Abstract
Caching at the wireless edge can be used to keep up with the increasing demand for high-definition wireless video streaming. By prefetching popular content into memory at wireless access points or end-user devices, requests can be served locally, relieving strain on expensive backhaul. In addition, using network coding allows the simultaneous serving of distinct cache misses via common coded multicast transmissions, resulting in significantly larger load reductions compared to those achieved with traditional delivery schemes. Most prior works simply treat video content as fixed-size files that users would like to fully download. This work is motivated by the fact that video can be coded in a scalable fashion and that the decoded video quality depends on the number of layers a user receives in sequence. Using a Gaussian source model, caching and coded delivery methods are designed to minimize the squared error distortion at end-user devices in a rate-limited caching network. The framework is very general and accounts for heterogeneous cache sizes, video popularities and user-file play-back qualities. As part of the solution, a new decentralized scheme for lossy cache-aided delivery subject to preset user distortion targets is proposed, which further generalizes prior literature to a setting with file heterogeneity.
Introduction
The explosive growth of high-definition video streaming poses a significant challenge for wireless networks. Traditional caching strategies often treat video files as monolithic, fixed-size blocks that must be delivered in their entirety. However, modern video is typically encoded using Scalable Video Coding (SVC), where content is divided into a base layer and multiple enhancement layers. This structure allows for a more flexible approach: instead of focusing solely on bit-delivery, networks can optimize for the user's actual viewing experience—measured by visual distortion.
This paper addresses the challenge of heterogeneous caching networks, where users possess varying storage capacities, device capabilities, and content preferences. The goal is to determine how to populate local caches and design delivery messages to minimize average video distortion across the network, given a limited broadcast bandwidth.
System Architecture and Methods
The researchers model the network as a central server (sender) connected to $K$ receivers via a shared broadcast link with capacity $R$. Each receiver $k$ has a local cache of size $M_k$. The library contains $N$ files, each modeled as a "successively refinable" source (e.g., a Gaussian source), which mirrors the behavior of SVC-encoded video.
The study compares two primary strategies:
- Local Cache-aided Unicast (LC-U): A baseline where users cache content independently. The server delivers unique, uncoded updates to satisfy individual requests.
- Cooperative Cache-aided Coded Multicast (CC-CM): A novel approach that coordinates caching across users. During the delivery phase, it employs Network Coding to create "coded multicast" packets. These packets are single transmissions that simultaneously satisfy different requests for multiple users.
To implement CC-CM in complex, heterogeneous environments, the authors propose RF-GCC (Random Fractional caching with Greedy Constrained Coloring). This algorithm constructs a "conflict graph" of requested video packets and applies graph coloring techniques to optimize the bundling of multicast messages.
Key Insights: Quality and Cooperation
A central theme of this work is the shift from "quantity" to "quality." By optimizing for distortion rather than raw bitrate, the network can prioritize the delivery of critical video layers.
Furthermore, the research highlights the "Power of Overlap." While traditional caching avoids redundancy, this framework utilizes Random Fractional Caching. By allowing users to store random pieces of files, the system creates more opportunities for coded multicasting. For instance, if User A has piece 1 and needs piece 2, while User B has piece 2 and needs piece 1, the server can broadcast a single XORed packet $(1 \oplus 2)$ that both users can decode using their local cache.
The framework also leverages heterogeneity. Using a "reverse water-filling" approach, it optimally allocates cache space across different files based on their popularity and the specific memory constraints of each device.
Performance Evaluation
The proposed schemes were evaluated through numerical simulations involving a library of $N=100$ files and $K=20$ receivers. File popularity followed a Zipf distribution, and distortion-rate functions were modeled using Gaussian sources.
Key results include:
- Significant Distortion Reduction: In standard scenarios (Zipf $\alpha=0.6$), the CC-CM scheme achieved a 2.1x to 5.4x reduction in expected distortion compared to the unicast baseline.
- Gains in Uniform Popularity: In cases where file popularity is uniform—typically the most difficult scenario for caching—the improvements reached up to 14x.
- Asymptotic Optimality: The RF-GCC algorithm was shown to be "asymptotically order-optimal," performing near theoretical limits as file sizes increase.
Conclusion and Future Directions
This work demonstrates that integrating rate-distortion theory with coded caching can drastically improve video quality in bandwidth-constrained networks. While the current model assumes an error-free broadcast link and Gaussian sources, it provides a robust foundation for future research. Future work could extend these findings to real-world wireless channels with fading and packet loss, as well as more complex, non-linear distortion models characteristic of modern codecs like H.265/HEVC. Additionally, addressing the computational complexity of the optimal CC-CM problem for very large-scale networks remains an open challenge.
Summary updated