Spatio-Temporal Edge Caching for Wireless Networks

This direction is merged into another direction: dynamic content popularity. Its last version stays readable and citable here.

Version 1 of 118 papers18 conceptsfrozen 2026-10-04

In modern wireless networks, the bottleneck is rarely the "last mile" of wireless spectrum, but rather the "backhaul"—the wired connection linking local base stations to the core internet. As mobile video traffic explodes, traditional reactive caching (storing content only after it is requested) fails to keep up.

A growing body of research suggests a shift toward proactive spatio-temporal caching. Instead of waiting for requests, base stations use machine learning to predict what content will be popular at specific locations and times, pre-loading it during off-peak hours to virtually eliminate backhaul congestion.

The Failure of Static Caching

Traditional caching algorithms like Least Recently Used (LRU) assume a relatively stable "global" popularity. However, as Ejder Baştuǧ, Mehdi Bennis, and colleagues argue in Big data meets telcos: A proactive caching perspective, mobile demand is highly volatile.

Content popularity is not just a function of what is trending, but where and when. A video that is viral in a business district at 10:00 AM might be irrelevant in a residential suburb at 8:00 PM. Static models fail to capture these "spatio-temporal dynamics," leading to low cache hit rates as users move between base stations and their preferences shift throughout the day.

Learning the Space-Time Map

To solve this, researchers have moved toward learning-based frameworks that treat content popularity as a dynamic process. Alireza Sadeghi and Georgios B. Giannakis propose modeling user requests as Markov processes in Optimal and Scalable Caching for 5G Using Reinforcement Learning of Space-Time Popularities.

The intuition is that the network doesn't need a perfect crystal ball; it just needs to learn the underlying patterns of demand. By using Reinforcement Learning (RL), base stations can:

  1. Observe local request patterns.
  2. Estimate the "value" of caching a specific file based on its predicted future demand in that specific cell.
  3. Update their strategy as popularity drifts over time.

Because the number of possible files is massive (the "curse of dimensionality"), these systems often use Deep Q-Networks (DQN) or linear function approximation to make the learning scalable. In Reinforcement Learning for Caching with Space-Time Popularity Dynamics, the authors demonstrate that a "Hyper-DQN" can manage hierarchical networks where a parent node learns to coordinate with multiple leaf nodes, adapting to their unknown local caching policies.

Scaling to Ultra-Dense Networks

As we move toward 5G and 6G, the density of base stations increases. Managing thousands of small cells individually becomes computationally impossible.

To address this, Hyesung Kim, Jihong Park, and others applied Mean-Field Game (MFG) theory in Mean-Field Game Theoretic Edge Caching in Ultra-Dense Networks. Instead of modeling every interaction between every base station, MFG simplifies the environment into a single "mean-field" interaction. This allows the system to achieve a distributed equilibrium where each base station makes optimal caching decisions based on the average behavior of the rest of the network, reducing redundant data by up to 42%.

A final challenge is balancing local niche interests with global viral trends. In Learning-based Caching in Cloud-Aided Wireless Networks, Syed Tamoor-ul-Hassan and Mehdi Bennis propose a decentralized strategy using "regret learning."

By clustering content into similarity groups, the edge nodes can satisfy local user preferences while the cloud provides "global" popularity hints. This hybrid approach ensures that if a video suddenly goes viral globally, the local edge caches are alerted to pre-fetch it before the local demand spike even hits, achieving a 40% improvement in network utility.

By treating the network as a living, learning map of human behavior across space and time, these proactive strategies transform the edge from a simple pass-through pipe into an intelligent buffer that anticipates user needs.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_spatio_temporal_edge_caching_for_wireles_a2086416,
  title = {Spatio-Temporal Edge Caching for Wireless Networks},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spatio-temporal-edge-caching-for-wireless-networks/rdr_a2086416e1a148f39e0357a8db85ca01},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_spatio_temporal_edge_caching_for_wireles_a2086416_v1,
  title = {Spatio-Temporal Edge Caching for Wireless Networks},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spatio-temporal-edge-caching-for-wireless-networks/rdr_a2086416e1a148f39e0357a8db85ca01/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}