Non-orthogonal multiple access (NOMA) technology allows multiple users to use the same time-frequency resource to send signals, which can improve spectral efficiency and throughput. We study the problems of user grouping and power allocation in the downlink of a multi-carrier NOMA system. The sum rate is the optimization goal. A step-by-step optimization is adopted. Users are grouped first and then power is allocated. For user grouping, the user grouping method based on the maximum channel gain difference is improved to prevent users with similar channel gains from being grouped together. For power allocation, the deep learning power allocation algorithm is used among subcarriers. Then, the closed-form expressions of power allocation between multiplexed users are derived according to the minimum transmission rate constraint. The simulation results show that compared with the fractional transmit power allocation method and fixed power allocation method, the deep learning power allocation method improves the system sum rate by about 2.2% and 19%, respectively. The power allocation methods we propose improve the system sum rate by about 10% compared to the fractional transmit power allocation method used among subcarriers and between multiplexed users.

The orthogonal frequency division multiplexing (OFDM) technique has received wide attention because of its high spectrum utilization. However, the drawback of inter-subcarrier interference in OFDM systems makes the channel estimation and signal detection performance of OFDM systems with few pilots and short cyclic prefixes (CP) poor. In this paper, we use deep learning to assist OFDM in recovering nonlinearly distorted transmission data. Specifically, we use a self-normalizing network (SNN) for channel estimation, combined with a convolutional neural network (CNN) and a bidirectional gated recurrent unit (BiGRU) for signal detection, thus proposing a novel SNN-CNN-BiGRU network structure (SCBiGNet). The simulation results show that the SCBiGNet model outperforms the existing techniques for the different numbers of pilots and lengths of CPs. The BER performance is improved by 0.2~9 dB.

In a cognitive radio network (CRN), spectrum sensing is an important prerequisite for improving the utilization of spectrum resources. In this paper, we propose a novel spectrum sensing method based on deep learning and cycle spectrum, which applies the advantage of the convolutional neural network (CNN) in an image to the spectrum sensing of an orthogonal frequency division multiplex (OFDM) signal. Firstly, we analyze the cyclic autocorrelation of an OFDM signal and the cyclic spectrum obtained by the time domain smoothing fast Fourier transformation (FFT) accumulation algorithm (FAM), and the cyclic spectrum is normalized to gray scale processing to form a cyclic autocorrelation gray scale image. Then, we learn the deep features of layer-by-layer extraction by the improved CNN classic LeNet-5 model. Finally, we input the test set to verify the trained CNN model. Simulation experiments show that this method can complete the spectrum sensing task by taking advantage of the cycle spectrum, which has better spectrum sensing performance for OFDM signals under a low signal-noise ratio (SNR) than traditional methods.

Reliable signal detection plays an essential role in enhancing the quality of signal transmission in wireless communication systems. In this paper, we combine signal detection theory with a deep learning model and propose a novel signal detection scheme based on adaptive ensemble long short term memory (AE-LSTM) neural network to handle wireless single carrier frequency domain equalization (SC-FDE) systems in an end-to-end manner. The feature information used for offline training of the deep learning model is extracted from the received signal containing channel state information (CSI) after the multi-path channel and fast Fourier transform (FFT), and the labels are assigned according to the constellation map adopted at the transmitter. To improve the adaptability of the system, we utilize the received power under different delays as the adaptive factor to integrate the output of each sub-network. Then the original data generated by the channel model is recovered by using the trained model instead of channel estimation and frequency domain equalization. Comparative experiments on SC-FDE symbol detection demonstrate that the proposed scheme achieves better performance in terms of reliability than the traditional scheme and the similar deep learning scheme.

IQ modulation enjoys great popularity in wireless communication. Its main advantage is the symmetric convenience of combining independent signal components into a single composite signals in the transmitter. And it splits the composite signal into its independent components in the receiver. However, due to the limitations of communication facilities and imperfect channel environments, the symbols will be distorted. In this paper, we propose several signal processing schemes based on deep learning (DL) neural networks (i.e., long short-term memory (LSTM)) to process signal in an end-to-end manner, known as DLA, EDLA and PDNet. These schemes combine with advanced DL architectures and data-driven models for complex-valued signal processing. Enlightened by IQ demodulation ideas, we adopt LSTMs to develop several schemes to implement different functions (i.e., signal detection and peak to average power ratio (PAPR) reduction). The simulation results show that the schemes have significant performance improvement in bit error rate (BER) while reducing the probability of high PAPR.

Adaptive transmission (AT) is considered as one of the critical technologies to enhance the effectiveness of communication systems. In this article, we propose a model-driven deep learning (DL) scheme for AT in multiple-input multiple-output single-carrier frequency-domain equalization (MIMO-SCFDE) systems, in which the adaptive modulation network (AMNet) and adaptive demodulation network (ADNet) are adopted to complete the modulation of the signal and the modulation recognition of the receiver. Under the target bit error rate (BER), the adaptive modulation (AM) scheme can adjust the modulation mode selection of different transmitting antennas adaptively according to the estimated channel information to improve the throughput. The features required by the AMNet are extracted from the received signal, and the labels are assigned according to the optimal modulation scheme got by analyzing the signal detection performance. Since the spectral correlation function has a powerful ability to suppress noise and the cyclic spectrum varies with the modulation mode, we take the preprocessed cyclic spectrogram as the input of ADNet to achieve the adaptive modulation recognition (AMR). Comparative experiments demonstrate that the proposed scheme gets better performance in terms of throughput and reliability in MIMO-SCFDE systems than the traditional scheme and the existing DL scheme.