Automatic Modulation Classification for Composite Radar Signals using Enhanced Time-Frequency Image Processing

Abstract

Automatic modulation classification (AMC) plays a crucial role in non-cooperative contexts, a significance further amplified in contemporary cognitive radios, electromagnetic war fare, and spectrum surveillance with the advent of composite radar signals. Recently, deep learning (DL) has emerged as a prominent technology for AMC across various types of signals. This paper proposes a DL-based AMC for composite radar signals using enhanced time-frequency image (TFI) in non-cooperative contexts. To achieve this, we first identify the informative region (IR) within the TFIs of the received radar signals through a sequence of processing steps. This process differentiates key signal components, mitigates the influence of extraneous noise, and improves separability across modulation schemes. We then classify the modulation schemes of the composite radar signals by DL models extracting localized features from the IR, and analyze their classification performance. To validate the proposed method, we mathematically analyze the effect of the enhanced TFI processing on the classification. Furthermore, through comprehensive computer simulations on various DL models, we demonstrate that the pro posed method effectively classifies 37 types of single and composite radar modulation schemes, outperforming conventional approaches in terms of classification accuracy.