DeepAlloc: Deep Learning Approach to Spectrum Allocation in Shared Spectrum Systems
Mohammad Ghaderibaneh, Caitao Zhan, Himanshu Gupta
Abstract
Shared spectrum systems facilitate spectrum allocation to unlicensed users without harming the licensed users; they offer great promise in optimizing spectrum utility, but their management (in particular, efficient spectrum allocation to unlicensed users) is challenging. To allocate spectrum efficiently to secondary users (SUs) in general scenarios, we fundamentally need to have knowledge of the signal path-loss function. In practice, however, even the best-known path-loss models have unsatisfactory accuracy, and conducting extensive surveys to gather path-loss values is infeasible. Thus, the current allocation methods are either (i) too conservative in preventing interference that they sacrifice performance, or (ii) are based on imperfect propagation models and/or spectrum sensing with insufficient spatial granularity. This leads to poor spectrum utilization, the fundamental objective of shared spectrum systems. In this work, we thus propose to <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">learn</i> the spectrum allocation function directly using supervised learning techniques. Such an approach has the potential to deliver near-optimal performance with sufficient and effective training data. In addition, it has the advantage of being viable even when certain information is unavailable; e.g., in settings where PUs’ information is not available, we make use of a crowdsourced sensing architecture and use the spectrum sensor readings as features. In general, for spectrum allocation to a single SU, we develop a CNN-based approach (called DeepAlloc) and address various challenges that arise in our context; to handle multiple SU requests simultaneously, we extend our approach based on recurrent neural networks (RNNs). Via extensive large-scale simulation and a small testbed, we demonstrate the effectiveness of our developed techniques; in particular, we observe that our approach improves the accuracy of standard learning techniques and prior work by up to 60%.