Abstract
Traditional MIMO theory often assumes infinite blocklengths, yet emerging 5G mission-critical applications rely on short data packets where this assumption fails. This paper develops new finite-blocklength bounds to analyze the efficiency of MIMO systems under these constraints. We demonstrate that the "estimation tax"—the overhead required to learn channel coefficients—can make using fewer antennas more efficient than using many. Our findings reveal a fundamental tradeoff between reliability, throughput, and latency, showing that traditional metrics like ergodic and outage capacity significantly overestimate performance in short-packet scenarios.
Introduction
The "more is better" mantra of Multiple-Input Multiple-Output (MIMO) systems is challenged when applied to the short data packets typical of 5G mission-critical machine-type communications (MTC). Applications such as industrial automation or remote surgery require ultra-low latency, necessitating packets of approximately 100 bits with extremely high reliability (e.g., $10^{-9}$ error rate).
In these regimes, standard metrics like ergodic capacity and outage capacity break down because they ignore two critical factors:
- Finite-Blocklength Penalty: Coding over short sequences is inherently less efficient than coding over long ones.
- Estimation Overhead: Effective multi-antenna use requires channel knowledge. Learning these coefficients via pilot symbols consumes precious space in a short packet, creating a prohibitively expensive "estimation tax."
Methodology
The authors employ finite-blocklength information theory to derive nonasymptotic upper and lower bounds on the maximum coding rate.
- Achievability (Lower Bound): A "Dependence-Testing" (DT) bound is derived based on Unitary Space-Time Modulation (USTM). This assumes no transmitter channel knowledge but utilizes isotropically distributed matrices, which are optimal at high signal-to-noise ratios (SNR).
- Converse (Upper Bound): A "Meta-Converse" (MC) theorem establishes the absolute physical limit of achievable coding rates.
- Comparison: These bounds are compared against practical schemes like the Alamouti code (for 2x2 MIMO) and Frequency-Switched Transmit Diversity (FSTD).
The theoretical bounds are evaluated using Monte Carlo simulations based on a standard Rayleigh block-fading channel model. Simulations utilize a blocklength of $n=168$ symbols (a 1ms LTE-like slot), an SNR of 6 dB, and target error probabilities of $10^{-3}$ and $10^{-5}$.
Results and Discussion
The most significant insight is the existence of an "estimation tax" that dictates an optimal number of transmit antennas. As the channel becomes more dynamic (increasing diversity branches), the cost of learning the channel for each antenna eventually outweighs the reliability gains.
As shown in Figure 1, the newly derived upper (MC) and lower (DT) bounds are remarkably tight, even for packets as short as 100 symbols. The figure highlights that outage capacity ($C_{out}$) can overestimate the achievable rate by a factor of two as diversity increases. Furthermore, there is a clear "sweet spot" for diversity; for a fixed packet size, spreading a codeword over too many frequency or time slots requires switching off some transmit antennas to minimize overhead. For instance, in a 2x2 system with 168 symbols, it is only optimal to use both antennas if the channel remains constant for at least 8 symbols ($n_c \ge 8$).
The Alamouti scheme performs near-optimally for low diversity. However, as diversity increases, its performance diverges from the theoretical bounds because it lacks a mechanism to reduce the number of active antennas to mitigate estimation costs.
Conclusion
This research provides a precise tool for predicting MIMO performance in the short-packet regime. It demonstrates that relying on infinite-blocklength theories can lead to rate losses of up to 30% due to suboptimal antenna configurations. While the current model assumes Rayleigh fading, future work could extend these bounds to Line-of-Sight (LoS) or millimeter-wave channels and bridge the gap between theoretical bounds and specific pilot-aided modulation formats.
Summary updated