Automatic Modulation Classification And Channel Coding Based On Neural Network

Neural Network is an important branch of artificial intelligence machine learning.Neural network gain“experience”through learning,and can solve and simplify many problems with complex calculations,such as image recognition,natural language processing,and machine translation.In the field of communications,there are also many problems with complex calculations,such as automatic modulation classification,channel coding and so on.Therefore,this article aims to study the combination of neural network and communication system,and use neural network to solve communication problems.First,the research based on the detection of modulation signal parameters of the neural network involves the detection of the signal-to-noise ratio of the modulation signal and the detection of the carrier frequency offset of the modulation signal.The research requires that the performance index is within the allowable error range,and the detection accuracy rate should reach more than 90%,this is,the total number of test samples whose parameter detection error is less than the allowable error range should account for more than 90%of the total number of test samples.The allowable error range specified here is10%of the maximum value of the corresponding parameter.The research method adopted is an algorithm model based on neural network feature extraction,feature generation,and feature processing for signal-to-noise ratio detection and a neural network algorithm model based on extensive simulations to carrier frequency offset.The communication channel environment required by the parameter detection is basically the same as the communication channel environment specified by the automatic modulation classification.The only difference is the fixed signal-to-noise ratio is changed to a random number of 0-30d B in the signal-to-noise ratio detection,and the rest of the parameters will not be repeated here.Finally,based on the neural network,the detection accuracy of the modulation signal signal-to-noise ratio detection is within the allowable error range of 97.33%;for the modulation signal carrier frequency offset detection,the detection accuracy is within the allowable error range detection accuracy rate 92.22%.Secondly,the research on the automatic modulation classification based on neural networks.The modulation types involved include 14 common digital modulation types.The research requires that the performance index is to achieve an average recognition accuracy of more than 90%under the specified communication channel environment,and it is required that normal recognition can be achieved for each modulation type,that is,the number of correctly recognized test samples must be greater than the incorrectly recognized test Number of samples.The designated channel here is a channel with a signal-to-noise ratio of 10d B,a carrier frequency offset of±5000Hz,a carrier phase offset of 0?2?,and a channel with 0-15 fixed time offsets for sampling points.The algorithm scheme based on the coordination of modulation recognition neural network and modulation signal carrier frequency detection is adopted.Under the premise of ensuring relatively simple network structure and high network operation efficiency,the average accuracy of the network under the specified communication channel environment is94.98%.And through software radio simulation to verify the effectiveness of the algorithm for actual communication signals.Cooperating with a modulation type recognition network with a software test performance of 99.07%and a carrier frequency shift detection network can achieve an average recognition accuracy of 96.71%for the actual communication signal modulation type between two software radio devices under certain conditions.Finally,the research on channel coding based on neural network.The research performance index is that in the Gaussian white noise channel,when the signal-to-noise ratio is 5d B,the bit error rate of the coding network can reach 10^-5.Use neural network learning to obtain channel coding algorithms,breaking the constraints of existing channel coding methods,and replacing complete data set preparation by randomly generating 0and 1 sequences.And cascade the trained channel coding network with a simple parity check module to obtain the final coding algorithm.When the algorithm error rate performance can reach a signal-to-noise ratio of 2.5d B,the error rate of the coding and decoding network can reach 10^-5,and the complexity of the decoding algorithm is relatively low.In summary,this article mainly studies the neural network to solve the modulation signal parameter detection,automatic modulation classification and channel coding problems in the communication system,and has reached a certain performance index.