Delay-Optimal Power and Subcarrier Allocation for OFDMA Systems via Stochastic Approximation

TL;DR

This paper develops a delay-optimal power and subcarrier allocation method for OFDMA systems using stochastic approximation, providing a convergent online algorithm that adapts to channel and queue states with reduced complexity.

Contribution

It introduces an online stochastic value iteration approach for delay-optimal resource allocation in OFDMA, with convergence guarantees and reduced computational complexity.

Findings

Proposed a multi-level water-filling power control algorithm.

Proved almost sure convergence to the optimal solution.

Achieved reduced complexity with linear scalability in system size.

Abstract

In this paper, we consider delay-optimal power and subcarrier allocation design for OFDMA systems with $N_F$ subcarriers, $K$ mobiles and one base station. There are $K$ queues at the base station for the downlink traffic to the $K$ mobiles with heterogeneous packet arrivals and delay requirements. We shall model the problem as a $K$-dimensional infinite horizon average reward Markov Decision Problem (MDP) where the control actions are assumed to be a function of the instantaneous Channel State Information (CSI) as well as the joint Queue State Information (QSI). This problem is challenging because it corresponds to a stochastic Network Utility Maximization (NUM) problem where general solution is still unknown. We propose an {\em online stochastic value iteration} solution using {\em stochastic approximation}. The proposed power control algorithm, which is a function of both the CSI and the QSI, takes the form of multi-level water-filling. We prove that under two mild conditions in Theorem 1 (One is the stepsize condition. The other is the condition on accessibility of the Markov Chain, which can be easily satisfied in most of the cases we are interested.), the proposed solution converges to the optimal solution almost surely (with probability 1) and the proposed framework offers a possible solution to the general stochastic NUM problem. By exploiting the birth-death structure of the queue dynamics, we obtain a reduced complexity decomposed solution with linear $\mathcal{O}(KN_F)$ complexity and $\mathcal{O}(K)$ memory requirement.