>
>I follow that H1 and H2 are the channel responses to the two input
>channels, and for a comm application X1 and X2 would both be recovered
>diversity inputs of the transmitted y (if that holds for what you're
>doing).   Since X1 and X2 have to be combined coherently in order to
>provide an improved estimate of y (rather than just taking X1 or X2
>independently), this suggests that synchronization is done
>independently on each channel prior to combining.
>
>I think my confusion is that since H1*X1 and H2*X2 have to be
>independently synchronized prior to combining, anyway, is there an
>advantage to making a single EQ work on both simultaneously (which
>seems to be what you're suggesting) or just run an independent EQ on
>each channel (to estimate H1 and H2 independently), and then combine
>the results?
>
>It sounds intriguing and if there's really an advantage to making one
>big MISO EQ rather than two SISO EQs that's something in which I'd be
>interested.
>
>Eric Jacobsen
>Minister of Algorithms
>Abineau Communications
>http://www.ericjacobsen.org
>
>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>
I don't know what you mean by combining X1 and X2 coherently or what you
mean be synchronization prior to combining. What I am doing is taking y,
X1, and X2 and wiener filtering y with X1 and X2. MISO wiener filtering
involves calculation cross correlation coefficients between the inputs. So,
if X1 and X2 are correlated, it is taken into account. If you didn't take
into account the correlation between X1 and X2 and just SISO wiener
filtered y with X1 and SISO filtered y with X2 you would have a big
problem.
y~=H1*X1+H2*X2+residual
X1~=G*X2+residual (since X1 and X2 are correlated)
X2~=H*X1+residual
then
y~=(H1+H2*G)X1+residual
y~=(H2+H1*H)X2+residual
and then SISO wiener filtering y with X1 would give you an estimate of
(H1+H2*G) instead of an estimate of H1
and SISO wiener filtering y with X2 would give you an estimate of
(H2+H1*H)
instead of an estimate of H2.

I believe SISO adaptive filtering y with X1 and SISO adaptive filtering y
with X2 would suffer the same problem if X1 and X2 were correlated.
That is why I want a MISO adaptive filter

>
>I follow that H1 and H2 are the channel responses to the two input
>channels, and for a comm application X1 and X2 would both be recovered
>diversity inputs of the transmitted y (if that holds for what you're
>doing).   Since X1 and X2 have to be combined coherently in order to
>provide an improved estimate of y (rather than just taking X1 or X2
>independently), this suggests that synchronization is done
>independently on each channel prior to combining.
>
>I think my confusion is that since H1*X1 and H2*X2 have to be
>independently synchronized prior to combining, anyway, is there an
>advantage to making a single EQ work on both simultaneously (which
>seems to be what you're suggesting) or just run an independent EQ on
>each channel (to estimate H1 and H2 independently), and then combine
>the results?
>
>It sounds intriguing and if there's really an advantage to making one
>big MISO EQ rather than two SISO EQs that's something in which I'd be
>interested.
>
>Eric Jacobsen
>Minister of Algorithms
>Abineau Communications
>http://www.ericjacobsen.org
>
>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>
I don't know what you mean by combining X1 and X2 coherently or what you
mean be synchronization prior to combining. What I am doing is taking y,
X1, and X2 and wiener filtering y with X1 and X2. MISO wiener filtering
involves calculation cross correlation coefficients between the inputs. So,
if X1 and X2 are correlated, it is taken into account. If you didn't take
into account the correlation between X1 and X2 and just SISO wiener
filtered y with X1 and SISO filtered y with X2 you would have a big
problem.
y~=H1*X1+H2*X2+residual
X1~=G*X2+residual (since X1 and X2 are correlated)
X2~=H*X1+residual
then
y~=(H1+H2*G)X1+residual
y~=(H2+H1*H)X2+residual
and then SISO wiener filtering y with X1 would give you an estimate of
(H1+H2*G) instead of an estimate of H1
and SISO wiener filtering y with X2 would give you an estimate of
(H2+H1*H)
instead of an estimate of H2.

I believe SISO adaptive filtering y with X1 and SISO adaptive filtering y
with X2 would suffer the same problem if X1 and X2 were correlated.
That is why I want a MISO adaptive filter

On Fri, 15 Aug 2008 16:19:28 -0500, "munchkin"
<munchkin@its.caltech.edu> wrote:

>>Gotta admit, you guys lost me.
>>
>>I gotta ask, what is the MISO filter doing?   Is the idea that the
>>coherent combining is done in this filter?   Is it just adapting to
>>each channel independently?
>>
>>Eric Jacobsen
>>Minister of Algorithms
>>Abineau Communications
>>http://www.ericjacobsen.org
>>
>>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>>
>
>So, I'm new to DSP but I'll try to explain myself.
>I'm currently MISO wiener filtering a signal y where y is correlated with
>two or more signals X1,X2,...
>   y~= H1*X1+H2*X2+....+residual
>(* represents convolution)
>My understanding is that miso wiener filtering y with X1, X2,... will give
>me FIR approximations of H1, H2,...
>However, these FIR filter coefficients need to be updated every few
>minutes because the environment is not strictly static.
>Instead of taking the processing time to recalculate the filters
>H1,H2,...
>I would like to use adaptive filtering to update H1,H2,...
>Does that make sense?

I follow that H1 and H2 are the channel responses to the two input
channels, and for a comm application X1 and X2 would both be recovered
diversity inputs of the transmitted y (if that holds for what you're
doing).   Since X1 and X2 have to be combined coherently in order to
provide an improved estimate of y (rather than just taking X1 or X2
independently), this suggests that synchronization is done
independently on each channel prior to combining.

I think my confusion is that since H1*X1 and H2*X2 have to be
independently synchronized prior to combining, anyway, is there an
advantage to making a single EQ work on both simultaneously (which
seems to be what you're suggesting) or just run an independent EQ on
each channel (to estimate H1 and H2 independently), and then combine
the results?

It sounds intriguing and if there's really an advantage to making one
big MISO EQ rather than two SISO EQs that's something in which I'd be
interested.

Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php

>Gotta admit, you guys lost me.
>
>I gotta ask, what is the MISO filter doing?   Is the idea that the
>coherent combining is done in this filter?   Is it just adapting to
>each channel independently?
>
>Eric Jacobsen
>Minister of Algorithms
>Abineau Communications
>http://www.ericjacobsen.org
>
>Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
>

So, I'm new to DSP but I'll try to explain myself.
I'm currently MISO wiener filtering a signal y where y is correlated with
two or more signals X1,X2,...
   y~= H1*X1+H2*X2+....+residual
(* represents convolution)
My understanding is that miso wiener filtering y with X1, X2,... will give
me FIR approximations of H1, H2,...
However, these FIR filter coefficients need to be updated every few
minutes because the environment is not strictly static.
Instead of taking the processing time to recalculate the filters
H1,H2,...
I would like to use adaptive filtering to update H1,H2,...
Does that make sense?

On Fri, 15 Aug 2008 13:03:40 -0700 (PDT), julius <juliusk@gmail.com>
wrote:

>On Aug 15, 2:44 pm, "munchkin" <munch...@its.caltech.edu> wrote:
>> Can you recommend any literature?
>
>For LMS / RLS I like Munson Hayes' book:
>
>  M. Hayes, "Digital Signal Processing and Modeling," Wiley 1996.
>
>It comes with MATLAB scripts for the basic algorithms, but almost
>all of then is SISO.  Adapting to MISO just to get some functionally
>correct is not that hard.

Gotta admit, you guys lost me.

I gotta ask, what is the MISO filter doing?   Is the idea that the
coherent combining is done in this filter?   Is it just adapting to
each channel independently?

Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php

On Aug 15, 2:44 pm, "munchkin" <munch...@its.caltech.edu> wrote:
> Can you recommend any literature?

For LMS / RLS I like Munson Hayes' book:

  M. Hayes, "Digital Signal Processing and Modeling," Wiley 1996.

It comes with MATLAB scripts for the basic algorithms, but almost
all of then is SISO.  Adapting to MISO just to get some functionally
correct is not that hard.