Trinaural decoding equations for 3 speaker stereo matrix ?

[Elias]

Originally Posted by markus76

Interesting observation: the center doesn't seem to be particularly stable with head movements. The stability of phantom sources don't seem to benefit from an additional speaker.

What coefficients have you tried ? Sounds like totally opposite to my findings.

[markus76]

Originally Posted by Elias

What coefficients have you tried ? Sounds like totally opposite to my findings.

Well, phantom center sources don't "snap" to the left or the right speaker when moving like they do with normal 2-speaker stereo. But the upmixed center speaker creates a considerable amount of ambiguity when moving.

[Elias]

Originally Posted by markus76

Well, phantom center sources don't "snap" to the left or the right speaker when moving like they do with normal 2-speaker stereo. But the upmixed center speaker creates a considerable amount of ambiguity when moving.

You should try coefficient of 0.5. It maximises the separation between speakers. Did I suggest it already ?

If you are after the "snap" effect, that is the image locked into the speaker, then you will need much more directional speakers than FRS8.

On the other hand I find the "snap effect" one of the worst features (symptoms) of conventional 2 speaker stereo. It ruins the the rest of the little realism that would possibly have been there in ideal conditions


- Elias

[markus76]

Originally Posted by Elias

You should try coefficient of 0.5. It maximises the separation between speakers. Did I suggest it already ?

0.5 is the default in my script...

Originally Posted by Elias

If you are after the "snap" effect, that is the image locked into the speaker, then you will need much more directional speakers than FRS8.

On the other hand I find the "snap effect" one of the worst features (symptoms) of conventional 2 speaker stereo. It ruins the the rest of the little realism that would possibly have been there in ideal conditions


- Elias

I'm not after the "snap" effect. But in my current configuration it sounds more natural despite the spatial distortion. Pick your poison.

You know that Gerzon's techniques are based on assumptions that not everyone agrees with. See papers by Theile.

[Elias]

You might also try placing the 3 speakers in line instead of arch.

In my FDTD simulations it produces better wavefront for both center and off-center listeners:
Elias Pekonen Home Page - FDTD simulations of stereo speaker configurations in small room








- Elias

[markus76]

Did you try placing L and R directly at the wall? The opening angle from the listening position should be about 45°. Adjust toe-in of speakers. This will influence the level of the contralateral reflections. How does that sound to you?

[Elias]

Yes I did try speakers at the side walls many years ago. All the sound was at the side and nothing much at the center. So it was not good at all. It was about the same time I understood there must be a speaker at the center to be able to generate stable center images. It also eventually lead me to my single speaker stereo SSS concept.


- Elias

[markus76]

Originally Posted by Elias

Yes I did try speakers at the side walls many years ago. All the sound was at the side and nothing much at the center. So it was not good at all.

You have to get the toe-in right then "everything falls into place". Of course this is just a one seat solution but it works remarkably well.

[markus76]

Originally Posted by Elias

You might also try placing the 3 speakers in line instead of arch.

Gerzon states in his 1991 AES convention paper #3180 that "loudspeakers [that] lie along a straight line [...] generally give[s] less good results".

[markus76]

Here's my implementation of a energy-preserving 3x2 matrix decoder:

//=================================================
desc: Energy-preserving 3x2 Matrix Decoder //based on Gerzon


slider1:35.26<-0,90,0.01>Angle parameter (0-90)

//=================================================
@slider
angle = slider1*0.017453;

//=================================================
@sample
l_in = spl0; r_in = spl1;


//create MS from Left/Right
m = 0.7071*(l_in+r_in);
s = 0.7071*(l_in-r_in);


//Energy Preservation
m_c = m*cos(angle);
m_lr = m*sin(angle);


//Re-matrix to Left/Right
l_out = 0.7071*(m_lr+s);
r_out = 0.7071*(m_lr-s);
c_out = 0.7071*m_c;

//output
spl0=l_out; // Left
spl1=r_out; // Right
spl2=c_out; // Center

//=================================================

I'm not sure if the mid signal for the center (after energy preservation) needs to be attenuated by 3dB or not. Anybody?

Schematic: