I'm puzzled by this line: "My issue with SQ is the heavy bias of frontal left / right information making the detail of the rear information hidden within a few dB."
If the Involve decoder doesn't use pairs of 0 and 90-degree phase shifters, no L/R separation in the two rear channels can be extracted from (ideal) SQ at all. The standard 4 -> 2 CBS encoders had some phase drift in the sidewall locations, which could translate into some L/R separation for a QS/EV4 style of decoder (which don't rely on pairs of 0 and 90-degree phase shifters to extract F/R information). An ideal SQ encoder, though, will give zero L/R separation in the rear channels for a QS/EV4 style of decoder.
On the other hand, if the decoder is an actual SQ decoder with the mandatory pairs of 0 and 90-degree phase shifters, and uses full-logic gain-riding, TATE cancellation methods, or a Shadow Vector dynamic matrix, there's 25 to 35 dB of separation in the rear channels ... from each other, from the adjacent front channel, or from the opposing front channel. I've measured it myself, it's not an illusion. Separation in the SQ system is mostly limited by the precision of the phase shifters; with each degree of departure from 90 degrees eating up several dB. If the phase shifters are accurate to 1 degree, 40 dB is available in all the cardinal points.
The Shadow Vector retains this separation as the signal is swept across the intermediate locations; I don't know if the TATE does this or not, since I can't unravel the math in the TATE patent. I think it might not, but I don't really know.
I think the shared CB localization in both QS and SQ can give the illusion the two systems are partially compatible. Center Back is actually the only place in the back quadrant of the Scheiber sphere where the two systems overlap. From the point of view of an ideal QS or EV4 decoder, the SQ Left Back and Right Back encodings are in never-never land, with 90-degree phase shifts between the channels. A QS or EV4 decoder, whether dynamic or not, interprets either condition as a 4-speaker sum, with no localization at all, and an odd phase relationship between the 4 channels.
The best description of SQ would be a "phase matrix", and in a deeper sense than QS or EV4. In QS or EV4, the Z axis is represented by the proportion of in-phase to out-of-phase content. The 90-degree phase shifters on the encode end of QS let the recording engineer make an overhead pan from front to back with no sudden jumps in phase. (EV4 does not use 90-degree phase shifters for either encode or decode.)
SQ uses phases all the way from -90 degrees to 180 degrees to +90 degrees to accomplish a pan from LB -> CB -> RB. Remarkably, as this pan happens, there is NO change in L/R ratios; both channels stay exactly equal throughout the pan. A pan across one of the sidewalls, from back to front, gradually reduces the level of the opposing channel to zero (the phase angle no longer has meaning when only one channel is present). A pan across the front is simply a conventional L/R pan in stereo, with the phase angle between the two channels now at 0 degrees.
I agree, this is a screwy matrix, when you look at all the crazy things happening in the sidewalls and rear. That's why Shadow Vector is several times as complex as a Vario-Matrix, but ultimately similar in operation. However, Vario-Matrix could get away with opto-couplers doing the blend functions, but this is too inaccurate for Shadow Vector. Precision VCA's are required to get the phase angles where they should be. I wasn't thrilled with Cliff Moulton designing switching FETs to do the VCA function, but he wanted reasonable distortion combined with precision and speed, which can't be done with opto-couplers. No problem in the digital era, so long as 32-bit precision, combined with independent dither for each VCA, is employed.
But ... if the Involve decoder doesn't have the required set of four allpass phase shifters, with a precision of 3 degrees or better across the audio band, the SQ decoding will be an artifact of encoder errors, which I grant are abundant in the standard CBS encoder. The standard 4 -> 2 CBS encoder isn't "wrong" per se, it just mis-encodes the sidewall signals. The corners are exact, and a pan across the rear is accurate. The "forward-looking" SQ encoder (widely used by EMI) was quite a bit more accurate, the Ghent microphone pretty much spot-on, and the "Position Encoder" right on the money, although it couldn't do a smooth pan, just click-stops at certain locations.
I'll be honest; although the control signals in a Shadow Vector can be skewed to offset the standard encoder errors, I never did this. First, too damn complex when you do this in an all-analog environment, and also, the decoding was accurate enough so the skew was audible for what it was, simply a mis-location of the signal as it traversed the sidewalls. In that sense, it was a good tool for the studio, since you could easily hear when the decoder was mis-locating things a little bit. If the encoding was 100% accurate, the signal stayed pinned to the sidewall, and if the decoder was off a bit, the signal drifted into the room (a little) as it was panned from front to back. I think Malcolm has probably noticed this with his prototype; you can hear the encoder errors as slight mislocations during pans.
I also admit I never built a multi-system SQ/QS version of the Shadow Vector. If it had gone to production, absolutely, QS would have been in there. In QS mode, the -90/+90 sensing axis is not used, the signals going into the VCA's are simpler, and only a pair of VCA's are needed for each decoded channel. The extra precision of VCA's, combined with higher speed, would give a more precise and stable image than the Vario-Matrix.
