WSJT-X/lib/ldpc/encoding.html
Steven Franke 5ac886855d Add ldpc sandbox folder.
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@6437 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
2016-01-25 00:04:21 +00:00

284 lines
13 KiB
HTML
Executable File

<HTML><HEAD>
<TITLE> Encoding Message Blocks </TITLE>
</HEAD><BODY>
<H1> Encoding Message Blocks </H1>
To use a code to send messages, we must define a mapping from a bit
vector, <B>s</B>, of length <I>K</I>, representing a source message,
to a codeword, <B>x</B>, of length <I>N</I>&gt;<I>K</I>. We will
consider only linear mappings, which can be written in the form
<B>x</B>=<B>G</B><SUP><SMALL>T</SMALL></SUP><B>s</B>, where <B>G</B>
is a <I>generator matrix</I>. For a code with parity check matrix
<B>H</B>, whose codewords satisfy <B>Hx</B>=<B>0</B>, the generator
matrix must satisfy <B>HG</B><SUP><SMALL>T</SMALL></SUP>=<B>0</B>.
This software assumes that the number of rows in the parity check
matrix, <I>M</I>, is equal to <I>N-K</I>, as would normally be the
case.
<P>This software deals only with <I>systematic</I> encodings, in which
the <I>K</I> bits of <B>s</B> are copied unchanged to some subset of
the <I>N</i> bits of <B>x</B> (the <I>message bits</I>), and the
remaining <I>M=N-K</I> <I>check bits</I> of <B>x</B> are then set so
as to make the result a codeword. For a linear code, a systematic
encoding scheme always exists, for some choice of which bits of a
codeword are message bits. It is conventional to rearrange the order
of the bits in a codeword so that the message bits come first. The
first <I>K</I> columns of the <I>K</I> by <I>N</I> generator matrix
will then be the identity matrix.
<P>However, this software does <I>not</I> assume that the message bits
come first, since different encoding methods prefer different
locations for the message bits. Instead, a vector of indexes of where
each message bit is located within a codeword is recorded in a file
along with a representation of the part of the generator matrix that
produces the check bits. More than one such generator matrix file can
be created for a single parity check file, in which the locations of
the message bits may be different. Decoding of a received message
into a codeword (with <A
HREF="decoding.html#decode"><TT>decode</TT></A>) does not depend on
knowing which are the message bits, though this does need to be known
in order to reconstruct the original message (with <A
HREF="decoding.html#extract"><TT>extract</TT></A>).
<P>This software stores representations of generator matrices in files
in a format that is not human-readable (except by using the <A
HREF="#print-gen"><TT>print-gen</TT></A> program). However, these
files <I>are</I> readable on a machine with a different architecture
than they were written on.
<A NAME="gen-rep"><H2>Generator matrix representations</H2></A>
<P>For simplicity of exposition, it will be assumed for the next few
paragraphs that the message bits are located at the <I>end</I> of the
codeword, so a codeword can be divided into <I>M</I> check bits,
<B>c</B>, followed by <I>K</I> message bits, <B>s</B>.
<P>On the above assumption, the parity check matrix, <B>H</B>, can be divided
into an <I>M</I> by <I>M</I> matrix <B>A</B> occupying
the first <I>M</I> columns of <B>H</B> and an <I>M</I> by <I>K</I> matrix
<B>B</B> occupying the remaining columns of <B>H</B>. The requirement that
a codeword, <B>x</B>, satisfy all parity checks (ie, that <B>Hx</B>=<B>0</B>)
can then be written as
<BLOCKQUOTE>
<B>Ac</B> + <B>Bs</B> = <B>0</B>
</BLOCKQUOTE>
Provided that <B>A</B> is non-singular, it follows that
<BLOCKQUOTE>
<B>c</B> = <B>A</B><SUP><SMALL>-1</SMALL></SUP><B>Bs</B>
</BLOCKQUOTE>
<B>A</B> may be singular for some choices of which codeword bits are message
bits, but a choice for which <B>A</B> is non-singular always exists if the
rows of <B>H</B> are linearly independent. It is possible, however, that the
rows of <B>H</B> are not linearly independent (ie, some rows are redundant).
This is an exceptional
and not particularly interesting case, which is mostly ignored in the
descriptions below; see the discussion of <A HREF="dep-H.html">linear
dependence in parity check matrices</A> for the details.
<P>The equation <B>c</B> = <B>A</B><SUP><SMALL>-1</SMALL></SUP><B>Bs</B>
defines what the check bits should be, but actual computation of these
check bits can be done in several ways, three of which are implemented
in this software. Each method involves a different representation of
the generator matrix. (Note that none of these methods involves the
explicit representation of the matrix <B>G</B> mentioned above.)
<P>In the <I>dense representation</I>, the <I>M</I> by <I>K</I> matrix
<B>A</B><SUP><SMALL>-1</SMALL></SUP><B>B</B> is computed, and stored
in a dense format (see the <A HREF="mod2dense.html">dense modulo-2
matrix package</A>). A message is encoded by multiplying the
source bits, <B>s</B>, by this matrix to obtain the required check bits.
<P>In the <I>mixed representation</I>, the <I>M</I> by <I>M</I> matrix
<B>A</B><SUP><SMALL>-1</SMALL></SUP> is computed and stored in a dense
format, and the <I>M</I> by <I>K</I> matrix <B>B</B>, the right
portion of the parity check matrix, is also stored, in a sparse format
(see the <A HREF="mod2sparse.html">sparse modulo-2 matrix package</A>).
To encode a message, the source vector <B>s</B> is first multiplied on
the left by <B>B</B>, an operation which can be done very quickly if
<B>B</B> is sparse (as it will be for LDPC codes). The result is then
multiplied on the left by <B>A</B><SUP><SMALL>-1</SMALL></SUP>. If
<I>M</I>&lt;<I>K</I>, the total time may be less than when using the
dense representation above.
<P>The <I>sparse representation</I> goes further, and avoids
explicitly computing <B>A</B><SUP><SMALL>-1</SMALL></SUP>, which tends
to be dense even if <B>H</B> is sparse. Instead, a <I>LU
decomposition</I> of <B>A</B> is found, consisting of a lower
triangular matrix <B>L</B> and an upper triangular matrix <B>U</B> for
which <B>LU</B>=<B>A</B>. The effect of multiplying <B>Bs</B>=<B>z</B> by
<B>A</B><SUP><SMALL>-1</SMALL></SUP> can then be obtained by
<BLOCKQUOTE>
Solving <B>Ly</B>=<B>z</B> for <B>y</B> using forward substitution.<BR>
Solving <B>Uc</B>=<B>y</B> for <B>c</B> using backward substitution.
</BLOCKQUOTE>
Both of these operations will be fast if <B>L</B> and <B>U</B> are
sparse. Heuristics are used to try to achieve this, by rearranging the
rows and columns of <B>H</B> in the process of selecting <B>A</B> and
finding its LU decomposition.
<P><A NAME="make-gen"><HR><B>make-gen</B>: Make a generator matrix from
a parity check matrix.
<BLOCKQUOTE><PRE>
make-gen <I>pchk-file gen-file method</I>
</PRE>
<BLOCKQUOTE>
where <TT><I>method</I></TT> is one of the following:
<BLOCKQUOTE><PRE>
sparse [ first | mincol | minprod ] [ <I>abandon-num abandon-when</I> ]
dense [ <I>other-gen-file </I> ]
mixed [ <I>other-gen-file </I> ]
</PRE></BLOCKQUOTE>
</BLOCKQUOTE>
</BLOCKQUOTE>
<P>Finds a generator matrix for the code whose parity check matrix is
in <TT><I>pchk-file</I></TT>, and writes a representation of this
generator matrix to <TT><I>gen-file</I></TT>. The remaining arguments
specify what representation of the generator matrix is to be used (see
the <A HREF="#gen-rep">description above</A>), and the method to be
used in finding it. A message regarding the density of 1s in the
resulting representation is displayed on standard error. For a sparse
representation, a smaller number of 1s will produce faster encoding.
<P>All representations include a specification for how the columns of
the parity check matrix should be re-ordered so that the message bits
come last. References to columns of the parity check matrix below
refer to their order after this reordering. For the <I>dense</I> and
<I>mixed</I> representations, an <TT><I>other-gen-file</I></TT> may be
specified, in which case the ordering of columns will be the same as
the ordering stored in that file (which must produce a non-singular
<B>A</B> matrix; redundant rows of <B>H</B> are not allowed with this
option). Otherwise, <TT>make-gen</TT> decides on an appropriate
ordering of columns itself. Note that the column rearrangement is
recorded as part of the representation of the generator matrix; the
parity check matrix as stored in its file is not altered.
<P>The <I>dense</I> representation consists of a dense representation
of the matrix <B>A</B><SUP><SMALL>-1</SMALL></SUP><B>B</B>, where
<B>A</B> is the matrix consisting of the first <I>M</I> columns (after
reordering) of the parity check matrix, and <B>B</B> is the remaining
columns. If <B>H</B> contains redundant rows, there is an additional
reordering of columns of <B>A</B> in order create the same effect as
if the redundant rows came last.
<P>The <I>mixed</I> representation consists of a dense representation
of the matrix <B>A</B><SUP><SMALL>-1</SMALL></SUP>, where <B>A</B> is
the matrix consisting of the first <I>M</I> columns (after reordering)
of the parity check matrix. The remaining columns of the parity check
matrix, making up the matrix <B>B</B>, are also part of the
representation, but are not written to <TT><I>gen-file</I></TT>, since
they can be obtained from <TT><I>pchk-file</I></TT>. As for mixed
representations, an additional reordering of columns of <B>A</B> may
be needed if <B>H</B> has redundant rows.
<P>A <I>sparse</I> representation consists of sparse representations
of the <B>L</B> and <B>U</B> matrices, whose product is <B>A</B>, the
first <I>M</I> columns of the parity check matrix (whose columns and
rows may both have been reordered). The matrix <B>B</B>, consisting
of the remaining columns of the parity check matrix, is also part of
the representation, but it is not written to <TT><I>gen-file</I></TT>,
since it can be obtained from <TT><I>pchk-file</I></TT>.
<P>If a sparse representation is chosen, arguments after
<TT>sparse</TT> specify what heuristic is used when reordering columns
and rows in order to try to make <B>L</B> and <B>U</B> as sparse as
possible. The default if no heuristic is specified is
<TT>minprod</TT>. If the <TT><I>abandon-num</I></TT> and
<TT><I>abandon-when</I></TT> options are given, some information is
discarded in order to speed up the process of finding <B>L</B> and
<B>U</B>, at a possible cost in terms of how good a result is
obtained. For details on these heuristics, see the descriptions of <A
HREF="sparse-LU.html">sparse LU decomposition methods</A>.
<P><B>Example:</B> A dense representation of a generator matrix for the
Hamming code created by the example for <A
HREF="pchk.html#make-pchk"><TT>make-pchk</TT></A> can be created as follows:
<UL><PRE>
<LI>make-gen ham7.pchk ham7.gen dense
Number of 1s per check in Inv(A) X B is 3.0
</PRE></UL>
<P><A NAME="print-gen"><HR><B>print-gen</B>: Print a representation of a
generator matrix.
<BLOCKQUOTE><PRE>
print-gen [ -d ] <I>gen-file</I>
</PRE></BLOCKQUOTE>
<P>Prints in human-readable form the representation of the generator
matrix that is stored in <TT><I>gen-file</I></TT>. The <B>-d</B>
option causes the matrices involved to be printed in a dense format,
even if they are stored in the file in a sparse format. See the <A
HREF="#gen-rep">description above</A> for details of generator matrix
representations. Note that the <B>L</B> matrix for a sparse representation
will be lower triangular only after the rows are rearranged, and the <B>U</B>
matrix will be upper triangular only after the columns are rearranged.
The matrix <B>B</B> that is part of the sparse
and mixed representations is not printed, since it is not stored
in the <TT><I>gen-file</I></TT>, but is rather a subset of columns
of the parity check matrix.
<P><B>Example:</B> The generator matrix for the
Hamming code created by the example for <A
HREF="#make-gen"><TT>make-gen</TT></A> can be printed as follows:
<UL><PRE>
<LI>print-gen ham7.gen
Generator matrix (dense representation):
Column order:
0 1 2 3 4 5 6
Inv(A) X B:
1 1 1 0
1 1 0 1
0 1 1 1
</PRE></UL>
For this example, the columns did not need to be rearranged, and hence the
message bits will be in positions 3, 4, 5, and 6 of a codeword.
<P><A NAME="encode"><HR><B>encode</B>: Encode message blocks as codewords
<BLOCKQUOTE><PRE>
encode [ -f ] <I>pchk-file gen-file source-file encoded-file</I>
</PRE></BLOCKQUOTE>
Encodes message blocks of length <I>K</I>, read from
<TT><I>source-file</I></TT>, as codewords of length <I>N</I>, which
are written to <TT><I>encoded-file</I></TT>, replacing any previous
data in this file. Here, <I>N</I> is the number of columns in the
parity check matrix in <TT><I>pchk-file</I></TT>, and
<I>K</I>=<I>N-M</I>, where <I>M</I> is the number of rows in the
parity check matrix. The generator matrix used, from
<TT><I>gen-file</I></TT>, determines which bits of the codeword are
set to the message bits, and how the remaining check bits are
computed. The generator matrix is created from
<TT><I>pchk-file</I></TT> using <A HREF="#make-gen"><TT>make-gen</TT></A>.
<P>A newline is output at the end of each block written to
<TT><I>encoded-file</I></TT>. Newlines in <TT><I>source-file</I></TT>
are ignored.
<P>If the <B>-f</B> option is given, output to <TT><I>encoded-file</I></TT>
is flushed after each block. This allows one to use encode as a server,
reading blocks to encode from a named pipe, and writing the encoded block
to another named pipe.
<HR>
<A HREF="index.html">Back to index for LDPC software</A>
</BODY></HTML>