Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Site-Specific Finetuning of Neural Receivers with Real-World 5G NR Measurements

Nuri Berke Baytekin, Reinhard Wiesmayr, Sebastian Cammerer, Chris Dick, Christoph Studer

TL;DR

This work empirically study site-specific finetuning of neural receivers using real-world 5G NR physical uplink shared channel (PUSCH) data collected with an over-the-air testbed at ETH Zurich across three scenarios: a small laboratory, a large office floor, and a high-mobility outdoor environment.

Abstract

Finetuning wireless receivers to a specific deployment scenario can yield significant error-rate performance improvements without increasing processing complexity. However, site-specific finetuning has so far only been demonstrated on synthetic channel data and lacks real-world benchmarks. In this work, we empirically study site-specific finetuning of neural receivers using real-world 5G NR physical uplink shared channel (PUSCH) data collected with an over-the-air testbed at ETH Zurich across three scenarios: (i) a small laboratory, (ii) a large office floor, and (iii) a high-mobility outdoor environment. Our results confirm substantial error-rate performance improvements from site-specific finetuning, consistent with earlier findings based on synthetic channel data. Moreover, we demonstrate that these improvements generalize across different user-equipment hardware and deployment scenarios.

Site-Specific Finetuning of Neural Receivers with Real-World 5G NR Measurements

TL;DR

This work empirically study site-specific finetuning of neural receivers using real-world 5G NR physical uplink shared channel (PUSCH) data collected with an over-the-air testbed at ETH Zurich across three scenarios: a small laboratory, a large office floor, and a high-mobility outdoor environment.

Abstract

Finetuning wireless receivers to a specific deployment scenario can yield significant error-rate performance improvements without increasing processing complexity. However, site-specific finetuning has so far only been demonstrated on synthetic channel data and lacks real-world benchmarks. In this work, we empirically study site-specific finetuning of neural receivers using real-world 5G NR physical uplink shared channel (PUSCH) data collected with an over-the-air testbed at ETH Zurich across three scenarios: (i) a small laboratory, (ii) a large office floor, and (iii) a high-mobility outdoor environment. Our results confirm substantial error-rate performance improvements from site-specific finetuning, consistent with earlier findings based on synthetic channel data. Moreover, we demonstrate that these improvements generalize across different user-equipment hardware and deployment scenarios.
Paper Structure (24 sections, 1 equation, 6 figures)

This paper contains 24 sections, 1 equation, 6 figures.

Table of Contents

  1. Introduction
  2. Site-specific Deployment Scenarios
  3. Indoor Small Laboratory Scenario
  4. Indoor Large Office Floor Scenario
  5. Outdoor High-Mobility Scenario with on
  6. Measurement Setup for Data Extraction
  7. System Configuration
  8. Measurement Procedure
  9. Data Extraction
  10. Real-World Site-Specific Finetuning
  11. Training Data Collection with 5G NR Processing
  12. Finetuning Training Pipeline
  13. Performance Evaluation
  14. Performance Baselines
  15. Evaluation Protocol
  16. ...and 9 more sections

Figures (6)

  • Figure 1: Site-specific finetuning pipeline comprising (i) 5G NR over-the-air measurements in the deployment scenarios shown in fig:deployment_scenarios, (ii) training data extraction, (iii) offline site-specific finetuning, and (iv) offline real-world block error rate (BLER) performance evaluation on a measured test dataset.
  • Figure 2: Floor plans (top) and photos (bottom) of the measured deployment scenarios used for site-specific finetuning.
  • Figure 3: performance improvements across scenarios.
  • Figure 4: performance improvements vs. varying complexity for varying number of iterations and inference latency.
  • Figure 5: BLER performance vs. number of finetuning training iterations for Shallow (2 iter.) finetuned on UE0 (Samsung Galaxy S23).
  • ...and 1 more figures