A Pointer Network Based Deep Learning Algorithm for User Pairing in Full-Duplex Wi-Fi Networks

We propose a novel deep learning based method to solve the user pairing problem in a full-duplex (FD) Wi-Fi network which consists of one FD Access Point and multiple half-duplex users where the number of users is dynamic. With the objective of maximising the total sum rate of all uplink and downlink user pairs, a user pairing problem is formulated. A traditional method such as Hungarian algorithm for this problem requires high computational complexity, which is impractical when the number of users increases. In this paper, we solve this problem using pointer network with lower computational complexity. Specifically, a pointer network with beam search technique and collision avoidance policy is developed and applied to the user pairing problem of different scaled scenarios. We compare our prediction performance with other two low-complexity methods called greedy assignment and random assignment algorithms. Simulation results show that our proposed method can effectively and efficiently solve the user pairing problems for both fixed and dynamic number of users. Meanwhile, our proposed method outperforms the two benchmark methods in terms of optimal ratio, average bias ratio and average total sum rate.