SQ Shadow Vector Soundfield Mapping

[Lynn Olson]

With all of these connections between two overlapping systems (a 2-channel all-triode system, and an all-transistor Marantz surround pre/pro and power amp), the exaSound e38 Mark II, the Involve Audio evaluation board, and the upcoming Shadow Vector decoder, I've realized I need a proper preamp/linestage to control this mess.

One of my neighbors about 30 miles to the north is John Broskie of TubeCAD Journal, at: TubeCAD Journal, who also makes an interesting Aikido linestage module. It's a clever SRPP + buffer module, with internal balancing to cancel power-supply noise and most distortion products (zero-feedback circuit). He also sells just the buffer (output) part of the Aikido circuit, which delivers slightly less than unity gain.

Aikido Linestage Modules
Aikido Buffer Modules

Anyway, I'm considering a 2+6 Stereo/Surround linestage, with 1 Aikido linestage (for 2-channel content) and 3 Aikido buffers (for 6-channel surround content). Channels 1 and 2 of the 6-channel input is split between the stereo Aikido linestage and the first Aikido buffer, while channels 3 through 6 go straight to Aikido buffers 2 and 3. There's a 2-channel stereo volume control with 100K impedance for the 2-channel stage (following the input selector), and a 6-channel volume control with 10K impedance for the 6-channel part of the linestage.

The input selector for the 2-channel part selects between Surround DAC (channels 1 and 2 of the 6-channel input), PHONO (the RIAA preamp is external), AUX 1, and AUX 2. An array of DPDT switches on the back selects between 6 Chan A and 6 Chan B set of inputs, followed by the 6-channel volume control which then feeds the 3 stereo buffer modules.

The signal split of the 6-channel Chan 1 and Chan 2 inputs are for multichannel DACs, which operate in 2-channel mode when fed 2-channel digital content. The higher-gain Aikido linestage is intended to be compatible with analog signal sources, external phono stages in particular, while the 6-channel section is only fed from DACs, Universal/SACD players, or quadraphonic decoders ... all of which are 2V rms devices.

The 6-channel outputs feed either an AVR, a pre/pro, or 6 channels of solid-state amplification. The 2-channel outputs feed the power amps of a high-quality stereo system with an analog focus ... phono stage, tape deck, tuner, or an unusual stereo DAC (R2R types).

Since all of these various source devices could present challenges with ground loops, I'm considering making all of the input selectors switch the grounds as well as the signals. That way, the preamp is only dealing with one signal source at a time, and for that matter, is only switched on when listening to music in surround or stereo format. Movies and TV are handled directly by the AVR with no intervention from the preamp.

I need to see how big this thing is going to be, with 3 Aikido buffers (2 tubes each), one Aikido linestage (4 tubes), and a B+ power supply using a pair of damper diodes (these have substantially less switching noise than conventional tube rectifiers and far less than solid-state diodes). So, 12 tubes in all, and shafts from the controls will need to go to the back of the chassis to minimize wiring distances. Grounding will require special attention, since there are so many possible devices that might be connected.

 

[gene_stl]

Really great prices on the boards and parts kits.

 

[BJMarsh]

I am a member, but I don't think I have posted. I have a quick couple of questions since I am fascinated by this thread. I haven't seen a post by malcolmlear in a while but was wondering if there is an SQ Shadow Vector product available. My preference would be software but hardware would be an option for me. For Lynn: I wanted to mention that I have a BT-2 preamp and CC-3 amp. Nice stuff but the right channel of the phono section is shot! Oh, well...

 

[malcolmlear]

Hi there,
No product yet, but most of the development has been completed and prototype is up and running. I'm a bit of a perfectionist, every detail, be it firmware, FPGA, functionality and performance must do justice to the shadow vector method of decoding matrix sound. As can be seen most matrix modes are supported including the rather strange BBC Matrix H which has the shadow vectors accurately mapping the curved lotus thus decoding recordings correctly, possibly for the first time.
Note on the rear panel the ADAT ports are supplemented by HDMI I2S used by some hi end DAC's. This is an optional subpanel that is so far is untested, however there's no reason this should not work.

 

[jonmchugh]

Looks very cool, Can we use the optical port to connect the SPDIF output of a CD player? Will reduced the digital to analogue to digital conversion loss for the input.
How does Shadow Vector decoding compare to more conventional decoding methods for SQ and QS? Is it a big improvement?

 

[malcolmlear]

At the present time, only ADAT optical in, however if timing constraints can be maintained in the FPGA it may be possible to implement SPDIF (48kHz) as well or in place of.

 

[malcolmlear]

A few firmed up specs:
24-bit / 48kHz Codecs
64 bit double float precision on all signal path maths.
8 decoders @ >16kHz 8kHz 4kHz 2kHz 1kHz 500Hz 250Hz <125Hz
Hilbert filter error immeasurable from 20kHz - 10Hz
Linear FIR filtering on all signal paths

 

[proufo]

A few firmed up specs:
24-bit / 48kHz Codecs
64 bit double float precision on all signal path maths.
8 decoders @ >16kHz 8kHz 4kHz 2kHz 1kHz 500Hz 250Hz <125Hz

Hello malcomlear.

How does the above relates to the original specs?

Hardware is basically the MCU and 32-bit analogue convertors running at 48k samples per second.

Thanks in advance.

 

[proufo]

What type of phono cartridge are you using in your development?

Would something called "induction phase-shift" --reportedly inherent to moving-magnet cartridges-- affect decoding?

I understand low-output moving coils have no induction phase shift.

 

[malcolmlear]

Original specs have not really changed. Codec's are still 32bit with all 32 bits passed to the audio processor, but in reality only 19 to 24 bits are probably noise free. Note that studio ADAT is 24 bit.
The final audio processing engine was close to the edge on performance, hence the change from 10 bands to 8. This was achieved by having the lower band encompass 125Hz down to nearly DC. This covers the lowest frequency that humans can discern sound location and gives 20% spare capacity which I'm much happier with.

 

[malcolmlear]

No phono cartridge has been directly used yet. All tests so far are vinyl rips and CD's. It's true that cartridge misalignment can seriously affect matrix decoding, however minimal adjustment on the decoders azimuth and balance has been found sufficient to correct most issues.

 

[MidiMagic]

A problem in the cartridge should affect the decoding only if it affects the two channels in different ways.

 

[malcolmlear]

Yes, indeed this is correct. The SV decoder has compensation for the two most common cartridge problems that affect the two channels in different ways, gain imbalance and azimuth (+- crosstalk). Note the compensation is not great and serious issues need resolving at source, ie the cartridge.

 

[malcolmlear]

By the way, many thanks Larry for the excellent info on your midimagic website, much appreciated ))

 

[BJMarsh]

Malcolm,

Is there any possibility of a Mac OS X application rather than hardware? Unfortunately, I have no multichannel analog inputs on my Yamaha RX-A2080. My original 2010 got trashed by an electrical storm and lo and behold, the 2080 dropped the multichannel analog inputs! I would be your first customer!!!

 

[Owen Smith]

This thread has been one of my best reads in ages.

I note Lynn has nice things to say about DTS Neo 6. I never got on with it on my Arcam AVR350, it would throw odd sibilants and other high frequency sounds into seemingly random speakers and my brain would go "what the heck?". So I settled on Pro Logic II for decoding stereo audio when watching TV, it might be dull and predictable but that's preferable to schizoid. That doesn't mean I like it, it's just my least worst option.

But I'm told DTS Neo 6 evolved over the years and there are quite a few versions. So unless you know which version you have, trying to compare it to anything else is pretty hard given people can't necessarily replicate your results with their DTS Neo 6 decoder.

 

[Lynn Olson]

Sorry for the long silence ... I had the CV19 gloomies for a while, since Karna and I self-isolated on March 5. Very much wish I was in New Zealand (or Oz) now, but Colorado is doing reasonably well for the USA, with the state turning the corner the last week or so. Still, it's a very good idea to wear facemasks outdoors, particularly for the 70+ age group, which I am in. Playing it safe with home-delivered groceries and only venturing outside to walk the doggie.

As for DTS Neo:6, my decoder is a ten-year-old Marantz pre/pro with balanced outputs, and I suspect this is one of the better implementations. It's almost impossible to find any info on DTS decoding; the only rumor I've heard is that it is a multiband unit, and pretty different than DPII (music mode). It's obviously much faster, but some recordings catch it out, which makes me wonder if it has lookahead or not, and also raises some questions about the direction-sensing logic.

Since Malcom's unit has lookahead, that should get rid of any glitches. I'm actually surprised it has 8 bands ... from my perspective, 6 should be plenty. The original Dolby A noise reduction system only had 4 bands, but of course that was in the all-analog days, when multiband signal processing was extremely complex, so there was an incentive to keep the product's complexity within reason.

It makes sense to have a single band from 0 Hz to 125 Hz. The drawback with lower-frequency bands is the whole logic sensing has to slow down to match the frequency (it does no good to direction-sense a fractional waveform), so that just increases the required lookahead time and increases computer processing requirements. In addition, if there is leakage between microphones or if the recording uses an array of spaced mikes, trying to interpret direction from low frequencies get problematic (spaced mikes have more or less random phase relations at low frequencies).

That was the reason the filter that preceded the logic-sensing of the original all-analog Shadow Vector only covered 500 Hz to 8 kHz, with an inverted Fletcher-Munson curve to match the sensitivity of the ear. It was already complex enough with 10 circuit boards and a hand-wired backplane, and of course did not have lookahead. As a result, the actual decoding was always 2~3 mSec behind the audio signal ... not ideal but better than most of the competition. Glitches could happen but were pretty rare, but like many decoders, it's didn't like phono cartridge mistracking.

As for azimuth adjusting, yes, that was an important feature of the original prototype. It even made it into one of the Audionics preamps, but that was after I left Audionics, so I have no idea how they implemented it. I discovered that most cartridges were reasonably well behaved up to 10 kHz, but above that, there was a considerable phase rotation. This turned out to be the result of asymmetrical mass distribution of the stylus cantilever, with a big heavy diamond on one end, and a symmetrical generator system on the other end. Since the logic section was pre-filtered (as mentioned above), the high frequency phase rotation didn't affect the function of the decoder. But you did have to make a one-time adjustment to match the cartridge you were using, since the cantilever very often has a 3 degree or more rotation, quite independent of the X/Y generator alignment. Just visually aligning the diamond in the groove did *not* assure good electrical azimuth, but twisting the headshell to achieve max separation is also a bad idea, since the diamond is then misaligned in the groove, which is OK for spherical styli, but really destructive for elliptical or fine-line (Shibata) profiles.

In Malcolm's digital implementation, unless he has a secondary azimuth correction above 5~10 kHz, it might be desirable to suppress logic action above 5~10 kHz for analog sources, since HF azimuth errors happen with both phono cartridges and magnetic tape. I wouldn't be too surprised that this error can linger in the CD transfers, if the studio playback of the 2-channel master didn't exactly match the azimuth of the original tape machine.

Long story short, with analog originals, HF azimuth is not guaranteed, regardless of source. It hardly matters in 2-channel 2-speaker playback, unless the error is really severe. It was mostly a matter of mono compatibility, where it *does* matter, but mono playback was typically on AM radio, on a limited-bandwidth car radio. So phase integrity in the all-analog days in the 5~10 kHz band was not a major consideration ... it was hard to maintain, phono cartridges were kind of dodgy (moving-coils being the best, if I recall correctly), and tape machines not only had to be kept aligned, the tapes they made had to match the reference Ampex alignment tape ... and that didn't always happen.

 

[Lynn Olson]

Might as well have a little digression on the topic of azimuth errors. Another name would be “axial tilt”, which is what it was called on the SV prototype. On a phono cartridge, it’s caused by the 45/45 generators inside the cartridge not matching the 45/45 axis of the original cutter head.

What is the effect of this? In 2-speaker playback, not much. Although the separation spec looks bad, subjectively, not much happens. The outer limit of the image moves a little outside one speaker (extra-width effect), and it moves a little inside for the other speaker. Hardly noticeable, which is why it wasn’t considered a big deal at the time. You have to remember loudspeakers of the late Sixties had pretty poor imaging (unless they were BBC monitors).

Azimuth errors are more interesting in matrix playback. For QS, it has the weird effect of rotating the whole image either clockwise or anticlockwise. Same for EV4, which is a similar matrix. SQ, as usual, is weirder. An azimuth error has NO effect on LB or RB. LF moves either towards Center Back, or moves a bit towards Center Front. Same for RF, with the error going in the other direction as LF. Not at all intuitive, and probably not very nice sounding in a corner-optimized SQ decoder. The symmetry of the Shadow Vector has equal separation in all directions, so there’s no corner-forcing effect.

How to deal with this? It’s traditional in LP playback to put the cartridge on a mirror surface and line it up on that. Unfortunately, the manufacturers of phono cartridges don’t do a very good job of axial accuracy on the cantilever, and 2 to 3 degree errors are common (I understand Ortofon and Denon have the best QC).

With a spherical stylus, an error isn’t too serious, although it certainly screws up tracking. It is catastrophic with an elliptical, or even worse, a fine-line stylus. This will really dig up and damage the grooves, and lead to very unpleasant tracking errors and groove distortion.

The mirror method lines up the cartridge body, but it’s very unlikely the cantilever is dead-on within a degree or so. So the best method is to find a mono LP with female vocals (I recommend “Ella Fitzgerald and Louis Armstrong” on Verve) and very carefully rotate the headshell until you hear the clearest and most natural sound on the vocals. Or you can do it the hard way with a test record and a distortion meter. Your choice.

So you can see that electrically correct azimuth is NOT the same as visually aligning the cartridge, unless you have a perfectly aligned cantilever. And in the real world, you won’t. In practice, you gotta do this by ear, although it’s not as tedious as twiddling with rake angle.

So the ideal solution with the kind of cartridges you can actually buy, is aligning the cantilever so the diamond accurately fits the 45/45 groove (least distortion and record wear), and a circuit in the preamp or decoder that electrically aligns the 45/45 generators in the cartridge. Do both, and 35 dB separation and very sweet sound are within reach.

 

[malcolmlear]

Hi Lynn,
Also been suffering the from the CV19 lockdown (( However it has had the benefit of a lot of time available between university commitments. Over the last few weeks many small problems have been resolved and ADAT optical in and out seem to work. Unfortunately only loop-back has been tested due to a delivery delay for a ADAT PCI board. Now working on an auto sensing SPDIF (48kHz) input.
I like 'Axial Tilt', perhaps a rename is in order ))
I agree 6 bands would probably be more than sufficient however the mathematics used to create the FIR filtering require bands at each octave, although its worth noting that there is a very noticeable improvement in image location over a 3 band version I briefly played with.
Last weekend a very annoying bug that randomly crashed the processor was tracked down. It was due to the x,y,z vector rotation rate filters producing a denormalized value if no audio input was present after several hours. This should have resulted in a 'flush to zero' and carried on without issues. However it turned out Linux switches this option off at boot time !! A bit of inline assembler fixed it ))

 

[proufo]

[QUOTE ]How to deal with this? It’s traditional in LP playback to put the cartridge on a mirror surface and line it up on that. Unfortunately, the manufacturers of phono cartridges don’t do a very good job of axial accuracy on the cantilever, and 2 to 3 degree errors are common (I understand Ortofon and Denon have the best QC).[QUOTE ]

And there are comments elsewhere about the incorrect alignment of the stylus in the cantilever tip.

Many thanks for the tip on mono female vocals.