Spectrum Signal Similarity-Based Segmentation and Illegal Transmission Monitoring

With the rapid advancement of wireless communication technologies， spectrum resources have become increasingly scarce. Illegal occupancy behaviors—such as abnormal frequency bursts， ultra-wideband interference， and high-power malicious transmissions—have further exacerbated spectrum congestion， severely threatening the security and reliability of legitimate communication systems. In practical spectrum monitoring， owing to blurred parameter boundaries between legitimate and illegal signals and the difficulty in constructing a comprehensive prior knowledge base of legitimate signals， existing fixed-rule detection methods （for example， local outlier factor and long short-term memory （LSTM）） face limitations in applicability. To address these challenges， this study proposes a spectrum signal similarity segmentation and illegal frequency usage monitoring algorithm. The method first separates signals in dense signal environments through time-frequency trajectory clustering. Then， the isolation distributional kernel （IDK） is introduced to measure the distributional similarity between signals more accurately， and the Pettitt change-point detection algorithm is employed for adaptive segmentation of the wideband spectrum to identify abrupt changes in the frequency usage behavior. On this foundation， similarity clustering is applied to automatically construct and update the legitimate signal library， significantly reducing reliance on manual labeling. Finally， the copula-based outlier detection （COPOD） algorithm is utilized to model complex dependencies among features， enabling an efficient anomaly analysis of real-time wideband radio frequency usage behaviors. Simulation results demonstrated that the legitimate signal library constructed by the proposed method achieved high accuracy and effectively identified legitimate signal patterns. In illegal signal detection tasks， compared with traditional methods such as LSTM， one-class support vector machine （OSVM）， and generative adversarial network （GAN）， the proposed algorithm exhibited superior performance in both false alarm rate and missed detection rate， offering an effective solution for building adaptive and intelligent spectrum monitoring systems.