Multifunction Radar Signal Sequence Prediction Via Trend Filtering and Similarity Comparison

Abstract

Multifunction radar signal sequence prediction can provide a profound understanding of radar intentions, thereby playing a pivotal role in radar signal analysis. However, traditional methods are impeded by noise and outliers in observed sequences. This article introduces a two-stage framework for predicting multifunction radar signal sequences by integrating robust enhanced trend filtering and signal sequence similarity comparison. In the preprocessing stage, the framework models the noise and outliers using a Gaussian mixture model, and extracts trend information through maximizing a posteriori estimation and interpolation. In the prediction stage, it simplifies signal sequence representation by leveraging extracted trend information. The method calculates similarity between current and historical signal sequences, generating predictions of radar pulse sequences using prototypes derived from the subsequent sequences of the most similar historical segments. Distinctively, the framework streamlines data representation in prediction, reduces storage and operational demands, and operates independent of prior pulse pattern knowledge. Experimental results demonstrate substantial enhancements in predicting pulse repetition interval of multifunction radar signal sequences compared to existing methods.