Deep Reinforcement Learning for Wireless Communications and Networking: Theory, Applications and Implementation by unknow

Author:unknow
Language: eng
Format: epub
Publisher: John Wiley & Sons, Incorporated
Published: 2023-08-01T00:00:00+00:00

6.2.2 MAC for Massive Access in IoT

Unlike the IEEE 802.11-based WLAN, where the number of concurrently connected devices is not very large, an Internet of Things (IoT) network may have thousands of devices, each with different data transmission intervals and limited resources (e.g. computing and energy). Thus, channel access control approaches for IoT need to address the massive access problem while keeping the complexity at a reasonable level. In addition, the access demand in IoT networks is not only much greater than that of conventional WLANs but also highly dynamic and uncertain, which makes designing the MAC protocols more challenging. In this context, DRL can be used to tackle emerging challenges in designing IoT MAC protocols [14–17].

In IoT networks, due to the nature of irregular communication and the requirement of channel utilization efficiency, it is preferable to use probabilistic MAC protocols (e.g. -persistent slotted ALOHA [7]) over deterministic MAC protocols, e.g. time-division multiple access (TDMA). However, probabilistic MAC protocols are prone to collisions when two or more devices access a shared channel simultaneously. This problem becomes more severe in an IoT network consisting of a very large number of devices. To this end, as illustrated in Figure 6.3, several DRL frameworks were proposed to address this problem of probabilistic MAC protocols [14–16]. In [14], the authors studied an MAC problem for wireless sensor networks (WSNs) that consists of multiple devices sensing the surrounding environment and sending collected data to a core network via a shared channel. In particular, the considered WSN adopts an irregular repetition slotted ALOHA (IRSA) method that can address the collision problem by leveraging successive interference cancelation (SIC) [18]. The main idea of SIC is that when two or more packets arrive at a receiver simultaneously, the device first decodes the stronger signal and then subtracts it from the combined signal before decoding weaker signals. To achieve the time-domain diversity in IRSA, devices send multiple replicas of a packet to a receiver. If the receiver successfully decodes this packet, it then leverages the SIC technique to cancel its interference in other time slots, thereby increasing the ability to decode more packets and resolving collisions successfully. In IRSA, the number of replicas is derived by sampling from a probability distribution, which is statically optimized by a differential evolution algorithm. Therefore, IRSA is prone to poor performance in dynamic environments, e.g. changes in network topology and channel condition. In addition, the lack of fully observing the network's state is another challenge when adopting IRSA for WSNs. To address these challenges, the authors propose to leverage the decentralized partial observation Markov decision process (Dec-POMDP) framework [19] to optimize the number of replicas for each sensor to maximize the total system throughput. Unlike the conventional MDP which involves one agent that can fully observe the system state, the Dec-POMDP consists of multiple agents; each can only observe a different part of the system state. In this work, time is divided into frames, and each consists of equal-size time slots. Each sensing device sends at most one packet per frame.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.