The era of ubiquitous, affordable wireless connectivity has opened doors to countless practical applications. In this context, ambient backscatter communication (AmBC) stands out, utilizing passive tags to establish connections with readers by harnessing reflected ambient radio frequency (RF) signals. However, conventional data detectors face limitations due to their inadequate knowledge of channel and RF-source parameters. To address this challenge, we propose an innovative approach using a deep neural network (DNN) for channel state information (CSI) estimation and signal detection within AmBC systems. Unlike traditional methods that separate CSI estimation and data detection, our approach leverages a DNN to implicitly estimate CSI and simultaneously detect data. The DNN model, trained offline using simulated data derived from channel statistics, excels in online data recovery, ensuring robust performance in practical scenarios. Comprehensive evaluations validate the superiority of our proposed DNN method over traditional detectors, particularly in terms of bit error rate (BER). In high signal-to-noise ratio (SNR) conditions, our method exhibits an impressive approximately 20% improvement in BER performance compared to the maximum likelihood (ML) approach. These results underscore the effectiveness of our developed approach for AmBC channel estimation and signal detection. In summary, our method outperforms traditional detectors, bolstering the reliability and efficiency of AmBC systems, even in challenging channel conditions.

Monostatic backscatter (MBS) networks provide connectivity for ultra-low power and low-cost tags for numerous applications. However, the reader operates in the full-duplex (FD) mode and experiences self-interference (SI) levels much higher (e.g., 160 dB) than the desired signal. However, hardware-based SI cancellation (SIC) techniques can remove SI partially only. Therefore, residual SI (RSI) dramatically degrades system performance. To remedy this problem, we develop a sum-rate maximization algorithm that suppresses the RSI and ensures that the tags harvest sufficient energy. It jointly optimizes the reader precoder and combiners and the tags reflection coefficients. Because of the non-convexity of this problem, we utilize alternating optimization (AO) to split it into three parts. They are then solved using successive convex approximation (SCA) and semidefinite relaxation (SDR) techniques to yield the precoder, a generalized Rayleigh quotient-based closed-form solution for the combiners, and geometric programming (GP) to get the reflection coefficients. Simulation results validate the fast convergence of the algorithm and show significant sum rate improvements (more than 21%) over the baselines.

In this paper, we consider a non-orthogonal multiple access cognitive radio network, where a full-duplex (FD) multi-antenna relay assists transmission from an access point (AP) to a cognitive far user, while at the same time, the AP transmits to a cognitive near user. Our objective is to maximize the rate of the near user under a constraint that the rate of the far user is above a certain threshold. To this end, a non-convex joint optimization problem of relay beamforming and the transmit powers at the AP and FD relay is solved as a semi-definite relaxation problem, in conjunction with an efficiently solvable line-search approach. We also consider a low complexity fixed beamformer design, where the optimum power allocation between the AP and FD relay is solved. Several fixed beamforming designs based on the zero-forcing criterion are proposed for which exact and asymptotic outage probability expressions corresponding to the near and far users are derived. Our results demonstrate that the proposed joint optimization can significantly reduce the self-interference impact at the FD relay and inter-user interference in the near user case.

In this paper, we consider a non-orthogonal multiple access cognitive radio network, where a full-duplex multi-antenna relay assists transmission from a base station (BS) to a cognitive far user, whereas, at the same time, the BS transmits to a cognitive near user. Our objective is to enlarge the far-near user rate region by maximizing the rate of the near user under a constraint that the rate of the far user is above a certain threshold. To this end, a non- convex joint optimization problem of relay beamforming and the transmit powers at the BS and cognitive relay is solved as a semi-definite relaxation problem, in conjunction with an efficiently solvable line-search approach. For comparisons, we also consider low complexity fixed beamformer design, where the optimum power allocation between the BS and cognitive relay is solved. Our results demonstrate that the proposed joint optimization can significantly reduce the impact of the residual self-interference at the FD relay and inter-user interference in the near user case.