⚡ electrical engineering

Asynchronous Multi-Channel USF: Modified CRT for Modulo Unfolding

This paper proposes an asynchronous Multi-Channel Unlimited Sampling Framework that replaces synchronization-dependent Chinese Remainder Theorem unfolding with a graph-based smoothness strategy, enabling robust signal reconstruction from temporally misaligned measurements.

Original authors: Ruiming Guo, Ayush Bhandari

Published 2026-06-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ruiming Guo, Ayush Bhandari

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Problem: The "Clipping" Ceiling

Imagine you are trying to record a sound with a microphone. If the sound is too quiet, you can't hear it. If it's too loud, the microphone "clips" or distorts the sound, turning a loud crash into a flat, useless noise.

In the digital world, this is the trade-off between dynamic range (how quiet to how loud you can hear) and resolution (how clear the sound is). Standard digital recorders (ADCs) usually force you to choose: either you can hear very loud sounds but lose the fine details, or you hear the details but the loud sounds get distorted.

The "Unlimited Sensing" Solution:
This paper builds on a technology called the Unlimited Sensing Framework (USF). Think of this like a special camera that doesn't just record a photo; it records a photo, and if the image is too bright, it "wraps" the brightness around like a clock hand going past 12 back to 1. It folds the signal. Later, a computer can "unfold" it to see the original, massive signal perfectly.

The New Challenge: The "Out-of-Sync" Orchestra

The authors wanted to take this "folding" technology and use it with multiple channels (like a choir of microphones) to capture even more complex signals.

In a perfect world, all these microphones would start recording at the exact same split-second. This is called synchronization.

The Old Way: Previous methods required all microphones to be perfectly synchronized. If one microphone was even a tiny bit late (due to a slow battery, a loose wire, or just "jitter"), the math to reconstruct the sound would break. It was like trying to solve a puzzle where the pieces only fit if they are perfectly aligned; if one is slightly off, the whole picture falls apart.

The Paper's Breakthrough: The "Rubber Band" Trick

This paper introduces a new way to handle asynchronous (out-of-sync) channels. The authors realized that even if the microphones aren't perfectly aligned in time, the sound wave itself is smooth. It doesn't jump around randomly; it flows.

The Analogy: The Rubber Band
Imagine you have a long, smooth rubber band representing a sound wave.

The Old View: You need to pinch the rubber band at the exact same spot with 5 fingers at the exact same time to measure it. If your fingers are slightly off, you can't measure it.
The New View: The authors realized that even if your 5 fingers pinch the rubber band at slightly different times, the rubber band is so smooth that the distance between any two fingers is predictable.

They treated the different microphones as points on a graph (a network of dots connected by lines). Because the signal is smooth, the "distance" between the measurements from different channels is limited. This creates a smoothness constraint.

How It Works (The Magic Math)

The Folded Mess: Each microphone records a "folded" (wrapped) version of the signal. Because they are out of sync, these folded versions look different and messy.
The Graph Puzzle: The computer looks at all the measurements at one specific moment. It treats them as a network. It knows that because the signal is smooth, the difference between any two measurements cannot be too huge.
The Integer Solution: The computer solves a math puzzle (an integer optimization problem). It asks: "What is the only set of 'unfolding' numbers that makes all these measurements fit together smoothly on the graph?"
The Result: Even though the microphones were out of sync, the "smoothness" of the signal acts like a guide rail. It forces the computer to find the one unique, correct answer.

The "Clock" Analogy

Think of the different channels as clocks with slightly different times.

Old Method: You can only tell the time if all clocks show the exact same minute.
New Method: You know the clocks are all ticking at the same speed and are only a few minutes apart. Even if Clock A says 12:00 and Clock B says 12:03, you can still figure out the exact time because you know the "drift" between them is small and smooth. You don't need them to be perfectly synced; you just need to know they are close enough.

What They Proved

The authors proved mathematically that as long as the "drift" (the time difference between channels) isn't too huge compared to how fast the signal changes, the computer can exactly reconstruct the original signal.

No Perfect Sync Needed: You don't need expensive, perfect hardware synchronization.
Robustness: The system works even if the channels drift, jitter, or are designed to be irregular (like a "multi-coset" pattern).
Efficiency: They showed this works even with millions of data points and noisy (low-quality) measurements.

The Bottom Line

This paper takes a powerful signal processing trick (Unlimited Sensing) and makes it practical for real-world hardware. Real-world hardware is messy, clocks drift, and wires are imperfect. This new method says, "We don't need perfect timing. As long as the signal is smooth, we can use the relationship between the channels to fix the timing errors and recover the perfect signal."

They tested this with computer simulations using millions of data points and showed that it recovers the signal with high precision, even when the sampling rate is low and the data is messy.

Technical Summary: Asynchronous Multi-Channel USF via Modified CRT for Modulo Unfolding

Problem Statement
The Unlimited Sampling Framework (USF) addresses the traditional trade-off between dynamic range (DR) and digital resolution by employing controlled non-linearities (signal folding) in the analog domain, followed by digital reconstruction. While Multi-Channel (MC) extensions of USF have been proposed to enable reconstruction from multiple folded measurements at critical sampling rates, existing methods rely heavily on the Chinese Remainder Theorem (CRT). A critical limitation of current CRT-based approaches is their strict requirement for channel-level sampling synchronization (i.e., time-aligned or equispaced samples across channels). In practical scenarios, this assumption is frequently violated due to clock drift, jitter, imperfect synchronization, or the inherent design of asynchronous systems (e.g., multi-coset architectures with irregular sampling). When time offsets ( $T_l$ ) are non-uniform, the standard CRT structure breaks down, rendering existing recovery guarantees inapplicable and limiting the real-world deployment of MC-USF.

Methodology
This paper proposes an asynchronous MC-USF architecture that eliminates the need for strict synchronization by leveraging the inherent smoothness of bandlimited signals. The core methodology involves the following steps:

Graph-Theoretic Modeling: The authors model the multi-channel samples at a fixed time instant as a complete graph where nodes represent the folded measurements from different channels. The key insight is that the bandlimited nature of the underlying signal $g(t)$ imposes a "smoothness constraint" on this graph. Specifically, the difference between any two channel samples is bounded by the signal's maximum frequency ( $\Omega$ ), its dynamic range ( $\|g\|_{L\infty}$ ), and the maximum time offset deviation ( $t_c^n$ ) between channels.
Constrained Integer Programming: Instead of solving a standard combinatorial optimization problem (which is computationally intractable for large $N$ and $L$ ), the authors formulate the recovery of the integer residues ( $\gamma_l^n$ ) as an integer programming problem. The objective is to minimize the $\ell_\infty$ norm of the graph differences:
$\min_{\tilde{\gamma}^n \in \mathbb{Z}^L} \| B y^n + 2 B \Lambda \tilde{\gamma}^n \|_{\ell_\infty}$
subject to constraints derived from the graph smoothness condition. Here, $B$ is the incidence matrix of the complete graph, $y^n$ represents the folded measurements, and $\Lambda$ is a diagonal matrix of folding thresholds.
Theoretical Guarantees: The paper establishes Theorem 1, which proves that exact and unique recovery of the signal is possible even with non-uniform time offsets, provided:
- The folding thresholds $\lambda_l$ are related by pairwise co-prime integers ( $\lambda_l = \lambda \kappa_l$ ).
- The signal's dynamic range satisfies a specific bound involving the least common multiple (LCM) of the $\kappa_l$ values and the maximum time offset deviation.
- The graph smoothness condition holds, ensuring the solution to the integer minimization coincides with the ground truth.

Key Contributions

Asynchronous Extension of CRT: The work generalizes the CRT-based USF framework to asynchronous settings, removing the requirement for $T_l = 0$ (time alignment).
Smoothness-Driven Recovery: By viewing spatial-temporal signal lifting as inducing smoothness over a graph of sensing channels, the authors develop a reconstruction strategy robust to temporal misalignment, jitter, and drift.
Backward Compatibility: The proposed formulation remains compatible with existing multi-channel architectures, such as multi-coset sampling, enabling CRT-style reconstruction in scenarios previously considered out of reach due to irregular sampling patterns.
Efficient Algorithm: The method transforms a potentially intractable combinatorial problem into a constrained integer minimization that can be solved efficiently, even for million-scale datasets.

Experimental Results
Numerical experiments were conducted using a dual-channel multi-coset architecture with 10-bit quantization and bandlimited signals (3.2 kHz to 6.5 kHz).

Precision: The method successfully recovered samples down to the precision of the modulo-ADCs ( $\max \lambda_l / (2^B - 1)$ ), validating the theoretical bounds.
Performance: In medium oversampling regimes, the approach achieved high-precision signal estimates (Mean Squared Error $< 5 \times 10^{-4}$ ) with up to 20 times upsampling.
Robustness: Even when pushed close to the Nyquist rate where temporal correlation diminishes, the method demonstrated reasonable signal reconstruction and significant dynamic range expansion, outperforming traditional non-linear filtering or Fourier-partitioning methods in these low-correlation regimes.

Significance
The paper claims that this work significantly broadens the applicability of CRT-based recovery to practical sensing systems where clock mismatch or asynchronous acquisition is inevitable. By unifying synchronous and asynchronous MC sensing within a common USF pipeline, the authors enable high-dynamic-range, high-resolution sensing without the hardware burden of strict synchronization. The results validate that exact recovery is theoretically guaranteed and practically achievable under mild conditions, offering a robust solution for real-world multi-channel unlimited sensing implementations.