A new method to solve the problem of facing less learning samples in signal modulation recognition

Abstract

In machine learning method, the number of training samples is an exceedingly important factor determining the learning system's robustness. In our previous researches (Liu et al.,

A Two-Stage Hybrid Feature Selection Method of Communication Signals Based on Combined Scoring and Improved Binary Whale Optimization Algorithm

“…In order to fully describe the signal from different angles and better recognize various signal modulation formats, this paper extracts 21 signal features. As shown in Table 1, these features come from seven different feature domains, including two entropy features [13], one frequency domain feature, one fractal feature [24], three cyclic spectrum features [25], five instantaneous features, seven higher order cumulant features [26] and two higher power spectrum features. We summarize them in Table 1.…”

Intrapulse Recognition of Radar Signals Via Bicubic Interpolation WVD

“…However, it is usually difficult to use a simple spectrogram to distinguish radar intra-pulses. Highorder spectra and cyclic spectra [30]…”

“…The above radar intra-pulse modulations include binary phase shift keying (BPSK), frequency encoding (FE), LFM, nonlinear frequency modulation (NLFM), CP and quaternary phase shift keying (QPSK). The PWs of the above intra-pulses are 10 µs, the pulse period is 30 between the modulated carrier and the unmodulated carrier.  Relative code means the phase difference between the previous code and the following symbol.…”