PC Based SQ DECODING - ALMOST DONE!

[imploder]

Only quick (I will write more later):
The formerly mentioned Wendy Carlos site goes through almost every aspect dealing with Quad systems. I recommend reading all the stuff, really amazing informations!
My excerpt is: SQ cannot do phantom channels between front and rear, because the 70's decoders could not deal with it. It would be possible (a bit...) now with the digital decoding method, but there exists no SQ record with that information, because it was avoided as far as possible to encode something that will confuse analog logic circuits. Later there were mixing consoles in the studios with locked out mixing variations that no mistake could be made....sa Wendy Carlos writes: SQ is the worst of all Matrix systems...
BUT: We can now extract as much as possible from that recordings !!

Tab, I have a simular decoding method for QS, it works with the same separation as the method for SQ, but is more complicated. As promised, I will post ist soon the next days.

And not to forget: Most Quad recordings were not mixed for room simulation but for the mentioned ping-pong effects, unfortunately!

Also try out the different speaker setup as Wendy Carlos describes (rears on left and right side, not behind, fronts in a smaller angle) and the "complete surround effect" will increase dramatically !! That in combination with the PC decoding method can squeeze out everything possible from SQ, I think.

OK, that post was not so short but it is difficult to express those feelings not in my mother tongue...

Andy

 

[winopener]

Imploder,

i've read the last post about the sq decoding and the problem of the centerleft-centerright images.
I think you're right that in real sq recordings there will be few things to extract there; i've wondered a bit about it and a way to extrapolate these side images could be this:

let's name the channels that way
FRONT LEFT: 1
FRONT RIGHT:2
REAR LEFT: 3
REAR RIGHT: 4

normally we have two stereo files with 1+2 and 3+4.
The idea can be this: le't combine the channels so we have 1+3 and 2+4 (both left and both right togheter).
From these special files let's do a Channel Mixer / LR to MS (left right to mid side).
Now we have on the L part of the stereo files the supposed "center left" (doing it with 1+3) or "center right" (doing it with 2+4); the R part has the side channels and we don't need this.
Channel Mixer again, BothLeft.
Now we have the same Center on both.
Let's go back to the original 1+3 and do a MixPaste/Overlap with the original 1+3 ad the file we created.
In theory... it should work... in pratice... need to see the phase relationship, delays, how much to add and other stuff.
I have stil no time for test this week, next one i can do something again.
Let me know what do you think.

I suppose there's something similar (1+3/2+4) going for QS...

 

[kempfand]

So...
is everybody lost in decoding out there?
Can somebody write something about the results and personal impressions? I would be interested!

A feedback to use the SQ decoding process on a non-Quad source:

Sting - Soul Cages

Credits for the idea to to this goto Cai Campbell's threat: Q-Sound Recordings (thanks !)

The result was really really excellent. I shared the conversion it with friends and the feedback was very positive as well.

Time to say "Thank You Andi" for coming up with the process, and sharing it.
Might sound like "flattering", but I wanted to mention it here.

Cheers,
Andreas

 

[QUADradial]

Now wait! I told Cai about Q-Sound!!!

I actually stumbled onto that while messing around with an old Q sound plug-in during my first attempt at SQ encoding. It worked...then I came to the conclusion that I don't need to ENCODE SQ!

 

[kempfand]

Now wait! I told Cai about Q-Sound!!!

I actually stumbled onto that while messing around with an old Q sound plug-in during my first attempt at SQ encoding. It worked...then I came to the conclusion that I don't need to ENCODE SQ!

I did didn't about this. Honest apologies for any misleading statement I might have made ! I'm just reading the public posts and deriving my thinking.

In any case: The Q-sound recording I mentioned converts very, very well (for my two ears) using the described SW-only proces, and wanted to mention this.

Kind regards,

Andreas

 

[QUADradial]

I was just kidding. I don't post that regularly and this was a long time ago on one of the older quad sites. I was just rippin' on Cai. I'm one of the ooooold timers.

 

[proufo]

I finally got to sit and try to understand this. I need reassurance on these:

1) The only two signals that get to be alone at 90 degress to the other signals in the matrix operations are Left front and Right front, B and C in Carlos' notation. So those two can be extracted by removing the components at 90 and/or -90 degrees. I guess that's what the CCE filter does and what is difficult to do in hardware so at the time dynamic processing had to be used instead.

2) After the matrix calculations, A and C are at 180 degrees and D and B are at 0 degrees. To each other, not the matrix angles.

3) So there should be no common components in A and C so there won't be cancellations (see 4 too). Those are Left Back and Right Front. I guess it's an "artistically" acceptable limitation.

4) D and B are summed, so the only way you can extract the original signals is if there are no common components in them. As above, those are "across" channels and I'd guess even Enoch Light survived that.

5) So after one round of processing one can extract a fair aproximation of the original signals for the front channels, B and C, and also for the combined signals A-C and B+D.

6) Then you add C to A-C to get A and substract B from B-D to get -D.

Thanks in advance.

 

[Cai Campbell]

I was just kidding. I don't post that regularly and this was a long time ago on one of the older quad sites. I was just rippin' on Cai. I'm one of the ooooold timers.

Huh? What? Um, snrrrk!

Snore... :zzzzz

 

[imploder]

2) After the matrix calculations, A and C are at 180 degrees and D and B are at 0 degrees.

No. That is wrong and so all following conclusions. See my Post:

https://www.quadraphonicquad.com/forums/showpost.php?p=30983&postcount=66

After basic matrix decoding we get the decoded REAR channels, the front remains untreated!

The scheme says (see my "drawing") that A and D (the rears !) stay in 0 deg phase AFTER basic mixing and B and C remain phase shifted, but every remaining front channel has a +90 or -90 deg counterpart in the opposite rear channel. That is the information the CCE of AA1.5 can cancel out:.

The CCE then cancels out every phase that is on 90, 180 and 270 deg. I confess that I also do not completely understand HOW that works, but IT WORKS.
Simular the cancellation of the rear channels in the front summary signal works: We cancel out every phase on 90, 180 and 270 degrees. Nothing more.

The CCE compares BOTH channels for cancellation, that is why also the remaining 0 deg B channel is cancelled out.

These are the limitations of SQ: No signal is allowed to be at those phases because they belong to the opposite channels!
BUT: This happens exactly if you record in a room with two or more microphones and mix it together, there are NATURAL phase shiftings in every recording that tries to reproduce the ambience of the recording room.That's why most Quad recordings are channel separated ping-pongs, that's more easy to do!
NEXT BUT: Pink Floyd managed in their DSOTM album to do ambience very, very well! It can work even with SQ but the audio technicians had to be well trained (they mostly still aren't...).

Finally: Best Quad is discrete Quad, the matrix has limitations we can not "fill up" while decoding by making crazy crossmixes, because there is just nothing there. I prefer the "High End" version, just to reproduce the channels as they are, no extra tricks, because otherwise it would be better to hear stereo in front and rear parallel, also that produces a nice effect and is almost not to separate from a 3db basic SQ decoder.

Again long posting, but I begin to feel that we begin to repeat ourselves here.
Perhaps I will do a homepage soon with all the excerpts from this long, long forum...

Andy

 

[imploder]

Imploder,

i've read the last post about the sq decoding and the problem of the centerleft-centerright images.
I think you're right that in real sq recordings there will be few things to extract there; i've wondered a bit about it and a way to extrapolate these side images could be this:

let's name the channels that way
FRONT LEFT: 1
FRONT RIGHT:2
REAR LEFT: 3
REAR RIGHT: 4

normally we have two stereo files with 1+2 and 3+4.
The idea can be this: le't combine the channels so we have 1+3 and 2+4 (both left and both right togheter).
From these special files let's do a Channel Mixer / LR to MS (left right to mid side).
Now we have on the L part of the stereo files the supposed "center left" (doing it with 1+3) or "center right" (doing it with 2+4); the R part has the side channels and we don't need this.
Channel Mixer again, BothLeft.
Now we have the same Center on both.
Let's go back to the original 1+3 and do a MixPaste/Overlap with the original 1+3 ad the file we created.
In theory... it should work... in pratice... need to see the phase relationship, delays, how much to add and other stuff.
I have stil no time for test this week, next one i can do something again.
Let me know what do you think.

I suppose there's something similar (1+3/2+4) going for QS...

Hmm, sounds logically correct, should work, but will destroy the original source, it is nothing for archiving, I think.
Principally You do a good ol' gain riding on the side phantom channels, that is 400% of a Tate(which "only" cares about the real channels)...but the bleeding in the opposite channel is not removed by this method.

As I wrote before:
Try to change the speaker setup, put the rears to the left and right of the listening position and move the front speakers more together as you do for a conventional stereo setup. I tested it yesterday with DSOTM, there are cruisin' around effects and panning sounded much smoother.
Try it!

 

[proufo]

Hello Andy and thanks for your reply.

What I meant is exactly what you have as the result of "Right-Back channel extract" and "Left Back channel extract". You obtain two cases of a sole signal at 90 degrees to the other two; in one case (C with D and B) the other two are in phase and in the other (B with A and C) the other two are at 180 degrees.

The front channels remain untreated but those are the matrix front channels. not the original ones.

No. That is wrong and so all following conclusions. See my Post:

https://www.quadraphonicquad.com/forums/showpost.php?p=30983&postcount=66

After basic matrix decoding we get the decoded REAR channels, the front remains untreated!

The scheme says (see my "drawing") that A and D (the rears !) stay in 0 deg phase AFTER basic mixing and B and C remain phase shifted, but every remaining front channel has a +90 or -90 deg counterpart in the opposite rear channel. That is the information the CCE of AA1.5 can cancel out:.

 

[proufo]

The CCE compares BOTH channels for cancellation, that is why also the remaining 0 deg B channel is cancelled out.

This is the key point, CCE requires two channels, not one. It can't extract out of phase signals from a single channel. And I wonder if that's possible, given that there are multiple frequencies at different phases in all the original channels by themselves.

From the CCE help file: "The Center Channel Extractor effect keeps or removes frequencies that are common to both the left and right channels".

 

[imploder]

This is the key point, CCE requires two channels, not one. It can't extract out of phase signals from a single channel. And I wonder if that's possible, given that there are multiple frequencies at different phases in all the original channels by themselves.

From the CCE help file: "The Center Channel Extractor effect keeps or removes frequencies that are common to both the left and right channels".

OK, then CCE cancels out the one selected phase, because it has a comparison with the other channel where in every case there is NOTHING, for example channel C in mixdown D' has no counterpart in mixdown A' and so on.

But then we have a problem: why is channel B then also removed from D' even if it's in phase with D?
The same exists for the front channels, D and B are also in the same phase there, this would mean that if our method would not work we would have a crosstalk between front-left and back-right, but there isn't.
That is the thing I do not understand at the CCE, but everything is removed correctly, that is clear after the tests I made with Tab's SQ testfiles at the very beginning.

Any ideas?

 

[imploder]

The front channels remain untreated but those are the matrix front channels. not the original ones.

Yes, that is the same at non-logic SQ matrix decoders!

 

[imploder]

I am actually working on a web site explaining the SQ and QS decoding method, but it will take about two weeks, because I am very busy at work at the moment.

Regards

Andy

 

[proufo]

I am actually working on a web site explaining the SQ and QS decoding method, but it will take about two weeks, because I am very busy at work at the moment.

Regards

Andy

Hello Andy.

Looking forward to this.

Many thanks in advance.

 

[sam77aus]

I am actually working on a web site explaining the SQ and QS decoding method, but it will take about two weeks, because I am very busy at work at the moment.

Regards

Andy

What will the web site address be (if known)?

Sam

 

[sam77aus]

I have automated a bit the processing; here's how i do it:
1) place the wave file sq-encoded on c:\sq
2) open it in AA, convert (if not already) to 32-bit res. and normalize at -6dB
3) save it as front.wav
4) save it again as rear.wav
5) run the script below; after the script complete, raise front.wav by 3dB

Thanks for the script, it makes decoding SQ wave files a lot easier.
I was wondering what settings the numbers 1 to 18 in your automated script referred to, so I can tweak the settings in the center channel extractor.

Thanks
Samuel

 

[imploder]

What will the web site address be (if known)?

Sam

Hi!
The site will be (what else...) http://www.imploder.de (sorry, nothing to see there yet!)

I am still working on it on my test server at home and I plan to lauch the site at about 12th of may, that week my family will be "out of town" and then there is finally more time for it...sorry for the delay!

Regards

Andy

 

[sam77aus]

Hi!
The site will be (what else...) http://www.imploder.de (sorry, nothing to see there yet!)

I am still working on it on my test server at home and I plan to lauch the site at about 12th of may, that week my family will be "out of town" and then there is finally more time for it...sorry for the delay!

Regards

Andy

Excellent!, I will be looking forward to it.

Thanks
Sam