for practical deployments. The results show WH-domain learning can achieve reliable, high-throughput wideband communication with strong energy efficiency gains, making it a viable path for hardware-efficient neural coded modulation in future networks.

Abstract

This paper investigates how end-to-end (E2E) channel autoencoders (AEs) can achieve energy-efficient wideband communications by leveraging Walsh-Hadamard (WH) interleaved converters. WH interleaving enables high sampling rate analog-digital conversion with reduced power consumption using an analog WH transformation. We demonstrate that E2E-trained neural coded modulation can transparently adapt to the WH-transceiver hardware without requiring algorithmic redesign. Focusing on the short block length regime, we train WH-domain AEs and benchmark them against standard neural and conventional baselines, including 5G Polar codes. We quantify the system-level energy tradeoffs among baseband compute, channel signal-to-noise ratio (SNR), and analog converter power. Our analysis shows that the proposed WH-AE system can approach conventional Polar code SNR performance within 0.14dB while consuming comparable or lower system power. Compared to the best neural baseline, WH-AE achieves, on average, 29% higher energy efficiency (in bit/J) for the same reliability. These findings establish WH-domain learning as a viable path to energy-efficient, high-throughput wideband communications by explicitly balancing compute complexity, SNR, and analog power consumption.

Efficient Channel Autoencoders for Wideband Communications leveraging Walsh-Hadamard interleaving

TL;DR

The paper tackles energy-efficient, wideband communication under hardware constraints by integrating Walsh-Hadamard domain interleaved converters with end-to-end autoencoder-based coded modulation. It develops a comprehensive WH-domain transceiver model, derives a hardware-aware power/energy framework, and trains WH-domain autoencoders for short block lengths to jointly optimize modulation, coding, and HW effects. The key contributions include a detailed power model, hyperparameter optimization demonstrating near-polar performance with substantially improved energy efficiency (average ~29% over TI-AE and up to 4.8× over CNN-AE), and a Pareto analysis guiding architecture selection for practical deployments. The results show WH-domain learning can achieve reliable, high-throughput wideband communication with strong energy efficiency gains, making it a viable path for hardware-efficient neural coded modulation in future networks.

Abstract

This paper investigates how end-to-end (E2E) channel autoencoders (AEs) can achieve energy-efficient wideband communications by leveraging Walsh-Hadamard (WH) interleaved converters. WH interleaving enables high sampling rate analog-digital conversion with reduced power consumption using an analog WH transformation. We demonstrate that E2E-trained neural coded modulation can transparently adapt to the WH-transceiver hardware without requiring algorithmic redesign. Focusing on the short block length regime, we train WH-domain AEs and benchmark them against standard neural and conventional baselines, including 5G Polar codes. We quantify the system-level energy tradeoffs among baseband compute, channel signal-to-noise ratio (SNR), and analog converter power. Our analysis shows that the proposed WH-AE system can approach conventional Polar code SNR performance within 0.14dB while consuming comparable or lower system power. Compared to the best neural baseline, WH-AE achieves, on average, 29% higher energy efficiency (in bit/J) for the same reliability. These findings establish WH-domain learning as a viable path to energy-efficient, high-throughput wideband communications by explicitly balancing compute complexity, SNR, and analog power consumption.
Paper Structure (19 sections, 15 equations, 17 figures, 6 tables, 1 algorithm)

This paper contains 19 sections, 15 equations, 17 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Wideband transceiver model
  3. System model
  4. Orthogonal Walsh-Hadamard domain conversion
  5. Autoencoder learned coded modulation
  6. Autoencoder system model
  7. Autoencoder architecture
  8. Baselines for benchmarking
  9. Training and evaluation
  10. Hyperparameter selection
  11. System-level power and energy model
  12. Converter power consumption
  13. Baseband power consumption estimation
  14. Energy efficiency model
  15. Pareto optimal hyperparameter selection
  16. ...and 4 more sections

Figures (17)

  • Figure 1: Proposed wideband transceiver architecture including high-rate interleaved DAC & ADC and intelligent AI baseband processing. The energy efficiency model takes into account the interleaving converter power, the baseband power, the baseband throughput and the channel SNR.
  • Figure 2: Example of the first 8 sequency-ordered Walsh-Hadamard basis functions in (a) time and (b) frequency domain.
  • Figure 3: Diagram of Walsh-Hadamard domain ADC (a) and DAC (b), highlighting the differences with respect to time interleaved converters.
  • Figure 4: Block diagram of the proposed Walsh-Hadamard domain autoencoder, with trainable layers having a shaded background.
  • Figure 5: Example BLER curves for autoencoders with $N=32$, $k=16$. The horizontal line indicates the BLER threshold $P_e=0.1\%$. The data is from Section \ref{['subsec:hyperparam']}, with L2 regularization enabled. The threshold SNR for each model is indicated with a diamond on the figure.
  • ...and 12 more figures