Prev: COMP.DSP 2010 update
Next: Normalised Cross-correlation FFT
From: Superfish on 15 Mar 2010 23:05
I'm trying to implement a fast transversal RLS algorithm and want it
to for now simply associate a binary input vector with a desired
value. For example x = [1 0 1 1 0] with d = 10, and then get the
desired value out later by op = x' * w.

I'm able to get this to work easily with the conventional RLS
algorithm, but I cannot get it to work with the FT-RLS algorithm. I
read in a text book that the input vector x for FT-RLS should be a
vector of consecutively delayed samples of the same signal, so is the
FT-RLS algorithm even capable of doing what I want?
From: Jerry Avins on 16 Mar 2010 11:32
Superfish wrote:
> I'm trying to implement a fast transversal RLS algorithm and want it
> to for now simply associate a binary input vector with a desired
> value. For example x = [1 0 1 1 0] with d = 10, and then get the
> desired value out later by op = x' * w.
>
> I'm able to get this to work easily with the conventional RLS
> algorithm, but I cannot get it to work with the FT-RLS algorithm. I
> read in a text book that the input vector x for FT-RLS should be a
> vector of consecutively delayed samples of the same signal, so is the
> FT-RLS algorithm even capable of doing what I want?

Have you read http://www.dechene.ca/papers/report_658b.pdf ?

Jerry
--
Discovery consists of seeing what everybody has seen, and thinking what
nobody has thought. .. Albert Szent-Gyorgi
�����������������������������������������������������������������������
From: Superfish on 16 Mar 2010 18:22
On Mar 17, 4:32 am, Jerry Avins <j...(a)ieee.org> wrote:
> Superfish wrote:
> > I'm trying to implement a fast transversal RLS algorithm and want it
> > to for now simply associate a binary input vector with a desired
> > value. For example x = [1 0 1 1 0] with d = 10, and then get the
> > desired value out later by op = x' * w.
>
> > I'm able to get this to work easily with the conventional RLS
> > algorithm, but I cannot get it to work with the FT-RLS algorithm. I
> > read in a text book that the input vector x for FT-RLS should be a
> > vector of consecutively delayed samples of the same signal, so is the
> > FT-RLS algorithm even capable of doing what I want?
>
> Have you readhttp://www.dechene.ca/papers/report_658b.pdf?
>
> Jerry
> --
> Discovery consists of seeing what everybody has seen, and thinking what
> nobody has thought.    .. Albert Szent-Gyorgi
> ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Actually I used that paper to implement the algorithm. The problem is
that it uses delayed input signals U(n) = [ u(n) u(n-1) ... u(n - M +
1) ]' as the input vector, whereas i'd like an arbitrary binary input
vector to be used. Is there no way to convert this algorithm to do
what I want? If not are there any alternative fast-rls algorithms that
might work?

Thanks
From: Dan Dechene on 7 Apr 2010 13:26
On Mar 16, 6:22 pm, Superfish <projectilef...(a)gmail.com> wrote:
> On Mar 17, 4:32 am, Jerry Avins <j...(a)ieee.org> wrote:
>
>
>
>
>
> > Superfish wrote:
> > > I'm trying to implement a fast transversalRLSalgorithm and want it
> > > to for now simply associate a binary input vector with a desired
> > > value. For example x = [1 0 1 1 0] with d = 10, and then get the
> > > desired value out later by op = x' * w.
>
> > > I'm able to get this to work easily with the conventionalRLS
> > > algorithm, but I cannot get it to work with the FT-RLSalgorithm. I
> > > read in a text book that the input vector x for FT-RLSshould be a
> > > vector of consecutively delayed samples of the same signal, so is the
> > > FT-RLSalgorithm even capable of doing what I want?
>
> > Have you readhttp://www.dechene.ca/papers/report_658b.pdf?
>
> > Jerry
> > --
> > Discovery consists of seeing what everybody has seen, and thinking what
> > nobody has thought.    .. Albert Szent-Gyorgi
> > ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
>
> Actually I used that paper to implement the algorithm. The problem is
> that it uses delayed input signals U(n) = [ u(n) u(n-1) ... u(n - M +
> 1) ]' as the input vector, whereas i'd like an arbitrary binary input
> vector to be used. Is there no way to convert this algorithm to do
> what I want? If not are there any alternative fast-rlsalgorithms that
> might work?
>
> Thanks

Did you end up figuring out a solution?

--Dan
 | 
Pages: 1
Prev: COMP.DSP 2010 update
Next: Normalised Cross-correlation FFT