>>>>> "gm" == gREG mCcOURT  writes:

 gm> Regarding impulse responses for use with convolve:

 gm> Q.1  What is the best source for an impulse signal when taking
 gm> the impulse response of:

 gm>      a)  An electronic device (like a reverb unit, etc)?

 gm>      b)  An natural acoustic space (like a cave, concert hall,
 gm>          etc)?

The MLSSA system uses an MLS (Maximum Length Sequence) as a source of
signal. The great thing about this signal is that you can get long
sequences of relatively random but allways present changes and that
when doing correlation with the incoming signal you have reduced the
multiplication part into a simple sign change since the MLS signal is
a digital signal.

I have used this system to measure both electrical systems, speakers,
PA rigs as well as troubled halls and arenas. Once you got the impulse
response you may cut out the room reflexes and by this create a
simulated dead room, which is usefull for speaker tests.

There are some pretty good AES articles about it.

 gm> Q.2  If an impulse signal x[n] is put into a system (electronic
 gm> device or natural acoustic space) and the measured (recorded)
 gm> output y[n] is the convolution of the impulse x[n] with the
 gm> response of the system, then doesn't y[n] contain the spectral
 gm> elements of the initial impulse x[n] as well?  Wouldn't another
 gm> signal p[n] when convolved with the y[n] impulse response also be
 gm> indirectly convolved with the initial impulse signal x[n]?

 gm> If this is so, wouldn't the resulting impulse response measurment
 gm> need some kind of de-convolution in order to seperate out the
 gm> excitation signal from the true impulse response of the system,
 gm> the later being what is kept and used to convolve other signals
 gm> with?

 gm> I am looking at some theory lit. on homomorphic filters for
 gm> cepstrum analysis and deconvolution and the above thoughts
 gm> occured to me.  There is an example of application to vocal
 gm> sounds where homomorphic processing is used to seperate out the
 gm> vocal cord excitation signal from the impulse response of the
 gm> vocal tract.  Isn't this the same situation as the one above? 
 gm> Anyone have any enlightning comments?

When you measure an system with an signal x[n] and get y[n] as output
you then correlate this into h[n]. In this process you remove the
properties of x[n] and get the impulse response of the measure
system. If a extra signal (say hum) gets into y[n] as well then this
will not correlate with the x[n] signal and it's effect may be reduced
by a longer sequence of x[n] and y[n] (assuming that x[n] does not
have a common frequence with the extra signal in y[n]).

Cheers,
Magnus