Research On Transient And Steady Atate Signal Recognition Based On Deep Learning And Machine Learning

Transient signal and steady-state signal recognition is important research direction in the field of signal recognition,which plays an important role in non-communication scenarios,such as electronic reconnaissance and electronic countermeasures.By identifying the modulation method of the steady-state modulation signal,we can further understand more of enemy combat communication equipment,then demodulation and parameter estimation are performed to obtain more information,so that our army is able to make timely and accurate decisions.In the civil field,some criminals occupy valuable frequency band resources that can interfere with normal communication.By detecting the outdoor communication environment,government departments can analyze the interference signals and track the legal responsibilities of relevant personnel in a timely manner through the modulation type to maintain a secure and stable communication order.Transient signals also play an important role in identifying individual radiation sources.For example,we can evaluate the reliability of complex electronic systems to find broken down machines in time,so that the stability of the overall system and the ability to handle faults in production can be improved.In the end the system can be carried out smoothly and orderly.So,this paper proposes recognition schemes for transient open signals based on machine learning and steady-state modulation signals based on deep learning.The main contents and innovations of this article are as follows:(1)Due to the randomness and instantaneity of the transient open signals,how to accurately determine the starting point has become a difficulty in the current research.In order to solve this problem,this paper proposes a method based on a sliding window.Adapt to the effective segment intercept algorithm of variance.To solve this problem,this paper proposes an effective variance interception algorithm based on sliding window adaptive variance,which can take advantage of the variance characteristics of transient signals.The threshold of the starting point changes when the noise in the environment changes,so it can separate the transient power-on signal from the original signal in the case of high noisy conditions,So we can extract purer features in the next steps.(2)A 38-dimensional feature map is proposed which include time domain,frequency domain and space domain.Then feature importance analysis is performed.In the offline situation based on the characteristics of transient signals,we created six machine learning models and one model fusion,improving the recognition accuracy of the four electronic devices from 96.21%to 97.91%,which also improved the stability of the model.(3)Set up a ResNet 34-layer residual network for the identification of 16 kinds of steady-state modulated signals.Using Hilbert transform and adding AM-Softmax layers to improve the recognition rate.We also build a real communication environment to make experiments to verify the Effectiveness of the network.