Since the emergence of 5G technology, the wireless communication system has had a huge data throughput, so the joint development of artificial intelligence technology and wireless communication technology is one of the current mainstream development directions. In particular the combination of deep learning technology and communication physical layer technology is the future research hotspot. The purpose of this research paper is to summarize the related algorithms of the combination of Automatic Modulation Recognition (AMR) technology and deep learning technology in the communication physical layer. In order to elicit the advantages of the modulation recognition algorithm based on deep learning, this paper firstly introduces the traditional AMR method, and then summarizes the advantages and disadvantages of the traditional algorithm. Then, the application of the deep learning algorithm in AMR is described, and the identification method based on a typical deep learning network is emphatically described. Afterwards, the existing Deep Learning (DL) modulation identification algorithm in a small sample environment is summarized. Finally, DL modulation is discussed, identifying field challenges, and future research directions.

Automatic modulation recognition is a critical step between signal detection and signal demodulation, and it is a critical technology for ensuring proper communication. Since the introduction of 5G technology, wireless communication systems have had massive data throughput, so combining deep learning technology with modulation recognition technology is now one of the most mainstream development directions in the field of communication. In order to effectively improve the robust recognition accuracy of modulation signals in low SNR condition, this paper proposes an end-to-end AMR model based on deep learning, called residual convolution long short memory Improved Transformer-Encoder deep neural network model (RLITNN). First, convolutional network and LSTM network extract the initial features from the communication signals of different modes. Second, the proposed Improved Transformer-Encoder module is used to capture the global and local focus of the extracted features. Then, these features are fused. Finally, the high quality features are further captured from the fused feature vector to enhance the feature representation ability of the model. Experiments on RML2016.10A and RML2016.10B datasets show that the proposed RLITNN has better feature learning ability and better recognition accuracy than other advanced (SOTA) techniques.

Radio spectrum resources are very limited and have become increasingly tight in recent years, and the exponential growth of various frequency-using devices has led to an increasingly complex and changeable electromagnetic environment. Wireless channel complexity and uncertainty have increased dramatically, and automated modulation recognition (AMR) performs poorly at low signal-to-noise ratios. It is proposed to use convolutional bidirectional long short-term memory deep neural networks (CNN-BiLSTM-DNNs) as a deep learning framework to extract features from single and mixed in-phase/orthogonal (I/Q) symbols in modulated data. The framework combines the capabilities of one- and two-dimensional convolution, a bidirectional long short-term memory network, and a deep neural network more efficiently, extracting characteristics from the perspective of time and space to enhance the accuracy of automatic modulation recognition. Modulation recognition experiments on the benchmark datasets RML2016.10b and RML2016.10a show that the average recognition accuracies of the proposed model from −20 dB to 18 dB are $64.76\%$ and $62.73\%$, respectively, and the improvement ranges of modulation recognition accuracy are 0.29−5.56% and 0.32−4.23% when the signal-to-noise ratio (SNR) is −10 dB to 4 dB, respectively. The CNN-BiLSTM-DNN model outperforms classical models such as MCLDNN, MCNet, CGDNet, ResNet, and IC-AMCNet in terms of modulation type recognition accuracy.

In order to improve the spectrum sensing (SS) performance under a low Signal Noise Ratio (SNR), this paper proposes a supervised learning spectrum sensing method based on Geometric Power (GP) feature. The GP is used as the feature vector in the supervised learning spectrum sensing method for training and testing based on the actual captured data set. Experimental results show that the detection performance of the GP-based supervised learning spectrum sensing method is better than that of the Energy Statistics (ES) and Differential Entropy (DE)-based supervised learning spectrum sensing methods.