SQ Shadow Vector Soundfield Mapping

[malcolmlear]

Hi there,
Whilst developing an accurate shadow vector decoder I plotted out of curiosity the encoding map for various decoded sound field locations. Its probably the first time the input/output relationships of the ideal full separation SQ system have been plotted so may be of interest. I might even use the info as a basis for a new digital panning encoder, however the hardware decoder is the priority at the moment.
Cheers,
Malcolm

 

[Soundfield]

Hi Malcom - it would be interesting to learn of your plans for the hardware decoder - what basic approach are you taking? Are you still working towards a multisystem decoder? Why adopt the shadow vector technique? I would have thought that you would end up with some very complex circuitry along the lines that our friend "national-kid" in Tokyo has produced - certainly a labour of love if that's the case!

 

[malcolmlear]

Hi Soundfield, guess that's not your real name )) The plan is cracking SQ decoding using Olson's shadow vectors approach. The reason I'm choosing this method is purely logic, as an engineer a 3-axis variable matrix makes so much sense and if you consider the best decoding vectors for all 4 channels for any predominant sound source, the shadow vector method is the only option that gives a perfect decode. As for complexity, I'm planning on doing this all in software running on a 500MHz Philips ARM MCU. Hardware is basically the MCU and 32-bit analogue convertors running at 48k samples per second. Prelimary calculations suggest a 10 band decoder could work at this speed or possibly 96k samples at 5 bands. So far I've been working on simulations to get a mathematically perfect shadow vector decode (digital offers a quite different and more accurate decode than the analogue methods employed in the 70's).
As for other systems, yes already a shadow vector QS decode has been simulated along with other less common systems such as matrix H.
I do have one thing in common with our 'old school' 'national-kid' and that is a labour of love )))

 

[malcolmlear]

I've been looking at this problem for several years trying to find a way to produce a hardware solution that provides accurate decodes in hardware but kept falling short. Key for a hardware processor based multiband decode was an excellent band splitter and decoder with low overheads in terms of multiplications/divides. Fortunately both problems are now solved and decoding can now all be done in the digital domain.

 

[malcolmlear]

Whilst SQ is a terrible system (other expletives here), it does however redeem itself somewhat by being perfectly decoded by the shadow vector method (Olsen should be praised for his vision).

 

[proufo]

As for complexity, I'm planning on doing this all in software running on a 500MHz Philips ARM MCU. Hardware is basically the MCU and 32-bit analogue convertors running at 48k samples per second. Prelimary calculations suggest a 10 band decoder could work at this speed or possibly 96k samples at 5 bands.

So looking forward to the outcome of this!

My wallet is ready.

Query: won't all the work done for SQ get you most of the way to decode QS? hint hint nudge nudge.

 

[malcolmlear]

True, the vector modification routine is now able to cope with most matrix systems, although surprisingly slightly more difficult with 2 - axis systems such as QS and BMX.
I've also managed to add dynamic phase correction to the primary sound source, thus resolving the old issue of the 180 degree 'join' somewhere in the Soundfield. Essentially the primary sound source is positioned correctly and all channels are in phase.

 

[Soundfield]

....As for complexity, I'm planning on doing this all in software running on a 500MHz Philips ARM MCU. Hardware is basically the MCU and 32-bit analogue convertors running at 48k samples per second.

Oh, I’m massively disappointed now! When you said a hardware shadow vector decoder I envisaged an even more elaborate contraption than that which our Japanese friend has created – and at the very least would be a couple of 19” rack mounting units with stuffed with dozens of boards covered in 741’s! Doing it in software is just cheating (says someone who wouldn’t know where to start!). Even better if you had done it using valves. Although I recognise that for some strange reason thermionic quad never took off (!) - unless you count Jim Fosgate’s wonderfully mad solid state / valve hybrid FAPV1 Dolby ProLogic II decoder of course:

Joking aside, this is a significant undertaking and as a former engineer I know how much work will have to go into this (I still haven’t built my bog standard CBS SQ decoder that I’ve had the bits for for years as I can't find the time let alone anything on this scale)! – a real time shadow vector decoder would be remarkable achievement . Please keep us informed of your progress.

 

[malcolmlear]

Ha ha, will consider 4 EF80 valves (I like the look of the grid type anodes) on the output channels. Will keep updates coming ))

 

[proufo]

For inspiration:

Exotic tube gear

 

[proufo]

True, the vector modification routine is now able to cope with most matrix systems, although surprisingly slightly more difficult with 2 - axis systems such as QS and BMX.
I've also managed to add dynamic phase correction to the primary sound source, thus resolving the old issue of the 180 degree 'join' somewhere in the Soundfield. Essentially the primary sound source is positioned correctly and all channels are in phase.

You are the king!

Call it Cordelia.

 

[DuncanS]

Joking aside, this is a significant undertaking and as a former engineer I know how much work will have to go into this (I still haven’t built my bog standard CBS SQ decoder that I’ve had the bits for for years as I can't find the time let alone anything on this scale)! – a real time shadow vector decoder would be remarkable achievement . Please keep us informed of your progress.

Well here is one I built in my 2nd year at Uni 1977-78 complete with home designed PCB and bought Cartridge pre-amp, so go on its never too late! I never could throw it out

 

[malcolmlear]

That brings back memories, Faranti transistors ))

 

[Soundfield]

Well here is one I built in my 2nd year at Uni 1977-78 complete with home designed PCB and bought Cartridge pre-amp, so go on its never too late! I never could throw it out
View attachment 34464

Ah, now that's proper engineering! Yes, I will make mine (honest, guv) - in fact during the winter I designed a PCB for it but the summer's been too nice to have done any more since then. As the autumn nights draw in I'll search out my Ferric Chloride!

 

[DuncanS]

That brings back memories, Faranti transistors ))

I've done quite a lot of designs with the ZTX series of Ferranti transistors...…...quite some time ago now though!

 

[DuncanS]

Ah, now that's proper engineering! Yes, I will make mine (honest, guv) - in fact during the winter I designed a PCB for it but the summer's been too nice to have done any more since then. As the autumn nights draw in I'll search out my Ferric Chloride!

My board was designed around the Motorola part. I used Hydrochloric Acid with Hydrogen Peroxide as the catalyst to etch that board, though I doubt you could go to the Chemist and buy it anymore as I did then!

 

[proufo]

I looked at the Colson patent and it mentions "variable gain".

No pumping is expected, right?

 

[malcolmlear]

The variable gain elements are used to vary the decoding coefficients for the 4 output channels. For example the gain of the zero degrees left input (Lt0) may be reduced but the 90 degree version.of the same signal (Lt90) would have it gain increased proportionally thus the combined effect is a phase shift of the signal with constant gain. All output channels of the shadow vector decoder always have fixed gain with only the decoding coefficients changing.

 

[Soundfield]

My board was designed around the Motorola part. I used Hydrochloric Acid with Hydrogen Peroxide as the catalyst to etch that board, though I doubt you could go to the Chemist and buy it anymore as I did then!

The pcb I’ve drawn up a layout for is for the most basic of the CBS “Full Logic” decoders (CBS circuit L1) using the three Motorola chips MC1312, MC1214 and MC1215. I haven’t etched a printed board at home for at least 30 years so the results may be a bit problematic even if I can find the Ferric Chloride (never tried the HCl method). Yes – you used to be able to get all kinds of useful stuff at the Chemist’s. If the manager was on duty in mine he’d make you sign the “Poisons Book”, but generally they didn’t bother and even a sixteen year old could walk out with bottles of stuff for a couple of quid that would be deemed extraordinarily dangerous now. Happy days!

 

[malcolmlear]

An update on progress ))
The decoding software is now working well and I'm very happy with a decode of several DSOTM tracks performed yesterday. The sound is nothing like a conventional single or 3 band decode with complete absence of pumping artifacts and solid positioning of instruments. It turned out best to filter vector modification across all 10 bands to around 5-7 positions/sec. Decoding accuracy is better than a few decimal places from 10Hz - 20kHz although that is with test tones and real life especially with vinyl is never that good. I've also implemented dynamic phase correction, which eliminates the traditional 180 degree 'hole' thus all sound sources between speakers are in phase.
Next job is to eliminate a very low level noise from the vector modification filters and implement QS followed by Matrix H.