Shannon Meets Nyquist: Capacity of Sampled Gaussian Channels

Yuxin Chen, Yonina C. Eldar, Andrea J. Goldsmith

Introduction

In classical communication theory, the Nyquist-Shannon sampling theorem dictates that a signal must be sampled at twice its bandwidth to ensure perfect reconstruction. However, modern wideband systems, such as cognitive radio, often operate across massive bandwidths where Nyquist-rate sampling is power-prohibitive or exceeds the physical limits of current Analog-to-Digital Converter (ADC) hardware. This paper bridges the gap between information theory and sampling theory by determining the capacity of Gaussian channels when sampled at sub-Nyquist rates. The authors explore how to optimize sampling structures—such as filter banks and modulators—to maximize information flow under these hardware constraints.

Methodology: Aliasing as a MIMO System

The authors employ a rigorous mathematical framework based on the asymptotic spectral properties of Toeplitz operators to analyze the transition from continuous-time analog channels to discrete-time sampled sequences. They investigate three primary sampling architectures: single filter samplers, multi-branch filter banks, and modulation-based filter banks.

A central innovation of this work is the mapping of "aliasing"—the overlap of frequency components caused by undersampling—to a Multiple-Input Multiple-Output (MIMO) channel model in the frequency domain. By treating each aliased frequency component as a separate virtual input branch, the authors can apply standard "water-filling" power allocation strategies to determine the maximum achievable capacity.

Frequency-domain representation of the sampling process, modeling aliasing as a multi-branch system.

Key Theoretical Insights

The research reveals several profound connections between sampling and information theory:

  • Optimal Alias Suppression: Contrary to some compressed sensing approaches that favor "mixing" frequencies, this paper proves that maximizing capacity requires suppressing aliasing. The optimal receiver should selectively filter for frequencies with the highest Signal-to-Noise Ratio (SNR) within each aliased set.
  • Capacity Non-Monotonicity: A surprising finding is that sampled capacity is not always a non-decreasing function of the sampling rate. Because different rates create different aliasing patterns, increasing the sampling rate can sometimes cause high-SNR bands to collapse onto each other more unfavorably than at a lower rate.
  • Unification of Capacity and Estimation: The authors demonstrate that the filters which maximize Shannon capacity are identical to those that minimize the Mean-Squared Error (MSE). This reconciles two traditionally distinct fields: information theory and estimation theory.

Key Results and Capacity Behavior

The paper provides the first precise closed-form capacity expressions for undersampled Gaussian channels. A significant result is that for multiband signals, the full analog capacity can be achieved at the Landau rate (the sum of the widths of the active bands), provided the sampling architecture is correctly designed.

Comparison of channel characteristics and the resulting capacity curves showing non-monotonicity.

Furthermore, the study shows that modulation-based sampling can achieve the same capacity as complex filter banks while using significantly fewer analog branches. This offers a practical roadmap for hardware designers to build more efficient wideband receivers that maintain high information throughput despite sub-Nyquist sampling.

Limitations and Future Directions

While the paper establishes a foundational theory, it assumes perfect Channel State Information (CSI) at the receiver, which may not be available in applications like blind spectrum sensing. The current analysis is also restricted to Linear Time-Invariant (LTI) channels and periodic modulation sequences. Future work is needed to extend these capacity results to time-varying channels, random modulation schemes, and multiuser interference scenarios.

Summary updated

BibTeX
@inproceedings{chen2011shannon,
  author = {Yuxin Chen and Yonina C. Eldar and Andrea J. Goldsmith},
  title = {{Shannon Meets Nyquist: Capacity of Sampled Gaussian Channels}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2011},
  doi = {10.1109/icassp.2011.5946352},
}
Figure