This work was supported in part by the National Natural Science Foundation of China under Grant 61701462 and Grant 41527901, in part by the Qingdao National Laboratory for Marine Science and Technology under Grant 2017ASKJ01, in part by the Qingdao Science and Technology Plan under Grant 17-1-1-7-jch, in part by the Fundamental Research Funds for the Central Universities under Grant 201713018, in part by the Shandong Province Natural Science Foundation under Grant ZR2017BF023, in part by the China Postdoctoral Science Foundation funded project under Grant 2017M612223, and in part by the Shandong Province Postdoctoral Innovation Project under Grant 201703032.

In wireless communication, modulation classification is an important part of the non-cooperative communication, and it is difficult to classify the various modulation schemes using conventional methods. The deep learning network has been used to handle the problem and acquire good results. In the deep convolutional neural network (CNN), the data length in the input is fixed. However, the signal length varies in communication, and it causes that the network cannot take advantage of the input signal data to improve the classification accuracy. In this paper, a novel deep network method using a multi-stream structure is proposed. The multi-stream network form increases the network width, and enriches the types of signal features extracted. The superposition convolutional unit in each stream can further improve the classification performance, while the shallower network form is easier to train for avoiding the over-fitting problem. Further, we show that the proposed method can learn more features of the signal data, and it is also shown to be superior to common deep networks.

Automatic modulation recognition has successfully used various machine learning methods and achieved certain results. As a subarea of machine learning, deep learning has made great progress in recent years and has made remarkable progress in the field of image and language processing. Deep learning requires a large amount of data support. As a communication field with a large amount of data, there is an inherent advantage of applying deep learning. However, the extensive application of deep learning in the field of communication has not yet been fully developed, especially in underwater acoustic communication. In this paper, we mainly discuss the modulation recognition process which is an important part of communication process by using the deep learning method. Different from the common machine learning methods that require feature extraction, the deep learning method does not require feature extraction and obtains more effects than common machine learning.

Automatic modulation recognition (AMR) is one of the essential parts in the intelligent communication system. In the underwater acoustic communication, it is a challenging work that promptly and easily recognizes the signal modulation schemes by conventional methods. The deep neural network method is a good solution to the problem, which creates a better recognition effect. The packets of data that are fed to the familiar neural network is constant. However, the packets of signal data on the communication course consistently change, which seriously reflects on the signal recognition veracity. A novel deep learning network with the sequence convolutional network in this paper is proposed, which is composed of one-dimensional sequence convolution of residual network modules and the variable convolution kernel range. By extracting the time-domain signal characteristics, the affection of various signal packets can be mitigated. In experiments, the employed network not only has more concentrated on the modulation recognition veracity, but also owns a lower parameter quantity and a shorter training time, which indicates ideal recognition results in the underwater communication environment. Moreover, it is more valuable to the real underwater communication system.