The pre-log of Gaussian broadcast with feedback can be two

Michèle A. Wigger, Michael Gastpar

Abstract

A generic intuition says that the pre-log, or multiplexing gain, cannot be larger than the minimum of the number of transmit and receive dimensions. This suggests that for the scalar broadcast channel, the pre-log cannot exceed one. By contrast, in this note, we show that when the noises are anti-correlated and feedback is present, then a pre-log of two can be attained. In other words, in this special case, in the limit of high SNR, the scalar Gaussian broadcast channel turns into two parallel AWGN channels. Achievability is established via a coding strategy due to Schalkwijk, Kailath, and Ozarow.

Introduction

In wireless communication, a fundamental "rule of thumb" suggests that the number of independent data streams (multiplexing gain or pre-log) is strictly limited by the number of transmit antennas. For a single-antenna (scalar) transmitter broadcasting to multiple receivers, this limit was widely believed to be one. This paper challenges that intuition by demonstrating how the interplay between feedback and noise correlation can effectively double the capacity scaling at high signal-to-noise ratios (SNR).

System Model and the Broadcast Channel

The authors investigate a two-user Additive White Gaussian Noise (AWGN) broadcast channel. In this model, a single transmitter (ENC) sends a signal $X$ to two receivers, which observe $Y_1$ and $Y_2$ respectively. Crucially, the transmitter receives perfect, causal feedback of these channel outputs, allowing it to adjust its transmission based on what the receivers have already seen.

A two-user AWGN broadcast channel where the transmitter (ENC) receives causal feedback from both receivers.

As shown in Figure 1, the feedback loop includes a delay element ($D$). The noise terms $Z_1$ and $Z_2$ are added to the transmitted signal $X$. The core discovery is that if these noises are perfectly anti-correlated ($\rho_z = -1$, meaning $Z_1 = -Z_2$), the pre-log jumps from one to two.

The SKO-based Coding Strategy

The achievability of this result relies on a specialized version of the Schalkwijk-Kailath-Ozarow (SKO) scheme. This iterative process works as follows:

  1. Initial Phase: The transmitter sends the messages $M_1$ and $M_2$ directly to establish baseline estimates at the decoders.
  2. Error Tracking: Using the causal feedback, the transmitter calculates the current estimation errors at both receivers.
  3. Refinement: The transmitter then broadcasts a weighted linear combination of these errors.

When the noises are perfectly anti-correlated, the transmitter can craft a single signal that simultaneously corrects the errors for both users. Because the transmitter knows the noise via feedback, it can effectively "cancel" the interference each user would otherwise experience from the other's message, turning one physical channel into two parallel, independent pipes.

Key Results and Extensions

The paper establishes several critical theoretical bounds:

  • Theorem 1: For the two-user AWGN broadcast channel with full causal feedback and $\rho_z = -1$, the pre-log is exactly two.
  • Discontinuity: This gain is highly sensitive; if the correlation is anything less than perfect ($|\rho_z| < 1$), the pre-log reverts to one.
  • Feedback Sufficiency: Interestingly, feedback from both users isn't strictly necessary. If the noise is perfectly anti-correlated, feedback from just one receiver allows the transmitter to deduce the state of the entire system.
A two-user AWGN interference channel with feedback, where the pre-log of two is also achievable under specific conditions.

Beyond the broadcast scenario, the authors extend this result to the two-user interference channel (Figure 2). Provided the channel gains are equal and the noise is anti-correlated, two separate transmitters can achieve a total pre-log of two through feedback-driven coordination.

Conclusion and Limitations

While mathematically groundbreaking, the result serves primarily as a theoretical existence proof. The requirement for perfect anti-correlation is physically motivated (e.g., two receivers affected by the same interferer at opposite phases) but extremely sensitive to deviations. Furthermore, the pre-log does not scale indefinitely; for $K > 2$ receivers, the maximum pre-log remains capped at two. Nonetheless, the work highlights that dimensionality is not destiny when structured noise and feedback are available to the transmitter.

Summary updated

Abstract

A generic intuition says that the pre-log, or multiplexing gain, cannot be larger than the minimum of the number of transmit and receive dimensions. This suggests that for the scalar broadcast channel, the pre-log cannot exceed one. By contrast, in this note, we show that when the noises are anti-correlated and feedback is present, then a pre-log of two can be attained. In other words, in this special case, in the limit of high SNR, the scalar Gaussian broadcast channel turns into two parallel AWGN channels. Achievability is established via a coding strategy due to Schalkwijk, Kailath, and Ozarow.
BibTeX
@misc{wigger2008pre,
  author = {Michèle A. Wigger and Michael Gastpar},
  title = {{The pre-log of Gaussian broadcast with feedback can be two}},
  year = {2008},
  eprint = {0805.0990},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/0805.0990},
}
Figure