Dynamic Partial Cooperative MIMO System for Delay-Sensitive Applications with Limited Backhaul Capacity

TL;DR

This paper introduces a flexible partial cooperative MIMO scheme that balances backhaul use and interference mitigation, optimizing delay and power with a distributed online learning approach under practical constraints.

Contribution

It proposes a novel partial cooperative MIMO scheme with dynamic power and rate control, formulated as a CPOMDP, and develops a low-complexity distributed online solution.

Findings

Achieves delay minimization under backhaul and power constraints.

Provides a distributed online algorithm with proven convergence.

Demonstrates robustness and asymptotic optimality of the proposed method.

Abstract

Considering backhaul consumption in practical systems, it may not be the best choice to engage all the time in full cooperative MIMO for interference mitigation. In this paper, we propose a novel downlink partial cooperative MIMO (Pco-MIMO) physical layer (PHY) scheme, which allows flexible tradeoff between the partial data cooperation level and the backhaul consumption. Based on this Pco-MIMO scheme, we consider dynamic transmit power and rate allocation according to the imperfect channel state information at transmitters (CSIT) and the queue state information (QSI) to minimize the average delay cost subject to average backhaul consumption constraints and average power constraints. The delay-optimal control problem is formulated as an infinite horizon average cost constrained partially observed Markov decision process (CPOMDP). By exploiting the special structure in our problem, we derive an equivalent Bellman Equation to solve the CPOMDP. To reduce computational complexity and facilitate distributed implementation, we propose a distributed online learning algorithm to estimate the per-flow potential functions and Lagrange multipliers (LMs) and a distributed online stochastic partial gradient algorithm to obtain the power and rate control policy. The proposed low-complexity distributed solution is based on local observations of the system states at the BSs and is very robust against model variations. We also prove the convergence and the asymptotic optimality of the proposed solution.