Energy Efficiency Optimization Based on Unsupervised Learning in Wireless Communication Systems

In this paper, we propose a transmission power control scheme based on deep learning (DL) in wireless communication systems (WCS). This scheme aims to solve the problem of maximizing energy efficiency (EE) in fading multi-user interference channels. Traditional iterative power allocation algorithms have high computational complexity, making them difficult to implement in practical applications. With the development of DL, it has been proved that DL schemes have excellent learning ability and lower computational complexity. We use an unsupervised learning (UL) scheme in DL, and there is relatively little research on using UL schemes to achieve power control to maximize EE. First, we implement a conventional double iterative algorithm to control the transmitted power and use the performance of this algorithm as a benchmark. Then in our proposed scheme, the EE of the system model is maximized by autonomously learning the transmission power in the WCS model through deep neural networks (DNN) while considering the constraint of minimum transmission rate per communication link. Compared to traditional iterative algorithm power control schemes, which require complex and tedious iterative processes to find the optimal solution within the feasible domain, our proposed scheme achieves similar performance with lower computation time, as shown by simulation results.