room impulse response and deconvolution

[Magnus Danielson]

From: Martin Czech <martin.czech at intermetall.de>
Subject: Re: room impulse response and deconvolution
Date: Mon, 24 Jan 2000 12:01:29 +0100 (MET)

> :::When I use this I use a signal which autocorrelation is a single pulse, thus
> :::basically acting in for the dirac delta. Now, knowing the autocorrelation is
> :::so nice to us, we can use the cross-correlation of the signal source and the
> :::resulting signal from the output of the linear system. A room can be
> 
> 
> SNIPPED
> 
> Ok, just to be sure:
> 
> On one hand you have the balloon sequence, B.
> Your hope is that the autocorrelation B##B is a unit inpulse.

Right. Or rather, I assumed for the sake of the argument. I do not have a
sample of a poping balloon and no balloons around to actually make a sample.

> On the other hand is the room sequence R, which is 
> the convolution of the ideal room sequence Ri and the balloon
> sequence B, ie.
> 
> R=Ri**B
> 
> you say that if you apply the correlation with B on both sides
> 
> B##R=Ri**B##B
> 
> the autocorrelation of B appears, and this is the unit impulse UI,
> so 
> 
> B##R=Ri**UI=RI
> 
> so the correlation of the Balloon response B with the measured
> room response R is the desired ideal room response Ri.
> All by linearity and time invariance.
> 
> Great!

Yeap. You got me correct allright.

> Unfortunately B looks very strange, this may be due to the microphones.
> It looks like a damped sinusoid, with very low frequency, some 10Hz.
> 
> So B##B will NOT look like UI.

If this is how you claim, then you are right.

> If you are interested, I could send you a gnuplot style data file.

Yes... I would also not mind the sample etc. BTW. I run Linux and has Octave
and Gnuplot which I am know to use for various experiments in time/frequency
analysis ;)

> I guess I need better microphones, phase linear.

So, you don't have that then ... SHAME ON YOU ;) ;)

Hmm... dampend sinus. You could actually filter that property out with a
suitable placement of a pair of zeros, right ;)

> :::Well, diffusion is nice, when you want it. If you want a clearer responce noise
> :::could be an issue.
> 
> It is a noise floor, low level, this will translate into a very low
> diffusion level, and this is not audible, because of masking.
> Convolution means also gating, no signal , no noise.

Well, yes... I was basically refering to when you have a poor S/N ratio so that
the responce noise gets high enougth to make the diffusion really audioble.
Notice that I said "...noise could be and issue".
                             ^^^^^

> And: a sine signal (critical in all other situations) convolved with some
> noisy filter sequence (long enough, so that the noise becomes audibl
> as such) is really a filter, only noise arround the sine will get through,
> but this is masked, also. 

Right.

Cheers,
Magnus