Gradient-free Online Resource Allocation Algorithms for Dynamic Wireless Networks

Alexandre Marcastel, Elena Veronica Belmega, Inbar Fijalkow, Panayotis Mertikopoulos

Abstract: Future communication networks will be faced with supporting highly mobile, heterogeneous (including aerial) devices, which calls for new and efficient resource allocation policies that are able to adapt on-the-fly to the network dynamics while relying on little and possibly outdated information. The aim of this paper is twofold: to explicitly take into account the device mobility, their network connectivity patterns and behavior (which may be completely arbitrary and unpredictable); and to greatly reduce the information required at the transmitter. For this, we exploit the framework of online optimization and exponential learning to derive a provably efficient and gradient-free online power allocation algorithm relying only on a scalar-worth of feedback.