Thank you for the extensive reply. I always know when I see a new post by you there will be some good meat to chew on.
I do have the Audionics patent you mentioned. I have, say, 18" if I stack them up of patents printed out going back to ~'74. I find it's a good way to find out how something works with out all the fluff of ad hype or dumbed down audio mag reviews.
So, without going downstairs & double check, if IIRC the direction sensing was based on comparing front L/R, rear L/R & center front to center back. I understand each of those can have +- control voltages. Much has been said here about how the Shadow Vector does not favor directions; no speaker hugging effects it is both precise & full at the same time. So what I don't get is how that can be done using what seems to me limited direction sensing. Seems like it should also be sensed on the diagonals since SQ can do diagonal pans also? Or maybe an omni direction where all signals are summed & compared to the corner signals? And is what you did similar to Fosgates's servo feed back logic that also claimed to have accurate total power with no favoring of speaker feeds?
Here's a deep dive into the Shadow Vector directional sensing. There are 6 fullwave precision rectifiers (using op-amps as precision rectifiers) for each cardinal point of SQ ... LF, CF, RF, LB, CB, and RB. There's a bit of lowpass filtering for the output of each fullwave rectifier, followed by 3 differential amplifiers for LF/RF, CF/CB, and LB/RB. There's a single tight AGC loop with at least 40 dB dynamic range wrapped around the array of precision rectifiers, lowpass filters, and differential amplifiers. The logic only relaxes for low-level inputs that approach the noise floor of an LP, around -50 dB or lower.
Following that, the three bidirectional (+/-) control lines go through three identical nonlinear circuits (in practice, an array of resistors and diodes) that shape the control response so the variable-gain amplifiers track the Scheiber sphere precisely. The nonlinear shaping circuit was determined by connecting a pair of analog oscillators 1~5 Hz apart, which generates a spinning-phase signal that goes from LF to CF to RF to CB to LF again. By observing the output of LB and RB on one axis of the scope, and using the LF/RF control line on the other axis, the nonlinear circuit can be trimmed so the each of the back channels precisely follows the ideal SQ pan-locus. LF to CF to RF to CB to LF isn't a standard SQ pan (in fact it is an ideal QS pan), but it calibrates the response of each back channel exactly, so separation is maintained as the signal sweeps across the front, and then smoothly moves back to the default LB or RB position as the signal sweeps across the back quadrant. Just to check, the spinning LF/RF input signal can be re-matrixed into LB/RB, or CF/CB, to confirm the other axis of the decoders have symmetrical responses.
The array of variable-gain amplifiers needed 1% or better accuracy on the gain control. Cliff Moulton made the decision to use high frequency time-slicing on FET switches, and we spent several months chasing out DC offsets and LF bumps. It also made the prototype very susceptible to RFI, which ruined the East Coast debut of the product.
I'm not too sure where feedback would be useful in a system like this, unless there were errors in the gains of the variable-gain amplifiers. The AGC loop on the directional-sensing array can always be improved, of course, but tracking directionality down into the noise isn't a good idea with LP playback. You don't want each little tick and pop to be hard-localized; better to just ignore them. There's also room for improvement in the ballistics of the control lines, initial localizing transients need a faster response than slow sounds, or phase conflicts between instruments. The Shadow Vector mimicked the attack and decay response of Dolby systems, around 1~3 mSec for attack, and 20~50 mSec for release ... this was implemented in the lowpass filters for each of the control lines, with diodes and lowpass filters. In a multiband system, of course, the attack and release times have to be optimized for each band .. slow for LF, and fast for HF.
Shadow Vector has the same dynamic separation on a diagonal split as a pan on the L or R sides. The engineers at EMI were quite surprised to hear a clear side pan when they played the SQ master of "Rhymin' Simon"; apparently their CBS decoder couldn't do that. I think Malcolm could chime here: side pans and diagonal splits are excellent with Shadow Vector; the crudely built prototype had no problem at all with those pans.
I'm not sure what you mean by "limited direction sensing". All 3 axes of a Scheiber sphere are sensed and responded to, but there's no corner optimization because it was felt that would seriously degrade the spatial impression. As it was, the analog prototype, with no feedback, had a measured separation between 30 and 35 dB, which is about all that could be expect with a 6-pole phase shifter array. We didn't think chasing 40 dB or more was worthwhile, considering that normal stereo cartridge separation was around 30 to 35 dB with very careful adjustment (using an axial tilt circuit and a mono LP to calibrate it).
What baffles me is how TATE works. I'm mystified by the 6 control lines, which are presumably unidirectional, and maybe this lets them hard-optimize for the corners (this would follow CBS's lead on maximizing corner separation). I personally think hard corners and "detenting" (image compression) around the speaker are very objectionable, and it certainly doesn't sound better in quadraphonic than in stereo. In quad, I'd say it was worse, since the whole goal of quad, or surround in general, is to make all the speakers completely disappear, leaving the listener surrounded by nothing but sound. That's kind of the point; as soon as you localize the physical speaker, the illusion is destroyed. In stereo, you kind of expect the soundstage to be just that, a stage, with definite limits. In quad, there shouldn't be an impression of a stage, more like being there in the presence of the musicians, whether they are a frontal array or all around.
But CBS was using the big JBL studio monitors popular back them, and those have almost no ability to form a stable image (thanks to a primitive crossover and multiple unsuppressed peaks in the drivers). The folks at Audionics, including myself, were heavily influenced by British design, since we were the authorized importer of Radford amplifiers and speakers, and the Brits at that time were far ahead of the Americans in smoothness of response, crossover sophistication, and image quality. This was important in the design of Shadow Vector: the image quality, spaciousness, and stability had to be as good as stereo at its best. That was an unstated but real goal for the project.