Difference-Based Recovery for Modulo Sampling: Tightened Bounds and Robustness Guarantees

Wenyi Yan, Zeyuan Li, Lu Gan, Honqing Liu, Guoquan LiarXiv 2025 · DOI · arXiv

TL;DR

This paper significantly lowers the barrier for "unlimited sensing" by tightening the theoretical oversampling requirements for modulo ADCs from $17.1\times$ Nyquist down to just $\pi \approx 3.14\times$. The authors introduce a robust, second-order difference recovery method (RSoD) that handles both measurement noise and—for the first time—timing jitter, validated by an FPGA prototype achieving a massive $108\times$ dynamic range expansion.

What Problem Does This Paper Address?

Standard Analog-to-Digital Converters (ADCs) have a "ceiling": if a signal's amplitude exceeds the ADC's voltage range, the peaks are clipped, destroying information. Modulo sampling (or "unlimited sensing") solves this by "folding" the signal back into the range $[-\lambda, \lambda)$ whenever it exceeds the threshold, much like how a phase wraps around $2\pi$.

While mathematically elegant, modulo sampling has faced two major hurdles in the real world:

  1. High Sampling Rates: Previous proofs suggested you needed to sample at least 17 times the Nyquist rate to reliably "unwrap" the folds.
  2. Hardware Imperfections: Real-world electronics suffer from thermal noise and timing jitter (imperfect clock cycles). Most existing algorithms assume a perfect, uniform clock, which makes them fragile in practice.

Methods

The researchers revisit the High-order Difference (HoD) recovery framework. The core idea is that while the modulo signal is discontinuous, its $N$-th order finite differences remain bounded and related to the original signal's derivatives.

  • Tightening the Bounds: By applying the Mean Value Theorem for divided differences instead of the looser Stirling-type approximations used in prior work, the authors proved that the "extra" factor of $e \approx 2.718$ in previous oversampling bounds was an artifact of the proof, not a physical requirement.
  • Revised Second-Order Difference (RSoD): The authors focus on the $N=2$ case as a "sweet spot." Higher orders ($N > 2$) reduce the required sampling rate further but amplify noise exponentially. RSoD provides the best trade-off for practical hardware.
  • Jitter-Aware Analysis: They extended the mathematical model to account for non-uniform sampling intervals (jitter). They derived a new "gap-dependent" bound that ensures the signal can still be unwrapped even if the clock isn't perfectly steady.
  • Hardware Implementation: They built an FPGA-based prototype to move beyond software simulations and prove the algorithm works with real analog front-ends.

Datasets

The paper uses two primary sources for evaluation:

  1. Synthetic Simulations: Bandlimited signals generated from shifted sinc functions and Gaussian noise to test "phase transitions" (where the algorithm succeeds vs. fails).
  2. Hardware Measurements: Data captured from a custom FPGA-based modulo ADC prototype. The system was tested with signals in the kHz range, using an 8-bit Tektronix oscilloscope to capture the "ground truth" and folded outputs.

Key Insights and Ideas

  • The "e" is gone: The reduction of the oversampling factor from $2\pi e$ to $\pi$ is a major theoretical win. It means modulo ADCs can operate much closer to the Nyquist limit than previously thought.
  • Noise vs. Order: There is a "Goldilocks" zone for the difference order $N$. While $N \to \infty$ theoretically lets you sample at exactly the Nyquist rate, the noise amplification ($2^N$) makes it unusable. $N=2$ or $N=3$ are the practical winners.
  • Jitter Robustness: The paper proves that timing jitter effectively acts as a penalty on the oversampling budget. As long as you sample slightly faster than the "noiseless" theoretical limit, the system remains stable under modest jitter.
  • Computational Efficiency: Unlike iterative solvers (like LASSO or ITER-SIS) which can take tens or hundreds of milliseconds to converge, the RSoD method is a direct, "one-pass" algorithm taking less than 2ms, making it ideal for real-time applications.

Key Results

  • Dynamic Range: The FPGA prototype successfully reconstructed signals with an amplitude expansion factor of $\rho = 108$. This means the system could digitize a signal 108 times larger than the ADC's native range.
  • Oversampling Efficiency: For a signal with $10\times$ amplitude expansion, the RSoD method required an oversampling factor (OF) of only ~8-10, whereas previous theories suggested an OF of >17 was mandatory.
  • Noise Resilience: In head-to-head tests, RSoD achieved reliable reconstruction at an input SNR of 11 dB, while the classical HoD method required at least 21 dB.
  • Resolution Gain: Using a modulo ADC with RSoD recovery provided a gain of ~37.5 dB in SINAD (Signal-to-Noise and Distortion ratio) compared to a conventional ADC at the same bit depth.

Limitations and Future Work

  • High-Frequency Artifacts: At very high frequencies (>450 kHz), the analog folding circuitry introduces non-idealities like hysteresis and slew-rate limits. These create "impulsive noise" that the current RSoD algorithm cannot yet handle perfectly.
  • Error Propagation: Because the method relies on anti-differencing (integration), a single mistake in "unwrapping" a fold can cause an error that propagates through the rest of the signal.
  • Future Directions: The authors suggest developing "impulsive noise" detectors to scrub hardware artifacts before the unwrapping stage and exploring how this technique could be applied to compressed sensing.

Summary updated

Abstract

Conventional analog-to-digital converters (ADCs) clip when signals exceed their input range. Modulo (unlimited) sampling overcomes this limitation by folding the signal before digitization, but existing recovery methods are either computationally intensive or constrained by loose oversampling bounds that demand high sampling rates. In addition, none account for sampling jitter, which is unavoidable in practice. This paper revisits difference-based recovery and establishes new theoretical and practical guarantees. In the noiseless setting, we prove that arbitrarily high difference order reduces the sufficient oversampling factor from $2πe$ to $π$, substantially tightening classical bounds. For fixed order $N$, we derive a noise-aware sampling condition that guarantees stable recovery. For second-order difference-based recovery ($N=2$), we further extend the analysis to non-uniform sampling, proving robustness under bounded jitter. An FPGA-based hardware prototype demonstrates reliable reconstruction with amplitude expansion up to $ρ= 108$, confirming the feasibility of high-performance unlimited sensing with a simple and robust recovery pipeline.
BibTeX
@misc{yan2025difference,
  author = {Wenyi Yan and Zeyuan Li and Lu Gan and Honqing Liu and Guoquan Li},
  title = {{Difference-Based Recovery for Modulo Sampling: Tightened Bounds and Robustness Guarantees}},
  year = {2025},
  eprint = {2509.12971},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.12971},
  howpublished = {arXiv},
}
Figure