[论文解读] Optimal User-Cell Association for Massive MIMO Wireless Networks

Dilip Bethanabhotla, Ozgun Y. Bursalioglu+2 more|arXiv (Cornell University)|Jul 24, 2014

Advanced MIMO Systems Optimization参考文献 37被引用 10

一句话总结

本文提出了一种集中式凸优化框架和一种去中心化用户中心算法，用于在大规模MIMO异构网络中实现最优用户-小区关联，通过长期平均用户速率最大化网络效用。主要贡献在于证明了最优解可通过调度实现物理可实现性，并且在公平效用函数下，去中心化方案收敛至接近最优的纯策略纳什均衡。

ABSTRACT

The use of a very large number of antennas at each base station site (referred to as "Massive MIMO") is one of the most promising approaches to cope with the predicted wireless data traffic explosion. In combination with Time Division Duplex and with simple per-cell processing, it achieves large throughput per cell, low latency, and attractive power efficiency performance. Following the current wireless technology trend of moving to higher frequency bands and denser small cell deployments, a large number of antennas can be implemented within a small form factor even in small cell base stations. In a heterogeneous network formed by large (macro) and small cell BSs, a key system optimization problem consists of "load balancing", that is, associating users to BSs in order to avoid congested hot-spots and/or under-utilized infrastructure. In this paper, we consider the user-BS association problem for a massive MIMO heterogeneous network. We formulate the problem as a network utility maximization, and provide a centralized solution in terms of the fraction of transmission resources (time-frequency slots) over which each user is served by a given BS. Furthermore, we show that such a solution is physically realizable, i.e., there exists a sequence of integer scheduling configurations realizing (by time-sharing) the optimal fractions. While this solution is optimal, it requires centralized computation. Then, we also consider decentralized user-centric schemes, formulated as non-cooperative games where each user makes individual selfish association decisions based only on its local information. We identify a class of schemes such that their Nash equilibrium is very close to the global centralized optimum. Hence, these user-centric algorithms are attractive not only for their simplicity and fully decentralized implementation, but also because they operate near the system "social" optimum.

研究动机与目标

• 解决大规模MIMO异构网络中用户密度非均匀且基站能力各异时的高效用户-小区关联挑战。
• 将关联问题建模为网络效用最大化问题，以优化长期平均用户吞吐量。
• 证明最优活动比例分配是凸的，并且可通过调度实现物理可实现性。
• 设计一种去中心化用户中心关联方案，使其收敛至近似最优均衡。
• 建立去中心化博弈收敛至接近全局最优的纯策略纳什均衡的条件。

提出的方法

• 将用户-小区关联问题建模为带有用户-基站活动比例约束的凸网络效用最大化问题。
• 使用集中式次梯度算法求解凸优化问题，以获得最优活动比例。
• 证明可行域的每个极点均对应一种整数调度配置，从而确保物理可实现性。
• 引入一种去中心化用户中心方案，用户在预期效用提升时以概率方式切换关联。
• 将关联过程建模为非合作博弈，并证明存在接近全局最优的纯策略纳什均衡。
• 利用KKT条件和拉格朗日对偶性，推导出比例公平和最大最小公平情况下的最优活动比例显式解。

实验结果

研究问题

• RQ1 大规模MIMO异构网络中的最优用户-小区关联能否被建模为凸优化问题？
• RQ2 最优解是否可通过实际调度实现物理可实现性？
• RQ3 去中心化用户中心关联方案是否收敛至近似最优均衡？
• RQ4 在何种条件下，全局最优解也是去中心化博弈中的纯策略纳什均衡？
• RQ5 基于公平性的效用函数如何影响去中心化方案的收敛性和最优性？

主要发现

• 最优用户-小区关联问题为凸问题，可通过集中式次梯度方法求解，确保全局最优性。
• 最优解具有物理可实现性，因为可行域的每个顶点均对应一种整数调度配置。
• 在比例公平和最大最小公平条件下，去中心化用户中心方案以概率1收敛至纯策略纳什均衡。
• 当集中式解唯一（即无用户关联多个基站）时，该解也是去中心化博弈的纯策略纳什均衡。
• 通过KKT条件显式推导出最优活动比例，其解为基于用户速率和基站资源的阈值形式。
• 该方案实现了接近系统最优的性能，兼具去中心化实现的简洁性与强大的理论收敛保证。