Reconfigurable intelligent surfaces (RIS)-based communications with reflection modulation (RM) is a novel area of research that opens up a range of unconventional modulation techniques. Existing literature primarily focuses on specific applications where the RIS encodes its own information onto its reflection pattern. Quadrature reflection modulation (QRM) and reflection pattern modulation (RPM) are two promising reflection pattern designs that effectively deliver local data available at the RIS. This paper explores a more general application of RIS-based information transfer for a single-user downlink system via jointly mapped RM (JRM), where the RIS and the access point (AP) jointly deliver the information available at the AP. The data symbols are mapped to a constellation of tuples, each tuple containing a transmit signal and a reflection pattern. Two JRM constellation designs are proposed, namely jointly-mapped QRM (JQRM) and jointly-mapped RPM (JRPM). The proposed constellation design employs a smaller transmit signal set size compared to a generic modulation scheme, increasing the separation among adjacent constellation points. A jointly active and passive beamforming design is adopted for a multiple-input-single-output (MISO) downlink system. The simulation results analyze and compare the bit-error-rate (BER) performance of the proposed JQRM and JRPM schemes, with their respective separately-mapped counterparts and theoretical upper bounds as benchmarks.

A deep learning (DL)-based joint pilot and data power control algorithm that solves the sum rate maximization problem in a cell-free massive multiple-input multiple-output (MIMO) system is proposed. The sum rate optimization problem for the uplink is formulated subject to per-user total transmit energy budget constraints, where user pilot and data power allocations are optimized to maximize the system sum rate. Instead of solving the non-convex problem using mathematical optimization theory, we utilize a data-driven solution approach to learn the optimal solutions. Specifically, we model a deep neural network (DNN) and train it via unsupervised learning using a custom loss function that captures the sum rate optimization objective and transmit energy constraints in the optimization problem. This unsupervised learning approach has a simpler and more flexible model training stage since it does not require labeled data for model training as in supervised learning. Simulation results show that the proposed DNN-based joint pilot and data power control algorithm improves the system sum rate compared to equal power and equal energy allocation heuristics and data power control-only approach. Furthermore, the joint power allocation results in significant energy savings (around 60 %) compared to fixed power allocation schemes.

A deep learning (DL)-based power control algorithm that solves the max-min user fairness problem in a cell-free massive multiple-input multiple-output (MIMO) system is proposed. Max-min rate optimization problem in a cell-free massive MIMO uplink setup is formulated, where user power allocations are optimized in order to maximize the minimum user rate. Instead of modeling the problem using mathematical optimization theory, and solving it with iterative algorithms, our proposed solution approach is using DL. Specifically, we model a deep neural network (DNN) and train it in an unsupervised manner to learn the optimum user power allocations which maximize the minimum user rate. This novel unsupervised learning-based approach does not require optimal power allocations to be known during model training as in previously used supervised learning techniques, hence it has a simpler and flexible model training stage. Numerical results show that the proposed DNN achieves a performance-complexity trade-off with around 400 times faster implementation and comparable performance to the optimization-based algorithm. An online learning stage is also introduced, which results in near-optimal performance with 4-6 times faster processing.

End-to-end learning of a communications system using the deep learning-based autoencoder concept has drawn interest in recent research due to its simplicity, flexibility and its potential of adapting to complex channel models and practical system imperfections. In this paper, we have compared the bit error rate (BER) performance of autoencoder based systems and conventional channel coded systems with convolutional coding (CC), in order to understand the potential of deep learning-based systems as alternatives to conventional systems. From the simulations, autoencoder implementation was observed to have a better BER in 0-5 dB Eb/N0range than its equivalent half-rate convolutional coded BPSK with hard decision decoding, and to have only less than 1 dB gap at a BER of 10-5. Furthermore, we have also proposed a novel low complexity autoencoder architecture to implement end-to-end learning of coded systems in which we have shown better BER performance than the baseline implementation. The newly proposed low complexity autoencoder was capable of achieving a better BER performance than half-rate 16-QAM with hard decision decoding over the full 0-10 dB Eb/N0range and a better BER performance than the soft decision decoding in 0-4 dB Eb/N0range.