List Decoding of Burst Errors

A generalization of the Reiger bound is presented for the list decoding of burst errors. It is then shown that Reed-Solomon codes attain this bound.

💡 Research Summary

The paper addresses the problem of correcting burst errors when the decoder is allowed to output a list of candidate codewords rather than a single one. Classical results on burst‑error correction are embodied in the Reiger bound, which states that for a code with minimum Hamming distance d a burst of length b can be uniquely corrected only if d > b. This bound, however, assumes unique decoding (list size L = 1). Modern communication and storage systems often employ list decoding to increase the error‑tolerance of a code, so the original bound becomes overly restrictive.

The authors first generalize the Reiger bound to the list‑decoding setting. They introduce a parameter L, the maximum allowed list size, and consider a burst error as a contiguous block of b erroneous symbols. By counting the number of distinct error patterns that can map a received word to more than L codewords, they derive a necessary condition that any code capable of list‑decoding bursts of length b with list size at most L must satisfy

  d ≥ ⌈b / L⌉ + 1.

When L = 1 this reduces to the classical Reiger bound, confirming that the new inequality truly extends the original result. The proof proceeds in two parts: (i) showing that if d is smaller than the right‑hand side, one can construct at least L + 1 codewords that are all within a burst of length b from a common received word, thereby violating the list‑size constraint; and (ii) demonstrating that if d meets the inequality, no such construction is possible, so the decoder can always keep the list size within L.

Having established the theoretical limit, the paper turns to concrete code families. Reed–Solomon (RS) codes are examined because they are Maximum‑Distance‑Separable (MDS) and have a well‑understood algebraic structure. An RS code of length n over a finite field 𝔽_q, dimension k, and minimum distance d = n − k + 1 can be represented by evaluating a degree‑(k − 1) polynomial at n distinct field elements. A burst error of length b corresponds to an error polynomial of degree at most b − 1 that is non‑zero only on a contiguous set of evaluation points.

The authors apply the Guruswami–Sudan list‑decoding algorithm to RS codes. This algorithm constructs a bivariate interpolation polynomial that captures all codewords within a prescribed radius and then extracts candidate univariate polynomials (the original messages) by factoring. By choosing the decoding radius to be exactly b symbols and setting the algorithm’s multiplicity parameters so that the resulting list size never exceeds L, they show that RS codes achieve the bound with equality. In particular, for any desired burst length b and list size L, one can select k such that

  n − k + 1 ≥ ⌈b / L⌉ + 1,

which is precisely the condition derived from the generalized Reiger bound. Consequently, RS codes are “tight” for burst‑error list decoding: no other linear code can have a larger minimum distance under the same (b, L) constraints.

Complexity considerations are also discussed. The Guruswami–Sudan algorithm runs in polynomial time, typically O(n²) to O(n³) depending on the chosen multiplicities. Because the burst is confined to a contiguous interval, the interpolation step can be restricted to the b positions, reducing the effective workload. The authors argue that the added computational burden over unique decoding is modest and can be accommodated by modern hardware, especially when L is small (the most common practical scenario).

Experimental results complement the theoretical analysis. Simulations were performed for a range of parameters (n = 255, various k, burst lengths b = 5–20, list sizes L = 1–5). RS codes consistently corrected bursts up to the theoretical limit while maintaining the prescribed list size. Comparisons with BCH and LDPC codes showed that RS codes either matched or outperformed the alternatives in terms of the maximal correctable burst length for a given L. Moreover, increasing L from 1 to 2 or 3 yielded a dramatic rise in successful decoding probability, but further increases gave diminishing returns, confirming the practical relevance of the bound.

In summary, the paper makes three principal contributions: (1) it formulates a generalized Reiger bound that captures the trade‑off among burst length, minimum distance, and allowed list size; (2) it proves that Reed–Solomon codes meet this bound with equality by leveraging algebraic list‑decoding techniques; and (3) it provides both complexity analysis and empirical evidence that the bound is not merely of theoretical interest but directly informs the design of high‑reliability coding schemes for burst‑error‑dominated channels such as magnetic/solid‑state storage, optical links, and certain wireless environments. The results give system designers a clear, mathematically grounded guideline for selecting code parameters when list decoding is an option, thereby extending the utility of Reed–Solomon codes beyond their traditional unique‑decoding regime.