Font Size: a A A

Research On Recognition Method Of Intra-Pulse Modulation Type Of Radar Signal Based On Deep Learning

Posted on:2022-10-04Degree:MasterType:Thesis
Country:ChinaCandidate:W Y WangFull Text:PDF
GTID:2518306353476414Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
The recognition of the pulse modulation type of the radar signal is an important part of the radar reconnaissance system.Accurate recognition of the signal type helps to determine the enemy radar type,threat level and function.However,with the continuous development of radar technology,new system radars represented by low intercept radars dominate the modern battlefield,and traditional manual feature extraction algorithms have poor recognition results.Scholars have proposed deep learning to automatically extract signal features and classify them.In recent years,such algorithms have continued to emerge,but there are still two problems.First,the cost of marking radar signals,which is already expensive,has doubled with the optimization of the algorithm;In addition,the network scale is not controlled,the algorithm recognition performance has not been improved to the same extent.This paper attempts to solve the above two problems while ensuring the performance of algorithm recognition.The main work of this paper is summarized as follows:1.Summarize the types of radar application modulation in the actual battlefield,and analyze the performance of typical modulation type radar signals combined with mathematical models and simulation results.Simulate Short-Time Fourier Transform(STFT),Wigner-Ville Distribution(WVD)and Cohen's Time-Frequency Distribution with common radar signals as objects.Test and analysis of results,and finally summarize the characteristics of time-frequency images of various types of radar signals.2.Propose a radar signal recognition algorithm based on Active,Incremental Fine-Tuning(AIFT).Through a query strategy based on sample uncertainty,the network independently selects samples with training value.After the experts mark,fine-tune the pre-training network to realize the active learning of the network.The simulation results show that when the cost of labeling by experts of this algorithm is reduced to 63% of other algorithms,when the SignalNoise Ratio(SNR)is-9 d B,the correct rate of identifying twelve kinds of signals exceeds 90%.While greatly saving the cost of labeling,it has the same or even better recognition performance than current algorithms.3.A radar signal recognition algorithm based on a lightweight network is proposed,which uses two different lightweight ideas,Fire Module and group convolution,respectively.Two network models for radar signal recognition are designed when the network parameters are the same.The simulation results of the two network structures are analyzed in terms of direct indicators,indirect indicators and recognition performance.In the end,the two lightweight networks can effectively reduce the network scale while ensuring the generalization and noise resistance of the algorithm.The simulation results show that the combination of AIFT and the radar signal recognition algorithm of lightweight network can reduce the amount of network parameters and the cost of radar signal labeling.At the same time,the signal-to-noise ratio is-9 d B,the overall signal recognition accuracy rate can reach 95.5%.
Keywords/Search Tags:Radar signal, Pulse modulation type recognition, Time-frequency analysis, Active learning, Lightweight network
PDF Full Text Request
Related items