6
AAECC 8, 415420 (1997) On New Perfect Binary Nonlinear Codes A. C. Lobstein1, V. A. Zinoviev2 1 Centre National de la Recherche Scientifique, Ecole Nationale Supe´rieure des Te´le´communications, 46 rue Barrault, F-75634 Paris ce´dex 13, France (e-mail: lobstein@inf.enst.fr) 2 Institute for Problems of Information Transmission, Russian Academy of Sciences, Bol’shoi Karetnyi, 19, GSP-4, Moscow, 101447, Russia (e-mail: zinov@ippi.ac.msk.su) Dedicated to Aimo ¹ieta ( va ( inen on the occasion of his 60th birthday Received: November 26, 1996; revised version: March 14, 1997 Abstract. We present a new construction of binary nonlinear perfect codes with minimum distance 3 and lowerbound the number of nonequivalent such codes. Keywords: Generalized concatenated codes, Binary nonlinear perfect codes. 1 Introduction A block code C over field F q (whose elements we denote 0, 1, . . . , q!1), with length n, size qk (k71), and minimum distance d is denoted by (n, k, d ). If C is linear, it is denoted by [n, k, d ]. For any a3Fn q , the set C#a"Mc#a : c3CN is a coset of C.A q-ary code C with minimum distance d"2e#1, length n, and size qk is perfect if and only if : qk · e + i/0 ( q!1)i A n i B "qn. It is known ([8], [9], [13], [14]) that a nontrivial perfect code C over F q necessarily has the same parameters as either a Hamming code (n"( qm!1)/ ( q!1), n!m, 3) or the binary [23, 12, 7] or ternary [11, 6, 5] Golay codes. Two binary codes C 1 and C 2 are called equivalent if one can be obtained from the other by permutations of positions and translations. The problem of evaluating the number of nonequivalent binary perfect Hamming codes is still open (see, among numerous papers, [10], [4], [5], [3], [7], [1]). Here, using the idea of general- ized concatenation, we describe a general construction of perfect nonlinear codes Correspondence to: A. C. Lobstein

On New Perfect Binary Nonlinear Codes

Embed Size (px)

Citation preview

AAECC 8, 415—420 (1997)

On New Perfect Binary Nonlinear Codes

A. C. Lobstein1, V. A. Zinoviev2

1Centre National de la Recherche Scientifique, Ecole Nationale Superieure desTelecommunications, 46 rue Barrault, F-75634 Paris cedex 13, France(e-mail: [email protected])2Institute for Problems of Information Transmission, Russian Academy of Sciences,Bol’shoi Karetnyi, 19, GSP-4, Moscow, 101447, Russia (e-mail: [email protected])

Dedicated to Aimo ¹ieta( va( inen on the occasion of his 60th birthday

Received: November 26, 1996; revised version: March 14, 1997

Abstract. We present a new construction of binary nonlinear perfect codes withminimum distance 3 and lowerbound the number of nonequivalent such codes.

Keywords: Generalized concatenated codes, Binary nonlinear perfect codes.

1 Introduction

A block code C over field Fq

(whose elements we denote 0, 1, . . . , q!1), withlength n, size qk (k71), and minimum distance d is denoted by (n, k, d ). If C islinear, it is denoted by [n, k, d]. For any a3Fn

q, the set C#a"Mc#a : c3CN is

a coset of C. A q-ary code C with minimum distance d"2e#1, length n, and sizeqk is perfect if and only if :

qk ·e+i/0

(q!1)iAn

i B"qn.

It is known ([8], [9], [13], [14]) that a nontrivial perfect code C over Fq

necessarily has the same parameters as either a Hamming code (n"(qm!1)/(q!1), n!m, 3) or the binary [23, 12, 7] or ternary [11, 6, 5] Golay codes. Twobinary codes C

1and C

2are called equivalent if one can be obtained from the other

by permutations of positions and translations. The problem of evaluating thenumber of nonequivalent binary perfect Hamming codes is still open (see, amongnumerous papers, [10], [4], [5], [3], [7], [1]). Here, using the idea of general-ized concatenation, we describe a general construction of perfect nonlinear codes

Correspondence to: A. C. Lobstein

and then bound the number of nonequivalent codes obtained through thisconstruction.

2 Perfect Codes by Concatenation

Our construction is inspired by generalized concatenated codes, which were intro-duced by Zinoviev [11] and are also the basis of the construction in [12]. LetnA"2u and n

B"2m. Let A

1be a binary extended perfect code (n

A, n

A!1!u, 4),

A2

be an nB-ary (n

A, n

A!1, 2) code, and A

3be a q

3-ary (n

A, n

A, 1) code, where

q3"2nB~1~m. Let B be a binary (n

B, n

B, 1) code which can be represented as an

union,

B" Zi/0,1

Bi,

where, for i"0, 1, Biis an (n

B, n

B!1, 2) code. Consider for each B

ithe partition

Bi"

nB~1Zj/0

Bi, j

, (2.1)

where, for j"0, 1, . . . , nB!1, the B

i,j’s are disjoint binary (n

B, n

B!1!m, 4)

codes. The Bi,j

’s partition B, hence any b3B belongs to exactly one Bi,j

and, ifb has index k in B

i,j, we see that (i, j, k)3M0, 1N]M0, . . . , n

B!1N]M0, . . . ,

2nB~1~m!1N characterizes vector b; we note: b"b (i, j, k).Consider, for l"1, 2, 3, a codeword a(l)"(a(l)

1, . . . , a(l)

nA)3A

l. For s"1, . . . , n

A,

the triple (a(1)s

, a(2)s

, a(3)s

) designates a codeword b"b (a(1)s

, a(2)s

, a(3)s

)3B. Definea new code

C"M (b(a(1)1

, a(2)1

, a(3)1

) D . . . Db (a(1)nA

, a(2)nA

, a(3)nA

)) : a(l)3Al, 16l63N.

Codes Ai, B and C are called, respectively, outer, inner and concatenated codes.

Theorem 2.1 C is a binary extended perfect (n"2m`u, n!(m#u)!1, 4) code.

Special cases of this basic construction were obtained in [12] and [6]. In turn thisconstruction is a special case of a construction by Phelps [5], described ina different way, without mentioning concatenation construction.

3 Modified Concatenation Constructions

First we note that the extended perfect binary code which we denote by B0,0

hasexactly n

Bcosets of weight 1: they are all codes B

1, j, j"0, . . . , n

B!1. Denote by

v(h) the binary vector of length nB

whose only nonzero element is on its h-thposition: any subcode B

1,jis uniquely defined by some vector v(h). Without loss of

generality, we can assume that h"j. Further, let the code B0, j

with j'0 bedefined by the vector w(j) with two nonzero positions, 0 and j. Therefore, partition(2.1) can be expressed as follows:

B0"B

0,0ZG

nB~1Zj/1

(B0,0

#w(j))H ; B1"

nB~1Zj/0

(B0,0

#v(j)).

416 A. C. Lobstein, V. A. Zinoviev

Denote by SB

the symmetric group of order nB! (i.e., the permutations of

M0, 1, . . . , nB!1N). Suppose that we numbered all the elements of S

B. Denote by

qi3S

Bthe permutation of ‘‘even’’ subcodes and by n

j3S

Bthe permutation of ‘‘odd’’

subcodes (i, j"1, 2, . . . , nB! ). It is clear that any permutation n

jof subcodes

B1, i

can be expressed as a permutation of vectors v(i) defining the cosets. Similarly,any permutation q

iof subcodes B

0,j( j71) can be expressed as a permutation of

vectors w(j).Let G"M1, . . . , n

AN be the set of coordinate positions of A

2. Denote by E

i"

(Ei,1

, . . . , Ei,k

) the i-th partition of G for even subcodes and byO

j"(O

j,1, . . . , O

j,s) the j-th partition of G for odd subcodes. Two partitions of

G are said to be equivalent if they differ only by the order of their elements.It is known (see, e.g., reference in [13]) that any binary (extended) perfect code is

self-complementary. Assume that for any pair of complementary vectors (a, a6 )3A1,

the vector a has 0 on its first position. For any pair (a, a6 )3A1

we choose twoarbitrary partitions:

E (a, a6 )"E (a)"(E1(a), . . . , E

k(a)) and O(a, a6 )"O (a)"(O

1(a), . . . , O

s(a))

and two tuples of arbitrary integers (i1(a), . . . , i

k(a)), ( j

1(a), . . . , j

s(a)) such that

16i1(a)(· · ·(i

k(a)6n

B! and 16j

1(a)(· · ·(j

s(a)6n

B!.

Now we define a code

C(ME(a(1)); i1(a(1) ), . . . , i

k(a(1)); O (a(1)); j

1(a(1)), . . . , j

s(a(1) ) : a(1)3A

1N)

"M(b(a(1)1

, a(2)1

, a(3)1

) D · · · Db (a(1)nA

, a(2)nA

, a(3)nA

)) : a(l)3Al, 16l63N, (3.2)

where the encoding function b( . , . , . ) now is more complicated: for j"1, . . . , nA,

b(a(1)j , a(2)

j , a(3)j )"G

b (a(1)j , q

it (a(1)) (a(2)j ), a(3)

j ) if a(1)j "0 and j3E

t(a(1) ),

b (a(1)j , n

je (a(1)) (a(2)j ), a(3)

j ) if a(1)j "1 and j3O

e(a(1) ).

(3.3)

We see that if all partitions for all pairs (a, a6 )3A1

are trivial, i.e., ifE(a)"O(a)"(G) and q

i1(a) and nj1(a) are identical permutations, then it gives

exactly the construction described in the previous section. Obviously the para-meters of this new code are the same as C, since the permutations of the symbols ofthe alphabet of A

2do not change the distances between its codewords. So we have

the following statement.

Theorem 3.1 For any partitions E(a), O(a) of G"M1, 2, . . . , nAN and for any

integers i1(a), . . . , i

k(a), j

1(a), . . . , j

s(a), where 16i

1(a)( · · ·(i

k(a)6n

B!,

16j1(a)( · · ·(j

s(a)6n

B! and a runs over all codewords of A

1, the resulting code

C"C (ME (a); i1(a), . . . , i

k(a); O (a); j

1(a), . . . , j

s(a) : a3A

1N)

is a binary extended perfect code with d"4.

Denote by C(a) a subcode of C obtained by fixing a pair (a, a6 ) of A1:

C" Z(a, a6 )3A

1

C(a).

On New Perfect Binary Nonlinear Codes 417

Each C (a) is defined by two partitions E (a) and O(a) and two tuples of integersi1(a), . . . , i

k(a) and j

1(a), . . . , j

s(a). Our aim is to count the number M(C) of

mutually nonequivalent codes C. In order to do it, first we want to count thenumber M (a) of mutually nonequivalent codes C(a). Now we need a simpleobservation, which we formulate as a lemma. Denote by ¹q (respectively, ¹n) a setof permutations of even subcodes B

0,j(respectively, odd subcodes B

1,j) induced by

all possible translations in FnB2

.

Lemma 3.1 ¸et ¹rdenote a regular permutation representation of the abelian group

type (2] · · · ]2) of order nB. ¹hen ¹q"¹n"¹

r.

From now on, we assume that nB"4. Denote by ¸ the set of the following

forbidden partitions of G"M1, 2, . . . , nAN: E"(E

1, . . . , E

k)3¸ if and only if the

following two properties are satisfied: (i) k6nB"4; (ii) all the numbers

DE1D, . . . , DE

kD are even.

As a criterion for nonequivalence of codes, we use the following simpleresult.

Lemma 3.2 ¸et C1

and C2

be two binary codes with the same parameters, anddistance d(C

1)"d(C

2)73. Suppose that for any two positions i, j, iOj, there exist

codewords a1, a

23C

1, b

1, b

23C

2such that a

1"b

1, a

2"b

2#v(i)#v(j), and

a1, i

Oa1,j

. ¹hen C1

and C2

are not equivalent.

Proof. We cannot obtain C2

from C1

using any combination of translations andpermutations of positions. The only possibility is if they coincide. But this contra-dicts d(C

1)'2. K

Note that code C (a) is defined by two partitions and two permutations, foreven and odd subcodes. But when we consider the equivalence of two codes C(a)and C(b), it is clear that only one type of partitions and permutations is necessary(for example, even partitions and even permutations). Now as a direct corollary ofthe previous two lemmas, we have

Lemma 3.3 ¹wo codes C(a) and C(b) defined by the partitions E (a)"(Ei1, . . . , E

ik)

and E@ (b)"(Ej1, . . . , E

js) and the permutations q

i1, . . . , q

ikand q

j1, . . . , q

js, respect-

ively, are equivalent if and only if either

(i) E"E @ and (i1, . . . , i

k)"( j

1, . . . , j

s), or

(ii) E, E @3¸ and all permutations qi1, . . . , q

ik, q

j1, . . . , q

js3¹

r.

Denote by Na the set of all mutually nonequivalent codes C(a).

Lemma 3.4. ¹he cardinality of Na is at least

M (a)7AnA#23

23 B!3+i/0A(n

A!2)/2

i BA4

i B. (3.4)

Proof. If we have no limitations for the partitions and the permutations we use (seeLemma 3.3), then the number of all possible codes is equal to (nA`23

23). So we have to

count the number of forbidden codes (equivalent to each other). The cardinality of

418 A. C. Lobstein, V. A. Zinoviev

¸ is easy to count:

D¸ D"3+i/0A(n

A!2)/2

i B.Now we have to take into account the number of possible ways to choose thepermutations from the set ¹

rfor the partition consisting of i subsets. Here

¹rconsists of 4 permutations, because n

B"4. So for the set with i subsets we have

(4i) such choices, yielding (3.4). K

Now using M(a) of the nonequivalent codes Ci(a), we build the nonequivalent

extended perfect codes C. Such a code consists of 2nA~2~u codes Ci(a). Note that if

two codes Ci(a) and C

j(b) are nonequivalent, then for any binary vector c of length

nA, C

i(a) and C

j(b#c) are also nonequivalent. Let us call shortly the code

Cj(b#c), which is equivalent to C

j(b) (as well as to C

j(c)), a code of type C

j(b). We

define now the following class of mutually nonequivalent codes NC. This class isa union of sets NC

s, s"1, 2, . . . , where NC

sconsists of the codes of s-th type. For

simplicity, let M(s)"DNCsD and !"2nA~2~u. ¹he codes of s-th type NC

sconsist of

t1codes of type C

1(a

1), t

2codes of type C

2(a

2), and so on . . . , t

s"!!t

1! · · · !

ts~1

codes of type Cs(a

s), where t

1, . . . , t

s~1are arbitrary integers the sum of which is

less than !, C1(a

1), . . . , C

s(a

s)3Na, and all C

i’s are distinct. ¹hen we have

M (s)"AM(a)

s BA!!1

s!1 B , where s6minMM(a), !N.

For the case M(a)7!/2, we clearly have the following lower bound.

Theorem 3.2 ¸et the value M(a) in ¸emma 3.4 be such that M (a)72nA~3~u, wherenA"2u74. ¹hen our construction provides at least M(C) binary extended perfect

nonlinear codes of length n"4nA, where

M (C)'22c (n~-0'2n) and c'0.25.

For nB"n

A"4, the outer code A

1consists of two codewords at distance 4.

Therefore any code C (a) is an extended perfect code of length 16. So Lemma 3.4gives the following lower bound on the number of nonequivalent extended perfectnonlinear codes of length 16:

M(C)7A4#24!1

24!1 B!(24#6)"17520.

Note that Phelps [4] showed that there are at least 31021 nonequivalent perfectcodes of length 15. It means that our estimation is quite rough.

For nA"8, we have according to Lemma 3.4 that M(a)77 888 690, which

gives the following lower bound for the number of nonequivalent binary extendedperfect codes of length n"32: M (C)72167.98.

Similarly, for n"64, we have M(a)737 711 260 695, and M(C)7227208.49.It is clear that for arbitrary n

B'4, the value M(a) is approximately equal to

(nB!`nAnA

). Then in the best case, when the values nA

and nB

are growing so thatnB!Kn

A, our estimation gives only the following number of nonequivalent ex-

tended perfect codes of length n"nAnB: M (C)722cn where c'1/n

B. This means

On New Perfect Binary Nonlinear Codes 419

again that our estimation is too rough. Note that in the recent paper [1] thefollowing lower bound for the number of different (not nonequivalent) perfectcodes of length n was obtained:

22n`12 ~-0'(n`1)

· 62 n`5

4 ~-0' (n`1)· (1!o (1)).

References

1. Avgustinovich, S., Solov’eva, F.: Construction of perfect binary codes by the sequentialtranslations of the i-components. Proc. 5th Internat. Workshop on Algebraic and Combina-torial Coding Theory, pp. 9—14, Sozopol, Bulgaria (1996)

2. Etzion, T., Vardy, A.: Perfect binary codes: Constructions, properties, and enumeration.IEEE Trans. Inf. Theory 40, 754—763 (1994)

3. Mollard, M.: A generalized parity function and its use in the construction of perfect codes.SIAM J. Algebraic Discr. Meth. 7, 113—115 (1986)

4. Phelps, K.: A combinatorial construction of perfect codes. SIAM J. Algebraic Discr. Meth. 4,398—403 (1983)

5. Phelps, K.: A general product construction for error correcting codes. SIAM J. AlgebraicDiscr. Meth. 5, 224—228 (1984)

6. Solov’eva, F.: A class of binary perfect codes generated by q-ary codes. Methodi Diskr.Analiza 48, 70—72 (1989) [in Russian]

7. Solov’eva, F.: A combinatorial construction of perfect binary codes. Proc. 4th Internat.Workshop on Algebraic and Combinatorial Coding Theory, pp. 171—174, Novgorod, Russia(1994)

8. Tietavainen, A.: On the nonexistence of perfect codes over finite fields. SIAM J. Appl. Math.24, 88—96 (1973)

9. Tietavainen, A., Perko, A.: There are no unknown perfect binary codes. Ann. Univ. Turku,Ser. A. I, 148, 3—10 (1971)

10. Vasiliev, J.: On nongroup close-packed codes. Problemy Kibernetiki 8, 337—339 (1962) [inRussian]

11. Zinoviev, V.: On generalized concatenated codes. Colloquia Mathematica Societatis JanosBolyai 16, Topics in Information Theory, pp. 587—592, Keszthely, Hungary (1975)

12. Zinoviev, V.: Combinatorial methods of construction and analysis of nonlinear error-correct-ing codes. Doctor of Sciences Diss., Computer Centre of Russian Academy of Sciences,Moscow (1988) [in Russian]

13. Zinoviev, V., Leontiev, V.: On perfect codes. Problems Inf. Trans. 8-1, 17—24 (1972)14. Zinoviev, V., Leontiev, V.: The nonexistence of perfect codes over Galois fields. Problems

Control Inf. Theory 2-2, 16—24 (1973)

.

420 A. C. Lobstein, V. A. Zinoviev