Massive MIMO and small cells: How to densify heterogeneous networks

Kianoush Hosseini, Jakob Hoydis, Stephan ten Brink, Mérouane Debbah

TL;DR

This paper proposes a novel network architecture that combines massive MIMO macrocells with dense small cells (SCs) to boost wireless network capacity. By leveraging Time Division Duplex (TDD) and channel reciprocity, the system can effectively suppress interference from SCs to the macrocell, significantly improving the performance of SCs without substantial loss to the macrocell.

What Problem Does This Paper Address?

The ever-increasing demand for mobile data necessitates significant network densification, primarily through the deployment of small cells (SCs). However, integrating these SCs with existing macrocell infrastructure presents a major challenge: cross-tier interference. While SCs offer localized capacity boosts, they can't replace macrocells for broad coverage. Simultaneously, the concept of "massive MIMO" — using a large number of antennas at base stations — promises increased capacity and improved signal quality by focusing energy and reducing interference. However, the overhead of channel training in traditional Frequency Division Duplex (FDD) systems limits the practical number of antennas. This paper tackles the problem of efficiently integrating these two promising technologies (massive MIMO and dense SCs) to achieve higher overall network capacity while managing the inherent interference challenges.

Methods

The core of the proposed solution lies in a TDD-based network architecture. TDD is crucial because it exploits channel reciprocity, meaning the channel used for uplink transmission is the same as the one for downlink. This allows the massive MIMO macrocell base station (BS) to:

  • Estimate Interference Covariance: By observing the uplink signals from SCs, the BS can estimate the covariance matrix of the interference it imposes on the SCs.
  • Covariance-Based Precoding: Using this interference covariance information, the BS designs its downlink precoding vectors. These vectors are engineered to be orthogonal to the dominant interference directions originating from the SCs. Essentially, the BS "sacrifices" some of its massive antenna array's degrees of freedom to actively cancel out interference to the SC tier.

The paper also explores a variation called "Reverse TDD" (R-TDD), where the uplink and downlink periods are swapped in one of the tiers. This can offer different interference management benefits depending on the network topology.

Datasets

Not discussed in the paper. The evaluation is based on simulations.

Key Insights and Ideas

  • TDD is Key for Massive MIMO + SCs: The paper strongly advocates for TDD in this context, highlighting its ability to reduce channel training overhead for massive MIMO and enable crucial interference estimation for SC coexistence.
  • Interference Cancellation via Antenna Sacrifice: The idea of using a portion of the massive MIMO antennas not for data transmission but for actively suppressing interference to other network tiers is a powerful insight.
  • Implicit Coordination: The proposed scheme achieves a form of coordination between the macro and SC tiers without explicit data exchange or complex centralized control, which is a significant advantage given the potential scale of SC deployments.
  • R-TDD Offers Flexibility: The exploration of R-TDD shows that the optimal duplexing strategy can be topology-dependent, offering an additional lever for performance optimization.
  • Indoor vs. Outdoor Performance: The paper reveals that the proposed scheme is particularly effective in indoor environments where cross-tier interference is more dominant. In outdoor scenarios, co-tier interference (interference between SCs) becomes a limiting factor, though the proposed method still offers improvements.

Key Results

The simulations demonstrate significant improvements:

  • Indoor SCs: The proposed TDD scheme with covariance-based precoding can boost SC downlink (DL) rates by up to 42% (with N=100 antennas) and 75% (with N=200 antennas) compared to a baseline without interference suppression. With N=200 antennas, the BS can completely eliminate cross-tier interference, achieving the same sum rate as FDMA-TDD/FDD (where the entire band is dedicated to SCs).
  • Outdoor SCs: The gains are more modest (around 8% for SC DL rates) due to the dominance of co-tier interference, but the macrocell performance loss is negligible.
  • Uplink (UL) Performance: The R-TDD scheme combined with covariance-based precoding shows substantial improvements in SC uplink rates, up to 61% with N=100 and 117% with N=200 antennas.
  • Comparison to Benchmarks: The proposed scheme generally outperforms benchmark architectures like FDMA-TDD/FDD and R-TDD-UD (uplink-downlink duality) in terms of achievable rates for the SC tier.

Limitations and Future Work

  • Stationary Channels Assumption: The paper assumes stationary and ergodic channels for covariance estimation. In reality, channels can fluctuate, potentially impacting estimation accuracy.
  • Single-Cell Focus: The current analysis is for a single-cell scenario. Future work will evaluate performance in a more realistic multi-cell setting.
  • Macrocell Performance Loss: While the paper states the macrocell performance loss is negligible, it's still a factor to consider, especially in scenarios with limited antennas.
  • Future Directions: The authors suggest extending the work to multi-cell scenarios, incorporating power control and other interference reduction techniques for further gains, and exploring similar precoding schemes for SCs equipped with multiple antennas. They also propose investigating the performance gains in a more realistic multi-cell setting where SCs might also employ multiple antennas.

Summary updated

Abstract

We propose a time division duplex (TDD) based network architecture where a macrocell tier with a “massive” multiple-input multiple-output (MIMO) base station (BS) is overlaid with a dense tier of small cells (SCs). In this context, the TDD protocol and the resulting channel reciprocity have two compelling advantages. First, a large number of BS antennas can be deployed without incurring a prohibitive overhead for channel training. Second, the BS can estimate the interference covariance matrix from the SC tier which can be leveraged for downlink precoding. In particular, the BS designs its precoding vectors to transmit independent data streams to its users while being orthogonal to the subspace spanned by the strongest interference directions; thereby minimizing the sum interference imposed on the SCs. In other words, the BS “sacrifices” some of its antennas for interference cancellation while the TDD protocol allows for an implicit coordination across both tiers. Simulation results suggest that, given a sufficiently large number of BS antennas, the proposed scheme can significantly improve the sum-rate of the SC tier at the price of a small macro performance loss.
BibTeX
@inproceedings{hosseini2013massive,
  author = {Kianoush Hosseini and Jakob Hoydis and Stephan ten Brink and Mérouane Debbah},
  title = {{Massive MIMO and small cells: How to densify heterogeneous networks}},
  booktitle = {ICC},
  year = {2013},
  doi = {10.1109/icc.2013.6655455},
}
Figure