Decentralized Delay Optimal Control for Interference Networks with Limited Renewable Energy Storage

TL;DR

This paper develops decentralized online learning algorithms for interference networks with renewable energy, optimizing delay and energy efficiency under partial observability, with proven convergence and significant performance improvements.

Contribution

It introduces novel decentralized policy gradient algorithms for delay optimization in renewable energy-powered interference networks, with convergence guarantees under DEC-POMDP and POSG frameworks.

Findings

Substantial delay reduction compared to baseline schemes

Energy savings achieved through proposed control policies

Algorithms demonstrate robustness to model variations

Abstract

In this paper, we consider delay minimization for interference networks with renewable energy source, where the transmission power of a node comes from both the conventional utility power (AC power) and the renewable energy source. We assume the transmission power of each node is a function of the local channel state, local data queue state and local energy queue state only. In turn, we consider two delay optimization formulations, namely the decentralized partially observable Markov decision process (DEC-POMDP) and Non-cooperative partially observable stochastic game (POSG). In DEC-POMDP formulation, we derive a decentralized online learning algorithm to determine the control actions and Lagrangian multipliers (LMs) simultaneously, based on the policy gradient approach. Under some mild technical conditions, the proposed decentralized policy gradient algorithm converges almost surely to a local optimal solution. On the other hand, in the non-cooperative POSG formulation, the transmitter nodes are non-cooperative. We extend the decentralized policy gradient solution and establish the technical proof for almost-sure convergence of the learning algorithms. In both cases, the solutions are very robust to model variations. Finally, the delay performance of the proposed solutions are compared with conventional baseline schemes for interference networks and it is illustrated that substantial delay performance gain and energy savings can be achieved.