Adaptive Subcarrier, Bit and Power Allocation Based on Hopfield Neural Network for Multiuser OFDM

Abstract

A kind of adaptive subcarrier, bit and power allocation method utilizing Hopfield neural network (HNN) to minimize the overall transmit power of multiuser OFDM is studied in this paper. In order to find the power optimal subcarrier, bit and power allocation under the constraints that one subcarrier can only be allocated to one user and all users are allocated the same numbers of subcarrier, the number of bits of each subcarrier is finite, bit data can be allocated to each subcarrier, two kinds of new energy constrained functions are constructed for the HNN. It is shown through numerical simulations that the proposed methods can find the optimal allocation with less complexity compared with the exhaustive method.

A Study on Deep Learning for Latency Constraint Applications in Beyond 5G Wireless Systems

“…In [16] Zhou et al have proposed an efficient DNN for resource allocation in cognitive radio networks aiming at the real-time performance to maximize the energy and spectral efficiency of the network. In [17], Li et al have proposed a model that utilizes a Hopfield neural network to predict the bit and power allocation in a multi-user OFDM system. In [18], the authors propose a supervised DNN model for subcarrier assignment in an OFDMA/NOMA downlink video transmission system.…”

On Deep Learning for Radio Resource Management in A Non-stationary Radio Environment

“…However, their work does not highlight the importance and the need for retraining, and thus the performance of the model could degrade in the long run without any retraining. In [14] Li et al have proposed a model that utilizes a Hopfield neural network (HNN) to predict the bit and power allocation in a multi-user OFDM system. Compared to the exhaustive method, the proposed technique is computationally efficient in finding the optimal solution, but the study does not analyze the change in performance with complexity and the ability of the model to perform in a timely manner in the presence of a dynamic environment.…”

“…The studies in [1]- [14] have neglected the significant practical limitations related to training requirements of the learning model and the impact of ageing of the learning model due to evolution of the wireless environment. Hence, we need to understand the requirements for retraining of the model in the presence of ageing environment and the trade-off between the complexity and the computational efficiency to achieve the target service requirements.…”

“…e) Relative sum rate of DQN during training: The performance of the DQN is compared with the optimal performance given by the WMMSE algorithm. Furthermore, the reward (sum rate) from random power allocation and maximum power allocation strategies are also considered for comparison 14 . The variation of the reward throughout the training process is shown in Fig.…”

“…"Warmup" phase is defined by the first 100 episodes followed by the training phase. Note that at the warmup phase, the relative performance of the 14 In random power allocation strategy, power is allocated to each user randomly. In the maximum power allocation strategy, the maximum power of each base station is assigned DQN is nearly equal to the random power allocation method.…”