1:
2: \documentclass[journal]{IEEEtran}
3: \usepackage{graphicx,epsfig,color,graphics}
4: \newtheorem{theorem}{Theorem}
5: \newtheorem{lemma}{Lemma}
6: \newtheorem{remark}{Remark}
7: \newtheorem{problem}{Problem}
8: \newtheorem{cor}{Corollary}
9: \newtheorem{claim}{Claim}
10: \newtheorem{observation}{Observation}
11: \newtheorem{definition}{Definition}
12: \hyphenation{op-tical net-works semi-conduc-tor IEEEtran}
13: %\bibliographystyle{unsrt}
14:
15: \begin{document}
16:
17: % PAPER TITLE
18:
19: \title{On the Complexity of finding Stopping Distance in Tanner Graphs
20: \thanks{A preliminary version of this paper was presented at CISS 2006,
21: March 2006}}
22:
23: % AUTHOR BEGINS HERE
24:
25:
26:
27: \author{
28: \authorblockN{K. Murali Krishnan , Priti Shankar \\
29: \authorblockA{Department of Computer Science and Automation \\
30: Indian Institute of Science, Bangalore - 560012, India.\\
31: Email: \{kmurali,priti\}@csa.iisc.ernet.in}}}
32:
33:
34: \maketitle
35:
36:
37: % ABSTRACT
38: \begin{abstract}
39: Two decision problems related to the computation of stopping sets in
40: Tanner graphs are shown to be NP-complete. NP-hardness of the
41: problem of computing the stopping distance of a Tanner graph follows
42: as a consequence.
43: \end{abstract}
44:
45:
46:
47: \IEEEpeerreviewmaketitle
48:
49:
50: \section{Introduction}
51: \label{intro}
52:
53: Stopping sets were introduced in \cite{Di} for the analysis of erasure
54: decoding of LDPC codes. It was shown that
55: the iterative decoder fails to decode to a codeword if and only
56: if the set of erasure positions is a superset of some stopping set in the
57: Tanner graph \cite{Tan} used in decoding.
58: Considerable analysis has been carried out on the
59: distribution of stopping set sizes in LDPC code ensembles, giving valuable
60: insight into the asymptotic performance
61: of message-passing decoding on LDPC ensembles --- see for example
62: \cite{Di2,Or}.
63: Since small stopping sets are directly responsible for
64: poor performance of iterative decoding algorithms, it is of interest to
65: determine the size of the smallest stopping set in a Tanner graph,
66: called the {\em stopping distance} of the graph. Construction of
67: codes for which there are Tanner graphs that do not contain small
68: stopping sets has been studied --- see for example \cite{Tian,Adi}.
69: The stopping distance of the
70: graph, is of interest as it gives the minimum number of erasures
71: that can cause iterative decoding to fail.
72:
73: The relationship between stopping distance
74: and other graph parameters like girth has been explored in \cite{Or2}
75: where it is shown that large girth implies high stopping distance.
76: Pishro-Nik and Fekri \cite{Nik} showed that by adding a suitable number
77: of parity checks the stopping distance of a Tanner graph for a code
78: can be increased to the maximum possible, viz., the minimum distance of
79: the code. Schwartz and Vardy \cite{Var} defines the {\em stopping
80: redundancy} of a code as the minimum number of rows in a parity check
81: matrix for the code such that the stopping distance of the corresponding
82: Tanner graph is equal to the minimum distance of the code
83: and proves some bounds on the stopping redundancy for
84: various classes of codes. Further investigations on stopping redundancy
85: may be found in \cite{Sgl}.
86:
87: In this correspondence, we show that the computational problems of
88: determining whether a given Tanner graph has a stopping set of a given size
89: or at most a given size are NP-complete.
90: These are shown by reductions from the well known NP-complete problems of
91: determining whether a given graph contains a vertex cover of a given size
92: (respectively at most a given size) to the above problems. NP-hardness
93: of the problem of finding the stopping distance of a Tanner graph
94: follows as a consequence of the latter result.
95:
96: \section{Background}
97: \label{background}
98:
99: Given a parity check matrix $H = [h_{ij}] \in GF(2)^{(n-k)\times n}$,
100: $1\leq k\leq n$ for an $(n,k)$ binary linear code, the
101: Tanner graph is the undirected bipartite graph $G=(L,R,E)$
102: where $L=\{ x_{i}, 1\leq i\leq n \}$, $R=\{ c_{j}, 1\leq j\leq n-k\}$
103: and $E=\{ (x_{i},c_{j}): h_{ji}=1, 1\leq i\leq n, 1\leq j\leq n-k\}$.
104: The set $L$ corresponds to the set of codeword elements
105: and $R$ corresponds to the set of parity checks.
106: We refer to the set $L$ and $R$ as the set of left and right vertices
107: respectively. For $S\subseteq L\cup R$, we define
108: $N(S)=\{ y : (x,y)\in E, x\in S\} $. $S\subseteq L$ is a
109: {\em stopping set} if for all $c_{j}\in N(S)$,
110: $|N(\{ c_{j}\} )\cap S|\geq 2$ ie., every vertex connected to some vertex
111: in a stopping set must have at least two neighbours in the the stopping set.
112: The {\em stopping distance} of a Tanner graph is the size of
113: the smallest stopping set in the graph. We define two
114: decision problems concerning stopping sets:
115:
116: \begin{problem}
117: STOPPING SET: Given a Tanner graph $G$ and positive integer $t$,
118: does $G$ have a stopping set of size $t$.
119: \end{problem}
120:
121: \begin{problem}
122: STOPPING DISTANCE: Given a Tanner graph $G$ and positive integer $t$,
123: does $G$ have a stopping set of size at most $t$.
124: \end{problem}
125:
126: Note that the corresponding decision problems arising out of the
127: problem of finding the minimum distance of a code
128: were shown to be NP-complete in \cite{Mac} and \cite{Var2}.
129:
130:
131: It is clear that if either STOPPING SET or STOPPING DISTANCE can
132: be solved in polynomial
133: time, then evoking the algorithm at most $|L|$ times, the problem of
134: actually finding the stopping distance of a Tanner graph can be
135: solved. Conversely, if there is a polynomial time algorithm for
136: finding the stopping distance of a given Tanner graph $G$, then
137: we can use the algorithm to solve STOPPING DISTANCE since $G$ has
138: stopping distance less than or equal to $t$ if and only if $G$ contains
139: is a stopping set of size less than or equal to $t$. Note
140: that it is not immediately clear how to solve STOPPING SET in polynomial
141: time even if a polynomial time algorithm for computing the stopping
142: distance of a Tanner graph is known.
143:
144: The notion of NP-completeness was introduced in \cite{Cook}, and
145: is well established in the computer science literature for the analysis of
146: the computational complexity of problems (see \cite{Cor, Gar} for
147: a detailed account). Typically, a problem is posed
148: as a decision problem, i.e., one where the solution consists
149: of answering it with a {\em yes} or a {\em no}.
150: All inputs for which the answer
151: is a {\em yes} from a set. We identify this set with the problem.
152: A decision problem $A$ belongs
153: to the class NP if there exists a polynomial time algorithm $\Pi $
154: such that, for all $x\in A$, there exists a string $y$ (called a
155: {\em certificate} for membership of $x$ in $A$), with $|y|$ polynomially
156: bounded in $|x|$, such that $\Pi$ accepts $(x,y)$, whereas, for all
157: $x\notin A$, $\Pi$ rejects $(x,y)$ for any string $y$ presented to the
158: algorithm. In other words, problems in NP are precisely those for which
159: membership verification is polynomially solvable.
160: We say a decision problem $A$ is {\em polynomial time
161: many-one reducible} to a decision problem $B$ if there exists a polynomial
162: time algorithm $\Pi'$ such that, given an instance $x$ of $A$, $\Pi' $
163: produces an instance $z$ of $B$ satisfying $z\in B$ if and only if $x\in A$.
164: In such case, we write $A\preceq_{p}B$.
165: A problem $A\in$NP is NP-complete if for every $X\in$NP, $X\preceq_{p}A$.
166: It is generally believed that NP-complete problems have no polynomial
167: time algorithms.
168:
169: Given an undirected graph (not necessarily bipartite) $G=(V,E)$,
170: $S\subseteq V$ is a {\em vertex cover} in $G$ if
171: for all $(u,v)\in E$ either $u\in S$ or $v \in S$ or both.
172: We will be using in our reductions the following decision problems
173: associated with the computation of vertex covers in a graph.
174:
175: \begin{problem}
176: VERTEX COVER: Given a graph $G$ and a positive integer $t$
177: does $G$ contain a vertex cover of size at most $t$.
178: \end{problem}
179:
180: The above problem is shown to be NP-complete in
181: \cite[p. 190]{Gar}. A variant of this problem referred to by
182: the same name and shown to be NP-complete
183: in \cite[pp. 949--950]{Cor} will be referred to
184: here as the following:
185:
186: \begin{problem}
187: VERTEX COVER(=): Given a graph $G$ and a positive integer $t$
188: does $G$ contain a vertex cover of size equal to $t$.
189: \end{problem}
190:
191: In the following section we show that both STOPPING DISTANCE
192: and STOPPING SET are NP-complete by establishing polynomial
193: time many-one reductions from VERTEX COVER and VERTEX COVER(=)
194: respectively to the above problems.
195:
196: \section{Hardness of STOPPING DISTANCE}
197: \label{stopping distance}
198:
199: Let $(G=(V,E),t)$ be an instance of the VERTEX COVER problem.
200: Let $|V|=n$, $|E|=m$. Excluding trivial cases of the problem
201: we may assume $1\leq t\leq n-1$. We shall make the further
202: assumption that $G$ is connected. It is not hard to show that both
203: VERTEX COVER and VERTEX COVER(=) remain NP-complete
204: even when restricted to connected graphs.
205:
206: The vertex-edge incidence graph of $G$ is the undirected bipartite graph
207: $G'=(L,R,E')$ with $L=V$, $R=E$ and edges $(e,u)$ and $(e,v)$ in $E'$
208: for each $e=(u,v)\in E$. Figure 1 shows the vertex-edge incidence
209: graph for a graph $G$ with $n=4$ and $m=3$.
210:
211:
212: % Figure 1
213: \begin{figure}
214: \begin{center}
215: % \input{cover1.ps}
216: \input{cover1.pstex_t}
217: % \input{cover.pstex_t}
218: % \input{covernopstex.pdf}
219: \caption{A graph G and its vertex-edge incidence graph}
220: % \caption{Construction of G'}
221: \label{figure1}
222: \end{center}
223: \end{figure}
224:
225:
226: The advantage of assuming that $G$ is connected
227: arises out of the following lemma:
228:
229: \begin{lemma}
230: \label{lemma1}
231: Let $G'=(L,R,E')$ be the vertex-edge incidence graph of a
232: connected graph $G=(V,E)$. Let $S$ be a stopping set in $G'$.
233: Then $S=L$.
234: \end{lemma}
235: \begin{proof}
236: Let $L\setminus S\neq \emptyset$. Then, as $G$ is connected
237: there exists $v\in L\setminus S$ and $u\in S$
238: such that $(u,v)\in E$. Let $e=(u,v)$. Then $e\in N(S)$.
239: Since $S$ is a stopping set $|N(\{e\})\cap S|\geq 2$. But
240: the only neighbours of $e$ in $G'$ are $u$ and $v$. Hence $v\in S$
241: contradicting $v\in L\setminus S$.
242: \end{proof}
243:
244:
245: We construct an undirected bipartite graph $G''=(L,R,E'')$ as follows:
246: $L=\bigcup_{i=0}^{m+1}L_{i}$, $R=\bigcup_{j=0}^{m+1}R_{j}$, where,
247: $R_{0}=\{z_{1},...,z_{m-1}\}$, $R_{j}=\{u_{j}^{r},u\in V\}$ for
248: $2\leq j\leq m+1$, $R_1=L_0=E$, $L_{i}=\{u_{i}^{l},u\in V\}$ for
249: $1\leq i\leq m+1$. Edges in $G''$ are connected as the following:
250:
251:
252: \begin{itemize}
253:
254: \item Connect $u_{i}^{l}\in L_{i}$ to $u_{i}^{r}\in R_{i}$,
255: $2\leq i\leq m+1$.
256:
257: \item Connect $u_{i}^{l}\in L_{i}$ to $u_{i+1}^{r}\in R_{i+1}$.
258: $1\leq i\leq m$
259:
260: \item For each $e=(u,v)$ in $E$, connect $e\in R_1$ to
261: $u$ and $v$ in $L_1$.
262:
263: \item For each $e\in E$ Connect $e\in L_0$ to $e\in R_1$.
264:
265: \item For the purpose of defining the edges between $R_0$ and $L_0$,
266: temporarily re-label vertices in $L_0$ as $e_1,e_2,..e_m$ in
267: some arbitrary way. Add the edges $(e_i,z_i)$ for $1\leq i\leq m-1$
268: and the edges $(e_i,z_{i-1})$ for $2\leq i\leq m$.
269:
270: \end{itemize}
271:
272:
273: \begin{figure}
274: \begin{center}
275: \input{cover3.pstex_t}
276: \caption{Construction of $G''$}
277: \label{figure 2 }
278: \end{center}
279: \end{figure}
280:
281:
282: The example in figure 3 illustrates the construction of $G''$ for
283: the graph in figure 1. The graph $G''$ consists of a copy of the
284: vertex-edge incidence graph of $G$ (vertex sets $L_{1}$ and $R_{1}$).
285: Additionally, there are $m$ copies of the vertex set $V$ on the
286: left ($L_2,L_3,...,L_{m+1}$) and right ($R_2,R_3,...R_{m+1}$).
287: The connections between $R_{0}$ and $L_{0}$ ensure that any stopping
288: set in $G''$ containing any one vertex in $L_0$ must contain the
289: whole of $L_0$. The vertex $u_{i}^{r}$ in $R_i$
290: has neigbours $u_{i-1}^{l}$ and $u_{i}^{l}$ for each $2\leq i\leq m+1$
291: and each $u\in V$. This ensures that if a stopping set $S$ in
292: $G''$ contains $u_{i}^{l}$ for some $i\in \{1,2,..m+1\}$ then all the
293: $m+1$ vertices $u_{1}^{l}, u_{2}^{l},....,u_{m+1}^{l}$
294: must be present in $S$. These observations summarized below
295: play a crucial role in the arguments that follow.
296:
297: \begin{observation}
298: \label{obs1}
299: A stopping set $S'$ in $G''$ satisfies $u_{i}^{l}\in S'$ for some
300: $1\leq i\leq m+1$ if and only if it satisfies
301: $u_{i}^{l}\in S'$ for {\em every} $1\leq i\leq m+1$.
302: Moreover either $L_0\subseteq S'$ or $L_0\cap S'=\emptyset$.
303: \end{observation}
304:
305: The following two claims establish the connection between vertex covers
306: in $G$ and stopping sets in $G''$.
307:
308: \begin{lemma}
309: \label{IF}
310: If $G$ contains a vertex cover $S$ of size $t$
311: for some $1\leq t\leq n-1$
312: then $G''$ contains a stopping set of size $t(m+1)+m$.
313: \end{lemma}
314: \begin{proof}
315: Consider the set $S'= L_0 \cup \{u_{i}^{l}: u\in S, 1\leq i\leq m+1\}$
316: in $G''$. Clearly $S'$ has $t(m+1)+m$ elements.
317: Let $w\in N(S')$. Then either $w=u_{i}^{r}$
318: for some $u\in S$, $i\in\{2,3,...m+1\}$ or $w\in R_1$ or $w\in R_0$.
319: In the first case, both $u_{i}^{l}$ and $u_{i-1}^{l}$ are neighbours
320: of $w$. If $w\in R_1$, then by construction, $w$ must correspond to
321: some edge $e=(u,v)$ in $E$. Since $L_0\subseteq S'$,
322: $e\in L_0$ is a neighbour of $w$. Since $S$ is a vertex cover in $G$,
323: either $u$ or $v$ or both must belong to $S$. Hence one or both of
324: $u_{1}^{l}$ and $v_{1}^{l}$ must be a neighbour of $w$ in
325: $S'$. Finally if $w\in R_0$, then both the neighbours of $w$ are
326: in $L_0$, and therefore in $S'$. Thus in all cases $w$ has at least
327: two neighbours in $S'$. Consequently $S'$ is a stopping set.
328: \end{proof}
329:
330: We now prove that every stopping set in $G''$ of size less than
331: $n(m+1)+m$ must correspond to some vertex cover of size $t$ in $G$
332: for some $1\leq t\leq n-1$ and must have size exactly $t(m+1)+m$
333:
334:
335: \begin{lemma}
336: \label{ONLYIF}
337: Let $S'$ be a stopping set in $G''$ of size less than $n(m+1)$.
338: Then the following must hold:
339: \begin{itemize}
340: \item $L_0\subseteq S'$,
341: \item $|S'|=t(m+1)+m$ for some $1\leq t\leq n-1$ and
342: $|S'\cap L_i|=t$ for every $1\leq i\leq m+1$
343: \item $S=\{u\in V$ : $u_{i}^{l}\in S'$ for some $1\leq i\leq m+1\}$
344: is a vertex cover of size $t$ in $G$.
345: \end{itemize}
346: \end{lemma}
347:
348: \begin{proof}
349: Suppose $L_0$ is not contained in $S'$. Then by Observation~\ref{obs1}
350: $L_0\cap S'=\emptyset$. Since $S'\neq \emptyset$, There must be
351: some $u\in V$ and $i\in\{1,2,..,m+1\}$ such that $u_{i}^{l}\in S'$.
352: By Observation~\ref{obs1} $u_{1}^{l}\in S'$. Since vertices in
353: the set $R_1$ are connected only to $L_1$ and $L_0$,
354: every neighbour of $S'$ in $R_1$ must have two neigbours in $S'\cap L_1$
355: in order for $S'$ to satisfy the conditions of a stopping set.
356: In other words, $S'\cap L_1$
357: must be a stopping set in the subgraph of $G''$ induced by the vertices
358: $L_1\cup R_1$. Note that this subgraph is the vertex-edge incidence
359: graph of $G$. Applying Lemma~\ref{lemma1} we get $S'\cap L_1=L_1$.
360: Hence Observation~\ref{obs1} shows that $S'=\bigcup_{i=1}^{m+1}L_i$. But
361: in that case $|S'|=n(m+1)$, a contradiction. Hence $L_0\subseteq S'$
362: and $|L_1\cap S'|<n$.
363: Let $|S'\cap L_1|=t$ for some $1\leq t\leq n-1$. Applying
364: Observation~\ref{obs1} once again, $|S'\cap L_i|=t$ for all
365: $1\leq i\leq m+1$. Hence $|S'|=t(m+1)+m$.
366:
367: To complete the proof of the lemma, it is sufficient to prove that
368: $S=\{u\in V : u_{1}^{l}\in S'\}$ is a vertex cover of $G$. Since
369: $L_0\subseteq S'$, $R_1\subseteq N(S')$. Since every vertex $e$ in $R_1$
370: has only one neighbour in the set $L_0$,
371: for $S'$ to satisfy the stopping set condition
372: $e$ must have a neighbour in $L_1\cap S'$.
373: Then, by construction $\{u\in V: u_{1}^{l}\in S'\}$ must be a vertex
374: cover in $G$ as required.
375: \end{proof}
376:
377: As a consequence of Lemma~\ref{IF} and Lemma~\ref{ONLYIF} we have:
378:
379: \begin{cor}
380: \label{cor1}
381: $G$ has a vertex cover of size $t$ if and only if $G''$
382: has a stopping set of size $t(m+1)+m$, $1\leq t\leq n-1$.
383: Hence $(G,t)\in$ VERTEX COVER(=) if and only if
384: $(G'',t(m+1)+m)\in$ STOPPING SET.
385: \end{cor}
386:
387: \begin{cor}
388: \label{cor2}
389: $G$ has a vertex cover of size at most $t$ if and only if $G''$
390: has a stopping set of size at most $t(m+1)+m$, $t\in\{1,2,..,n-1\}$.
391: Hence $(G,t)\in$ VERTEX COVER if and only if
392: $(G'',t(m+1)+m)\in$ STOPPING DISTANCE.
393: \end{cor}
394:
395: We are now ready to prove:
396:
397: \begin{theorem}
398: STOPPING DISTANCE and STOPPING SET are NP-complete
399: \end{theorem}
400: \begin{proof}
401: We have proved that $(G,t)\in$ VERTEX COVER if and only
402: if $(G'',t(m+1)+m)\in$ STOPPING SET.
403:
404: Since $G''$ can be constructed
405: from $G$ in polynomial time ($O(mn)$ time suffices), it
406: follows that VERTEX COVER(=) $\preceq_p$ STOPPING SET and
407: VERTEX COVER $\preceq_p$ STOPPING DISTANCE from
408: Corollary~\ref{cor1} and Corollary~\ref{cor2} respectively.
409: It is easy to verify whether a given set of left vertices of a bipartite
410: graph forms a stopping set in time linear in the size of the
411: graph. Hence both STOPPING DISTANCE and STOPPING SET belong to
412: the class NP.
413: \end{proof}
414:
415: As a consequence, we have:
416:
417: \begin{cor}
418: \label{Harddist}
419: Computing stopping distance in a Tanner graph is NP-hard.
420: \end{cor}
421:
422:
423: \section{Acknowledgment}
424: \label{ack}
425: The authors would like to thank Dr. L. Sunil Chandran for
426: helpful discussions.
427:
428:
429: \begin{thebibliography}{99}
430:
431: \bibitem{Di}C. Di, D. Proietti, I. E. Telatar, T. J. Richardson and
432: R. Urbanke, ``Finite length analysis of low-density parity-check
433: codes on the binary erasure channel,'' {\em IEEE Trans. Inform. Theory.},
434: vol. 48, no. 6, pp. 1570-1579, June 2002.
435:
436: \bibitem{Di2} C. Di, A. Montanari and R. Urbanke, ``Weight distribution
437: of LDPC code ensembles: Combinatorics meets statistical physics,''
438: in {\em Proc. IEEE Int. Symp. Inform. Theory}, Chicago, IL., July
439: 2004, p. 102.
440:
441: \bibitem{Or} A. Orlitsky, K. Viswanathan, and J. Shang, ``Stopping set
442: distribution of LDPC code ensembles,'' {\em IEEE Trans. Inform. Theory},
443: vol. 51, no. 3, March 2005, pp. 929-953.
444:
445: \bibitem{Tian} T. Tian, C. Jones, J. D. Villasenor, and R. D. Wesel,
446: ``Construction of irregular LDPC codes with low error floors,''
447: in {\em Proc. ICC 2003}, Seattle, Washington, May 2003, pp. 3125-3129.
448:
449:
450: \bibitem{Adi} A. Ramamoorthy, and R. Wesel, ``Construction of short block
451: length irregular LDPC codes,'' in {\em Proc. ICC 2004}, Paris, June 2004,
452: pp. 410-414.
453:
454:
455: \bibitem{Or2} A. Orlitsky, R. Urbanke, K. Viswanathan, and J. Shang,
456: ``Stopping sets and girth of Tanner graphs,'' in {\em Proc. IEEE Int. Symp. Inform. Theory},
457: Lausanne, June 2002, p. 2.
458:
459: \bibitem{Var} M. Schwartz and A. Vardy, ``On the stopping distance and the
460: stopping redundancy of codes,'' {\em IEEE Trans. Inform. Theory}
461: vol. 52, no. 3, pp. 922-932, March 2006.
462:
463: \bibitem{Tan} M. Tanner, ``A recursive approach to low-complexity
464: codes,'' {\em IEEE Trans. Inform. Theory}, vol. 27, no. 5
465: pp. 533-547, Sept 1981.
466:
467: \bibitem{Cor} T. H. Cormen, C. E. Leiserson, and R. L. Rivest,
468: {\em Introduction to Algorithms}, MIT Press, 1990.
469:
470: \bibitem{Gar} M. R. Garey, and D. S. Johnson, {\em Computers and
471: Intractability: A Guide to the Theory of NP-completeness},
472: W. H. Freeman, 1979.
473:
474: \bibitem{Cook} S. Cook, ``The complexity of theorem proving
475: procedures,'' in {\em Proc. Third ACM Ann. Symposium on Theory of Computing}, Shaker Heights, Ohio, May 1971, pp.151-158.
476:
477: \bibitem{Nik} H. Pishro-Nik and F. Fekri, ``On decoding of low-density
478: parity-check codes over the binary erasure channel,'' {\em IEEE
479: Trans. Inform. Theory}, vol. 50, no. 3, pp. 439-454, March 2004.
480:
481: \bibitem{Sgl} J. Han and P. Siegel, ``Improved upper bounds on stopping
482: redundancy,'' preprint available at: {\em http://www.arXiv.org},
483: cs.IT/0511056.
484:
485: \bibitem{Var2} A. Vardy, ``The intractability of computing the minimum
486: distance of a code,'' {\em IEEE Trans. Inform. Theory}, vol. 46 no. 6
487: pp. 1757-1766, Nov. 1997.
488:
489: \bibitem{Mac} E. R. Belekamp, R. J. McEliece and H. C. A van Tilborg,
490: ``On the inherent intractability of certain coding problems,''
491: {\em IEEE Trans. Inform. Theory}, vol. 24 no. 3,
492: pp. 384-386, May 1978.
493:
494: \end{thebibliography}
495:
496: \end{document}
497:
498:
499:
500:
501:
502:
503:
504:
505:
506:
507:
508:
509: