Research On Blind Recognition Method Of Convolutional Code Based On Deep Learning

With the continuous development of modern communications,both communication parties are faced with various challenges brought about by the complex changes in the channel environment,which will lead to increasingly frequent use of spectrum resources,resulting in a shortage of spectrum resources.The shortage of spectrum resources brings new challenges to the development of modern communications.Adaptive Modulation and Coding(AMC)technology is an effective technology to improve the utilization of spectrum resources.The receiver needs to frequently transmit parameters in the control channel according to the state of the transmitter for adaptive adjustment,which will inevitably consume additional channel resources.Therefore,it is necessary for the receiver to blindly identify the channel coding adopted by the transmitter,so as to reduce the utilization rate of the control channel and alleviate the spectrum resources.The traditional channel coding blind recognition technology usually needs to extract features manually and identify parameters through complex matrix operation in the case of specific channel and no error code.It is difficult to realize and the recognition accuracy is low.In recent years,due to the rapid development of deep learning,many efficient neural network models have emerged,which have the ability of automatic feature extraction and classification and recognition.Therefore,in order to solve the shortcomings of traditional channel coding blind recognition technology,this paper applies deep learning to the field of coding blind recognition.Combined with the characteristics of convolutional code encoder,two methods based on neural network model are proposed,and an effective preprocessing method is designed to extract the features of convolutional code,so as to improve the blind recognition performance of convolutional code.The main work is as follows:Firstly,a blind convolutional code recognition method based on TextCNN-LSTM model is proposed.The received sequence is processed with soft information as the sequence to be identified by the convolutional code,and then enters the preprocessing stage.At this stage,the soft information sequence is quantized at equal intervals in a certain interval,and the quantized data are combined into words.The dictionary matrix is obtained by processing the sentence matrix composed of these words with a word splitter,which is used as the input data of the network model.In addition,TextCNN-LSTM model is built according to the characteristics of Long Short-Term Memory(LSTM)network and convolutional encoder.The preprocessed dictionary matrix is input into the model for training,and the model is adjusted and optimized.The simulation results show that the method based on TextCNNLSTM model has good recognition performance,and can show good recognition effect for convolutional codes with long received sequence.Then,aiming at the great limitation of TextCNN-LSTM model,a blind recognition method of convolutional code based on Res Net model is proposed.This method proposes two preprocessing schemes,namely SPP mean calculation and normalization based on soft information,which greatly simplifies the complexity of data preprocessing.In addition,the Res Net network model is built to simplify the training objectives and improve the training speed on the premise of ensuring the data integrity.From the simulation results,the normalization method based on Res Net model has the best blind recognition performance for convolutional codes.It can maintain good recognition accuracy in Gaussian channel and single path Rayleigh fading channel,and has a certain degree of improvement compared with other methods.At the same time,the blind recognition performance of convolutional codes with different bit rates is tested,and it is found that under the same constraint length,the recognition performance of convolutional codes with lower bit rate is better.