Content delivery with coded caching and massive MIMO in 5G

Sheng Yang, Khac–Hoang Ngo, Mari Kobayashi2016 · DOI

TL;DR

This paper investigates whether combining massive MIMO and coded caching in 5G networks offers synergistic benefits for content delivery. The authors demonstrate that while each technology is often considered scalable on its own, their combination is crucial for achieving true scalability in practical wireless scenarios, especially when dealing with imperfect channel state information.

What Problem Does This Paper Address?

The explosive growth of mobile data traffic, particularly video content, presents a significant challenge for current and future wireless networks. Two promising technologies, massive MIMO and coded caching, have emerged as potential solutions. Massive MIMO, by employing a large number of antennas at the base station, aims to boost spectral efficiency. Coded caching, on the other hand, leverages user-side memory to prefetch popular content, reducing peak-hour traffic. A common perception is that both technologies are independently scalable with the number of users. However, this paper questions whether these individual scalability claims hold true in realistic wireless environments, particularly when considering imperfect channel state information at the transmitter. The core problem is to determine if and how these two powerful techniques complement each other to provide a robust and scalable content delivery solution for 5G.

Methods

The authors analyze a simplified MISO (Multiple-Input Single-Output) downlink channel with a base station equipped with $n_t$ antennas and $K$ single-antenna users. The channel is assumed to be quasi-static Rayleigh fading, meaning it remains constant for the duration of a coded block but varies between blocks. A key aspect of their methodology is the consideration of imperfect Channel State Information at the Transmitter (CSIT), modeled as an additive error to the true channel.

They define an "equivalent content delivery rate" as a unified metric to compare different strategies. This metric quantifies the total amount of demanded information (including cached content) delivered per unit of time. The paper then compares two primary strategies:

  1. Spatial Multiplexing (Unicast): This approach leverages massive MIMO to send private streams to individual users, aiming to exploit the multiplexing gain. This is analyzed under both perfect and imperfect CSIT.
  2. Coded Caching (Multicast): This strategy utilizes the Maddah-Ali-Niesen coded caching scheme to send common coded streams that are useful to multiple users simultaneously.

The analysis focuses on the asymptotic behavior as the number of users ($K$) approaches infinity, examining different regimes based on how per-user power and CSIT error scale with $K$. They derive a "coded caching gain" ($\gamma$) which represents the ratio of the content delivery rate achieved by coded caching to that achieved by spatial multiplexing.

Datasets

Not discussed in the paper. The evaluation is based on theoretical analysis and numerical simulations.

Key Insights and Ideas

  • Complementarity, Not Just Cumulation: The paper strongly suggests that massive MIMO and coded caching are not just additive but complementary. Their combined power unlocks scalability that neither can achieve alone in practical scenarios.
  • The Achilles' Heel of Massive MIMO: Imperfect CSIT: The scalability of massive MIMO heavily relies on accurate CSIT. When CSIT is imperfect (a realistic scenario), the gains of massive MIMO can vanish, making it less effective.
  • Coded Caching's Robustness: Coded caching, on the other hand, demonstrates remarkable robustness to imperfect CSIT. Its performance degrades much less significantly compared to spatial multiplexing when channel information is noisy.
  • The "Sweet Spot" for Coded Caching: The analysis reveals specific regimes where coded caching significantly outperforms spatial multiplexing. This is particularly true when the per-user power is limited or increases slowly with the number of users, and when the cache size is non-negligible.
  • Cache Size Matters: The benefit of coded caching scales with the amount of cache memory available at the users. A larger cache size leads to a proportionally larger gain over spatial multiplexing, especially in scenarios with imperfect CSIT.

Key Results

  • Scalability in Practical Scenarios: In a simple MISO downlink setting with Rayleigh quasi-static fading and imperfect CSIT, the combination of massive MIMO and coded caching provides a scalable solution, answering the paper's central question affirmatively.
  • Coded Caching Gain ($\gamma$): The paper quantifies the coded caching gain ($\gamma$) in various regimes:
  • Power-limited regime ($P = \Theta(1)$): Coded caching is beneficial unless the cache memory per user vanishes.
  • Fixed per-user power regime ($P = \Theta(K)$): The gain increases with $K$, as multicast rate scales better than unicast rate with imperfect CSIT.
  • Increasing per-user power regime ($p = \Theta(K^\eta)$): The gain depends significantly on $\eta$. For $\eta \in (-1, 1]$, the gain is logarithmic with $K$. For $\eta > 1$, the gain is substantial, especially with a non-zero cache size.
  • Numerical Illustrations: Figures 1-3 demonstrate these findings. For instance, Figure 3(b) shows that with imperfect CSIT, coded caching can be 8.4 times better than spatial multiplexing for a specific cache size and power level, whereas it's only twice as good under perfect CSIT.

Limitations and Future Work

  • Simplified Channel Model: The analysis is based on a simplified MISO downlink channel with i.i.d. Rayleigh fading. Future work could explore more complex channel models, including asymmetric channels and antenna correlation.
  • Number of Antennas: The current analysis assumes $n_t = K$ or $n_t = \Theta(K)$. Extending the analysis to consider the impact of a varying number of transmit antennas ($n_t$) is a promising direction.
  • Combined Strategies: The paper primarily compares pure spatial multiplexing with pure coded caching. Future research could investigate combining both techniques to optimize content delivery rate further, as suggested by prior work [11, 12].
  • Clustering for Asymmetric Channels: For asymmetric channels, the authors propose developing clustering algorithms to maximize the equivalent content delivery rate, with coded caching performed within each cluster.

Summary updated

Abstract

Do the gains from massive MIMO and coded caching cumulate? In this paper, we try to answer the question in a simple setting of downlink MIMO channel with Rayleigh quasi-static fading. While it is generally perceived that each of massive MIMO and coded caching is scalable alone with the number of users, we show that in this setting MIMO and coded caching are indeed complementary and the combination of both provides a scalable solution in most practical scenarios.
BibTeX
@misc{yang2016content,
  author = {Sheng Yang and Khac–Hoang Ngo and Mari Kobayashi},
  title = {{Content delivery with coded caching and massive MIMO in 5G}},
  year = {2016},
  doi = {10.1109/istc.2016.7593139},
}
Figure