ECRT Decoding: Robust Chinese Remainder Theorem

• ECRT decoding is a robust extension of the Chinese Remainder Theorem that enables the unique reconstruction of integers or polynomials, even when residues are noisy or unordered.
• It formulates precise error bounds and dynamic range trade-offs, ensuring reliable recovery despite bounded or adversarial errors.
• Multiple algorithmic frameworks—including quotient-recovery, EM-based statistical methods, and polynomial techniques—support applications in communication, coding, and synchronization.

The Chinese Remainder Theorem (CRT) provides a unique reconstruction of integers (or polynomials) from their residue classes modulo several pairwise coprime (or, more generally, possibly non-coprime) moduli, given an appropriate dynamic range. However, classical CRT decoding is highly sensitive to residue errors: even a single erroneous residue can result in catastrophic reconstruction errors. This problem motivates the development of Error-Correcting Chinese Remainder Theorem (ECRT) decoding—extensions of CRT that provide robust recovery of one or more unknowns in the presence of residue errors, possibly unordered and with unknown correspondences. ECRT decoding appears in a range of applications including frequency estimation, phase unwrapping, synchronization, and the design of algebraic codes in Hamming, rank, and sum-rank metrics.

1. Error Models and Problem Formulation

ECRT addresses decoding under various noise models and data association ambiguities. The principal settings are as follows:

• Single-integer ECRT: Given an unknown $x \in [0, M)$ and observed residues $\widetilde r_i = x \bmod M_i + e_i$ with $|e_i| \leq \tau$, reconstruct $x$ robustly even when some or all residues are perturbed. The moduli $M_i$ may be arbitrary (possibly non-coprime), and $M = \operatorname{lcm}(M_1, \ldots, M_k)$ defines the dynamic range (Xu, 2014, Xiao et al., 2017).
• Multiple-integer ECRT (GRCRT): Recover $N$ unknown integers $x_i \in [0, D)$ from $L$ observed, unordered residue sets per modulus, possibly with unknown correspondence and each residue corrupted by bounded or stochastic noise (Xiao et al., 2019, Li et al., 2015, Xiao et al., 2017).
• Structured noise models: Some schemes address stochastic ("wrapped Gaussian") noise (Xiao et al., 2019), while others focus on bounded or adversarial error models (Xu, 2014, Xiao et al., 2017).
• Polynomial and linearized polynomial settings: For codes over $F[x]$ or noncommutative rings $\widetilde r_i = x \bmod M_i + e_i$0, residue errors are measured via degree or subspace rank (sum-rank), and ECRT decoding provides approximate polynomial reconstruction (Xiao et al., 2017, Gaborit et al., 21 May 2025).

The core challenge is to identify the maximal error bound $\widetilde r_i = x \bmod M_i + e_i$1 (in integers) or degree bound (in polynomials) under which unique or approximate decoding is possible, and to develop efficient decoding algorithms that achieve this bound.

2. Error Bound and Dynamic Range Trade-offs

Robust CRT decoding is feasible only if the residue errors are sufficiently small relative to the moduli. The fundamental results are:

• Integer case: Define, for each $\widetilde r_i = x \bmod M_i + e_i$2, $\widetilde r_i = x \bmod M_i + e_i$3. Then for arbitrary moduli,

$\widetilde r_i = x \bmod M_i + e_i$4

is the maximal uniform error bound for which unique decoding is possible; this is a sharp bound (Xu, 2014). In particular, for moduli of the form $\widetilde r_i = x \bmod M_i + e_i$5 with all $\widetilde r_i = x \bmod M_i + e_i$6, the uniform error bound is $\widetilde r_i = x \bmod M_i + e_i$7.

• Multi-integer case (unordered residues): With $\widetilde r_i = x \bmod M_i + e_i$8 unknowns, $\widetilde r_i = x \bmod M_i + e_i$9 moduli $|e_i| \leq \tau$0 with pairwise coprime $|e_i| \leq \tau$1, errors bounded by $|e_i| \leq \tau$2 can be corrected, and the dynamic range is reduced accordingly (Xiao et al., 2017).
• Two-integer case: For $|e_i| \leq \tau$3, $|e_i| \leq \tau$4 coprime, robust reconstruction of two unordered unknown integers is guaranteed for $|e_i| \leq \tau$5, and the dynamic range for unique recovery is $|e_i| \leq \tau$6, where $|e_i| \leq \tau$7 (Li et al., 2015).
• Polynomial setting: For two moduli $|e_i| \leq \tau$8, the multi-level robust CRT provides a trade-off between correctly reconstructible degree and tolerable degree of residue errors $|e_i| \leq \tau$9 (Xiao et al., 2017). At each "level" (determined by Euclidean reduction steps), degree- and error-bounds simultaneously constrain robust decoding.

These limits encode inherent information-theoretic trade-offs: as error-tolerance increases, the achievable dynamic range for unique decoding must decrease.

3. Algorithmic Frameworks for ECRT Decoding

Several ECRT decoding paradigms have been developed for both integer and polynomial settings:

a) Quotient-Recovery and Direct-CRT Methods

• Compute, for each modulus, the quotient $x$0 using observed residues and error bounds.
• Employ CRT or the extended Euclidean algorithm to reconstruct $x$1 from the quotients.
• For specialized moduli structures ($x$2), modulo-$x$3 analysis and residue "extrema" can be exploited for even more efficient decoding.

Algorithmic complexity is $x$4 for $x$5 moduli and is theoretically optimal (one CRT call suffices) (Xu, 2014, Xiao et al., 2017).

b) Statistical and EM-based ECRT Decoding

• For multi-integer settings with unordered, noisy residues, a latent variable $x$6 models residue assignments.
• The joint posterior $x$7 is maximized using a two-stage MAP estimator: first assign clusters (latent) then estimate the integer values.
• The wrapped-Gaussian mixture EM algorithm alternates E-steps (clustering residues modulo a common $x$8) and M-steps (estimating centroids via closed-form minimization) (Xiao et al., 2019).
• Algorithmic steps are $x$9 per iteration, typically requiring fewer than 10 iterations for convergence. Residue error correction is achieved by voting among subsets of moduli to correct misclustered residues.

This approach achieves near-optimal performance even in high-noise regimes (SNR down to –20 dB) and supports robust reconstruction for $M_i$0 up to 10 (Xiao et al., 2019).

c) Symmetric-Polynomial and Newton–Vieta Approaches

• For unordered multi-integer recovery, symmetric polynomial decoding reconstructs integer roots from power sums computed over the moduli.
• Root-finding (factorization) over the integers is employed after lifting Newton sums via CRT, providing polynomial-time solutions for moderate $M_i$1 (Xiao et al., 2017).

d) Polynomial and Linearized Polynomial ECRT

• For polynomial codes with non-coprime moduli, multi-level robust CRT uses extended Euclidean reductions to expose "folding polynomials" and reconstruct the message up to a defined error degree (Xiao et al., 2017).
• For linearized polynomial codes (qCRT), decoding uses noncommutative CRT over $M_i$2, with support subspace recovery and syndrome-linear systems yielding explicit message reconstruction (Gaborit et al., 21 May 2025).
• These quantum-CRT decoders are especially relevant for rank- and sum-rank error models in network coding.

4. ECRT Decoding for Multiple Numbers and Unlabeled Residues

In scenarios where multiple unknowns are to be reconstructed from noisy, unlabeled residues:

• The problem is mapped into a clustering task of noisy points on the residue circle modulo a coarse modulus $M_i$3.
• Algorithmic steps include computation of circular gaps among residues, identification of the maximal "opposite" gap to split clusters, assignment of residues to clusters, estimation of the common remainder in each cluster, and reduction to a smaller CRT (mod $M_i$4) (Li et al., 2015, Xiao et al., 2017).
• For $M_i$5 unknowns, clustering can be accomplished via sorting and a single cyclic shift ($M_i$6 per sample in the EM formulation (Xiao et al., 2019)).
• Residue-error-correcting codes (generalized Reed-Solomon, in effect) are used to tolerate a constant fraction of clustering errors via majority voting and subset selection techniques (Xiao et al., 2019).

This framework allows robust decoding even when clustering errors affect up to a constant fraction of residue assignments.

5. Complexity, Performance, and Theoretical Guarantees

ECRT decoding achieves polynomial complexity in all major models:

• Integer quotient-recovery/CRT: $M_i$7 (Xu, 2014).
• Multi-integer EM/mixture-model: $M_i$8 per iteration, with $M_i$9 matching steps for clustering (Xiao et al., 2019).
• Multi-integer symmetric polynomial decoding: $M = \operatorname{lcm}(M_1, \ldots, M_k)$0 to $M = \operatorname{lcm}(M_1, \ldots, M_k)$1 for root finding (Xiao et al., 2017).
• Polynomial ECRT decoding: $M = \operatorname{lcm}(M_1, \ldots, M_k)$2 per two-modulus step, cascading to $M = \operatorname{lcm}(M_1, \ldots, M_k)$3 for $M = \operatorname{lcm}(M_1, \ldots, M_k)$4 moduli, with $M = \operatorname{lcm}(M_1, \ldots, M_k)$5 the cost of degree-$M = \operatorname{lcm}(M_1, \ldots, M_k)$6 multiplication (Xiao et al., 2017).
• Linearized polynomial (qCRT) decoding: dominated by matrix operations and right-Euclidean reductions, typically $M = \operatorname{lcm}(M_1, \ldots, M_k)$7 over $M = \operatorname{lcm}(M_1, \ldots, M_k)$8 (Gaborit et al., 21 May 2025).

Error bounds are provably optimal in the sense that for $M = \operatorname{lcm}(M_1, \ldots, M_k)$9, no decoder can guarantee unique recovery (Xu, 2014). For moderate system sizes, statistical ECRT decoders maintain success rates $N$0 in adversarial or heavy-noise scenarios (Xiao et al., 2019).

6. ECRT in Polynomial and Linearized Polynomial Codes

Error-correcting CRT decoding naturally extends beyond integer settings:

• Polynomial CRT: Multi-level robust CRT for polynomials provides explicit bounds on reconstructible degree and error tolerance, with the code minimum distance governing the number and degree of correctable errors. Closed-form decoders retrieve all but the top-$N$1 coefficients exactly, with complexity $N$2 per modular pair (Xiao et al., 2017).
• Linearized Polynomial CRT (qCRT): qCRT codes over $N$3 enable construction of rank and sum-rank metric codes. Decoding employs noncommutative CRT lifting, high-degree part support identification, syndrome-linear system solving, and right-division. Decoding radius achieves correction beyond $N$4 for large $N$5 (Gaborit et al., 21 May 2025).

This unifies ECRT decoding with Gabidulin-style rank-metric decoding and enhances error correction in algebraic coding frameworks.

7. Applications and Implications

ECRT decoding strategies are foundational in:

• Frequency and phase estimation under heavy undersampling or aliasing (Li et al., 2015, Xiao et al., 2019).
• Distributed data storage, sensor fusion, and synchronization where unordered or corrupted residue sets appear (Xiao et al., 2017).
• Code design in Hamming, rank, and sum-rank metrics, including emergent qCRT codes (Xiao et al., 2017, Gaborit et al., 21 May 2025).
• Ambiguity resolution in estimation problems demanding robust integer parameter recovery (Xiao et al., 2019).

By quantifying and attaining fundamental robustness limits, ECRT decoding informs both theory and system design in communications, signal processing, and coding.

---

References:

• "Statistical Robust Chinese Remainder Theorem for Multiple Numbers" (Xiao et al., 2019)
• "A Robust Generalized Chinese Remainder Theorem for Two Integers" (Li et al., 2015)
• "Robustness in Chinese Remainder Theorem" (Xiao et al., 2017)
• "On Solving a Generalized Chinese Remainder Theorem in the Presence of Remainder Errors" (Xu, 2014)
• "Robust Polynomial Reconstruction via Chinese Remainder Theorem in the Presence of Small Degree Residue Errors" (Xiao et al., 2017)
• "Linearized Polynomial Chinese remainder codes" (Gaborit et al., 21 May 2025)