Diversity versus Multiplexing at Finite Blocklength

Johan Östman, Wei Yang2016

TL;DR

This paper analyzes the fundamental tradeoff between diversity (reliability) and multiplexing (throughput) in wireless communication systems, but crucially, it does so for finite blocklengths (short packets). They derive new bounds that reveal how this tradeoff shifts for short packets, showing that the optimal strategy depends heavily on packet size, number of antennas, and channel conditions, offering insights for designing future ultra-reliable, low-latency communication links.

What Problem Does This Paper Address?

The classic diversity-multiplexing tradeoff, a cornerstone of MIMO wireless communication, has been well-understood for systems with infinitely long packets. This theory suggests that you can either gain reliability (diversity) or increase data rate (multiplexing), but not both simultaneously. However, emerging 5G applications (like IoT, traffic safety, industrial control) demand short packets with stringent latency and reliability requirements. The existing theory, based on infinite packet lengths, doesn't accurately capture the performance in these short-packet scenarios. This paper addresses the critical question: how should we use multiple antennas for diversity and multiplexing when packets are short, and what are the performance implications of this finite blocklength constraint?

Methods

The authors leverage recent advancements in finite blocklength information theory to derive non-asymptotic bounds on the maximum achievable channel coding rate. They analyze a block-memoryless Rayleigh-fading channel with $M_t$ transmit and $M_r$ receive antennas. Key to their approach is the use of:

  • Dependency Testing (DT) Bound: This bound, adapted for the finite blocklength regime, allows them to establish a lower bound on the achievable rate. They specifically use the Unitary Space-Time Modulation (USTM) input distribution, which is known to perform well in certain scenarios.
  • Meta-Converse (MC) Bound: This bound provides an upper limit on the achievable rate.

By deriving both a lower and an upper bound, they can tightly characterize the achievable rate for finite blocklengths. They also consider the impact of no CSI at the transmitter, which is a realistic assumption for high-mobility or time-critical applications where feedback is impractical.

Datasets

Not discussed in the paper. The evaluation is based on theoretical analysis and numerical simulations using parameters inspired by the 3GPP LTE standard.

Key Insights and Ideas

  • Finite Blocklength Matters: The most significant insight is that the optimal strategy for diversity versus multiplexing fundamentally changes when packet lengths are finite. The infinite blocklength assumption can be misleading.
  • Optimal Tradeoff is Non-Monotonic: For a fixed number of antennas and packet size, the achievable rate is not simply maximized by either pure diversity or pure multiplexing. Instead, there's an optimal balance (an optimal number of time-frequency diversity branches) that maximizes throughput for a given reliability target.
  • Cost of Channel Estimation: For short coherence intervals (many time-frequency branches), the "cost" of learning or accounting for the channel (even without transmitter CSI) becomes significant and can outweigh the benefits of spatial multiplexing.
  • Diversity is More Beneficial for Ultra-Reliability: When extremely low packet error rates (e.g., $10^{-5}$) are required, diversity-exploiting schemes become more advantageous, even when multiplexing gains are theoretically possible.
  • Outage Capacity is a Poor Indicator for Short Packets: The traditional outage capacity, which assumes infinite blocklengths, provides a good approximation only when the channel is quasi-static (long coherence interval). For shorter coherence intervals, it fails to capture the performance degradation due to channel estimation overhead.

Key Results

The paper presents numerical results for $2 \times 2$ and $4 \times 4$ MIMO systems with a packet length of $n=168$ symbols.

  • Optimal Diversity-Multiplexing Balance: Figures 1 and 2 show that for a target error rate of $10^{-3}$, there exists an optimal number of time-frequency diversity branches ($L^*$) that maximizes the achievable rate. For $M_t=M_r=2$, this optimal $L^*$ is around 7-14.
  • Comparison with Alamouti/FSTD: The paper demonstrates that while diversity schemes like Alamouti (for $2 \times 2$) or FSTD (for $4 \times 4$) are nearly optimal when time-frequency diversity is limited ($L \approx 1$), they become significantly suboptimal as the channel offers more diversity ($L$ grows). The gap between optimal multiplexing and these diversity schemes can be up to 0.5 bits/channel use.
  • Ultra-Reliability ($10^{-5}$): For very low error rates (Figures 3 and 4), the gap between optimal schemes and diversity-based schemes narrows, indicating that transmit diversity becomes more crucial for achieving ultra-high reliability.

Limitations and Future Work

  • Block-Memoryless Assumption: The analysis assumes a block-memoryless channel, meaning the channel is constant within a coherence interval but can change between intervals. Future work could explore more complex fading models.
  • No CSI at Transmitter: The analysis assumes no CSI at the transmitter. Extending this to scenarios with limited CSI at the transmitter could provide a more complete picture.
  • Specific Input Distribution: The DT bound relies on the USTM input distribution. While this is a good choice, exploring other input distributions could refine the bounds.
  • Numerical Stability: The authors mention that their derived expression for the output PDF is "easier to compute and more stable numerically" than previous ones, suggesting that numerical computation of these bounds is a practical concern.

The paper's findings directly inform the design of future wireless systems, particularly for 5G and beyond, where short packets and ultra-reliability are paramount. The work highlights the need to move beyond infinite-blocklength theory for these applications.

Summary updated

Abstract

A finite blocklenth analysis of the diversity-multiplexing tradeoff is presented, based on nonasymptotic bounds on the maximum channel coding rate of multiple-antenna block-memoryless Rayleigh-fading channels. The bounds in this paper allow one to numerically assess for which packet size, number of antennas, and degree of channel selectivity, diversity-exploiting schemes are close to optimal, and when instead the available spatial degrees of freedom should be used to provide spatial multiplexing. This finite blocklength view on the diversity-multiplexing tradeoff provides insights on the design of delay-sensitive ultra-reliable communication links. I.
BibTeX
@misc{ostman2016diversity,
  author = {Johan Östman and Wei Yang},
  title = {{Diversity versus Multiplexing at Finite Blocklength}},
  year = {2016},
}
Figure