vrAIn: A Deep Learning Approach Tailoring Computing and Radio Resources in Virtualized RANs

Abstract

The virtualization of radio access networks (vRAN) is the last milestone in the NFV revolution. The dependency between computing resources and radio channel conditions makes vRAN resource control particularly daunting. We present vrAIn, a vRAN dynamic resource controller that builds upon deep reinforcement learning, which we implement using using an open-source LTE stack over different platforms. First, vrAIn uses an autoencoder that projects the large context space (traffic and signal quality patterns) into a latent representation. Then, vrAIn uses a deep deterministic policy gradient (DDPG) algorithm, implemented by an actor-critic neural network structure, and a simple classifier to map (encoded) contexts into appropriate resource control decisions. Our results show that vrAIn successfully derives appropriate compute and radio control decisions irrespective of the platform and context: (i) it provides savings in computational capacity of up to 30% over a CPU-unaware approach while paying a very small price in performance; (ii) it improves the probability of meeting QoS targets by 25% over a static allocation policy that uses the same CPU resources; and (iii) it improves throughput performance by 25% over state-of-the-art approaches upon CPU capacity shortage. To the best of our knowledge, this is the first work that thoroughly studies the computational behavior of vRAN, and the first approach to a model-free solution that does not need to assume any particular vRAN platform or system conditions.