Deep Learning-Based Cross-Layer Resource Allocation For Wired Communication Systems

In this paper, a cross-layer resource allocation (RA) scheme based on deep learning is introduced for multi-tone multi-user wired communication systems such as, for instance, digital subscriber line (DSL) systems under the current G.fast standard. The upper layer reports time-varying user demands to the physical layer using proportional priority weights. Unlike the deep neural network (DNN)-based RA schemes available in the literature for wireless communication, here the dynamic part is in adjusting the priority weights which are then fed to a DNN to map these priority weights to an appropriate power allocation, maximizing the weighted sum rate (WSR). The DNN is trained employing an unsupervised strategy to increase its generalization capabilities. Through the simulations, we show that the proposed cross-layer scheme can scale very well, even for real-world systems with thousands of RA variables. Moreover, the proposed cross-layer scheme significantly outperforms the conventional WMMSE method in terms of computation speed, while achieving the same data rates. Hence, the proposed cross-layer scheme is more suitable for cross-layer RA than the conventional methods, as it increases the speed with which the physical layer can react to time-varying upper layer data rate demands.