Wireless networks QoS optimization using coded caching and machine learning algorithms

Khalil, Mohamed (2023) Wireless networks QoS optimization using coded caching and machine learning algorithms. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (2MB)

Abstract

Proactive caching shows great potential to minimize peak traffic rates by storing popular data, in advance, at different nodes in the network. We study three new angles of proactive caching that were not covered before in the literature. We develop more practical algorithms that bring proactive caching closer to practical wireless networks. The first angle is where the popularities of the cached files are changing over time and the file delivery is asynchronous. We provide an algorithm that minimizes files’ delivery rate under this setting. We show that we can use the file delivery messages to proactively and constantly update the receiver finite caches. We show that this mechanism reduces the downloaded traffic of the network. The proposed scheme uses index coding [1], and app. A to jointly encodes the delivery of different demanded files with the cache updates to other receivers to follow the changes in the file popularities. An offline and online (dynamic) versions of the scheme are proposed, where the offline version requires knowledge of the file popularities across the whole transmission period in advance and the online one requires the file popularities for one succeeding time slot only. The optimal caching for both the offline and online schemes is obtained numerically. The second angle is the study of segmented caching for delay minimization in networks with congested backhaul. Studies have mainly focused on proactively storing popular whole files. For certain categories of files like videos, this is not the best strategy. As videos can be segmented, sending later segments of videos can be less time-critical. Video is expected to constitute 82% of internet traffic by 2020 [2]. We study the effect of segmenting video caching decisions under the assumption that the backhaul is congested. We provide an algorithm for proactive segmented caching that optimizes the choice of segments to be cached to minimize delay and compare the performance to the whole file proactive caching. The third angle focuses on using reinforcement learning for coded caching in networks with changing file popularities. For such a dynamic environment, reinforcement learning has the flexibility to learn the environment and adapt accordingly. We develop a reinforcement learning-based coded caching algorithm and compare its performance to rule-based coded caching.

Actions (login required)

View Item