Dirichlet Regions and Perfect Codes in Additive Channel

In the present work, the class of metrics connected with subsets of the linear space on the field, GF(2), is considered and a number of facts are established, which allow us to express the correcting capacity of codes for the additive channel in terms of this metrics. It is also considered a partition of the metric space, B, by means of D-representable codes. The equivalence of D-representable and the perfect codes in the additive channel is proved.


Introduction
We consider the additive channel of communication [1][2][3][4] as a transformer of information, which is a generalization of the classical binary channel with limited number of distortions, . Many notions and facts in the present work take their origins in the classic coding theory and are the direct analogues of the well-known results [1][2][3][4][5].The "noise" generated by the additive channel leads to the fact that there appears a word at the outlet of the channel which is different from that at its inlet.In connection with this there rises a necessity of transforming (coding) information for conducting it through the given channel, as well as a necessity of retransforming (decoding) it at the channel outlet.This circumstance makes one introduce such standard notions in the coding theory as: error correcting code; transfer/decoding speed, etc.
On the other hand, as there are many additive channels, the problem of ordering and classification of such channels rises, taking into account the main difficulty, namely, the possibility of correcting the generated errors.
We consider the class of metrics connected with the subsets of the space on the field, GF (2), and establish a number of facts which allow us to express the correcting capacity of codes for the additive channel in terms of this metrics.Also, we consider the partition of the metric space, , through -representable codes.The equivalence of -representable and perfect codes in the additive channel is proved.
For convenience we take, , where x is the logic negation of x . 2 is the parity counter, then: where A is complement of the set, A , in .

then in the additive channel,
A , can occur no more than t 'errors' of the form, .Consequently,  is a sphere of the radius, t, having its centre at the point, x.
Thus, we get the classical case of the binary channel with limited number of errors.
or the one symmetrical to it: It is obvious that the preceding definitions are symmetric with respect to the pair,  , A V , and therefore, both "generation" and "correction" of errors have the same nature.
Statement 1 [3].If the code, , corrects the errors of the additive channel, V A , then the code, A , corrects the errors of the additive channel, .V To describe the "relations" of the additive channel, A , and the code, , correcting the errors of that channel it is convenient to introduce the following double-case predicate, Definition.We call any pair,   , additive if it is a solution of the following equation:

X A 
Taking this definition into account, the property of perfectness of codes can be written as follows: Note that there are as many additive channels, as there are Boolean functions, and a few of them do not essentially differ from each other.It is not clear how to classify such channels yet, but the following statements correspond the commonly accepted viewpoint.
Definition [3].The channels, A and C, are called equivalent if any code correcting the errors of the additive channel, A, corrects the errors of the channel, C, and vice versa.
Introducing the following relation of partial order, one can formally write: It follows from the preceding statements that one can consider-without loss of generality: A is a class of additive channels equivalent to A , then it is sufficient to solve the coding problem for any representative of that class.
b) The additive channel, A , includes the null vector, which can be interpreted as the possibility of errorless transfer of the signal through that channel.
As it follows from then an analogical statement is correct for the vector, , too, i.e. it is sufficient to discuss the codes including the null vector.
V Thus, it follows from that the sets, , can overlap only at zero, and the search of the code, , is to be organized in the set,  As the power of the neighbourhood of order t does not depend on the vector, , we make the following denotation, Note that for the additive pair,   , A V , the following limits take place [2]: It is clear that the upper limit is reached for the perfect pair,   , A V

Metrics and Codes
The standard and most used metrics in coding theory is Hamming's metrics, i.e. the following function: .
, in the following way: .
It is obvious that if another basis, , is chosen then another metrics is generated: .
The more general procedure of metrics generation in the above-mentioned way is as follows.For the given subset,   , where .
In particular, for , the MLM norms of vectors in take the following values: In terms of graph theory, the above situation is as follows.We give the following binary relation on the set of vortices, :


and the distance between two points in is equal to the minimum number of sides in the circuit connecting the corresponding vortices, or it is infinity, if there is no such circuit.n B Let be a basis for the additive channel, A. We consider any basis, , in the space, , where , and f is a linear reversible transformation, : n n f B  B , defined in the fol- lowing way: We denote the image of the set, , by . We denote the linear shell of the set, , by C   L C , and we denote by   L C the subset in satisfying the following condition: where + means the direct sum the subspaces.It is obvi- The following holds true.

 
, u v L M  , the following relations hold true: , , ; , .
The code, , corrects the errors of the additive channel, А, iff the following holds true: Example: 5) We consider the channel,   1 2 2 0 ,10 ,1 0 , ,1 , as an illustration."Physically", the channel A means that the "errors" of the form, , which take place either in the 1 st place, or in the 1 st and 2 nd places simultaneously, and so on.Thus, A n   and to build a maximum volume code correcting the errors of the given channel we use Lemma 3.
It is sufficient to consider all the subsets, , for which: It follows from the conditions of lemma that the vectors,   , x y L C  , are one to one represented in the form [7]: , , , , We assume-without loss of generality-that: where: , .
Consequently, taking (2) into account, we get: , , Lemma 5.For any additive pairs,   From the definition of the direct sum of sets it follows that: where 1 1 , , , , , , , .
Consequently, taking (3) into account, we get: That is, the pairs,   , A V are not ad- ditive. Q.E.D.

Partition of the Metric Space into Dirichlet's Regions
Let   It is obvious that: In fact, Dirichlet's region of the point, x , includes all points of the metric space,   , n С B  , which are not farther from x than from the other points in .
V It is easy to notice that it is sufficient given the cidence of the sets,  L С and , for (4).Nevertheless, this condition is not always necessary, which can be seen from the following example.
Then we have: that is, 4  .
x y   z and , , x y z D D D do not overlap in pairs.It follows from this example that the condition, is not necessary for the equality (4).
The following theorem 'connects' and V , giving the answer to the question: which are the conditions providing Equation (4).С Theorem 2. The equation: As for any vectors, , , y y y y L C   , holds the fol- lowing: Definition.The code, , is called -representable in the metrical space,  , if all Dirichlet regions of the points in do not overlap in pairs.V We note that D-representability of a code is connected with a certain metrics and, in general, this property does not preserve if the metrics is changed.
Example: 8) Let: then it is logical to consider the case,   .
Consequently, the code, Then there is such vector, , that It is not difficult to prove with these that in the space, This theorem can be formulated in another way.
Corollary.The metric space,   Step 1.We choose an arbitrary code,   V y , from the space,   , into Dirichlet's regions for which Step 2. The code obtained through the formula: partitions the metric space, into Dirichlet's regions.

 ,
field of two elements and be an n-dimensional vector space on that field. of , then the notion of the additive channel is connected with A, as follows.the channel, A , is transformed into one of the following vectors:  is the operation of addition (with respect to mod 2) in the space, .call the neighbourhood of order t with respect to C, the following set: This relation defines adjacency and, geometrically, the set of vortices of the N-dimensional unit cube corresponds to the metrical space,   , the point, x V  , in the following way:

4 Theorem 3 .
The code, , is D-representable in the V metrical space,   , n С B  , iff the code,

D
Let the code, V , be D-representable in where x V  .Then, according to Theorem 2, we have that is parti- tioned into Dirichlet's regions of the points of the code, , iff the metric space, V into Dirichlet's regions of the points,  .We describe the algorithm of building the code, V , partitioning the metric space,   , n C B  , into Dirichlet's regions.

4 .. 4 B 5 .
Let М any basis of the channel, A .Theorem If the code, V , is -representable in the metric space, It follows from Theorem 3 that the code, As M is a basis for   L M , then it follows from Theorem 4[8] (taking Lemma 2 and Lemma 3 into account) that the code,Theorem If the additive pair,   , A V is perfect, then V is -representable in the metric space, D .On the other hand, from perfectness of the code, V, it follows that the set,x D , for all , x V  disjoints.That is, V is D-representable.Copyright © 2013 SciRes.OJDM