Success Probability and Area Spectral Efficiency in Multiuser MIMO HetNets

Chang Li, Jun Zhang, Jeffrey G. Andrews, Khaled Ben Letaief

TL;DR

This paper mathematically proves that the "SIR invariance" property—the idea that adding more base stations doesn't hurt signal quality—found in simple cellular networks disappears when you introduce multi-antenna (MIMO) technology. The authors reveal a fundamental tradeoff: while more base stations maximize total network capacity (Area Spectral Efficiency), they can actually degrade individual link reliability, requiring a careful balancing act in 5G network design.

What Problem Does This Paper Address?

In the early days of heterogeneous network (HetNet) research, a surprising "invariance" property was discovered: in a single-antenna (SISO) network, adding more base stations (BSs) increases the total capacity without decreasing the Signal-to-Interference Ratio (SIR). Essentially, the increased interference from more BSs is perfectly balanced by the fact that users are closer to their serving BSs.

However, modern 5G networks rely on Multi-User MIMO (MU-MIMO), where BSs use many antennas to serve multiple users simultaneously. Does this "more is better" rule still hold? This paper addresses the gap in understanding how BS density, antenna counts, and multiplexing gains interact to affect both the reliability of a single connection and the overall capacity of the entire network.

Methods

The authors employ stochastic geometry, specifically modeling base station locations as a Poisson Point Process (PPP) across $K$ different tiers (e.g., macro, pico, and femtocells).

Key technical contributions include:

  • Toeplitz Matrix Representation: They derive a novel, closed-form expression for success probability using a lower triangular Toeplitz matrix. This makes the complex math of gamma-distributed channel gains (common in MIMO) much more tractable.
  • Asymptotic Analysis: They provide a simplified "large SIR" limit formula that remains highly accurate even at moderate SIR levels (above 5 dB).
  • Optimization Framework: To navigate the tradeoff between capacity and reliability, they develop two algorithms:
  • Algorithm 1: A simple greedy-style approach for "U-SDMA" networks where all tiers serve the same number of users.
  • Algorithm 2: A more complex approach using Dinkelbach’s algorithm and Sequential Convex Programming (SCP) to find optimal BS densities for general MIMO HetNets.

Datasets

As a theoretical paper in communications engineering, it does not use external datasets. Instead, the authors validate their mathematical derivations through extensive Monte Carlo simulations of a 3-tier HetNet (Macro, Micro, and Pico BSs) with varying antenna configurations ($M_k$) and user loads ($U_k$).

Key Insights and Ideas

  • The End of Invariance: Unlike SISO networks, MIMO HetNets are not SIR-invariant. Increasing the density of a specific tier can either increase or decrease the overall link reliability depending on that tier's antenna-to-user ratio.
  • The "One-Tier" Paradox: To achieve the absolute maximum link reliability, the network should actually deactivate all but one tier of base stations (specifically, the one with the best ratio of diversity gain to interference).
  • The ASE-Reliability Tradeoff: There is a direct conflict between Area Spectral Efficiency (ASE)—the total bits per second per unit area—and Link Reliability. Maximizing ASE requires activating every possible BS, but this creates a "floor" of interference that can make individual connections unreliable.
  • Ordering of Influence: Different tiers have different "strengths" of influence on the network. The authors show that as reliability requirements become stricter, the network should progressively deactivate tiers, starting with those that cause the most interference relative to the signal gain they provide.

Key Results

  • Mathematical Accuracy: The derived asymptotic results are shown to be nearly indistinguishable from exact numerical results and simulations for SIR thresholds above 10 dB.
  • Quantifiable Tradeoffs: In a simulated 3-tier network, the authors demonstrate that to move from a success probability of 0.6 to 0.8, a significant portion of the network's potential ASE must be sacrificed by deactivating specific tiers of BSs.
  • Density Sensitivity: The paper proves that in MIMO HetNets, the success probability is a monotonic function of BS density, providing a clear mathematical path for network operators to tune their deployments.

Limitations and Future Work

  • Perfect CSI: The study assumes perfect Channel State Information (CSI) at the base stations. In reality, CSI feedback is noisy and delayed, which would likely further degrade performance.
  • Full Buffer Model: The authors assume a "fully-loaded" network where every BS always has users to serve. A "lightly-loaded" model might show different interference dynamics.
  • Precoding Strategy: The analysis is specific to Zero-Forcing (ZF) precoding. While ZF is common, other strategies like Maximum Ratio Transmission (MRT) or interference alignment might yield different scaling laws.
  • Path Loss: The paper uses a standard single-slope path loss model; the authors suggest that multi-slope models or more sophisticated interference coordination should be investigated in future work.

Summary updated

BibTeX
@article{li2016success,
  author = {Chang Li and Jun Zhang and Jeffrey G. Andrews and Khaled Ben Letaief},
  title = {{Success Probability and Area Spectral Efficiency in Multiuser MIMO HetNets}},
  journal = {IEEE Trans. Communications},
  year = {2016},
  doi = {10.1109/tcomm.2016.2531719},
}
Figure