I am sorry for the double-posts - I don't know why this happens (perhaps a
little problem with my firefox?)

>In cases where signals are passed through multiple filters, I have
>achieved better results by combining as many filters as possible into
>one overall filter response, and letting FDLS approximate that, than
>by using FDLS to approximate each individual filter.
>
>Greg

I think this will be a good idea, otherwise there could be a lot of
problems designing a good phase response. Right now, I have to finish my
diploma, but I will come back to this in january.

One more question about your FDLS-algorithm: You say that the
pseudoinverse is (X^T*X)^{-1}*X^T. Is it save the say that the columns of
X are always independent? Otherwise it would become a little bit more
complicated. It seems to me that this should be the cast nearly always,
but I cannot figure out why exactly? 

On Oct 16, 8:46 am, "Narax" <soenke.trein...@unibw.de> wrote:

> I finished the filter so far. I still have
> to design the lowpass fiter, but I will realise this later.

In cases where signals are passed through multiple filters, I have
achieved better results by combining as many filters as possible into
one overall filter response, and letting FDLS approximate that, than
by using FDLS to approximate each individual filter.

> This board has been a great help for me :-)

Good people at comp.dsp.  Lately a little more noise than usual, but
still plenty of signal left.

Greg

>A little delay adds diversity to the phase response. Gives the least
>squares algorithm a little more to work with, resulting in improved
>convergence.
>
>In the article I mention the limiting case of this -- if the phase
>response of the prototype is an odd multiple of 90� at half the
sampling
>frequency, then the elements of the matrices corresponding to the
>sampled output waveform will be zeroes because the samples occur at the
>zero-crossings of the sine wave. You'll be trying to find the inverse
>of a matrix in its null-space; Ax=0. Adding a little delay to the
>system mitigates the problem.
>
>Greg


Thank you for the explanation. I finished the filter so far. I still have
to design the lowpass fiter, but I will realise this later.

This board has been a great help for me :-)

>A little delay adds diversity to the phase response. Gives the least
>squares algorithm a little more to work with, resulting in improved
>convergence.
>
>In the article I mention the limiting case of this -- if the phase
>response of the prototype is an odd multiple of 90� at half the
sampling
>frequency, then the elements of the matrices corresponding to the
>sampled output waveform will be zeroes because the samples occur at the
>zero-crossings of the sine wave. You'll be trying to find the inverse
>of a matrix in its null-space; Ax=0. Adding a little delay to the
>system mitigates the problem.
>
>Greg


Thank you for the explanation. I finished the filter so far. I still have
to design the lowpass fiter, but I will realise this later.

This board has been a great help for me :-)

Narax wrote:

   ...

> Things are beginning to work fine - I just now took a look at the results
> from the FDLS algorithm and they are good. As always I have another
> question: What is the artificial delay good for? I know that things don't
> work fine without it, but why?

How does the frequency-domain Least squares algorithm enter the picture? 
I don't know which delay you mean, but one commonly adds a delay to 
match an unavoidable delay in a parallel part of a process.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

>The deemphasis filter in the receiver is a simple RC. The preemphasis 
>filter should complement it, and the complement is also a simple RC. Why

>do you want a different response?
>
>Jerry

My mistake. I don't want another response.

Things are beginning to work fine - I just now took a look at the results
from the FDLS algorithm and they are good. As always I have another
question: What is the artificial delay good for? I know that things don't
work fine without it, but why?

Narax wrote:
> Sorry, I accidentally hit the "Send Message!" button.
> 
>> Notice that the pre-emphasis has about 15 dB or more boost at high
>> frequencies, make sure that this does not casue OVEVERMODULATION to
>> your modulator.
>>
>> Mark
> 
> Thank you for this hint - I didn't take this in concideration till now and
> I think this is the first thing the check out.
> 
> 
> One more question about the RC-curcuit: You said that I should take its
> phase response. But a simple RC-curcuit has a diffrent magnitude response
> than I want to achieve. Is a mix of the desired magnitude response and the
> RC-circuit's phase response a good design?

The deemphasis filter in the receiver is a simple RC. The preemphasis 
filter should complement it, and the complement is also a simple RC. Why 
do you want a different response?

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Narax wrote:
>> IIRC, on-air FM is only specified up to 15 KHz. Especially with
>> preemphasis, the modulation index gets too high above that. There were
>> few sources that could reach that high when the standard was first laid
>> down.
>>
>> Jerry
> 
> That is the case. So I don't have to worry about the last 5kHz. I have to
> cut above 15kHz anyway, because I need to secure the pilot tone at 19kHz
> which indicates a stereo transmission. For this reason there has to be a
> lowpass and I think I forgot about its phase response. I should be minimal
> phase as well when I correct?

You want the pilot-carrier blocking filter's influence on the phase of 
the passband to be as small as possible. A symmetric FIR is best. 
Remember that stereo is backwardly compatible with the older mono FM.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

On Oct 10, 2:31 am, "Narax" <soenke.trein...@unibw.de> wrote:

> But a simple RC-curcuit has a diffrent magnitude response
> than I want to achieve. Is a mix of the desired magnitude response and the
> RC-circuit's phase response a good design?

Not just "any" RC circuit; use the magnitude and phase responses from
the RC circuit that implements the preemphasis filter that you want.

Greg

Sorry, I accidentally hit the "Send Message!" button.

>Notice that the pre-emphasis has about 15 dB or more boost at high
>frequencies, make sure that this does not casue OVEVERMODULATION to
>your modulator.
>
>Mark

Thank you for this hint - I didn't take this in concideration till now and
I think this is the first thing the check out.


One more question about the RC-curcuit: You said that I should take its
phase response. But a simple RC-curcuit has a diffrent magnitude response
than I want to achieve. Is a mix of the desired magnitude response and the
RC-circuit's phase response a good design?