> Hi,
> I am implementing the MUSIC DOA estimator in TI DSP. I am searching for a
> better implementation of Singular value decomposition of a matrix
> algorithm
> in DSP. Can anyone throw some light on that?
> Is it efficient to implement the LAPACK routines in DSP or is there any
> better techniques?
> Also can anyone tell me the links where I can learn about SVD
> implementations.
@BOOK{meyer,
title = "{Matrix Analysis and Applied Linear Algebra}",
author = "{Carl~D.~Meyer}",
publisher = "Society for Industrial and Applied Mathematics",
year = "2000"}
> Thanks.
> Regards,
> Krish
--
% Randy Yates % "Bird, on the wing,
%% Fuquay-Varina, NC % goes floating by
%%% 919-577-9882 % but there's a teardrop in his eye..."
%%%% <yates@ieee.org> % 'One Summer Dream', *Face The Music*, ELO
http://www.digitalsignallabs.com
Reply by Rune Allnor●April 10, 20092009-04-10
On 10 Apr, 19:46, "krish_dsp" <nmkr...@gmail.com> wrote:
> Hi,
> I am implementing the MUSIC DOA estimator in TI DSP. I am searching for a
> better implementation of Singular value decomposition of a matrix
> algorithm
> in DSP. Can anyone throw some light on that?
Depends on what you mean by 'better'. It's not very
likely you will find faster general-purpose implementations
than LAPACK on general-purpose platforms.
> Is it efficient to implement the LAPACK routines in DSP or is there any
> better techniques?
Again, it depends on what you mean by 'better'. LAPACK
used to be available as FORTRAN source code (and presumably
it still is), so if you find a compatible fortran compiler, you might
be able to compile original LAPACK code for your target
platform. This might be 'more efficient' in terms of man-hours
needed to get the thing up and running. But run-time might
not be optimal, and there are no guarantees you can
compile the source code for your target. Even if you can,
there is the question if the library (the whole LAPACK is
large) fits into the available memory, or if you can identify
and isolate only the parts you need.
On the other hand, hand-crafted assembly code might be
faster and smaller, but might take ages to produce and test.
> Also can anyone tell me the links where I can learn about SVD
> implementations.
The book 'Matrix Computations' by Golub and van Loan
is mandatory. Start with that one, and work out from
the references there.
There was a guy, Steinbeck(?), who used to publish on fast
ESPRIT algorithms some ten years ago. Don't know the
details of those algorithms, though, or if they are convertible
to use with MUSIC.
Rune
Reply by krish_dsp●April 10, 20092009-04-10
Hi,
I am implementing the MUSIC DOA estimator in TI DSP. I am searching for a
better implementation of Singular value decomposition of a matrix
algorithm
in DSP. Can anyone throw some light on that?
Is it efficient to implement the LAPACK routines in DSP or is there any
better techniques?
Also can anyone tell me the links where I can learn about SVD
implementations.
Thanks.
Regards,
Krish