Non-orthogonal multiple access (NOMA) is a promising scheme to improve the spectral efficiency, user fairness, and overall throughput in visible light communication (VLC) systems. However, the error propagation problem together with linear and nonlinear distortions induced by multipath, limited modulation bandwidth, and nonlinearity of light emitting diode significantly limit the transmission performance of NOMA-VLC systems. In addition, having an accurate channel state information, which is important in the recovery of NOMA signal, in mobile wireless VLC is challenging. In this work, we propose a convolutional neural network (CNN) based demodulator for NOMA-VLC, in which signal compensation and recovery are jointly realized. Both simulation and experiment results show that, the proposed CNN based demodulator can effectively compensate for both linear and nonlinear distortions, thus achieving improved bit error ratio performance compared with the successive interference cancellation and joint detection based receivers. Compared to SIC, the performance gains are 1.9, 2.7, and 2.7 dB for User1 for the power allocation ratios (PARs) of 0.16, 0.25, and 0.36, respectively, which are 4, 4 and 2.6 dB for User2 for PARs of 0.16, 0.25, and 0.36, respectively.

This paper presents a parametric low-dimensional (LD) representation learning method that allows to reconstruct high-dimensional (HD) input vectors in an unsupervised manner. Under the assumption that the HD data and its LD representation share the same or similar local sparse structure, the proposed method achieves reconstructible dimensionality reduction via jointly learning dictionaries in both the original HD data space and its LD representation space. By regarding the sparse representation as a smooth function with respect to a specific dictionary, we construct an encoding-decoding block for learning LD representations from sparse coefficients of HD data. It is expected that this learning process preserves the desirable structure of HD data in the LD representation space, and simultaneously allows a reliable reconstruction from the LD space back to the original HD space. In addition, the proposed single layer encoding-decoding block can be easily extended to deep learning structures. Numerical experiments on both synthetic data sets and real images show that the proposed method achieves strongly competitive and robust performance in data DR, reconstruction, and synthesis, even on heavily corrupted data. The proposed method can be used as an alternative approach to compressive sensing (CS); however, it can outperform the traditional CS methods in: 1) task-driven learning problems, such as 2-D/3-D data visualization, and 2) data reconstruction at a lower dimensional space.

An optical signal suffers from irradiance and phase fluctuations when propagating through the free space optical (FSO) turbulence channel, thus resulting in the degradation of the bit error rate (BER) performance. The BER performance can be improved by adopting the multiple-input multiple-output (MIMO) scheme. In this paper, we propose a coherent binary polarization shift keying (BPOLSK) modulation scheme with MIMO employing maximum ratio combining and equal gain combining diversity techniques to mitigate the turbulence effect. The gamma-gamma statistical channel model is adopted for all the turbulence regimes. The BER performance for the proposed BPOLSK-MIMO FSO link is compared with the single-input single-output and ON-OFF-keying systems by means of computer simulation. The optical power gain is investigated and demonstrated under different turbulence regimes for a number of transmitters/receivers.

The Hybrid Non-Orthogonal Multiple Access (NOMA) is a promising candidate for multiple access techniques of future wireless communication, which integrates orthogonal multiple access and traditional NOMA. The performance of hybrid NOMA systems depends on resource allocation including power and channel. In this letter, we focus on the channel assignment. Since channel assignment needs to be adapted to a real-time changing environment and accomplished in a restricted time slot, we treat the optimization of the dynamic channel assignment problem as a deep reinforcement learning task, to achieve better environmental adaptability with low time complexity. Simulation results show that the proposed method achieves better performance in terms of sum rate and spectral efficiency, compared to conventional methods.