On the complementary roles of massive MIMO and coded caching for content delivery

Khac–Hoang Ngo, Sheng Yang, Mari Kobayashi, Kaibin Huang2016 · DOI

TL;DR

This paper investigates whether the benefits of massive MIMO and coded caching, two promising technologies for future content delivery, can be combined for even greater gains. The authors demonstrate that while neither technology is perfectly scalable on its own under practical conditions, their synergistic combination offers a robust and scalable solution, particularly when power is optimally allocated.

What Problem Does This Paper Address?

The ever-increasing demand for content delivery in mobile networks presents a significant challenge. Two key technologies have emerged as potential solutions: massive MIMO, which uses a large number of antennas at the base station to boost spectral efficiency, and coded caching, which leverages user-side memory to prefetch popular content, reducing peak-hour traffic. A common assumption is that both technologies are independently scalable with the number of users. However, this paper questions this assumption, particularly under realistic scenarios where ideal conditions (like perfect channel state information at the transmitter) may not hold. The core problem addressed is to understand if and how these two technologies can be combined to achieve a truly scalable and improved content delivery rate, and to quantify their complementary roles.

Methods

The authors consider a MISO (Multiple-Input Single-Output) downlink channel with Rayleigh quasi-static fading. They analyze a system that can transmit both common information (useful to all users) and private information (specific to individual users). This is achieved through a mixed transmission strategy, where a portion of the power is allocated to a common signal and the rest to private signals. Zero-forcing (ZF) precoding is employed for the private signals to mitigate inter-user interference, with the assumption that the number of users is less than the number of antennas.

The paper then defines an "equivalent content delivery rate" as the metric for evaluating performance. This rate considers the combined throughput from both spatial multiplexing (inherent in MIMO) and coded caching. They analyze three scenarios:

  1. Spatial multiplexing only: Sending private streams to individual users.
  2. Coded caching only: Sending common coded streams to all users.
  3. Mixed delivery: Combining both common and private transmissions.

A significant part of the paper focuses on the asymptotic behavior of these rates as the number of users ($K$) approaches infinity. This analysis involves exploring different "regimes" based on how total power, per-user power, and estimation error scale with $K$. Finally, they investigate optimal power splitting between common and private signals to maximize the mixed delivery rate.

Datasets

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

Key Insights and Ideas

  • Scalability is Not Guaranteed: The paper challenges the notion that massive MIMO and coded caching are inherently scalable on their own. Practical limitations, such as imperfect channel state information at the transmitter (CSIT) for massive MIMO and the need for a non-vanishing multicast rate for coded caching, can hinder scalability.
  • Complementarity is Key: The central finding is that massive MIMO and coded caching are indeed complementary. When combined, they can overcome the limitations of each other, leading to a scalable solution that outperforms either technology in isolation.
  • Power Allocation is Crucial: The effectiveness of the combined approach heavily relies on optimal power splitting between common and private signals. The paper shows that in many scenarios, allocating a significant portion of the power to the common signal is beneficial for maximizing the overall content delivery rate.
  • Cache Size Matters: The optimal strategy and the degree of complementarity depend on the cache size ($M$) relative to the library size ($N$). For small cache sizes, spatial multiplexing might be more effective, while larger cache sizes favor coded caching. The mixed approach offers a balance that is often superior.

Key Results

  • Asymptotic Analysis: The paper provides detailed asymptotic analysis of the content delivery rates for the three schemes (spatial multiplexing, coded caching, and mixed delivery) under various power and error regimes as the number of users ($K$) grows. This analysis reveals how the rates scale and which scheme dominates under different conditions.
  • Optimal Power Splitting: A proposition is presented that outlines the optimal power splitting strategy for the mixed delivery scheme in the large $K$ regime. It suggests allocating more power to the common signal when cache sizes are substantial.
  • Numerical Simulations: Figures 1, 2, and 3 illustrate the findings. Figure 1 shows that the mixed delivery rate increases with cache size and that optimal power splitting often favors the common signal. Figure 2 demonstrates that the mixed transmission consistently outperforms individual spatial multiplexing and coded multicasting. Figure 3 visually represents the optimal power splitting as a function of cache memory, confirming the trend of allocating more power to the common signal with increasing cache size. For instance, in one scenario, the mixed transmission achieved over 150% gain compared to other approaches.

Limitations and Future Work

  • Idealized Channel Model: The paper uses a simplified MISO downlink channel model with Rayleigh quasi-static fading and assumes perfect CSI at the receiver. More complex and realistic channel models, such as those with asymmetry due to user location or antenna correlation, are suggested for future work.
  • Centralized Caching Assumption: The analysis of coded caching primarily focuses on centralized caching. Extending this to decentralized caching scenarios would be valuable.
  • Clustering and Fairness: For asymmetric channel settings, the authors propose exploring clustering algorithms where coded caching is performed within clusters. Additionally, deriving rate regions and developing utility optimization frameworks to ensure fairness among users are identified as important future research directions.

Summary updated

Abstract

In this paper, we extend the answer for the question: Do the gains from massive MIMO and coded caching cumulate? in a simple setting of downlink MIMO channel with Rayleigh quasi-static fading. Under some practical assumptions, the scalability of massive MIMO and coded caching alone with the number of users does not hold. On the other hand, we show that in this setting, the combination of both provides a scalable solution in most practical scenarios. Therefore, MIMO and coded caching are indeed complementary, especially under an optimal power allocation.
BibTeX
@misc{ngo2016complementary,
  author = {Khac–Hoang Ngo and Sheng Yang and Mari Kobayashi and Kaibin Huang},
  title = {{On the complementary roles of massive MIMO and coded caching for content delivery}},
  year = {2016},
  doi = {10.1109/atc.2016.7764780},
}
Figure