Don't know if this info will help or not, but here goes:
For Sansui's QS with basic 2-channel stereo playback, Lt-Rt will have 7.7db of separation. When decoded, it's -3db all around with diagonally opposite channels being -20db or more down as Lucanu's graphic shows. The basic linear QS decode equations are:
Lf = Lt +.414Rt
Rf = Rt +.414Lt
Lb = -J(Lt - .414Rt)
Rb = J(Rt -.414Lt)
For variable matrix decoding, if you add to the sum of the two composite signals 2.41 times the DIFFERENCE between the two signals, then Lf will be totally separated from Rf. To extract Rf, you subtract 2.41 times the difference. If any back sounds are present, they will be increased 2.41 times.
The full equation for Lf extraction is:
(1.414Lf + 1.414Rf) + 2.41(.586Lf - .586Rf) =
1.414Lf +1.414Rf + 1.414Lf - 1.414Rf = 2.82Lf
The output from from the Lt channel now contains only the Lf with the original crosstalk from Rt cancelled out by the 2.41 coefficient.
For Lb extraction, the encoding is: jLb + j.414Rb.
The decoding is:
jLb + J.414Rb + jRb + j.414Lb +j.2.41Lb + jRb - j.2.41Rb - jLb = j2.82Lb
Or, simplified
(Lt - Rt) + 2.41(Lt + Rt)
Phase can be used to sense Front/Back since, in QS, the phase difference between the channels is directly proportional to the F/B ratio. It's 90 degrees when F/B ratio equals 1-to-1 and 180 degrees when all sounds are at back, no fronts.
As I said, I don't know if this info will be helpful or not since I have not tried PC based decoding, but perhaps one of you will be able to adapt it for improved QS decoding results.
Cheers!
Ty C.