Oversampling Increases the Pre-Log of Noncoherent Fading Channels

Meik Dörpinghaus, Günther Koliander, Giuseppe Durisi, Erwin Riegler, Heinrich Meyr

TL;DR

This paper demonstrates that a common method for analyzing wireless channel capacity, called symbol matched filtering, is suboptimal. By oversampling the received signal, the authors show it's possible to achieve a higher capacity pre-log, meaning more efficient data transmission at high signal-to-noise ratios, especially in noncoherent fading channels.

What Problem Does This Paper Address?

Wireless communication channels are often subject to fading, where the signal strength fluctuates over time. In many real-world scenarios, the exact characteristics of this fading are unknown to both the transmitter and receiver (a "noncoherent" setting). Analyzing the maximum achievable data rate (capacity) of such channels is crucial for designing efficient communication systems. A standard approach for analyzing these channels, particularly in the high signal-to-noise ratio (SNR) regime, involves simplifying the continuous-time channel into a discrete-time one by filtering the received signal with a matched filter and sampling at the symbol rate. This method, known as symbol matched filtering, has been shown to yield a capacity pre-log of $1 - Q/N$, where $N$ is the number of symbols within a fading block and $Q$ relates to the channel's temporal correlation. However, it's unclear if this simplification is truly optimal for the underlying continuous-time channel. This paper investigates whether this symbol matched filtering approach is indeed capacity-achieving or if alternative processing strategies can yield better results.

Methods

The core of this paper's contribution lies in analyzing the capacity of a continuous-time, time-selective, Rayleigh block-fading channel. The authors move beyond the standard symbol matched filtering approach by considering oversampling the received signal. Specifically, they propose sampling the continuous-time channel output at twice the symbol rate.

The analysis leverages techniques from information theory, particularly focusing on the high-SNR capacity pre-log. This pre-log represents the asymptotic growth rate of the channel capacity as the SNR becomes very large. The authors derive a lower bound for the capacity pre-log under the oversampling strategy, proving it to be at least $1 - 1/N$. This is a significant improvement over the $1 - Q/N$ achieved by symbol matched filtering, especially when $Q > 1$.

The proof relies on advanced mathematical tools, including:

  • Differential entropy calculations: Analyzing the information content of the received signal.
  • Matrix analysis: Utilizing properties of covariance matrices and determinants.
  • Jensen's inequality: A mathematical tool for bounding expectations of concave functions.
  • Jacobian analysis: Examining the transformation of random variables through nonlinear mappings, particularly in the context of proving that oversampling provides a "sufficient statistic" for channel estimation.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and does not involve experimental data or simulations on specific datasets.

Key Insights and Ideas

  • Symbol matched filtering is not capacity-achieving: The paper provides a theoretical proof that the common practice of sampling at symbol rate after matched filtering is suboptimal for achieving the maximum data rate in noncoherent fading channels at high SNRs.
  • Oversampling unlocks higher capacity: By simply increasing the sampling rate (specifically, by a factor of two), the system can achieve a higher capacity pre-log. This suggests that the "extra" samples provide valuable information for better channel estimation or decoding.
  • The pre-log improvement is significant: The new lower bound of $1 - 1/N$ is demonstrably better than $1 - Q/N$ when $Q > 1$, meaning the gains are more pronounced in channels with higher temporal correlation.
  • Oversampling provides a "sufficient statistic": The paper argues that oversampling allows the receiver to obtain a sufficient statistic for the fading channel, which is crucial for optimal detection and capacity achievement. This is in contrast to symbol-rate sampling, which may not capture all necessary information due to bandwidth expansion caused by the fading process.
  • Trade-off between simplicity and optimality: While symbol matched filtering is computationally simpler, oversampling, though requiring more complex processing (often nonlinear), leads to better theoretical performance bounds.

Key Results

The main quantitative result is the establishment of a lower bound for the capacity pre-log of a continuous-time, time-selective, Rayleigh block-fading channel in the noncoherent setting. The paper proves that by oversampling the received signal by a factor of two, a capacity pre-log of at least $1 - 1/N$ can be achieved. This is compared to the $1 - Q/N$ pre-log obtained with symbol matched filtering, where $N$ is the number of symbols per block and $Q$ is the rank of the channel covariance matrix. This result implies that symbol matched filtering is not capacity achieving at high SNR.

Limitations and Future Work

  • Oversampling factor: The paper specifically analyzes oversampling by a factor of two. While this demonstrates the principle, it would be interesting to explore if further oversampling yields even greater gains or if there's an optimal oversampling factor.
  • Nonlinear processing: The paper notes that acquiring the additional channel knowledge through oversampling often requires nonlinear processing, which can be computationally intensive. Future work could focus on developing more efficient and practical nonlinear estimation and decoding schemes.
  • Generalization: The analysis is for Rayleigh block-fading channels. Extending these results to other fading distributions or channel models (e.g., frequency-selective channels, channels with memory beyond a single block) would be valuable.
  • Practical implementation: While the theoretical gains are clear, the practical implementation challenges of oversampling and the associated nonlinear processing in real-world wireless systems remain an area for further research.
  • Other SNR regimes: The analysis focuses on the high-SNR regime. Understanding the impact of oversampling at lower SNRs would also be beneficial.

Summary updated

Abstract

We analyze the capacity of a continuous-time, time-selective, Rayleigh block-fading channel in the high signal-to-noise ratio (SNR) regime. The fading process is assumed stationary within each block and to change independently from block to block; furthermore, its realizations are not known a priori to the transmitter and the receiver (noncoherent setting). A common approach to analyzing the capacity of this channel is to assume that the receiver performs matched filtering followed by sampling at symbol rate (symbol matched filtering). This yields a discrete-time channel in which each transmitted symbol corresponds to one output sample. Liang & Veeravalli (2004) showed that the capacity of this discrete-time channel grows logarithmically with the SNR, with a capacity pre-log equal to $1-{Q}/{N}$. Here, $N$ is the number of symbols transmitted within one fading block, and $Q$ is the rank of the covariance matrix of the discrete-time channel gains within each fading block. In this paper, we show that symbol matched filtering is not a capacity-achieving strategy for the underlying continuous-time channel. Specifically, we analyze the capacity pre-log of the discrete-time channel obtained by oversampling the continuous-time channel output, i.e., by sampling it faster than at symbol rate. We prove that by oversampling by a factor two one gets a capacity pre-log that is at least as large as $1-1/N$. Since the capacity pre-log corresponding to symbol-rate sampling is $1-Q/N$, our result implies indeed that symbol matched filtering is not capacity achieving at high SNR.
BibTeX
@article{dorpinghaus2014oversampling,
  author = {Meik Dörpinghaus and Günther Koliander and Giuseppe Durisi and Erwin Riegler and Heinrich Meyr},
  title = {{Oversampling Increases the Pre-Log of Noncoherent Fading Channels}},
  journal = {IEEE Trans. Information Theory},
  year = {2014},
  url = {https://openreview.net/forum?id=FGfUckiA4H},
}
Figure