strcovauto.3

speech signal process tools

.\" Copyright (c) 1986-1990 Entropic Speech, Inc.
.\" Copyright (c) 1991 Entropic Research Laboratory, Inc.; All rights reserved
.\" @(#)strcovauto.3	1.2 02 May 1997 ESI/ERL
.ds ]W (c) 1991 Entropic Research Laboratory, Inc.
.TH STRCOV_AUTO 3\-ESPSsp 02 May 1997
.SH NAME
.nf
strcov_auto \- estimation of auto-correlation coefficients using structured covariance
.fi
.SH SYNOPSIS
.nf
.ft B

extern debug_level;

strcov_auto(data, lnt, r, order, matsiz, window_flag, alg, conv_test, max_iter)
float   *data;
int	lnt;
double  *r;
int	order;
int	matsiz;
int	window_flag
char	alg;
double	conv_test;
int	max_iter;

.ft
.fi
.SH DESCRIPTION
\fIstrcov_auto\fP computes the auto-correlation coefficients of
a given data sequence using the structured covariance method 
method.  
.PP
\fIstrcov_auto\fP takes data of size
.I lnt
in array
.I data
as input, computes the sample covariance matrix of size
\fImatsiz\fP, and then estimates the energy \fIr\fP[0] 
and \fIorder\fP auto-correlation coefficients \fIr\fP[i] 
using the structured covariance method (the computed data 
\fIr\fP has size \fIorder\fP+1).  The parameter \fImatsiz\fP should
be larger than \fIorder\fP; typically, \fImatsiz\fP = \fIorder\fP + 1;
.PP
If the parameter \fIwindow_flag\fP is non-zero, triangular window is
applied to the sample data vectors in the computation of sample
covariance matrix.
.PP
If the parameter \fIalg\fP is 'f' (the default), the structured
covariance computation is performed by \fIstruct_cov\fP
(3\-\s-1ESPS\s+1).  If the parameter \fIalg\fP is 'w', the structured
covariance computation is performed by \fIgenburg\fP
(3\-\s-1ESPS\s+1).  \fIstruct_cov\fP uses a fast, single channel
algorithm developed by John Burg and programmed by Bernard Fraenkel.
\fIgenburg\fP uses an older (but more general) algorithm
\fIgenburg\fP (3\-\s-1ESPS\s+1) that was programmed by Daniel Wenger.
.PP
If the case of \fIalg\fP == 'f', the estimator terminates after
\fImax_iter\fP iterations, or after the relative change in
autocorrelation values falls below the convergence threshold
\fIconv_test\fR.
.PP
If the external \fIdebug_level\fP is non-zero, various debugging 
messages are printed.  Higher values yield more verbose output.  
.SH BUGS
None known
.SH SEE ALSO
\fIget_auto\fP(3\-\s-1ESPS\s+1sp), \fIgenburg\fP(3\-\s-1ESPS\s+1sp),
\fIstruct_cov\fP(3\-\s-1ESPS\s+1sp), \fIget_vburg\fP(3\-\s-1ESPS\s+1sp), 
\fIestimate_covar\fP(3\-\s-1ESPS\s+1sp),
.br
\fIget_fburg\fP(3\-\s-1ESPS\s+1sp)
.SH FUTURE CHANGES
The restriction on the size of \fIorder\fP should be changed.
.SH REFERENCES
.TP
[1]
J.P.Burg, D.G.Luenberger, D.L.Wenger, "Estimation of Structured
Covariance Matrices" \fIProceedings of the IEEE\fP, Vol. 70, No. 9
September 1982
.TP
[2]
Shankar Narayan and J.P. Burg, "Spectral Estimation of Quasi-Periodic
Data", \fIProceedings ICASSP 87\fP, pp. 944-947.  
.SH AUTHOR
program by Shankar Narayan; man page by John Shore