1 Introduction

\footrulewidth

0pt \headrulewidth0.4pt \lhead[\fancyplain0]\fancyplain \rhead[\fancyplain]\fancyplain0 \lfoot \rfoot \chead \cfoot

Bounds on List Decoding Gabidulin Codes

Antonia Wachter-Zeh1 antonia.wachter@uni-ulm.de
Institute of Communications Engineering, Ulm University, Ulm, Germany and Institut de Recherche Mathématique de Rennes, Université de Rennes 1, Rennes, France

Abstract. An open question about Gabidulin codes is whether polynomial-time list decoding beyond half the minimum distance is possible or not. In this contribution, we give a lower and an upper bound on the list size, i.e., the number of codewords in a ball around the received word. The lower bound shows that if the radius of this ball is greater than the Johnson radius, this list size can be exponential and hence, no polynomial-time list decoding is possible. The upper bound on the list size uses subspace properties.

1 Introduction

Gabidulin codes [1] can be seen as the analogs of Reed–Solomon (RS) codes in rank metric. There are several efficient decoding algorithms up to half the minimum rank distance. However, in contrast to RS codes, there is no polynomial-time decoding algorithm beyond half the minimum distance. For RS codes, it can be shown that the number of codewords in a ball around any received word is always polynomial in the length when the radius of the ball is at most the Johnson radius. The Guruswami–Sudan algorithm provides an efficient polynomial-time list decoding algorithm of RS codes up to the Johnson radius.

For Gabidulin codes, there is no polynomial-time list decoding algorithm; it is not even known, whether such an algorithm can exist or not. An exponential lower bound on the number of codewords in a ball of radius around the received word would prohibit polynomial-time list decoding since the list size can be exponential, whereas a polynomial upper bound would show that it might be possible. Faure [2] and Augot and Loidreau [3] made first investigations of this problem.

In this paper, we provide a lower and an upper bound on the list size. The lower bound shows that the list size can be exponential in the length when the radius is at least the Johnson radius and therefore in this region, no polynomial-time list decoding is possible. The upper bound uses the properties of subspaces and gives a good estimate of the number of codewords in such a ball, but is exponential in the length and therefore does not provide an answer to polynomial-time list decodability in the region up to the Johnson bound.

2 Preliminaries

2.1 Definitions and Notations

Let be a power of a prime, let denote the finite field of order and let be the extension field of degree over . We denote for any integer , then a linearized polynomial, introduced by Ore [4], over has the form , with . If the coefficient , we call the q-degree of . For all and , the following holds: . The (usual) addition and the non-commutative composition (also called symbolic product) convert the set of linearized polynomials into a non-commutative ring with the identity element . In the following, all polynomials are linearized polynomials.

Given a basis of over , there exists a one-to-one mapping for each vector on a matrix . Let denote the (usual) rank of over and let denote the row space of over . The kernel of a matrix is denoted by and the image by . For an matrix, if , then . Throughout this paper, we use the notation as vector or matrix equivalently, whatever is more convenient. The minimum rank distance of a code is defined by

A Gabidulin code can be defined by the evaluation of degree-restricted linearized polynomials as follows.

Definition 1 (Gabidulin Code [1]).

A linear Gabidulin code of length and dimension over for is the set of all codewords, which are the evaluation of a -degree restricted linearized polynomial :

where the fixed elements are linearly independent over .

Gabidulin codes are Maximum Rank Distance (MRD) codes, i.e., they fulfill the rank metric Singleton bound with equality and [1].

The number of -dimensional subspaces of an -dimensional vector space over is the Gaussian binomial, calculated by

with the upper and lower bounds .

Moreover, denotes a ball of radius in rank metric around a word . The volume of is independent of its center and is simply the number of matrices of rank less than or equal to .

2.2 Problem Statement

Problem 1 (Maximum List Size).

Let the Gabidulin code over with and be given. Let . Find a lower and upper bound on the maximum number of codewords in the ball of rank radius around . Hence, find a bound on

For an upper bound, we have to show that the bound holds for any received word , whereas for a lower bound it is sufficient to show that there exists one for which this bound on the list size is valid.

Let with denote the list of all codewords in the ball of radius around , i.e., and , for .

3 A Lower Bound on the List Size

Faure showed with a probabilistic approach in [2] that the maximum list size of Gabidulin codes with is exponential in for . Our bound slightly improves this value and uses a different proving strategy, based on evaluating linearized polynomials. This approach is inspired by Justesen and Hoholdt’s [5] and Ben-Sasson, Kopparty, and Radhakrishna’s [6] approaches for bounding the list size of RS codes.

Theorem 1 (Lower Bound on the List Size).

Let the Gabidulin code over with and be given. Let . Then, there exists a word such that

(1)

and for the special case of : .

Proof.

Since , also holds. Consider all monic linearized polynomials of -degree exactly with a root space of dimension , where all roots are in . There are exactly (see e.g. [7, Theorem 11.52]) such polynomials. Now, let us consider a subset of these polynomials, denoted by : all polynomials where the -monomials of -degree greater than or equal to have the same coefficients. Due to Dirichlet’s principle there exist coefficients such that the number of such polynomials is

since there are possibilities to choose the highest coefficients of a monic linearized polynomial over .

Note that the difference between any two polynomials in is a linearized polynomial of -degree strictly less than and therefore the evaluation polynomial of a codeword of . Let be the evaluation of at a basis of over :

Further, let also , then has -degree less than . Let denote the evaluation of at . Then, is the evaluation of , whose root space has dimension and all roots are in . Thus, and . Therefore, for any , the evaluation of is a codeword from and has rank distance from . This provides the following lower bound on the maximum list size:

and for the special case follows. ∎

This lower bound is valid for any , but we want to know, which is the smallest value for such that this expression is exponential in . For , we can rewrite (1) by

where the first part is exponential in for any . The second exponent is positive for

(2)

Therefore, our lower bound (1) shows that the maximum list size is exponential in for . For , the value gives exactly the Johnson radius for Hamming metric.

This reveals a difference between the known limits to list decoding of Gabidulin and RS codes. For RS codes, polynomial-time list decoding up to the Johnson radius is guaranteed by the Guruswami–Sudan algorithm. However, it is not proven that the Johnson radius is tight for RS codes, i.e., it is not known if the list size is polynomial between the Johnson radius and the known exponential lower bounds (see e.g. [5, 6]). For Gabidulin codes, we have shown that the maximum list size is exponential for , which is asymptotically equal to the Hamming metric Johnson radius.

Example 1.

For the Gabidulin code with , the Bounded Minimum Distance decoding radius is , the lower bound by Faure (equivalent to the Hamming metric Johnson radius) is and (2) with gives . This means for this code of rate , no polynomial time list-decoding beyond is possible.

4 An Upper Bound on the List Size

The following lemma shows that the row spaces of and , , , have no -dimensional subspace in common.

Lemma 1.

Let and let . Let , for , be codewords of the Gabidulin code with minimum rank distance and let hold for all . Let and , . Then, the row spaces of and have no subspace of dimension at least in common, for .

Proof.

Assume, there exist and , with , , , such that their row spaces contain the same subspace of dimension at least . Then,

which is a contradiction to . ∎

This means in particular, if , they have no subspace of dimension at least in common. Based on this lemma, we obtain the following upper bound on the maximum list size.

Theorem 2 (Upper Bound on the List Size).

Let the Gabidulin code over with and be given. Let . Then, for any word and hence, for the maximum list size, the following holds

(3)
(4)
Proof.

We consider all words in with , therefore these words can be seen as matrices in an -dimensional space. For any , where , there are subspaces of dimension . Let be any fixed word in and all codewords in have . Each , for , of rank contains subspaces of dimension .

Due to Lemma 1, different have no -dimensional subspace in common and therefore there are at most possible codewords in rank distance to the word . We sum this up for from up to and we obtain (3).

With the bounds for the Gaussian binomial and since , the upper bound from (4) follows. ∎

Note that for the special case and even minimum distance , the upper bound from (3) is the bound from [3, Equation (4)], which is

Thus, we have proved an upper bound on the maximum list size of a Gabidulin code. Unfortunately, this upper bound is exponential in for any and therefore does not provide any conclusion if polynomial-time list decoding is possible or not in the region up to the Johnson bound.

Footnotes

  1. This work was supported by the German Research Council ”Deutsche Forschungsgemeinschaft” (DFG) under Grant No. Bo 867/21-2.

References

  1. E. M. Gabidulin, “Theory of Codes with Maximum Rank Distance,” Probl. Peredachi Inf., vol. 21, no. 1, pp. 3–16, 1985.
  2. C. Faure, “Average Number of Gabidulin Codewords within a Sphere,” in Tenth International Workshop on Algebraic and Combinatorial Coding Theory, Sept. 2006, pp. 86–89.
  3. D. Augot and P. Loidreau, “Johnson-like bounds for the rank metric,” preprint, 2011.
  4. O. Ore, “On a Special Class of Polynomials,” Transactions of the American Mathematical Society, vol. 35, pp. 559–584, 1933.
  5. J. Justesen and T. Hoholdt, “Bounds on list decoding of MDS codes,” Information Theory, IEEE Transactions on, vol. 47, no. 4, pp. 1604–1609, May 2001.
  6. E. Ben-Sasson, S. Kopparty, and J. Radhakrishnan, “Subspace Polynomials and Limits to List Decoding of Reed–Solomon Codes,” Information Theory, IEEE Transactions on, vol. 56, no. 1, pp. 113–120, Jan. 2010.
  7. E. R. Berlekamp, Algebraic Coding Theory, Revised Edition (M-6) (No. M-6), revised ed.   Aegean Park Pr, June 1984.
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
""
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
   
Add comment
Cancel
Loading ...
165854
This is a comment super asjknd jkasnjk adsnkj
Upvote
Downvote
""
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters
Submit
Cancel

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test
Test description