A Learning-Based Signal Parameter Extraction Approach for Multi-Source Frequency-Hopping Signal Sorting

Multi-source frequency-hopping (FH) signal sorting without prior information remains a challenging problem. Conventional multi-source FH signal sorting is a two-stage scheme based on parameter estimation and signal classification, in which the low accuracy of parameter estimation will degrade signal sorting performance. Existing parameter estimation methods rely heavily on prior information about the signal and are susceptible to noise. This letter proposes a data-driven context-level FH signal parameter extractor (FHExt) to alleviate the mentioned limits by considering the correlations between pixels within signal areas and learning-based signal detection. To verify the sorting performance of the FHExt-based framework, an accurately labeled FH signal parameter extraction and sorting (FHES) dataset is developed. Experiments reveal that the FHExt-based framework outperforms benchmarks in terms of accuracy and mean average precision (mAP) in fully-blind scenarios. In addition, the FHExtbased framework can be adapted to semi-blind scenarios by slightly adjusting the post-processing method.