Possible bugs in roomsim.c

[kaimex]

Hi, I have a question regarding code in roomsim.c :

		for (j=0; j<Nout; j++)
		{
			r = sqrt(1.0 - alpha[j]);
			/* r = 1.0 - alpha[j]; */
			pSimulation->logreflection        [i*Nout + j] = LOGDOMAIN(r);
			LOGLIMIT(pSimulation->logreflection        [i*Nout + j]);

			pSimulation->logabsorption        [i*Nout + j] = LOGDOMAIN(1.0 - r);
			LOGLIMIT(pSimulation->logabsorption        [i*Nout + j]);

			pSimulation->logdiffusion         [i*Nout + j] = LOGDOMAIN(d[j]);
			LOGLIMIT(pSimulation->logdiffusion         [i*Nout + j]);

			pSimulation->logspecularreflection[i*Nout + j] = LOGDOMAIN((1.0 - d[j]) * r);
			LOGLIMIT(pSimulation->logspecularreflection[i*Nout + j]);

			pSimulation->logdiffusereflection [i*Nout + j] = LOGDOMAIN(d[j] * r);
			LOGLIMIT(pSimulation->logdiffusereflection [i*Nout + j]);

			pSimulation->diffusioncoefficient [i*Nout + j] = d[j];
		}

I wonder, whether the line ending with "LOGDOMAIN(1.0-r);" is correct.
Before the initial change from r=1.0-alpha[j] to r=sqrt(1.0-alpha[j]) this was the same as LOGDOMAIN(alpha[j]).
After the change "1-r" is no longer equal to absorption, neither intensity/power-absorption nor wave-amplitude absorption.
My understanding is, that, depending on what logabsorption is meant to be (absorption of power or wave-amplitude) the right hand side should be LOGDOMAIN(alpha[j]) (for power-absorption) or 0.5*LOGDOMAIN(alpha[j]) (for wave-amplitude absorption).

2nd question: The absorption values in sampleroomsetup.m are understood as values of intensity absorption. I have not found a statement about the diffusion values. One would assume, that they describe diffusion of intensity. In the above code excerpt they are, however, used as factors for wave-amplitude reflection ((1.0-d[j]*r) and d[j]*r). So there they should be diffusion values for wave-amplitudes, not intensities. What is intended/correct ?

3rd issue: Some manufacturers of sound absorption materials specify NRC values >1. If such values are entered in the absorption specs of roomsetup.m sqrt(1-alpha) produces NaN results.

4th question: I wonder why nobody considers negative reflection values.

[robaru]

Hi Kaimex,
Thanks for going this deep in the toolbox and for opening this issue.

I believe that we inherited this part of roomsim.c from the original toolbox of Schimmel et al. 2009. I would need to dig a bit into this issue before making a conclusion and I have little bandwidth at the moment.

I will keep this problem in mind and try to address it. Meanwhile, if you could help by providing any reference or insight, it would be awesome.

[kaimex]

Hi robaru,
thank you for your positive response.
It is indeed a heritage from Schimmel et al. and possibly even from his predecessor Campbell et al..
In the meantime I have succeeded to recompile Schimmel's RoomSim with a modified version of roomsim.c with the alpha values limited to 1 ( a=alpha[j]; if (a>1) a=1; ) and logabsorption changed to 0.5*LOGDOMAIN(a) (i.e. = LOGDOMAIN(sqrt(a)), but to my astonishment the resulting impulse response is exactly the same. Looks as if the logabsorption values are not used for the actual calculation. I must confess, however, that I have not yet looked thoroughly through the subsequent processing, so I do not yet understand how the algorithm proceeds.

I have a further issue: I am somewhat disappointed, that apparently the incorporation of Peterson's interpolation scheme, that was already incorporated in Campbell's version, has been dropped by Schimmel and also most versions that build on his publication.
I have more an electromagnetic background than in acoustics, so I find it very strange/unbelievable, when resulting impulse responses have only positive values and reflection values are assumed to be positive real numbers and when the material values used as input are possibly a mixture of intensity and wave-amplitude properties or when intensity values are used mixed with amplitude values (as in diffusion*reflection after the change from r=1-alpha (r: intensity reflection) to r=sqrt(1-alpha) (r: amplitude reflection with ignored sign; alpha: intensity absorption). The difference between intensity versus amplitude properties has not been emphasized in any of the accompanying texts. The absorption values are synonymous for NRC values and thus clearly meant to be intensity properties. Nothing specific is said about the diffusion values. Only positive numbers appear. This nourishes the assumption that they are also meant to hold for intensity. But in the code of roomsim.c they are used as factors for r which, after the change to r=sqrt(1-alpha), is an amplitude property (which might have a negative sign or even a complex value, imho).
Regards
Kai

[kaimex]

Looking through the code, I stumbled upon another questionable handling of phases:
In sensor.c all cases of source.description and receiver.description in sampleroomsetup.m ( as e.g. omnidirectional, bidirectional, cardioid or dipole etc.) are handled by taking the log(abs()) of the directivity function. This results in transforming a dipole characteristic into a another bi-directional sensitivity function instead of keeping the negative branch of the figure 8 function.
Think of a pipe shaped 'room' with fully reflecting walls except one far open end: Now place a dipole source a quarter wave in front of the fully reflecting end wall of the 'pipe' with the +branch of the '8' directed to the open end and the -branch to the reflecting end. Contrary to reality a Null in the frequency response is produced at the quarter wave frequency instead of some enhancement.