Hello,
I've solved my problem.
It was a bug with the DAW I'm using (Ardour) which doesn't deal well with
multichannel imported tracks : While playing, all the channels were a copy of
the first one, so the HOA encoding was obviously wrong.
Thank you anyway,
Pierre
> Hello everyone,
>
> I'm trying to implement an HOA decoder for spherical microphones and I'm
> getting some numerical issues :
>
> The principle is the following : I have the temporal signals of all the
> capsules of the microphone : it's temporal vector of signals, let's call it
> "I" (size n x 1)
>
> I have to multiply this vector by a Matrix "M" (size m x n) to obtain my HOA
> components.
>
> So I have to do this operation at each discrete time t :
>
> B = M.I
>
> I've used the solution proposed by Aaron Heller to implement the matrix
> multiplication :
>
> // bus with gains
> gain(c) = R(c) with {
> R((c,cl)) = R(c),R(cl);
> R(1) = _;
> R(0) = !;
> //R(0) = !:0; // if you need to preserve the number of outputs
> R(float(0)) = R(0);
> R(float(1)) = R(1);
> R(c) = *(c);
> };
>
> // n = number of inputs
> // m = number of output
> matrix(n,m) = par(i,n,_) <: par(i,m,gain(a(i)):>_);
>
> a(0) = (0.5, -0.5, 1.1, 1.2,0);
> a(1) = (1,2,3,4,-1);
> a(2) = (0.5,0.5,0,0.25,1/3);
> a(3) = (0,0,0,.1, sqrt(2));
>
> process = matrix(5,4);
>
> It's working but I have got some unexplained numerical problems :
>
> In my matrix "M", the sum of the components of each rows = 0 except the
> first row. So if all inputs signals are exactly the same, the outputs
> will be 0 except the first output.
>
> But, if the signals are not the same, I should have m outputs signals not
> null. But I don't when I'm playing my input wav : only the first output is
> not null !
>
> So I've tried the following :
>
> I take a constant signal as input signal : the n samples at discrete time t
> (extract from my input wav signal)
>
> matrix(n,m)=
> (0.143993,0.147716,0.118169,0.141023,0.155442,0.112583,0.166597,0.178432,0.1
> 35545,0.186261,0.181142,0.168309,0.137996,0.14302,0.139751,0.157548,0.159706
> ,0.133034,0.168049,0.124931,0.160939,0.161888,0.146289,0.140364,0.188028,0.1
> 68121,0.160577,0.152144,0.156805,0.145675,0.196994,0.134682)<:
> par(i,m,gain(a(i)):>_);
>
> I measure the output and I've got something everywhere ! But when I play the
> wav entirely, no outputs except the first one !
>
> I've implemented this matrix operation in PureData (without the use of
> Faust) and it's working when I'm playing the input wav (the M matrix is
> exactly the same)
>
> My observations are the following with my Faust code : when the inputs
> signals are slightly different, I've got output signal everywhere, but when
> they are not exactly identical but close enough, I just have the first
> output
>
>
> Here is my Faust code for a M = 3*32 matrix :
>
> import("math.lib");
> import("music.lib");
>
> // bus with gains
> gain(c) = R(c) with {
> R((c,cl)) = R(c),R(cl);
> R(1) = _;
> R(0) = !;
> //R(0) = !:0; // if you need to preserve the number of outputs
> R(float(0)) = R(0);
> R(float(1)) = R(1);
> R(c) = *(c);
> };
>
> //vmeter(x) = attach(x, envelop(x) : vbargraph("[unit:dB]", -70,
> 10));
> //envelop = abs : max(db2linear(-70)) : linear2db :
> min(10) : max ~
> -(80.0/SR);
>
> vmeter(x) = attach(x, envelop(x) : vbargraph("[unit:dB]", -100,
> 10));
> envelop = abs : max(db2linear(-100)) : linear2db :
> min(10) : max ~
> -(80.0/SR);
>
> // n = number of inputs
> // m = number of output
>
> matrix(n,m) = hgroup("Inputs",par(i,n,vgroup("%i",
> vmeter)))<:par(i,m,gain(a(i)):>hgroup("Outputs",vgroup("%i",vmeter)));
>
> //matrixtest(n,m) = par(i,n,1) <: par(i,m,gain(a(i)):>_);
>
> //matrixtest2(n,m)=
> (0.143993,0.147716,0.118169,0.141023,0.155442,0.112583,0.166597,0.178432,0.1
> 35545,0.186261,0.181142,0.168309,0.137996,0.14302,0.139751,0.157548,0.159706
> ,0.133034,0.168049,0.124931,0.160939,0.161888,0.146289,0.140364,0.188028,0.1
> 68121,0.160577,0.152144,0.156805,0.145675,0.196994,0.134682)<:
> par(i,m,gain(a(i)):>_);
>
> a(0)=(0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.0312
> 5,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.
> 03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.03125,0.0312
> 5,0.03125,0.03125,0.03125,0.03125);
>
> a(1)=(0.0505636,0.0460428,0.0505636,0.0460428,0.028456,0.03125,0.0193136,0.0
> 3125,0.028456,0.03125,0.0193136,0.03125,0,0,0,0,-0.0505636,-0.0460428,-0.050
> 5636,-0.0460428,-0.028456,-0.03125,-0.0193136,-0.03125,-0.028456,-0.03125,-0
> .0193136,-0.03125,0,0,0,0);
>
> a(2)=(7.53456*10^-10,0.028456,0,-0.028456,0,0.03125,0.0505636,0.03125,0,-0.0
> 3125,-0.0505636,-0.03125,0.0193136,0.0460428,0.0460428,0.0193136,7.53456*10^
> -10,-0.028456,7.53456*10^-10,0.028456,0,-0.03125,-0.0505636,-0.03125,0,0.031
> 25,0.0505636,0.03125,-0.0193136,-0.0460428,-0.0460428,-0.0193136);
>
> process = matrix(32,25);
> //process=matrixtest2(32,25);
>
> Thanks for your help
>
> Pierre Lecomte
> PhD student
------------------------------------------------------------------------------
Comprehensive Server Monitoring with Site24x7.
Monitor 10 servers for $9/Month.
Get alerted through email, SMS, voice calls or mobile push notifications.
Take corrective actions from your mobile device.
http://p.sf.net/sfu/Zoho
_______________________________________________
Faudiostream-users mailing list
Faudiostream-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/faudiostream-users
