Abstract: Due to its superior anti-interception and inherent security, the wide application of frequency hopping (FH) signals in satellite communications, satellite measurement and control radio frequency links, satellite navigation systems and Link 16 data link has brought great challenges to space-based electronic reconnaissance. In non-cooperative scenarios, wideband FH communication reconnaissance including FH signal detection, parameter estimation, and network sorting under single-channel reception is challenging. The FH pattern contains the most information about FH signals. So it is the core of FH parameter estimation. Based on the task analysis, this paper proposed a blind prediction framework combining time and power domain features to improve the accuracy and efficiency of FH pattern estimation for mixed signals in the wideband spectrum. First, the spectrogram generated by the short-time Fourier transform (STFT) was used as the input of the signal detection network. The signals were detected on the multi-scale feature maps and then the time-equency (TF) characteristics of FH signals were predicted. After signal detection and localization, the relative power density characteristics were estimated based on the pixels in the signal area. Then the TF and power features were used to identify signal categories and predict the corresponding FH patterns. The unique advantage of this framework is that it requires no prior signal information and anchors but exploits the inherent TF and power properties of asynchronous FH signals. It has a strong generalization ability and can adapt to signals of any shape. The proposed framework can achieve a recognition accuracy of 98.77% for mixed signals with 2 FH signal radiation sources. Experimental results demonstrate the superiority of the proposed framework in fully blind detection, identification, sorting, and parameter estimation of hybrid FH signals.

Key words: neural network, multi-source feature fusion, hybrid asynchronous frequency hopping signals, signal detection, blind parameter estimation