Thanks Fons for you help,

> As you say the matrix inversion amounts to
> 
> H1 =  Hi / D       (1)
> H2 = -Hc / D       (2)
> 
> with D = Hi^2 - Hc^2
> 
> What you write makes me think you are trying to regularise
> this by replacing D by something like
> 
>  E + (1 - E) * D
> 
> ??

Not exactly: I was trying to add the regularization parameter to D, so that
Dregul = Hi^2 - Hc^2 + E

with 0 <= E <= 1

> 
> Then setting E = 1 would mean
> 
> H1 =  Hi
> H2 = -Hc
> 
> and since Hi ~= Hc at VLF, H1 + H2 will be near zero,
> removing the mono bass.
> 
> But this is the wrong way to regularise such an inversion.
> The Hi,Hc matrix is very ill-conditioned at VLF, but it
> still has rank 1, not 0. So you need to regularise only
> in one dimension.
> 
> Starting from (1),(2) we have
> 
> H1 + H2 = (Hi - Hc) / D = 1 / (Hi + Hc)   (3)
> H1 - H2 = (Hi + Hc) / D = 1 / (Hi - Hc)   (4)
> 
> There is no problem with (3) when Hi ~= Hc, only
> with (4), so only that one needs regularisation.
> 

Well, that was my mistake: I was dumbly trying to regularize both inversions 
with the same E value. If I understand well, H1 and H2 (or H1+H2 and H1-H2) 
have to be calculated independently, with different regularization parameter 
values (E(w) being frequency-dependent "alla Farina" to account for 
ill-conditioning in VLF or "notches" in the HRFT).

> Which means that H1 + H2 (the mono response) can just
> be (3), with no loss of bass at all.

Dub lovers say thanks!

Best,
FM
_______________________________________________
Sursound mailing list
[email protected]
https://mail.music.vt.edu/mailman/listinfo/sursound

Reply via email to