Distributed Coded Caching with Application to Content Delivery in Wireless Networks

Abstract: The amount of content downloaded to mobile devices, mainly driven by the demand for video content, threatens to completely congest wireless networks and the trend of ever increasing video traffic is expected to continue unabated for many years. A promising solution to this problem is to store popular content closer to end users, effectively trading expensive bandwidth resources for affordable memory, a technique known as caching. In this thesis, we study the use of erasure correcting codes (ECCs) to increase the amount of data that can be downloaded directly from the caches when content is cached in a distributed fashion across several base stations (BSs) or mobile devices. When content is cached in mobile devices, users may download coded packets directly from caching devices using device-to-device communication and, if necessary, from the BS at a higher communication cost. Devices moving out of range or turning off will cause a loss of cached content. To restore the initial state of reliability in the network, data is transmitted to available mobile devices in a process known as content repair. We compare the amount of data transmitted in the network due to content download and content repair using various ECCs when content is repaired at periodic times. We analyze the performance when mobile devices enter the network with or without usable cached content and show that increasing the time between repairs, so called lazy repairs, can be beneficial. Furthermore, we analyze content caching in mobile devices using maximum distance separable codes for scenarios where the density of devices is high. We optimize the number of mobile devices to involve in the caching of content and demonstrate the significant gains that can be achieved in terms of data downloaded from caching devices. We proceed to analyze how to optimally manage cached content over time when users request content according to a renewal process, i.e., a process with memory. Specifically, we consider the distributed coded caching of content at small BSs where coded packets may be evicted from the caches at periodic times. We prove that the problem of maximizing the amount of data that users can download from the caches is concave and that our problem formulation is a generalization of the previously studied cases where content is cached in a single cache and where content is not managed over time, so called static caching. We show that optimizing caching policies can offer considerable gains in the amount of data that can be downloaded from the caches, especially when the request process is bursty. Conversely, we prove that static caching is optimal for request processes without memory. Finally, we suggest a multi-agent reinforcement learning approach to learn cache management policies for spatially non-uniform renewal request processes. Our algorithm obtains cache management policies, substantially increasing the amount of data that can be downloaded from the caches.