Device-to-device (D2D) communications allow short-range communication devices to multiplex cellular-licensed spectrum to directly establish local connections for ultra-high number of terminal connections and greater system throughput. However, spectrum sharing also brings serious interference to the network. Therefore, a reliable and efficient resource allocation strategy is important to mitigate the interference and improve the system spectral efficiency. In this paper, we investigated spectrum access and power allocation in D2D communications underlay cellular networks based on deep reinforcement learning with the aim of finding a feasible resource allocation strategy to maximize data rate and system fairness. We proposed a value decomposition network-based resource allocation scheme for D2D communication networks. Our proposed scheme avoids frequent information exchanges among D2D users by centralized training, while allowing D2D users to make distributed joint resource allocation decisions. Simulation results show that the proposed scheme has stable convergence and good scalability, and can effectively improve the system capacity.

Video-based deception detection, which identifies lies through facial expressions and behaviors, has proven to be an effective approach in criminal interrogation. In this paper, a deception detection framework is proposed that incorporates a novel set of features and a unique deception detection method based on facial expressions, particularly micro-expressions. Two feature selection methods are applied to optimize these features. Specifically, facial action units (AUs), eye gaze, and head pose were extracted using the OpenFace toolkit, while micro-expression information was obtained via the SOFTNet model, trained on the CAS(ME) $$^{2}$$ data set. A sequential combination of the Fischer Score and Principal Component Analysis (PCA) was employed for feature selection, with a Support Vector Machine (SVM) used for classification. Feature importance analysis indicated that micro-expression (ME) information had a significant impact on the deception detection task. The proposed framework was evaluated on two publicly available data sets, achieving accuracies of 98.07% and 98.23% on the real-life and MU3D data sets, respectively, thus demonstrating its superiority over prior approaches in the literature.

Signal to interference plus noise ratio maximization (Max-SINR) algorithm are designed to increase the transmission rate of multiple-input multiple-output interference channels (MIMO IC). The algorithms are sensitive to channel estimation error, which makes the rate performance under imperfect channel state information (CSI) conditions become unsatisfactory. To deal with this, a relay assisted signal to interference plus noise ratio maximization (Relay-Max-SINR) algorithm is proposed in this work to improve the robustness of Max-SINR in MIMO ICs. Different from the traditional relay assisted using beamforming and forward transmission, we mathematically equivalent the signal processing process of relay to the receive filtering and transmit precoding, and design the relay processing matrix. In real communication process, the relay processing matrix is directly used for the signal processing at relay. Using the channel reciprocity, the receive filtering matrix at relay, the transmit precoding matrix at relay, the relay processing matrix, the receive filtering matrix at destination and the transmit precoding matrix at source are optimized iteratively according to the nature of generalized Rayleigh quotient. With the help of additional transmission links provided by the relay, interference is more effectively suppressed and the power loss of the desired signal is reduced. Simulations show that the proposed Relay-Max-SINR algorithm can significantly increase the capacity and the robustness of MIMO ICs.

A practical solution to the power allocation problem in ultra-dense small cell networks can be achieved by using deep reinforcement learning (DRL) methods. Unlike traditional algorithms, DRL methods are capable of achieving low latency and operating without the need for global real-time channel state information (CSI). Based on the actor–critic framework, we propose a policy optimization of the power allocation algorithm (POPA) for small cell networks in this paper. The POPA adopts the proximal policy optimization (PPO) algorithm to update the policy, which has been shown to have stable exploration and convergence effects in our simulations. Thanks to our proposed actor–critic architecture with distributed execution and centralized exploration training, the POPA can meet real-time requirements and has multi-dimensional scalability. Through simulations, we demonstrate that the POPA outperforms existing methods in terms of spectral efficiency. Our findings suggest that the POPA can be of practical value for power allocation in small cell networks.