SQ Shadow Vector Soundfield Mapping

[Lynn Olson]

Here’s a thought: dynamic decoders are usually set up so their static (logic sensing at zero) coefficients are the same as a “classical” static SQ or QS decoder. But it doesn’t have to be that way. The static coefficients can be different ... for SQ, the “tetrahedral” coefficients suggested some years back would be interesting, with 7.7 dB of separation between LF and RF, and 7.7 dB of separation between LB and RB. In a sense, this “primes” the decoder for further shifts in any direction, and should also require less steering overall, since the decoding coefficients are already partway there. (The less steering required, the better ... fewer transient artifacts, less sense of things shuffling around, etc.)

Of course, a QS decoder doesn’t need any changes of this sort, unless to weight it more towards an EV4 set of coefficients. (About 12 to 14 dB in the front, and 3 to 7.7 dB in the back.)

This changes the “resting” set of SQ or QS coefficients to a set that are better optimized for most recordings, and asks less of the logic. This could even be determined experimentally by looking at how much logic activity was demanded on different recordings ... basically, a long-term spectrum analysis of the control signals themselves, and comparing different sets of “resting” decoding coefficients. The coefficients (or default matrix) could be chosen on a basis of requiring the least logic activity over the course of a given recording.

A statistics-based approach to determining the optimum SQ, QS, or stereo-enhancement decoding coefficients would be the smart way to accomplish decoding. I suspect the EV4 set of coefficients would be close to optimum for stereo-enhancement purposes, since they were chosen on a subjective basis many decades ago.

P.S. I agree 100% that many discrete recordings sound unnatural, compared to Shadow Vector at its best. I have not heard any of the TATE units, so can’t comment on their spatial qualities. I think the spatial defects of lesser decoders (including Dolby Pro-Logic II) has given the public, and magazine reviewers, a poor impression of how good decoding can really sound.

 

[Sonik Wiz]

Fixed 12dB LF to RF and LB to RB channel separation,
"mild" Front/Back logic (similar in concept to what was
used in the Sony SQD-1000 SQ decoder), but providing
a maximum of 12dB (logic assisted) F/B channel separation.

Whether done in hardware or software I think it's intuitively obvious that fixing Lf/Rf & Lb/Rb blend to max 12dB would automatically increase F/B separation over basic decoding. What the improved F/B seperation would be I haven't the math smarts to nail down. But perhaps on its own you wouldn't even need any kind of logic assist.

 

[Lynn Olson]

The “tetrahedral” static SQ matrix has improved CF to CB separation, compared to the 0 dB separation of a classical SQ matrix. It can be thought of as a simple variable-matrix “half-logic” decoder with the blend partially activated, which naturally results in reduced LF/RF and LB/RB separation. Looked at another way, it brings QS-like default separation to SQ, while leaving the encoding coefficients alone.

The part I like is that it asks less of the dynamic matrix, in terms of speed and aggressiveness of cancellation logic (the vectors don’t have to move as far or as fast). Like QS, not as much is needed, which is why the original Sansui Vario-Matrix was pretty slow, not very precise, and non-aggressive in its action.

The Sansui used opto-couplers, if I recall right, which are very low distortion but not very accurate in terms of gain control. By contrast, with a dynamic SQ decoder, you really need 1% precision in the variable-gain amplifiers to make the thing work acceptably. This was a non-trivial requirement back in the early Seventies.

Nowadays, with digital implementation, we can make the decoder content-aware. If there are lots of spiky transients, they can get the full-separation treatment, but the rest can go much slower, or not move at all. Multiband makes this even easier, since each band is independently analyzed for transient content. Done right, it can become subjectively transparent, and resolve the full spatial content of all program material.

The fact that Dolby Pro-Logic II (music), or DTS Neo, or the modern upmixers can be heard *at all* indicates pretty serious design errors and the lack of a subjective listening process before design sign-off. I kind of wonder if *anyone* did any subjective listening, based on the artifacts that are clearly audible with these decoders.

By “artifacts” I mean the audible slowness and lack of responsiveness, as well as audibly closed-in sounding, for DPL-II, and HF artifacts and occasionally edgy sound for DTS Neo. These are defects in design that would have raised objections from recording engineers in the Seventies, and there’s no reason to tolerate them forty years later.

 

[proufo]

Nowadays, with digital implementation, we can make the decoder content-aware. If there are lots of spiky transients, they can get the full-separation treatment, but the rest can go much slower, or not move at all. Multiband makes this even easier, since each band is independently analyzed for transient content. Done right, it can become subjectively transparent, and resolve the full spatial content of all program material.

In real time?!!!

 

[Sonik Wiz]

The Sansui used opto-couplers, if I recall right, which are very low distortion but not very accurate in terms of gain control. By contrast, with a dynamic SQ decoder, you really need 1% precision in the variable-gain amplifiers to make the thing work acceptably. This was a non-trivial requirement back in the early Seventies.

In the very earliest Sansui Quadsonic decoders such as the QS-1, opto-couplers were used only in the rear ch output modulated by a 3Hz input that supposedly imparted a sense of spaciousness to the rear chs. In all other gear they used FET's for gain control. In earlier Vario-Matrix recievers discrete FET's were used, in later products such as QSD-1/2 & other receivers the HD3103 FET IC was used. It actually had 5 FET's inside with the 5th one controlling bias to the other f4 to achieve best dynamic range & linearity. The FET's were used as shunts to ground to control the mixing coefficients in the HA1328 matrix IC.

Another nifty thing I liked about Sansui's approach was the direction sensing. While everyone else was messing with full wave rectification & log amp differencing Sansui did it another way. The HA1327 chips were AGC limiter chips. Basicaly they totally squashed the dynamic range (60dB accurate) so there was no ampltude difference between the input chs. Then it was fed to a simple differencing circuit. In the case of sensing if a sound is in front or behind, if there is no amplitude difference then the only difference to detect between those chs is phase. It swings to one polarity, it's front. Swings the other way, it's rear. These are the control signals sent to the FET IC.

 

[Lynn Olson]

In real time?!!!

Sure, why not? Any digital implementation has latency; the decoding is not done in real-time ... the incoming audio goes into a buffer deep enough to accommodate the slowest part of a multiband decoder, so the steering of the decoding coefficient is done *after* the sensing logic is satisfied there is a real change in direction. This is luxury not present in all-analog decoder. Since a buffer *must* be present in a digital decoder, there’s enough time to do an analysis of the audio content. So why not?

Big thanks to SonikWhiz for the in-depth examination of the Sansui technology. I only owned the standalone, first-generation Sansui QSD-1 decoder, not any of the receivers. The 2nd-generation tech is really fascinating.

Interesting also they used a 60 dB AGC, which is similar to what we did with Shadow Vector, although we used precision full wave rectification inside the AGC loop.

 

[par4ken]

In the very earliest Sansui Quadsonic decoders such as the QS-1, opto-couplers were used only in the rear ch output modulated by a 3Hz input that supposedly imparted a sense of spaciousness to the rear chs. In all other gear they used FET's for gain control. In earlier Vario-Matrix recievers discrete FET's were used, in later products such as QSD-1/2 & other receivers the HD3103 FET IC was used. It actually had 5 FET's inside with the 5th one controlling bias to the other f4 to achieve best dynamic range & linearity. The FET's were used as shunts to ground to control the mixing coefficients in the HA1328 matrix IC.

Another nifty thing I liked about Sansui's approach was the direction sensing. While everyone else was messing with full wave rectification & log amp differencing Sansui did it another way. The HA1327 chips were AGC limiter chips. Basicaly they totally squashed the dynamic range (60dB accurate) so there was no ampltude difference between the input chs. Then it was fed to a simple differencing circuit. In the case of sensing if a sound is in front or behind, if there is no amplitude difference then the only difference to detect between those chs is phase. It swings to one polarity, it's front. Swings the other way, it's rear. These are the control signals sent to the FET IC.

This is deja-vu, this same topic came up before and Sonic-Wiz you provided (Lynn) the same reply. The phase modulation effect sounds does nice in a small room but in a more typical larger living or rec room the effect is totally unnoticeable. Only recently did I realize that the early QR receivers also had the phase modulation effect as well as the QS-1.

 

[par4ken]

Here’s a thought: dynamic decoders are usually set up so their static (logic sensing at zero) coefficients are the same as a “classical” static SQ or QS decoder. But it doesn’t have to be that way. The static coefficients can be different ... for SQ, the “tetrahedral” coefficients suggested some years back would be interesting, with 7.7 dB of separation between LF and RF, and 7.7 dB of separation between LB and RB. In a sense, this “primes” the decoder for further shifts in any direction, and should also require less steering overall, since the decoding coefficients are already partway there. (The less steering required, the better ... fewer transient artifacts, less sense of things shuffling around, etc.)

Of course, a QS decoder doesn’t need any changes of this sort, unless to weight it more towards an EV4 set of coefficients. (About 12 to 14 dB in the front, and 3 to 7.7 dB in the back.)

This changes the “resting” set of SQ or QS coefficients to a set that are better optimized for most recordings, and asks less of the logic. This could even be determined experimentally by looking at how much logic activity was demanded on different recordings ... basically, a long-term spectrum analysis of the control signals themselves, and comparing different sets of “resting” decoding coefficients. The coefficients (or default matrix) could be chosen on a basis of requiring the least logic activity over the course of a given recording.

A statistics-based approach to determining the optimum SQ, QS, or stereo-enhancement decoding coefficients would be the smart way to accomplish decoding. I suspect the EV4 set of coefficients would be close to optimum for stereo-enhancement purposes, since they were chosen on a subjective basis many decades ago.

P.S. I agree 100% that many discrete recordings sound unnatural, compared to Shadow Vector at its best. I have not heard any of the TATE units, so can’t comment on their spatial qualities. I think the spatial defects of lesser decoders (including Dolby Pro-Logic II) has given the public, and magazine reviewers, a poor impression of how good decoding can really sound.

While this is a novel idea and might have it's merits, it still seems like Sacrilege as SQ's main virtue is full left to right separation across the front and across the back all without logic! The original Audionics decoder had a blend setting but I never used it! It just seems wrong to convert SQ to something approaching QS, even if further logic action is easier.
My original S&IC had the Exar chips. Artifacts were rarely noticeable, mainly in the back channels and more noticeable when listening to them in isolation. The "Composer" also has a separation control to dial down the logic. I don't recall ever using it. My "new" Composer has the old National Chips and with that unit, on some material, artifacting (sort of a pumping effect) is audible, so I found that I actually had to dial down the separation. I have an Involve SQ evaluation board, and although it tests good on the bench and mops the floor with the older SQ logic decoders sonically it's still no match for the Tate.

 

[Lynn Olson]

My original S&IC had the Exar chips. Artifacts were rarely noticeable, mainly in the back channels and more noticeable when listening to them in isolation. The "Composer" also has a separation control to dial down the logic. I don't recall ever using it. My "new" Composer has the old National Chips and with that unit, on some material, artifacting (sort of a pumping effect) is audible, so I found that I actually had to dial down the separation. I have an Involve SQ evaluation board, and although it tests good on the bench and mops the floor with the older SQ logic decoders sonically it's still no match for the Tate.

Now that’s interesting. Your all-analog Audionics S&IC subjectively outperforms the mostly-digital Involve evaluation board, if I read your letter correctly. In what way? Spaciousness? Natural sound? Most 3D sounding?

Very curious about your impressions. I have no dog in this fight, since I had long left Audionics when they designed the S&IC, they did not ask for my input, and the decoding algorithms of both the TATE and the Involve are largely a mystery to me.

As for the special features of SQ as a matrix, I’m kind of neutral. It does some things very well, and other things not so well. QS has a different set of tradeoffs, and it was far easier to build a good dynamic decoder for it back in the analog era. SQ dynamic decoding was right on the limit of what could be done with all-analog tech (mostly due to the complexity and precision required in the signal processing).

In a way, the Audionics S&IC mirrored the difficulties of building an NTSC or PAL analog color television with all-vacuum-tube signal processing. Yes, it *could* be done, but it was very complex and expensive. Audionics did not have the resources of Dolby Labs, never mind RCA or Telefunken, so the S&IC must have been a very steep climb for one or two techs and Charley Woods to carry out. That it worked at all is a miracle.

P.S. In an alternate universe, I probably would have been called in as a consultant on the S&IC project. Audionics and Tektronix were only a few miles apart, and I could have worked on weekends helping out on the S&IC project. But the bridges were so thoroughly burned that was not going to happen, so I only heard about it years after it went on the market.

 

[par4ken]

Yes Lynn to all of the above (Spaciousness, Natural sound (on SQ), Most 3D sounding). The most immediate difference noticeable is much fuller bass.

I was without a properly working unit for a number of years, but before that (30+years) it was used for everything including stereo. It was so much better on movies than any Dolby decoder.

After my original Composer started acting up I was making do with lesser decoders, I was not overly impressed with the muddy sounding QSD-1 (on QS and stereo) either. I found a "Vista" (Photolume) decoder on eBay a number of years ago and loved it for stereo in surround mode, it is very natural sounding with the stereo image simply stretched in a U around the speakers and no audible artifacting. The one drawback is that the out of phase blend used to create the QS surround mode can at times noticeably reduce the level of center front, SQ enhancement doesn't have that same problem. The Photolume is a QSD-2 clone without the all-pass phase shift network.

The Composer enhance mode is similar to QS surround mode but doesn't pin much to the front corner speakers, but still produces a similar effect. Since I was last using my original Composer, I've improved my vinyl playback system with an Ariston turntable, moving coil cartridge and tube based preamplifier. My "new" Composer (old chips), sounds even more discrete than I remember, I think that it's due to my improved vinyl system and proper adjustment of the Axial Tilt. I'm finding that I have to reduce the Separation control a bit on some records so the old National chips do seem to have a less smooth but more aggressive decode action than the Exar chips?

For stereo the Photolume is a bit more natural sounding, for SQ the sound is damn near discrete with the composer. I never get the sense that the sound disappears from any speaker just because another channel is dominant, just a continuous unbroken soundfield. I don't get that same unbroken soundfield with other SQ decoders or even with discrete.

Lynn, working at Audionics sounds like a dream job to me, but unfortunately dream jobs never seem to last long.

 

[proufo]

Is all this knowledge (12 pages) being incorporated into your hardware decoder?

We are not worthy.

 

[Sonik Wiz]

Hi Lynn, I'll have a look out for Loggins & Messina "Full Sail" (if I can locate it), sounds like a good test )))
Just got auto selection of s/pdif and ADAT input (48kHz) working and it works and sounds great. Its just mesmerizing, selecting different decode modes on the front panel )). I'm very happy with the specs and performance now, so its time for more documentation and finalize the PCB and parts lists for pre-production.
Hopefully I'll have time after this to look into an advanced variable matrix SQ encoder just to see how good the system might have been.

Daring the risk of straying back on topic.... a shout out here to Malcolm. It's been almost 2 years & 12 pages since this thread was started. And your last post on progress was in April. I am very interested if you can share the latest update on how this all is going. Other than general over all stauts I am specifically wondering if you plan to put this into production or is it just a personal passion? Is there a stereo synthesis mode? And certainly the idea of multi-band decoding of QS with Shadow Vector is very intriguing. Looking forward to hearing how it's shaping up!

 

[kfbkfb]

12 dB channel separation = (virtually) infinite channel separation

https://www.quadraphonicquad.com/forums/threads/involve-encode-decode.25222/#post-381773
I encourage the development of an alternative ("mild logic")
(all software) SQ Matrix decoder using the parameters written
about in post #221 in this thread.

~6 dB of additional (logic assisted) channel separation would
bring the resulting separation to ~14dB, above 12 dB (and with
a small safety margin).


Kirk Bayne

 

[Lynn Olson]

Thanks, Par4ken, for the comments on the various decoders, some of which I’d never heard of before, like the Photolume, which was completely new to me.

Yes, the phono source makes a *big* difference to the spatial and naturalistic qualities of quadraphonic reproduction. It is *not* all the same ... some phono sources are flat and lifeless, and some are very vivid and filled with a “you are there” quality of presence. Unfortunately, in an era of AAC lossy-compression digital, it’s kind of rare to hear this, but I think 320 kbps is still tolerable (but right at the limit).

One thing that’s a potential quality loss for SQ is the extreme amount of phase shifting for the entire signal chain. If I recall right, the big CBS encoder had 10 poles of phase shifting, the amount of phase shifting varied from 6 to 10 poles in the consumer decoder, and conventional multi-way speakers have 2 to 8 poles in the crossover network. Note that all of this phase shifting is cumulative; it takes a digital gizmo to unwrap it, and then it has to be programmed with the precise inverse function of the original analog phase shifters.

Now, the audibility of allpass phase shifters remains controversial, but there is a general consensus (in the speaker-design community) that there are group delay limits that should not be exceeded. As for myself, all I hear on a direct A/B comparison is a slight timbral shift, but other audiophiles I have met are extremely sensitive to this. Since individual sensitivity to allpass phase shift seems to vary a lot, I feel it’s a desirable option for decoders to have an EV4 mode that bypasses all the phase shifters, and is basically an intelligent Dynaquad mode. Aside from the lack of CB localization, there is no real downside to this as a stereo-enhance mode.

 

[malcolmlear]

Yes very much agreed about conventional allpass phase shifters. I've found it not particularly unpleasant, but certainly audible. One of the great advantages of digital decoding is the option of linear phase FIR filtering. The filters I've implemented for the SV decoder applies a 90 degree shift across the entire audio band with reference to the raw audio stream, albeit delayed. Also accuracy is spot on all the way down to 10Hz at which point a slight reduction of amplitude occurs but phase still holds at 90 degrees.
I've toyed with the idea of equal separation fallback for SQ. However the coding gets messy and seems 'unnatural'. I may revisit this sometime but it all works nicely at the moment, however I will instrument the code to gather statistical data on periods it does fallback to standard decode coefficients.
Update:
I've now been running the prototype decoder almost continuously for months without issues and have nearly completed the build of 2 pre-production units which should be up and running in the next few weeks. Work is evenings and weekend, but most of the hard work is now done. It now supports both S/PDiff and ADAT @ 44.1/48kHz and ADAT out in addition to standard phono sockets. After optimizing threads to cores the design settled down with 9 decode bands with all cores running at 70% maximum clock speed (nice and cool).

 

[BJMarsh]

I have a question regarding the “almost discrete” statements with SQ. If the original Shadow Vector reverts to no logic when there is no dominant channel, am I correct that “almost discrete” is the result of multiband processing?

 

[BJMarsh]

Implementing Lynn’s SV patent in software has been a real challenge for me but I’m seeing the light at the end of the tunnel. I’m hoping to get there in a couple of weeks. Not sure what the next steps will be, likely multiband but I am FIR filter challenged!

 

[Lynn Olson]

You can imagine what this was like using discrete transistor electronics ... no microprocessors, no op-amps, just bipolar transistors and FETs for the time-switching VCA’s. Ten circuit boards and a hand-wired backplane to connect them all together.

Just as well that Cliff Moulton, Audionics’chief engineer and co-founder, worked at Tektronix for many years, since the physical Shadow Vector build was a lot like a scope with many plug-in boards and a common backplane.

it is interesting that the underlying static matrix is still audible in a dynamic decoder ... the Shadow Vector sounds very different than a Vario-Matrix, particularly in the spatial aspect. Some of that was the speed of the dynamics, but there is also a pronounced difference in spatial presentation. I haven’t spent much time with the Tate chipset, but it sounded fast but a little on the dry side, while Shadow Vector is a very “wet” sounding decoder, typically a little more spacious than the discrete original. Other dynamic decoders, particularly Dolby ProLogic II, sound more dry and closed-in. I imagine this has been Malcolm’s experience as well.

Regarding “almost discrete”. This can be taken two ways: snappiness of image localization ... crispness, in other words, and often an artifact of pairwise mixing, and spaciousness, which is something very different. And there are several kinds of spaciousness: synthetic bathtub reverb (late Fifties), a rock-n-roll special effect (early Seventies), or the impression of actually being in a large space. I aim for the latter. The data is actually there in most stereo recordings, it’s just not very audible in 2-speaker playback. The hard part is recovering it without creating noticeable artifacts.

 

[BJMarsh]

Lynn,

I have read all of your posts on the subject and come away with an incredible amount of respect for your concept and the fact that you actually built this with discrete components available in the early’70’s! Wow! With Malcolm’s test files I can confirm that my algorithms are correct. Thanks, Malcom. My immediate challenges are the shapers and the final gain stages that are used to adjust the four output channels that are used for the dominate channel. I am in the right church but the wrong pew at the moment. Good things take time. I think I have memorized your patent and have coded your figure 4 block diagram. From your posts I have gained insight into some of the subtlest not apparent in the original patent - thank you. My math stinks and there has been a lot of trial & error. It’s a labor of love. Back in the 70’s I never had an SQ decoder that was worth the money. By the time Tate and the Audionics products were out of my price range for a format that was orphaned! But I do have a number of SQ, QS and CD-4 albums that I’m burning to hear for the first time. My end goal is a software decoder that does SQ, QS and synthesis. Like SM-2 sans hardware. I’d have the SM-2 if it played via HDMI but we know how the licensing is cost prohibitive.

Wow, that’s a mouthful! Thanks to you and all of the other QQ members for help & encouragement.

Blaine

 

[Lynn Olson]

It was an interesting experience working on this gizmo for two solid years ... from LSD-vision insight, to researching the patent and filing the document of disclosure, moving 800 miles north to Portland, building the prototype, and then finally hearing it for the first time in 1974. I always had a sound in mind during that whole period, which came in handy later when designing speakers and amplifiers.

But actually hearing it for the first time was a revelation. It did everything I expected, but also did things I didn’t expect at all. It was far more spacious than I expected, and had a very memorable and distinct sense of presence and a vivid you-are-there quality to it. I didn’t expect that at all ... more spacious, sure, I expected that all along, but not the Cinerama-sized stage presence. I was frankly stunned and bowled over. I imagine Malcolm was equally stunned when he got the digital version going. That aspect was VERY different than the CBS Paramatrix, the Tate unit, or Vario-Matrix, or anything I’ve heard before or since. It pulls things out of the mix that you don’t expect, all kinds of little details, and even dry recordings sound very 3D. If there’s a Shadow Vector hallmark, that’s it. That’s how you can know if you got your version right; it will be obvious, and not sound like any decoder you’ve heard before.

Curiously, my speakers and amps sound that way too. I don’t intentionally design things that way, but they come out sounding like that. It just seems to end up that way. I couldn’t design a “mainstream” sounding product if I tried.