Collaborative Distributed Q-Learning for RACH Congestion Minimization in Cellular IoT Networks
Collaborative Distributed Q-Learning for RACH Congestion Minimization in Cellular IoT Networks
复制标题
DOI：
10.1109/lcomm.2019.2896929
复制
发表时间：
2019-02
期刊：
影响因子：
--
通讯作者：
S. Sharma;Xianbin Wang
中科院分区：
文献类型：
--
作者：
S. Sharma;Xianbin Wang
由于越来越多的机器类型通信(MTC)设备发出不频繁和海量的并发访问请求，现有的基于竞争的随机访问(RA)协议，如时隙ALOHA，在新兴的蜂窝物联网网络中面临着严重的随机访问信道(RACH)拥塞问题。为了解决这个问题，我们提出了一种新的协作分布式Q-学习机制，用于资源受限的MTC设备，使它们能够为其传输找到唯一的RA时隙，从而显著减少可能的冲突数量。与独立Q学习算法相比，该方法利用RA时隙的拥塞程度作为学习过程中的全局代价，可以显著降低低端MTC设备的学习时间。结果表明，所提出的学习方案可以显著降低蜂窝物联网网络中的RACH拥塞。
Due to infrequent and massive concurrent access requests from the ever-increasing number of machine-type communication (MTC) devices, the existing contention-based random access (RA) protocols, such as slotted ALOHA, suffer from the severe problem of random access channel (RACH) congestion in emerging cellular IoT networks. To address this issue, we propose a novel collaborative distributed Q-learning mechanism for the resource-constrained MTC devices in order to enable them to find unique RA slots for their transmissions so that the number of possible collisions can be significantly reduced. In contrast to the independent Q-learning scheme, the proposed approach utilizes the congestion level of RA slots as the global cost during the learning process and thus can notably lower the learning time for the low-end MTC devices. Our results show that the proposed learning scheme can significantly minimize the RACH congestion in cellular IoT networks.