Deep Learning for Optimized Multiuser OFDMA Energy-efficient Wireless Transmission

The energy-efficient multiuser transmission is applied to an orthogonal frequency division multiple access (OFDMA) system. The subchannel assignment and power allocation strategies are optimized in order to maximize the total energy efficiency while maintaining the information rate from the base station to each mobile user. Due to the non-convexity of the optimization, two individual Deep Neural Networks are trained to solve the optimization problem. With the channel gain as the input, two Deep Neural Networks can output the optimal power allocation and subchannel assignment, respectively. A Refined Exhaustive Search Algorithm is invented in order to generate training data efficiently. The neural networks are trained with simulated data offline but can be utilized online for an immediate reaction. The simulation results prove the Deep Neural Networks achieve a high precision with extremely high execution speed.

I. Introduction

Applying multiple user transmission over parallel frequency channels to orthogonal frequency-division multiple access (OFDMA) is a promising technique, since multiple parallel subchannels can resist the transmission inference [1], [2]. Adapting to different channel conditions, the transmission strategies on each subchannel can be adjusted in order to maximize the communication quality. In [3], the authors formed the optimization as maximizing a weighted sum of rates under the sum power and additional receiver-specific rate constraints for a system consisting of multiple parallel Gaussian broadcast channels.