On Fri, 29 Apr 2005, Gie78 wrote:

> Thanks for your prompt reply. When I insert an delay into the primary
> path the result would be much better. OK I think this problem is known
> as "causality problem".

Yes, you should have some time advanced signal to predict the "anti-noise"
to get good results. But if you have a feedback system, then you don't
have that predicion since you don't have a delay in the primary path which
is the case in the feedforward case. In the feedback case you have to rely
on the predictivity in the signal itself.  I hope you understood my
explanation.... :)


> And the crosscorrelation between the disturbance signal and the
> filterd-x-signal how should these crosscorrelation look like?
> I think the two signals should be well correlated for a good result,
> or?
> But could you explain me why?

I'm not shure if you are simulating a feedforward or feedback case but:
And I'm not shure if I can explain you exactly why (in a short way), but
it can be shown, for the feed-forward case, that

--------------------------------------------
See_min = Sdd - Srd./Srr

where,
See_min is the power spectral density of the minimum error signal,

Sdd is the fouriertransform of the autocorrelation of the disturbance
signal (not to be mixed with the reference signal x),

Srd is the fouriertransform of the crosscorrelation of the filtered x
signal and the disturbance signal,

Srr is the fouriertransform of the autocorrelation function of the
filtered x signal.

r = x*g, where x is the reference signal, g is the sec. path, * is conv.
operator
d = x*p, where p is the primary path

In the feedback case the minimum error should be:

See_min = Sdd - Sr'd./Sr'r'

where Sr'd is the crosscorrelation between the filtered x signal and the
disturbance signal,

Sr'r' is the autocorrelation for the filtered x signal

r' = d*g
-----------------------------------------------


I hope this explained a little, and someone correct me if I'm wrong. :)

Sverre

Thanks for your prompt reply. When I insert an delay into the primary
path the result would be much better. OK I think this problem is known
as "causality problem".
And the crosscorrelation between the disturbance signal and the
filterd-x-signal how should these crosscorrelation look like?
I think the two signals should be well correlated for a good result,
or?
But could you explain me why?

Thanks