Research On Modulation Recognition Based On CNN In Non-cooperative Underwater Acoustic Communication

At present,the marine field is an important economic and military strategic focus,which highlights the importance of underwater acoustic communication technology.In communication monitoring,research on modulation recognition of underwater communication is of great significance.The complexity of the marine communication environment brings about the Doppler effect and multipath effect of the underwater acoustic channel.At the same time,it has the characteristics of time and space variation due to the influence of the ocean dynamics.Therefore,the identification of the acquired unknown underwater acoustic signals faces many challenges.In recent years,the rapid development of deep learning technology has made breakthroughs in image recognition and text classification.This paper extracts the preferred features of the received signal,and based on deep learning network technology,introduces deep learning frameworks and models to achieve water blind recognition of acoustic signal modulation methods.The main work is as follows:1.In the underwater acoustic modulation recognition algorithm,different characteristic parameters are studied and extracted,and its theoretical analysis and performance simulation verification are carried out.Based on independent simulation and experimental results,a series of parameters are optimized.First,the maximum correlation coefficient of the spectrum and the instantaneous frequency variance are selected to classify the three major types of MASK,MFSK and MPSK signals.The modulation order of the specific MASK signal is identified by amplitude compactness.Exponent,the number of peaks estimated by the signal spectrum and the mean value of the envelope are used to distinguish 2FSK and 4FSK signals.At the same time,based on the characteristics of comb-shaped dispersion curves near the peak of the quadratic spectrum of the 4PSK signal,the maximum coefficient average of the quadratic spectrum is extracted.Distinguish the specific modulation order of the MPSK signal.2.In order to improve the classification and recognition rate of MFSK signals in a multipath environment,a power spectrum image feature recognition algorithm based on shorttime Fourier transform is proposed.The algorithm performs short-time Fourier transform on the received signal,and effectively extracts its power spectrum data,and classifies it through a convolutional neural network.The feature map can also reflect local frequency features with time as the dimension even in small sample data and in the actual pool channel with multiple obvious paths.The experimental results prove that the performance of this algorithm is improved by 6.25%compared with the modern spectrum estimation method to obtain specific parameters.3.Existing classification research results usually use a single feature to distinguish between two types of signals.This research proposes to use multi-feature fusion into image data to improve the noise resistance and robustness of the recognition algorithm,and then put it into a convolutional neural network.Learning and classifying,compared with the support vector machine as a classifier,has a higher recognition accuracy rate,and using two different deep learning network models AlexNet and VGG16 to verify and compare,obtain the model that best matches this algorithm.4.Aiming at the underwater acoustic signal recognition algorithm proposed in this paper,the actual pool channel,ocean channel and noise are experimentally explored.In order to verify the algorithm,a pool experiment was carried out,and the average recognition error of multiple types of signals is only 1.42%.At the same time,the high recognition rate of MPSK signals is also classified through sea trial experimental data.