Deep Learning-Based Interference Detection, Classification, and Forecasting Algorithm for ESM Radar Systems

doi:10.1109/ACCESS.2024.3475732

Return

Deep Learning-Based Interference Detection, Classification, and Forecasting Algorithm for ESM Radar Systems

2024-01-01

Hamda Bouzabia *

Georges Kaddoum

Tri Nhu

DOI：10.1109/ACCESS.2024.3475732

Original

Original request for help

Save

Abstract

En 中文

In this study, aiming to address the challenges posed by interference from communication systems and jammers, we investigate the application of deep learning (DL) in electronic support measures (ESM) radar systems. Our primary objective is to detect, classify, and forecast interference that can disrupt detection of low probability of intercept (LPI) and low probability of detection (LPD) signals. The proposed algorithm uses a time-frequency distribution (TFD) and received interference strength (RIS) to detect and predict interference. To ensure high precision,we develop a new DL-based outlier detection (OD) technique that is based on the relationship between true positive rate (TPR) and latent space. More specifically, the OD technique applies a new dual-threshold mechanism to the TFD representation for interference detection. We also introduce a DL-enabled classifier designed using the OD architecture to identify the source of interference. Finally, we forecast the RIS by proposing a new DL autoregressive (AR) model through a sliding window designed using the classifier's output. By integrating OD in classifier design and using its output for forecasting, our approach achieves superior accuracy as compared to independent models. Simulation results demonstrate that the proposed algorithm outperforms others, particularly in low signal-to-interference plus noise ratio (SINR) conditions. Specifically, in terms of interference detection, our algorithm achieves 0.9978 TPR, 0.9415 recall, and 0.0004 false positive ratio (FPR). With regard to classification, it records 0.9784 precision and 0.7847 recall. In forecasting, it achieves a 0.2100 mean average error (MAE), thus significantly enhancing ESM radar awareness. The TFD feature also proves to be more accurate than in-phase and quadrature features. These strengths, coupled with an optimal balance of cost and accuracy, make our framework robust and resistant to interference.

Keywords:

Radar systems

LPI/LPD

FMCW

interference detection and classification

forecasting

deep learning

forecasting

deep learning