Echo State Networks for Self-Organizing Resource Allocation in LTE-U with Uplink-Downlink Decoupling

TL;DR

This paper introduces a novel distributed resource allocation method for LTE-U networks with uplink-downlink decoupling using echo state networks, achieving significant performance improvements and reduced information exchange.

Contribution

It proposes a machine learning-based distributed algorithm using echo state networks for resource allocation in LTE-U with uplink-downlink decoupling, a novel approach in this context.

Findings

Up to 78% increase in sum-rate for the 50th percentile of users.

Convergence to a Nash equilibrium in the proposed game.

Significant reduction in information exchange compared to Q-learning.

Abstract

Uplink-downlink decoupling in which users can be associated to different base stations in the uplink and downlink of heterogeneous small cell networks (SCNs) has attracted significant attention recently. However, most existing works focus on simple association mechanisms in LTE SCNs that operate only in the licensed band. In contrast, in this paper, the problem of resource allocation with uplink-downlink decoupling is studied for an SCN that incorporates LTE in the unlicensed band (LTE-U). Here, the users can access both licensed and unlicensed bands while being associated to different base stations. This problem is formulated as a noncooperative game that incorporates user association, spectrum allocation, and load balancing. To solve this problem, a distributed algorithm based on the machine learning framework of echo state networks is proposed using which the small base stations autonomously choose their optimal bands allocation strategies while having only limited information on the network's and users' states. It is shown that the proposed algorithm converges to a stationary mixed-strategy distribution which constitutes a mixed strategy Nash equilibrium for the studied game. Simulation results show that the proposed approach yields significant gains, in terms of the sum-rate of the 50th percentile of users, that reach up to 60% and 78%, respectively, compared to Q-learning and Q-learning without decoupling. The results also show that ESN significantly provides a considerable reduction of information exchange for the wireless network.