"new" way to decode SQ

[Disclord]

I was recently given a Denon AVR-3808 receiver that has Neural surround (in addition to DTS-HD & Dolby True HD, plus their lower bitrate variants). Well, imagine my surprise to find that Neural decodes SQ almost perfectly - it's not always "Tate" perfect, but it's the next best thing - It sounds better than my SQD-2020 in terms of placing channels correctly. I'm really amazed by it. The opening of "Chase" CD via Neural places the trumpets correctly around the room and not just a blob of sound for the rear channels like Pro Logic II and DTS Neo:6. The SQ CD of "Annie" sounds like it's being decoded by a Tate but with totally discrete separation - from reading the patents, the Neural decoder splits the signal up into 50+ bands and steers each separately. It's really amazing.

If you have a decoder with Neural surround, give it a try with your favorite SQ album. I've been totally amazed. Neural must be using some kind of quadrature shifts to "watermark" their encoding, as they call it. And that "watermark" just happens to be very close to the "watermark" quadrature shift used by CBS SQ,

 

[proufo]

Cool!

Now let's wait for the ones that know to provide a full tech explanation.

TIA!

 

[Disclord]

I sure hope someone here knows how Neural works - other than the fact that it's multiband like DTS Neo:6, I don't know what else they are doing on encode/decode. I've tried to work my way through the patents and while I know a HUGE amount about so-called 'regular' matrixing and adaptive dematrixing, the Neural stuff seems just to be designed to confuse. Their info is a lot like that provided by HDCD in the early days when they didn't want people to know HDCD was basically a noise reduction system.

 

[jsrstereo]

I sure hope someone here knows how Neural works - other than the fact that it's multiband like DTS Neo:6, I don't know what else they are doing on encode/decode. I've tried to work my way through the patents and while I know a HUGE amount about so-called 'regular' matrixing and adaptive dematrixing, the Neural stuff seems just to be designed to confuse. Their info is a lot like that provided by HDCD in the early days when they didn't want people to know HDCD was basically a noise reduction system.

Disclord: would you please get one of these http://daysequerra.com/ViewManual.aspx, and then give us a full review?

 

[jefe1]

I just heard the sound of Tate prices dropping......

 

[ChristopherLees]

and to think we've had a great SQ decoder built into most of our home theatre amps for a few years now and not realised it...


well that's one less box on the shelf..

 

[inquadwetrust]

I had a close up demo of the Neural workflow for decode/encode at the NAB show in Vegas today with Brandon at the DTS booth - I was really blown away by how discrete it was. I asked whether this would work with matrix recording, yes was his resounding answer. I'm getting a demo license upon my return so I'll post some findings as soon as I can.

 

[jsrstereo]

Well, I wonder just how well Neural performs stereo-to-quad synthesis---??

 

[John Svensson]

"I wonder how well Neural performs stereo-to-quad synthesis..?" Not very well from past threads of folks who have demo'd it...including Disclord. No better than DPL2 for most music although there may be the occassional miracle. However, I'm pleased to hear it seems to have an SQ decode directionality to it. I will have to give this a try, as soon as I find out if I even have a Neural equiped receiver! More SQ LPs on the shelves than I have of decent conversions. John S.

 

[Disclord]

After doing more googling - and via the wayback machine, I've found out a bit more how Neural works - it is just an amplitude and phase encoder - BUT, it encodes/decodes over HUNDREDS of bands, breaking down the signal into small segments, kinda like AC-3, that are then given individual encoding - if two bands happen to coincide, the encoder recognizes this, because it has a built-in decoder to compare the results and then adapts the encoding so that both signals can be decoded properly even though they are in the same spectral band. I still don't know why/how it's decoding SQ so well - it does place the 'center left/right' sounds too far to the back (but then, only the Tate II decoded those correctly anyway!) and it seems to do weird things on the Quadraphile test tones, placing the 'pulse' sounds correctly as well as the voices, but the reference tones then sound slightly wrong in placement. It can't do the QS or UMX tests at all - QS sounds like mono on "Money" and UMX it images almost all sounds at Left Back and Right Front - "Money" sounds really weird that way - a diagonal stereo soundfield.

So, it looks like we have at least a passable way to hear SQ again without absolutely having to have a Tate. I just can't believe how well it decodes it - yes, it's not perfect, but it's really close.

For those of you with Neural Surround in your receivers, try some SQ decoding and post your findings - I'm interested in others results.

Oh, yeah, BTW, it sucks at stereo-to-surround synthesis. I use Neo or DPL-II or the Fosgate Tate, depending on my mood.

 

[Wunlow]

On a sidenote to this: I recently was reading about the DTS Neural upmix decoder, which may or may not be different than their consumer decoding. Of interest is the choice to choose where the surrounds are placed, which may effect the Neural decoding parameters.

http://www.google.com/url?sa=t&sour...r-2EBw&usg=AFQjCNGX-5IDvVH9DLeaz8soznXFckvwtQ

 

[neil wilkes]

On a sidenote to this: I recently was reading about the DTS Neural upmix decoder, which may or may not be different than their consumer decoding. Of interest is the choice to choose where the surrounds are placed, which may effect the Neural decoding parameters.

I am not at all sure if the chipset is the same as the software plugin application or not - but I know a man who will know, and I shall ask & post back.
The details - such as they are available - can be found at http://www.dts.com/DTS_Audio_Formats/DTS_Neural_Surround/DTS_Neural_Surround_UpMix.aspx
There also appear to be different optimizations of the codecs - one is specifically designed to upmix the dreadfully low bitrate audio in standard definition digital TV broadcasts which is, as we all know, piss-poor pathetically bad (The bitrate, not the process!!) and the other is for full bandwidth material.

 

[surroundophile]

After reading this I was anxious to try it out myself.

I hate to have to bust y'alls bubble, but..
I put on my CBS Labs SQ test disc, put my Onkyo preamp on THX Neural, and played through the channel identification bands.
And the results were, using a six speaker set up (LF,CF,RF,RB,CB,LB)............
CF from CF RF from RF, RS[ide] from RS, RB from RS, CB from all six, LB from LS, RS from RS, LF from LF. The LF had a little bleed-over into LB, and RF a little bleed-over into RB.
By now I was so dissapointed I skipped the last test, CT[op].

So, Neural does NOT decode the back channels (RB, CB, LB) correctly at all. However, it does seem to get the two side (phantom) channels correct.

 

[surroundophile]

After reading this I was anxious to try it out myself.

I hate to have to bust y'alls bubble, but..
I put on my CBS Labs SQ test disc, put my Onkyo preamp on THX Neural, and played through the channel identification bands.
And the results were, using a six speaker set up (LF,CF,RF,RB,CB,LB)............
CF from CF RF from RF, RS[ide] from RS, RB from RS, CB from all six, LB from LS, RS from RS, LF from LF. The LF had a little bleed-over into LB, and RF a little bleed-over into RB.
By now I was so dissapointed I skipped the last test, CT[op].

So, Neural does NOT decode the back channels (RB, CB, LB) correctly at all. However, it does seem to get the two side (phantom) channels correct.

Oops, the results sentence above should read:
CF from CF RF from RF, RS[ide] from RS, RB from RS, CB from all six, LB from LS, LS from LS, LF from LF

 

[DKA]

On a sidenote to this: I recently was reading about the DTS Neural upmix decoder, which may or may not be different than their consumer decoding. Of interest is the choice to choose where the surrounds are placed, which may effect the Neural decoding parameters.

http://www.google.com/url?sa=t&sour...r-2EBw&usg=AFQjCNGX-5IDvVH9DLeaz8soznXFckvwtQ

I know nothing as to decoding quad, so my response has nothing to do with that. You're referring to DTS Neural Upmix, though, and I have messed around with the program.

As far as I know, the program accepts stereo inputs only. While it does a reasonably good job at creating a solid soundfield compared to other "one-button" solutions, the interface and options are extremely limited. you have controls for width, depth, and that's it. Again, it does a good job for a program which only controls for both those things but, at close to $500, it is limited compared to what you can do with SPEC, which merely costs you the price of Plogue Bidule. That's a steep price for something that can't even beat a piece of a homebrew method.

Some of the older surround methods were developed with QS and SQ decoding in mind, but those had their own problems. I am very surprised that no one has attempted a conversation with Zeerround @ SBU on this. If anyone can help develop a better method for some folks, if that's even needed, that's the person.

Again, not my forte, but maybe this leads the conversation somewhere.

 

[Disclord]

After reading this I was anxious to try it out myself.

I hate to have to bust y'alls bubble, but..
I put on my CBS Labs SQ test disc, put my Onkyo preamp on THX Neural, and played through the channel identification bands.
And the results were, using a six speaker set up (LF,CF,RF,RB,CB,LB)............
CF from CF RF from RF, RS[ide] from RS, RB from RS, CB from all six, LB from LS, RS from RS, LF from LF. The LF had a little bleed-over into LB, and RF a little bleed-over into RB.
By now I was so dissapointed I skipped the last test, CT[op].

So, Neural does NOT decode the back channels (RB, CB, LB) correctly at all. However, it does seem to get the two side (phantom) channels correct.

Except for the back channel results, that's what mine pretty much were - except CB correctly decoded at two Surround Back L/R speakers (I have a 7.1 setup) and Rb and Lb seemed pulled a bit forward so they were 'kinda' side-walled. CT doesn't sound like anything - in fact, it seems "pulled apart" - not that many SQ albums have a true CT. And as I said, it seems to decode differently depending on the source - the Quadrafile test-tones were all messed up, but the pulsed tones were 'basically' correct as well as the SQ "Money" track. From reading various things I can find about Neural, it's different depending on the company - like Neural THX is slightly different than DTS Neural. I found one thing from DTS claiming that DTS Neural is somehow "improved" over the 'older' Neural systems.

Who knows? It IS a strange system and gives bizzare results sometimes on Dolby Surround encoded material - sounds will 'pop out' of the surrounds that clearly belong up-front, and other times it sounds really good. It has a strange fidelity sometimes too... which I can't put my finger on - almost like the midrange is messed with. Maybe a side-effect of decoding over hundreds of bands?

Still, for someone with NO SQ decoder, it works pretty well and better than any other 'active' decoder like PL-II or the various later non-Tate based Fosgate's.

 

[Disclord]

I know nothing as to decoding quad, so my response has nothing to do with that. You're referring to DTS Neural Upmix, though, and I have messed around with the program.

As far as I know, the program accepts stereo inputs only. While it does a reasonably good job at creating a solid soundfield compared to other "one-button" solutions, the interface and options are extremely limited. you have controls for width, depth, and that's it. Again, it does a good job for a program which only controls for both those things but, at close to $500, it is limited compared to what you can do with SPEC, which merely costs you the price of Plogue Bidule. That's a steep price for something that can't even beat a piece of a homebrew method.

Some of the older surround methods were developed with QS and SQ decoding in mind, but those had their own problems. I am very surprised that no one has attempted a conversation with Zeerround @ SBU on this. If anyone can help develop a better method for some folks, if that's even needed, that's the person.

Again, not my forte, but maybe this leads the conversation somewhere.

Is there even a Neural encoder from DTS now that accepts more than 2 channels for ENCODING into a "watermarked" (their stupid term!) 2 channel signal? i.e. an N:2:M encoder?

As a side-note, some of the (few) SQ albums I've heard, decoded with a PC based SQ decoder like those for Audition, seem to have phase problems, like with Center Front - it's all smeared between the 2 front speakers and not a hard CF image. In fact, all phantoms seem to have problems - but the corner speaker positions decode beautifully and sound incredibly discrete. But overall, I like the Tate II better since it keeps all phantoms in phase, especially all front sounds.

 

[proufo]

Well, we can only hope that Lou Dorren would eventually put to good use the non-RIAA output of his CD-4 demod.

 

[ChristopherLees]

Oops, the results sentence above should read:
CF from CF RF from RF, RS[ide] from RS, RB from RS, CB from all six, LB from LS, LS from LS, LF from LF

so with LF from LF , RF from RF, LB from LS,RB from RS....and if you then disconnect the rest of the speakers and perhaps move the LS and RS back a bit would that be acceptable?...

and does the CF really need to be on since it's not part of the sq quad system...

it also makes me wonder if some brands work better than others

cheers

 

[boojidad]

Couldn't wait to give this a try on my Denon. I think the mechanics are there, so the quality of the decode may have more to do with the source material. I tried Columbia's cast albums for My Fair Lady (20th anniversary production) and ThreePenny Opera (with Raul Julia) and got fantastic separation, very discrete. Next was Neville Marriner doing Eine Kleine Nachtmusik (Angel S-37443) and got some ambience but it was hard to tell if it was intentional or just reverb. Finally was an Italian pressing of Mozart's Symphony No 29 with Leopold Hager conducting (PDU ACSQ 60071) which had nothing. It was almost mono. Would love to hear a report from someone who can try a Vanguard or Project 3 release.