Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Co-Channel Interference Mitigation Using Deep Learning for Drone-Based Large-Scale Antenna Measurements

Kadyrzhan Tortayev, Oliver Falkenberg Damborg, Jònas À Hàlvmørk Joensen, Jonas Pedesk, Yifa Li, Fengchun Zhang, Zeliang An, Yubo Wang, Ming Shen

TL;DR

Co-channel interference severely degrades CW amplitude estimation in drone-based large-scale antenna measurements. The authors introduce a lightweight DC-CNN that operates on time-domain I/Q bursts to estimate CW amplitude under interference, outperforming FFT baselines and approaching heavier neural baselines in accuracy. They demonstrate an average MAE of about $7\%$ across a wide SIR range ($-33.3$ dB to $+46.7$ dB) with sub-$1$ dB accuracy for $ ext{SIR} \ge -30$ dB, using a compact model of fewer than $20{,}000$ parameters suitable for embedded UAV deployment. The work enables practical, interference-resilient in-situ antenna pattern characterization and outlines directions for multi-tone measurements and on-device adaptation on UAV platforms.

Abstract

Unmanned aerial vehicles (UAVs) enable efficient in-situ radiation characterization of large-aperture antennas directly in their deployment environments. In such measurements, a continuous-wave (CW) probe tone is commonly transmitted to characterize the antenna response. However, active co-channel emissions from neighboring antennas often introduce severe in-band interference, where classical FFT-based estimators fail to accurately estimate the CW tone amplitude when the signal-to-interference ratios (SIR) falls below -10 dB. This paper proposes a lightweight deep convolutional neural network (DC-CNN) that estimates the amplitude of the CW tone. The model is trained and evaluated on real 5~GHz measurement bursts spanning an effective SIR range of --33.3 dB to +46.7 dB. Despite its compact size (<20k parameters), the proposed DC-CNN achieves a mean absolute error (MAE) of 7% over the full range, with <1 dB error for SIR >= -30 dB. This robustness and efficiency make DC-CNN suitable for deployment on embedded UAV platforms for interference-resilient antenna pattern characterization.

Co-Channel Interference Mitigation Using Deep Learning for Drone-Based Large-Scale Antenna Measurements

TL;DR

Co-channel interference severely degrades CW amplitude estimation in drone-based large-scale antenna measurements. The authors introduce a lightweight DC-CNN that operates on time-domain I/Q bursts to estimate CW amplitude under interference, outperforming FFT baselines and approaching heavier neural baselines in accuracy. They demonstrate an average MAE of about across a wide SIR range ( dB to dB) with sub- dB accuracy for dB, using a compact model of fewer than parameters suitable for embedded UAV deployment. The work enables practical, interference-resilient in-situ antenna pattern characterization and outlines directions for multi-tone measurements and on-device adaptation on UAV platforms.

Abstract

Unmanned aerial vehicles (UAVs) enable efficient in-situ radiation characterization of large-aperture antennas directly in their deployment environments. In such measurements, a continuous-wave (CW) probe tone is commonly transmitted to characterize the antenna response. However, active co-channel emissions from neighboring antennas often introduce severe in-band interference, where classical FFT-based estimators fail to accurately estimate the CW tone amplitude when the signal-to-interference ratios (SIR) falls below -10 dB. This paper proposes a lightweight deep convolutional neural network (DC-CNN) that estimates the amplitude of the CW tone. The model is trained and evaluated on real 5~GHz measurement bursts spanning an effective SIR range of --33.3 dB to +46.7 dB. Despite its compact size (<20k parameters), the proposed DC-CNN achieves a mean absolute error (MAE) of 7% over the full range, with <1 dB error for SIR >= -30 dB. This robustness and efficiency make DC-CNN suitable for deployment on embedded UAV platforms for interference-resilient antenna pattern characterization.
Paper Structure (14 sections, 5 equations, 9 figures, 3 tables)

This paper contains 14 sections, 5 equations, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Architecture & Method
  3. Overview of DC‑CNN
  4. Baseline Models
  5. Training Procedure
  6. Measurement & Dataset Preparation
  7. Capture Setup
  8. Dataset and RF Configuration
  9. Complex-Gain Extraction
  10. Results & Discussion
  11. Time-Domain and Spectral Characteristics
  12. Carrier–Power Error versus Interference Level
  13. Carrier Power Estimation Accuracy Across the Full SIR Range
  14. Conclusion & Future Work

Figures (9)

  • Figure 1: Overview of the measurement scenario and signal model.
  • Figure 2: Time-domain I/Q samples and their corresponding magnitude spectra. (a) Time-domain $I/Q$ samples of Mixture, CW(sine), and QPSK burst with the 200 kHz carrier offset shifted back. (b) Corresponding magnitude spectra (Hann window, 4096-point FFT).
  • Figure 3: Three-layer CNN front-end (non-dilated). Kernel sizes {9, 7, 5} yield a receptive-field radius of 9 samples (19 total), i.e. ±0.9µs at the 10Ms rate. Adaptive-average pooling collapses the 1000-sample burst to a 64-dimensional embedding; a two-layer MLP then regresses $\Re\{g\}$ and $\Im\{g\}$. All layer dimensions are shown in blue.
  • Figure 5: Time-domain I/Q bursts for every CW–QPSK power combination (-10dBm to -50dBm in 10dBm steps).
  • Figure 6: Power spectral density (PSD) for every CW–QPSK power combination (-10dBm to -50dBm in 10dBm steps).
  • ...and 4 more figures