Peak-to-average power ratio of good codes for Gaussian channel

Yury Polyanskiy, Yihong Wu

Abstract

High-performance communication codes that approach the Shannon capacity of a Gaussian channel must inherently grapple with "spiky" signals. This research establishes that any code achieving a fast convergence to capacity necessarily has a Peak-to-Average Power Ratio (PAPR) that grows logarithmically with the codeword length. This finding suggests that high PAPR is an unavoidable "tax" on efficient communication systems.

Introduction

In digital communications, the Additive White Gaussian Noise (AWGN) channel serves as the fundamental model for analyzing data transmission. A primary goal for engineers is to design error-correcting codes that approach the Shannon capacity—the theoretical maximum data rate for a given power level. However, a significant practical hurdle remains: the Peak-to-Average Power Ratio (PAPR).

High PAPR indicates that a signal contains occasional power spikes significantly higher than its average level. To transmit these signals without distortion, radio frequency (RF) amplifiers must operate in a highly linear regime, which is inherently power-inefficient. While it has long been observed that high-performance codes (such as those used in 5G or WiFi) exhibit high PAPR, this paper investigates whether this is a mathematical necessity or a design limitation. Specifically, it asks: can we design codes that are both capacity-achieving and have low power peaks?

Methodology

The study employs a "converse" approach from information theory, using mathematical bounds to define the fundamental limits of code performance rather than proposing a specific new code. The analysis relies on three key pillars:

  1. Finite Blocklength Analysis: Moving beyond classical Shannon theory, which assumes infinite-length signals, the authors analyze the "backoff" from capacity for finite codeword lengths ($n$) using the concept of channel dispersion.
  2. I-Projection and Divergence: To maximize capacity, a channel's output must approximate a Gaussian distribution. The authors utilize I-projection—finding the closest distribution within a constrained set—and Pinsker’s inequality to quantify the rate loss incurred when signal peaks are artificially capped.
  3. Extremal Value Theory: The authors apply statistical results regarding the behavior of the maximum value of a sequence of random variables to characterize how signal peaks scale as the blocklength $n$ increases.

The Gaussianity Requirement and the Speed-Peak Trade-off

To maximize information transfer, a signal should ideally resemble Gaussian noise. However, a true Gaussian distribution possesses "fat tails," meaning it naturally produces very large values. The paper reveals a critical distinction in how codes approach capacity:

  • Slow Convergence: If a code approaches capacity extremely slowly as codewords lengthen, it is possible to maintain a low PAPR (Proposition 1).
  • Fast Convergence: If a code approaches capacity at the "fast" rate required by modern high-performance standards, power spikes become mandatory.

The research identifies that for high-performance codes, the peak amplitude must scale at a rate of $\Omega(\sqrt{\log n})$. This matches the observed behavior of Orthogonal Frequency Division Multiplexing (OFDM), the technology underpinning most modern high-speed wireless standards.

Key Results

The core of the research is formalized in Theorem 2, which proves that for any code where the rate $R$ approaches capacity $C(P)$ at a standard fast rate (specifically $O(n^{\alpha-1})$), at least half of the codewords must have a peak amplitude exceeding $\sqrt{2\delta \log n}$.

To illustrate the practical impact, the authors provide a numerical example (Remark 4): For a signal-to-noise ratio (SNR) of 20 dB and a blocklength of 10,000, any code achieving 99.9% of the Shannon capacity must have a PAPR of at least 3.85 dB. This universal converse holds regardless of the specific code structure or modulation technique used.

Conclusion and Limitations

This work establishes a fundamental lower bound on PAPR, demonstrating that high power peaks are a necessary consequence of high-speed, capacity-approaching communication. However, several areas remain for future exploration. As a converse result, the paper defines the "floor" of PAPR but does not provide a specific construction for codes that achieve this minimum. Additionally, while the paper establishes logarithmic scaling, the exact optimal trade-off constant between convergence speed and peak height remains an open question. Finally, the analysis focuses on PAPR at Nyquist sampling points; in real-world hardware, "peak regrowth" after filtering may result in even higher peaks than those predicted by these theoretical bounds.

Summary updated

BibTeX
@article{polyanskiy2014peak,
  author = {Yury Polyanskiy and Yihong Wu},
  title = {{Peak-to-average power ratio of good codes for Gaussian channel}},
  journal = {IEEE Trans. Information Theory},
  year = {2014},
  url = {https://openreview.net/forum?id=9qEFwDbEFsY},
}
Figure