Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fibbinary-Based Compression and Quantization for Efficient Neural Radio Receivers

Roberta Fiandaca, Manil Dev Gomony

TL;DR

The paper tackles the challenge of deploying neural receivers on hardware-constrained devices by reducing compute, area, and memory via quantization and compression. It introduces Fibonacci Codeword Quantization (FCQ) with Incremental Network Quantization (INQ) and two lossless compression algorithms based on Zeckendorf expansion and Fibbinary weight redundancy. Results show that FCQ with INQ cuts multiplier power by about 45% and area by about 44%, and that applying both FCQ-INQ and compression reduces memory footprint by about 63% while preserving BER close to the full-precision network as a function of $E_b/N_0$. The study offers insights into a fine-grained quantization schedule, subfractions to mitigate degradation, and a compression strategy tailored to Fibonacci-quantized weights, enabling practical neural radio receivers for 6G.

Abstract

Neural receivers have shown outstanding performance compared to the conventional ones but this comes with a high network complexity leading to a heavy computational cost. This poses significant challenges in their deployment on hardware-constrained devices. To address the issue, this paper explores two optimization strategies: quantization and compression. We introduce both uniform and non-uniform quantization such as the Fibonacci Code word Quantization (FCQ). A novel fine-grained approach to the Incremental Network Quantization (INQ) strategy is then proposed to compensate for the losses introduced by the above mentioned quantization techniques. Additionally, we introduce two novel lossless compression algorithms that effectively reduce the memory size by compressing sequences of Fibonacci quantized parameters characterized by a huge redundancy. The quantization technique provides a saving of 45\% and 44\% in the multiplier's power and area, respectively, and its combination with the compression determines a 63.4\% reduction in memory footprint, while still providing higher performances than a conventional receiver.

Fibbinary-Based Compression and Quantization for Efficient Neural Radio Receivers

TL;DR

The paper tackles the challenge of deploying neural receivers on hardware-constrained devices by reducing compute, area, and memory via quantization and compression. It introduces Fibonacci Codeword Quantization (FCQ) with Incremental Network Quantization (INQ) and two lossless compression algorithms based on Zeckendorf expansion and Fibbinary weight redundancy. Results show that FCQ with INQ cuts multiplier power by about 45% and area by about 44%, and that applying both FCQ-INQ and compression reduces memory footprint by about 63% while preserving BER close to the full-precision network as a function of . The study offers insights into a fine-grained quantization schedule, subfractions to mitigate degradation, and a compression strategy tailored to Fibonacci-quantized weights, enabling practical neural radio receivers for 6G.

Abstract

Neural receivers have shown outstanding performance compared to the conventional ones but this comes with a high network complexity leading to a heavy computational cost. This poses significant challenges in their deployment on hardware-constrained devices. To address the issue, this paper explores two optimization strategies: quantization and compression. We introduce both uniform and non-uniform quantization such as the Fibonacci Code word Quantization (FCQ). A novel fine-grained approach to the Incremental Network Quantization (INQ) strategy is then proposed to compensate for the losses introduced by the above mentioned quantization techniques. Additionally, we introduce two novel lossless compression algorithms that effectively reduce the memory size by compressing sequences of Fibonacci quantized parameters characterized by a huge redundancy. The quantization technique provides a saving of 45\% and 44\% in the multiplier's power and area, respectively, and its combination with the compression determines a 63.4\% reduction in memory footprint, while still providing higher performances than a conventional receiver.

Paper Structure

This paper contains 7 sections, 2 equations, 5 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background information
  3. Neural Receiver
  4. Neural Network Quantization
  5. Non-uniform Fibonacci Quantization
  6. Fibbinary Compression
  7. Conclusions

Figures (5)

  • Figure 1:
  • Figure 2:
  • Figure 4:
  • Figure 5:
  • Figure 7: Pictorial representation of the word-length and word-count compressions algorithms. UL stands for Uncompressed Length, i.e. the number of weights in the tensors before the compression. UB describes the Uncompressed Bitwidth. CB represents Compressed Bithwidth. CL is the Compressed Length, i.e. the tensors' length after the compression. The compression ratio (CR) is computed as in the picture.