Research On Intelligent Recognition Methods Of Wireless Signals

Intelligent signal recognition is one of the key technologies to enhance the antijamming ability of communication systems and maintain the security of the electromagnetic environment.It employs pattern recognition and machine learning algorithms to analyze and decode specific signal samples,accurately identifying their categories and features.This technology holds important research significance and application requirements in the field of electronic countermeasures,communication countermeasures,cognitive radio,and other fields.With the rapid development of wireless communication and Internet of Things(Io T)technology,there has been a significant increase in both the quantity and diversity of signals accessed within the electromagnetic environment,causing signal recognition tasks to become more complex and diverse.In complex electromagnetic environments,the collected wireless signals are typically characterized by high randomness,strong concealment,and non-cooperativeness.Consequently,the natural frequency of occurrence varies across different signal categories.This discrepancy remains unaffected by the robustness of the communication system but leads to sample redundancy,low sample quality,insufficient sample size,and missing sample category within the signal dataset.These problems have emerged as critical barriers impeding the further progress of signal recognition technology in both research and practical applications.Hence,it is imperative to enhance the model’s recognition capabilities to effectively address these novel signal recognition challenges.By analyzing the existing signal recognition models and objective phenomena of these signal datasets,this doctoral dissertation systematically summarizes several scientific problems that require targeted solutions.Furthermore,this dissertation conducts comprehensive research on wireless signal recognition methods from both theoretical and architectural perspectives to mitigate the adverse effects caused by aforementioned practical issues.The research content and main contributions of this dissertation are as follows:1.To address the issue of training resource waste caused by redundant samples in complex electromagnetic environments,this dissertation proposes a signal screening method based on sample importance evaluation.This method constructs a border sample selector based on the geometric distribution of signal features,effectively resolving the challenge of locating supporting data near decision boundaries of the signal classifier.Then,it utilizes the statistical information of each signal sample labeled as supporting data during initial stages of model training to assess the significance of individual training samples for recognition models,thereby facilitating removal of redundant samples and identification of representative subsets.This method reduces the size of the signal dataset,improves the training efficiency of the model,and lays a robust theoretical foundation for intelligent recognition tasks of wireless signals in resource-constrained scenarios.2.To address the issue of poor signal recognition performance in low SNR conditions,this dissertation proposes a novel signal recognition method based on real-complex domain joint feature extraction.The proposed approach integrates the strengths of traditional convolutional layers and LSTM layers,while introducing complex convolutional layers to optimize the architecture of the signal classifier.This addresses the limitation in feature information mining ability for low SNR signals caused by existing model designs.Moreover,discriminative signal features are constructed under joint constraints in both real and complex domains,effectively enhancing their distinguishability and noise robustness.As a result,the recognition accuracy of low SNR signals is improved significantly,providing an efficient solution for wireless signal recognition with low quality.3.To address the issue of model overfitting caused by a limited number of signal samples,this dissertation proposes a novel method for generating signal features based on sample class direction.With the aim of expanding discriminant information from signal features,a model for generating signal features is constructed to fit the data distribution in a manner similar to real signal features.This method relies on the generation model to realize the information constraints of signal feature categories,thereby generating supplementary signal features to compensate for the discriminant information deficiency in the recognition model.It effectively calibrates the decision boundary of the model and alleviates over-fitting phenomenon in the signal recognition model,providing a theoretical foundation for intelligent recognition tasks of wireless signals in limited sample scenarios.4.To address the issue of undetectable novel signals due to missing signal categories in the signal training set,this dissertation proposes a novel method for recognizing novel class signal based on unknown information inference and model parameter sharing,which is derived from extensive research on information inference.It infers the information of the known class signals to construct substitutes for the novel class signals,which provides the missing information about the new class to the model.This method relies on knowledge sharing and parameter adjustment to refine the decision boundary,enabling the model to accurately identify known class signals and effectively discover novel class signals.This method provides a robust solution for intelligent recognition tasks of wireless signals in open scenarios.This dissertation conducted comprehensive experiments and theoretical analysis on the above content.Simulation results demonstrated that these proposed signal recognition methods can effectively cope with various new signal recognition challenges within complex electromagnetic environments.Research on intelligent recognition methods of wireless signals can provide strong technical support for recognition tasks in the fields of electronic countermeasures,communication countermeasures or cognitive radio fields,thereby presenting broad application prospects and significant research value.