A novel context-aware caching
scheme for 5G networks
Dr. Noman Islam
13th International Conference - Mathematics, Actuarial, Computer Science & Statistics (MACS 13), IoBM, Karachi
Samsung’s 5G rainbow
1. Very high data rates
2. High spectral efficiency
3. Speed during mobility conditions
4. High data transmission rates even at the boundary of a cell
5. Maximum number of concurrent connections 6. Reduced delay in communication
Caching has been regarded as amongst the five most disruptive technologies for 5G
Introduction
• Most of the requests on cellular networks are
for videos
• Popular contents can be cached • Where to cache?
– Cache at the edge on small base stations or mobile
Objective
• The paper analyzes the current approaches
available for caching in 5G networks
• Discusses a context-aware collaborative
Increasing demand for data
• Rise in computing devices, connectivity
medium
• How to cope with soaring demands for data? • Capacity can be increased by installing more
access points or base stations
• There is a limit to the network densification • Most of the contents are videos or social
Caching in 5G network
• If the popular contents (such as videos) can be
proactively cached at the edges closer to the user, the backhaul network can be offloaded.
• Caching can not only avoid the network
Research gap
• Caching is a very important research problem
in 5G networks
• Only one study is found that considered
context parameters such as mobility of the nodes while performing caching.
• Context which is of paramount importance in
Collaborative filtering
• Collaborative filtering is based on the idea that
people who liked similar things in past would likely to have same opinion about future items.
• Collaborative filtering approaches are classified as
neighborhood based and latent factor approaches.
• The former finds a set of neighbors to a user or
item. While in latent factor approach, the rating of a user is decomposed using matrix factorizing
Bastug et al. (2014)
• Bastug et al. have proposed an approach
based on recommendation system for caching in 5G network. A popularity matrix is
calculated by solving a least square problem. The regularized singular value decomposition was chosen to decompose the popularity
Context-aware collaborative filtering
• In the established domain of CF,
context-aware approaches already exist
• They are classified as contextual pre-filtering,
contextual post-filtering and contextual modeling
• We suggest employing these techniques to
TF-based context-aware collaborative
filtering
• The demands for data items of a particular user are
predictable based on a popularity matrix P
• Each small cell base station is equipped with storage
capabilities M to cache the popular contents.
• As the storage available is small, a popularity matrix
P is used to decide what particular contents to be cached. It is also assumed that the user arrives
Popularity matrix
• We extend the popularity function P such
that:
P: Users × Item × Context → Ratings
• Hence, the popularity matrix is a function of
not only users and items, but also the context in which user issued the request for the
• Six contextual information has been identified
based on Schmidt et al. [27]: information about user, social information, user’s tasks, location, infrastructure and physical conditions.
• The contextual information can be obtained
from various sources such as sensors and logs.
• The popularity matrix is indexed according to a
Tensor factorization
• Using tensor factorization, a low rank version of
popularity matrix is constructed as shown in Figure.
• The tensor factorization decomposes the
popularity matrix into factors of users, items and context inferred from popularity.
• There are a number of tensor factorization
Performing caching
• The resultant popularity matrix can be used
for deciding the items to be cached proactively.
• The most popular files are greedily cached
until there is not enough storage available.
• This helps in offloading the network during
Conclusion
• A collaborative filtering based context-aware caching
scheme has been proposed for 5G networks.
• The proposed approach stores the popularity matrix
as a combination of user, item and context’s rating.
• The popularity matrix is used to decide about
caching the data items.
• The future work lies in the implementation of
References
• A. Karatzoglou, X. Amatriain, L. Baltrunas, and N.
Oliver, "Multiverse recommendation: n-dimensional tensor factorization for context-aware collaborative filtering," presented at Proceedings of the fourth ACM conference on Recommender systems, 2010.
• E. Bastug, M. Bennis, and M. r. Debbah, "Living on
the edge: The role of proactive caching in 5G wireless networks," IEEE Communications