WMMSE Resource Allocation for FD-NOMA

Abstract

Resource allocation in interference-limited systems is a key enabler for beyond 5G (B5G) technologies, such as multicarrier full duplex non-orthogonal multiple access (FD-NOMA). In FD-NOMA systems resource allocation is a computationintensive non-convex problem due to the presence of strong interference and the integrality condition on channel allocation. In this paper, we propose an iterative algorithm based on the combination of channel and power allocations aimed at the minimization of the weighted mean square error, which converges to a feasible allocation of the original problem. Experimental results show that the proposed algorithm has lower complexity than other state-of-the-art solutions for the same problem. Moreover, the presented results assess the validity of our approach showing performance close to the theoretical optimum.

Joint DDPG and Unsupervised Learning for Channel Allocation and Power Control in Centralized Wireless Cellular Networks

“…As can be seen from Figures 7, 8 and 9, the results of the proposed algorithm in both the energy efficiency and the transmit rate increase with the increase of the number of channels in the cell. At the same time, as the number of channels varies from 4 to 12, the minimum computation time of the proposed algorithm can reach 3 10  (sec) order of magnitude, which is not only obviously lower than the computation time of other deep learning algorithms such as DQN+DDPG algorithm [19], Unsupervised channel power control algorithm [10], but also significantly lower than the computation time of other traditional resource allocation algorithms such as ABC [9], Greedy + WMMSE power control algorithm [6], and Random + WMMSE power control algorithm [6]. This indicates that the proposed algorithm not only always obtains higher energy efficiency and transmit rate than other algorithms, but also has lower computation time and delay than other algorithms.…”

“…In order to train neural networks effectively, the learning rate for the power control DNN is set as 0.001, and the learning rate for the channel allocation DNN is set as 0.0003. After the neural networks are well trained, the proposed algorithm in this paper is compared with other methods, such as the artificial bee colony algorithm (ABC) [9], joint DQN and DDPG algorithm (DQN+DDPG) [19], the channel allocation and power control algorithm based on unsupervised learning (Unsupervised channel power control) [10], the centralized random/greedy channel allocation and WMMSE channel power control (Greedy/Random+ WMMSE power control) [6]. In order to reduce the impact of errors on the performance of the algorithm, the average value of 500 Monte Carlo calculations is used as the result in comparisons.…”

“…Traditional methods for resource allocation in wireless cellular networks mainly include iterative algorithms [5], [6], heuristic algorithms [7]- [9], and so on. These methods tend to have relatively high computational burden and computation time [10], and are not feasible to handle the time-varying wireless environments with real time.…”