1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: \section{Introduction}\label{S-intro}
3: \setcounter{equation}{0}
4: %%%%%%%%%%%%%%%%%%%%%%%%%%%
5: Convolutional codes (CC's) and block codes are the most widely
6: used types of codes in engineering practice, a fact which leads to a
7: continuing need for a thorough mathematical basis for the design
8: of useful codes. In consequence, coding theory has become one of the various
9: young branches of mathematics which are attractive because of the
10: active interplay between sophisticated engineering inventions and
11: high level mathematics. This is particularly true for the theory
12: of cyclic block codes.
13:
14: The algebraic theory of CC's was initiated mainly by the articles
15: of Forney~\cite{Fo70} and Massey et~al.~\cite{MaSa67,MaSa68}, and, as can
16: be seen from the books~\cite{JoZi99,Pi88b} and the
17: article~\cite{McE98}, a lot of material has been accumulated
18: since. In the last decade Rosenthal and co-workers began a
19: successful project, dedicated to a better and deeper
20: mathematical understanding of CC's by exploiting more
21: systematically the existing links to control theory,
22: see~\cite{RSY96,RoYo99,Ro00}. Yet, up to now the mathematical
23: theory of CC's is not nearly as developed as that of block
24: codes. This gap is even larger when it comes to the notion of
25: cyclicity. Despite the well-known and frequently exploited efficiency of
26: cyclic block codes, almost nothing is known about
27: cyclic structures for convolutional codes and their possible impact on
28: applications.
29:
30: In 1976 Piret showed in his fundamental paper~\cite{Pi76} how
31: cyclicity has to be understood for CC's and laid the basis
32: for a mathematical theory of cyclic CC's.
33: The first important discovery of Piret was that classical cyclicity ---
34: as common for block codes --- is trivial for CC's
35: (see Proposition~\ref{P-trivialCCC} in the next section).
36: He also showed that a more sophisticated ``graded cyclicity''
37: leads to interesting examples of good convolutional codes,
38: some of which can be found in~\cite[Sect.~IV]{Pi76} and
39: in~\cite{GSS02}.
40: His second main discovery was that irreducible graded cyclic CC's
41: can algebraically be described by one-sided principal ideals in a
42: noncommutative algebra $\Azs$.
43: This algebra will be introduced in the next section.
44: For the moment we only mention that $\Azs$ is equal to
45: $A[z]$ as a left $\F[z]$-module where
46: $A\cong \F[x]/(x^n-1)$,~$\F$ is a finite field and~$n$ is the length of
47: the code. Only the multiplication in the algebra $\Azs$ is quite different
48: from the ordinary one. It depends on an $\F$-automorphism~$\sigma$ of~$A$ and is
49: typically non-commutative.
50:
51: The results of Piret indicate a surprising analogy to the
52: theory of block codes where cyclic codes are described as
53: principal ideals, see~\cite{MS77,Be98}, with the only difference
54: that the latter are in the commutative ring~$A$.
55:
56: Shortly after Piret, a thorough
57: analysis of his results was undertaken by Roos~\cite{Ro79} in a module
58: theoretic framework, avoiding thereby cumbersome matrix
59: manipulations. At the same time Roos considerably extended
60: Piret's notion to what will be called $\sigma$-cyclicity later on in this paper.
61: But apart from this, no substantially new results could be added
62: and Piret's idea of a generating polynomial~\cite[Thm.3.10]{Pi76}
63: could not be incorporated. Furthermore, Roos' results are partly
64: non-constructive.
65:
66: After the work of Piret and Roos no substantial effort has been made
67: towards a concise mathematical description of cyclic CC's --- as far as we know.
68: This may partly be due to the limited mathematical readership of the
69: journals in question and to the circumstance that Piret's
70: article is quite cumbersome to read.
71:
72: Yet, we think that this topic is worth being investigated in more detail.
73: We realized that Piret's results may serve as a good basis for a theory of
74: $\sigma$-cyclic CC's, which we would like to re-initiate with this paper.
75: Although we do not consider distance properties of cyclic CC's,
76: we will present the exact free distance of all codes constructed in the examples.
77: This way we hope to indicate that the big class of $\sigma$-cyclic CC's contains
78: quite some good codes and, therefore, deserves to be investigated further.
79:
80: We proceed with an outline of the paper.
81: In Section~\ref{S-CCdef} we will trace the steps
82: which lead to the definition of $\sigma$-cyclicity for CC's.
83: We think it is worthwhile recalling also the original idea of Piret before going
84: into the more general setting initiated by Roos.
85: We will construct the (generalized) Piret algebra $\Azs$ and develop the
86: representation of $\sigma$-cyclic CC's as left ideals in $\Azs$.
87: As the Piret algebra is based on an automorphism~$\sigma$
88: of~$A$, we have to collect some information about the group of automorphisms of~$A$ and
89: about how the structure of the Piret algebra depends on~$\sigma$.
90: This will be done in Section~\ref{S-autos}.
91: In Section~\ref{S-idealgen} we give an algebraic and extended version of Piret's
92: main result which states that~$\sigma$-cyclic CC's are left principal ideals
93: in $\Azs$.
94: Thereafter we investigate as to what extent a generator of a left
95: ideal in $\Azs$ is unique and in Section~\ref{gencomp} we show how this unique
96: generator can be computed by means of a finite algorithmic procedure.
97: The basic algebraic tool for these sections
98: is a decomposition of the Piret algebra by idempotents of~$A$ and a reduction
99: procedure based on a monomial order of the skew polynomials.
100: In Section~\ref{S-circ} we introduce a new type of non-commuting $\sigma$-circulant
101: matrices along with a thorough investigation of their properties.
102: These matrices are just the proper medium for the interplay between
103: left ideals together with their principal generators on the one side and
104: CC's as submodules of $\F[z]^n$ along with their generating matrices on the other.
105: They also turn out to be a canonical, yet nontrivial, generalization of classical
106: circulants as they are common in the theory of cyclic block codes.
107: This becomes in particularly clear when we derive our results on generator and
108: control polynomials and dual codes in Section~\ref{S-gencontr}.
109: Indeed, we arrive at a scenario very similar to that of cyclic block codes.
110: The notion of a control polynomial is also included in this framework, it is obtained via
111: (right) annihilator ideals in the Piret algebra.
112: Beyond this algebraic structure, convolutional coding requires to also discuss some other
113: properties and invariants of the codes, as there are non-catastrophicity, minimal
114: generator matrices and the complexity of the given code.
115: All these issues can nicely be dealt with in our algebraic context.
116: As it turns out, the reduced principal generator polynomials for left ideals
117: in $\Azs$, as constructed in Section~\ref{S-idealgen} and~\ref{gencomp},
118: also suits well when it comes to the properties of the associate circulant matrix.
119: The latter leads in a canonical way to a basic minimal generator matrix of the
120: given code, and, consequently, the complexity can be computed in terms of the
121: generator polynomial.
122: In order to derive these results one has to combine the techniques for
123: circulant matrices with the algebraic methods from
124: Section~\ref{S-autos} --~\ref{gencomp}.
125: In the final Section~\ref{S-outlook} we give a short
126: outline of several future research topics.
127:
128: Throughout this paper we make an effort to motivate and
129: justify the main steps of our investigations by using the classical theory of
130: cyclic block codes as a guideline.
131: We also give explicit examples in order to show how the objects in question can be computed.
132: This is particularly so in Section~\ref{S-gencontr} and we hope
133: that this way any possible impact of our results for convolutional coding can be judged more easily.
134: