A Deep Learning based Resource Allocation Algorithm for Variable Dimensions in D2D-Enabled Cellular Networks

Optimization algorithms play an important role in resource allocation problems. However, the algorithms are difficult to be applied in practice due to the high complexity. Some deep neural networks (DNNs) are thus proposed to approach the traditional algorithms, which can realize real-time resource allocation. However, the DNNs is designed for invariable dimensions. Therefore, it remains unclear whether the neural network under variable dimensions can still approach the traditional algorithm. Furthermore, it still remains unclear how to train the neural network for variable dimensions. In this work, we propose a deep learning based power control scheme for variable D2D pairs, where low-dimensional inputs are preprocessed by zero-padding, and several hybrid training methods are proposed. Through a large number of experimental simulations, it is proved that the preprocessing method can better deal with the variable dimensions problem without introducing new interference. The fully connected DNN trained by different-dimensional data is proved to be the closest to the traditional algorithms.