Parallel Architectures for Wideband Sub-Nyquist Sampling

18 concepts14 papers1 proposal

The Nyquist-Shannon sampling theorem is the bedrock of digital signal processing, stating that a signal must be sampled at twice its highest frequency to be perfectly reconstructed. However, as modern applications like cognitive radio and ultra-wideband (UWB) communication push into the gigahertz range, the hardware requirements for Nyquist-rate sampling become prohibitively expensive and power-intensive.

A growing body of research proposes a shift from single, high-speed analog-to-digital converters (ADCs) toward parallel architectures that sample well below the Nyquist rate. By distributing the sampling load across multiple low-rate branches, these systems can capture wideband information at a rate proportional to the signal's actual occupancy—the Landau rate—rather than its highest frequency.

The Multi-Branch Filter Bank

The core intuition behind this direction is that a wideband signal is often "sparse" in the frequency domain, consisting of several narrow active bands separated by empty space. Instead of capturing the entire spectrum with one fast ADC, Chen et al. and Kipnis describe architectures that split the analog signal into parallel branches.

In a typical multi-branch filter bank, each branch applies a unique pre-filter or time delay before sampling at a sub-Nyquist rate. This process effectively creates different "views" of the signal. While a single sub-Nyquist sampler would suffer from aliasing—where high-frequency components fold over and overlap with low-frequency ones—the use of multiple branches with distinct transfer functions allows the system to preserve independent degrees of freedom. Chen et al. demonstrate that such architectures can achieve the full capacity of the analog channel as long as the aggregate sampling rate meets the Landau rate (the sum of the widths of all active bands).

Diversity Through Multi-Rate Sampling

A key challenge in sub-Nyquist sampling is resolving frequency ambiguities. If two different frequencies alias to the same baseband location, they become indistinguishable. To solve this, Xiong et al. and Chiu et al. propose Multi-rate Asynchronous Sub-Nyquist Sampling (MASS).

Instead of all branches sampling at the same low rate, each branch operates at a slightly different frequency. Huang et al. show that if these sampling ratios are chosen to be pairwise coprime, the aliasing patterns in each branch will be unique. By comparing the outputs of these diverse branches, the system can mathematically "unfold" the spectrum and uniquely identify the original signal frequencies. This approach prioritizes sample diversity over raw sampling speed, making it robust to timing synchronization errors that typically plague high-speed hardware.

Spectral Scrambling via Periodic Modulation

While multi-branch filter banks are theoretically optimal, implementing dozens of high-precision analog filters can be hardware-intensive. To simplify the front-end, Chen et al. and Chen et al. introduce the concept of periodic spectral scrambling.

In this architecture, the signal is multiplied by a periodic modulation sequence (often a pseudo-random bit stream) before it is filtered and sampled. This modulation spreads the spectral information across the entire baseband. The result is that a single sampling branch can capture information from multiple disparate frequency bands simultaneously. The authors prove that this modulation-based approach is information-theoretically equivalent to complex multi-branch filter banks, allowing hardware designers to trade off the number of physical branches for computational complexity in the reconstruction phase.

Bridging Information and Sampling Theory

Beyond hardware implementation, this research area establishes a deep link between sampling and Shannon’s capacity limits. Chen et al. and Chen et al. argue that the capacity of a sub-Nyquist sampled channel is determined by the bands with the highest signal-to-noise ratio (SNR).

They show that "alias-suppressing" strategies—which prioritize these high-SNR bands—are necessary to maximize data rates. Interestingly, Kipnis reveals a fundamental equivalence: the sampling architecture that maximizes information capacity is the same one that minimizes the mean-squared error (MSE) of the signal reconstruction. This suggests that efficient sub-Nyquist sampling is not just a hardware convenience, but a mathematically optimal way to process wideband information under resource constraints.

Applications in Cognitive and Optical Networks

These parallel sampling concepts have immediate practical utility:

  • Cognitive Radio: Liang et al. and Pohl et al. apply these architectures to wideband spectrum sensing, allowing devices to detect vacant channels across a massive range without needing expensive, high-speed ADCs.
  • Ultra-Wideband (UWB): Molisch et al. demonstrate that by performing correlation in the analog domain, receivers can sample at the symbol rate rather than the much higher chip rate, drastically reducing power consumption.
  • Optical Networks: Hsu et al. utilize hybrid sub-Nyquist rates to support heterogeneous users on a single fiber link, enabling low-cost and high-performance devices to coexist on the same network.

Key Foundational Work

Joint Compressive Sensing in Wideband Cognitive Networks (2010) — contributed 2 concept elements to this direction.
A low-cost time-hopping impulse radio system for high data rate transmission (2005) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_parallel_filter_bank_sampling_3cd00b91,
  title = {parallel filter bank sampling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/parallel-filter-bank-sampling/rdr_3cd00b910834402da35951e85718e05c},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_parallel_filter_bank_sampling_3cd00b91_v1,
  title = {parallel filter bank sampling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/parallel-filter-bank-sampling/rdr_3cd00b910834402da35951e85718e05c/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake