Can anyone tell me the REAL use for THX Neural Surround? I own the Onkyo 705 and according to the book and THX's website, Neural will matrix ANY 2 channel signal into 5.1 or 7.1.
Now I get a DD 2.0 signal for my digital cable box and when I select Neural Surround, I get NO SOUND from my surrounds at all.
Just wanted to see if anyone could help me understand what it really does...

Can anyone tell me the REAL use for THX Neural Surround? I own the Onkyo 705 and according to the book and THX's website, Neural will matrix ANY 2 channel signal into 5.1 or 7.1.
Now I get a DD 2.0 signal for my digital cable box and when I select Neural Surround, I get NO SOUND from my surrounds at all.
Just wanted to see if anyone could help me understand what it really does...

Don't know much about THX (or care for that mess since it is only good for theaters) but you can use DDPLII and get excellent faux 5.1/7.1 sound.

Hi there!

If I'm not mistaken, the Neural Surround function on your Onkyo receiver is to decode multichannel HD radio broadcasts from XM radio. If you visit the XM radio site, and earch for Neural Surround, you'll find info about it there. I found this out since I'm currently an XM radio subscriber, and am in the market for a receiver- looking at the Integra DTR 6.8 myself, which also has the same decoders built in.

bcarithers,

Sorry for the late reply. If you go to the Neural audio web site (neuralaudio.com) there is a group of white papers that directly address your questions. In short, yes; The Neural upmixer is actually for "N" channel rendering of any two dimensional (width and depth) content. Properly produced stereo demonstrates width and depth just as 5.1 encoded material. The Neural rendering process (different rom a matrix) can take a stereo source and render to as many channels as the industry will tolerate...essentially replacing "phantom images" with "hard sources". This makes the positioning of images less arbitrary, that is, more people in the listing area can hear the intended image correctly. The source could be original stereo or encoded 5.1.

rob r.

Hi all, here is a puzzling thing that happened to me with regard to Neural THX. I have an Onkyo 705 with Definitive Speakers, tuned using Audyssey.

1. I was on 5.1 channels and could preset listening modes for 2-channel PCM, Dolby Digital, DTS and D.F. 2-ch all to Neural THX 5.1. I was happy with the results.

2. Then I added two back surr speakers and retuned the speakers. I chose Neural THX 7.1 as my preset for the same sources as above.

3. Unfortunately I had to revert to 5.1 as I have been having clicking and loss of sound issues when using the Onkyo with my Homecast 8000 HD PVR [another forum suggested this could be related to back surrounds coming on]. Now I tried to set all the above sources back to Neural THX 5.1, it lets me do it only for two channel sources!!! I.e. only for 2 Channel PCM and D.F. 2-Channel. It does not show up as an option for Dolby Digital or DTS any more.

Background: I had read on other parts of this forum and similar ones that Neural THX 5.1/7.1 will work even with 5.1 [multichannel] sources and not just two channel ones. The Onkyo manual only talks about two channel sources for Neural THX, but I know it was working earlier with multichannel sources.

It is as if the receiver has decided to start following the manual :-) :-(.

Has anyone else had this experience or have any ideas as to what may be going on? I do like the Neural and would like to use it for Dolby Digital 5.1, DTS 5.1 and other multichannel sources. Should I try a factory reset? Or a Clear?

Thanks in advance.

I have an Onkyo HT-SR800 and an XM Radio connected to it through RCA stereo inputs. When Neural Surround is selected on the receiver, XM channels 76 and 113 are decoded automatically as 5.1 audio. The rest of the channels are still stereo.

I think there is some sort of special frequency codification on those channels that the receiver can separate to generate the extra audio channels... it really sounds like a natural 5.1 source, not a Dolby or DTS simulation.

Actually, I would love to know the answer to that too. Although Penteo is pretty much wide open about what we do and how we do it (www.penteosurround.com), I still can't exactly figure out what it is that Neural portends to do. Even if you start out with stereo material, the stereo material has to be stabilized and corrected because of physical tape or mag film azimuth stability and working-part level variances and before running it through the microscopic correlation process, so unless you're doing that first, the results will be unstable.

I've read their backgrounder (http://www.neuralaudio.com/downloads/NeuralSurroundBackgrounder.pdf) and I still don't get it.

Anyone who has any real insight, please post!

I still can't exactly figure out what it is that Neural portends to do.They watermark their 2-channel downmixes during encoding so that decoding back to 5.1 results in more "discreteness" (their word, not mine) than other matrix decoders like PLII or Neo:6. At least that's the jist of it.

If you wanted to transmit 5.1-channel music, you could do it a couple of ways. Send it as a 5.1 bitstream: it would remain discrete, but wouldn't be backwards compatible with 2-channel systems. You could alternatively send it as a stereo downmix: it would be compatible with 2-channel systems, but wouldn't have the channel separation of the discrete 5.1 original.

But what if you did something in between? Downmix to 2 channels using typical matrix techniques, but embed spatial watermarks that flagged what data went to which channel. The decoded result wouldn't be as discrete as the 5.1 original (you can never perfectly unscramble an egg), but it would be more discrete than typical matrix codecs (since the decoder will have more than just amplitude and phase cues to steer the sound).

Imagine having to transmit a movie as an interlaced video signal. Your video processor would rely on the image content and cadence in order to deinterlace the signal back to the original. However, suppose you could flag the signal to help the deinterlacer do a better/easier job of combining matching video fields and tossing out redundant ones. Neural portends to do something similar with multi-channel audio by watermarking the signal. Make sense?

Sanjay

Does anyone here prefer Neural THX over DPLII for two channel sources?

Personally, i've had better results with Neural THX. I have an Integra DTC-9.8.

(you can never perfectly unscramble an egg)
Sanjay

I unscramble eggs using Penteo every day... :-) Yolk Part 1 (http://www.penteosurround.com/Sample/LDS1.mpg) Yolk Part 2 (http://www.penteosurround.com/Sample/LDS2.mpg) White (http://www.penteosurround.com/Sample/LDS3.mpg)

So what you're saying is that XM is actually just transmitting it as another stereo channel, and somehow Neural is giving certain sounds certain qualities ("watermarking") which allow re-separation on the receiving end.

This sure sounds like the original DPL or an old quadraphonic matrix system like SQ or QS, that has a center-channel deriver on the receiving end.

I unscramble eggs using Penteo every day... :-)You do, but you still rely on phase and amplitude cues (correlated dual-mono info) to tell you what to extract to the centre output. Imagine if that specific information had been flagged in every 2-channel recording made and your decoder could recognize those flags. So what you're saying is that XM is actually just transmitting it as another stereo channel, and somehow Neural is giving certain sounds certain qualities ("watermarking") which allow re-separation on the receiving end.Exactly. The watermarked spatial info will help minimize crosstalk and maximize channel separation, making the results better than typical matrix decoding of 2-channel material and close enough to discrete to satisfy XM. For recordings that aren't Neural encoded (i.e., not watermarked), the Neural decoder can be used just like a regular stereo-to-5.1 matrix decoder.

Some more Neural white papers that may be of interest to you:

http://www.neuralaudio.com/literature.html

Sanjay

Imagine if that specific information had been flagged in every 2-channel recording made and your decoder could recognize those flags.
Sanjay

The only way to "watermark" something within the audio stream is with either phase or frequency manipulation. I still think that it sounds like a phase rotation thing, similar to the old DPL, SQ, QS, etc. Their patent is "pending" so I understand their reluctance to divulge any details. Their papers are pretty un-technical and very high on marketing language, which adds to the consumer confusion.

The only way to "watermark" something within the audio stream is with either phase or frequency manipulation.They're claiming to do something more precise than the usual phase and frequency manipulation of typical 5:2:5 matrix encode/decode schemes. But until they're more forthcoming in their white papers, we won't know the recipe of their special sauce. From the couple of demos I've heard, it does work.

Sanjay

.....
Their patent is "pending" so I understand their reluctance to ivulge any details. Their papers are pretty un-technical and very high on marketing language, which adds to the consumer confusion.


And, may I add, are yours. :)
No offense meant. Any contributions to the craft are appreciated, at least by myself.

I've been doing conversions for some time.
Many songs separate nicely using either phase or other methods.
Freqency domain is easy enough, but doesn't work well with some panning.
Frequency/Time domain can suffer the same.
I don't believe anyone can pull enough "discrete" info from pan position only, (try PF "Final Cut" sometime) as often multiple instruments that occupy the same frequency are panned to the same location, or are spread all over.
More complex algorithms demand intense processing power that is more in the provence of hardware for real-time decoding.

I'm not one that thinks DPLII is good, you see.
I don't see a magic bullet, but would be pleasantly surprised if one arises.
Just thinking out loud.........see y'all next week.

And, may I add, are yours. :)

Have you read http://www.penteosurround.com/pwp.pdf ? Of course I don't explain the algorithm itself, but every step that is done is in there. I'd love to see something equally deep...

multiple instruments that occupy the same frequency are panned to the same location

That's actually one of Penteo's prime selling points. If the original mixer panned several different tracks to the center, they should stay there. Just as if they panned something all the way hard left or hard right, it should tend to come from as far out to the side (rear) as possible, with all other pan positions falling into their corresponding location. The original mixer knew exactly what they were doing, and I respect that, and want to keep it that way.

Right now, I'm achieving >50db discrete separation between center and either side and >90db between left and right, more than enough for dramatic surround.

Penteo,

Sanjay has given an excellent link to the Neural Audio site. You should READ IT before firing off anymore "guesses" about what it is or how it works.

Comparing it to "quad" demonstrates that you have either not read or not understood the white papers.

I don't like accosting someone on a public forum but Neural has been more than forthcoming on the inner workings of the dowmixer and the upmixer within the link courteously provided by Sanjay.

The watermark is as it is described in the literature. It is positional information encoded into a stereo pair with both image and spectrum correction (as compared with the original), resulting in an infallable representation of the original image.

This 2-dimensional image may then be rendered into as many channels as you wish (2.0, 4.0, 5.1, 6.1, 7.1, 10.1, etc.). The beauty of this is that the content, once encoded in this format, will remain useful for a very long time, i.e. transportable through any stereo medium (radio, internet, ipod, vinyl, TV, etc. and recoverable to any multichannel format you may prefer. It is not hard to conceive the replacement of "phantom images" with "hard sources".

If the technology didn't work, there wouldn't be any adoption (believe me) and as it sits, there is more than 70% international adoption of the technology on the consumer side while the uptake in radio and television (over 300 encoding systems internationally within the first year of deployment) has been equally impressive.

My apologies to the forum for the rant...I'm just sooo weary of comments about "Neural not being forthcoming"...AFTER the accusing party is politely shown where to get the info they seek.

Thanks,

rob r.

Penteo,

Sanjay has given an excellent link to the Neural Audio site. You should READ IT before firing off anymore "guesses" about what it is or how it works.

I did read some of them, and that's what prompted my response. Don't get upset, I didn't mean anything harsh by what I was saying, but the parallels between encoding sounds (watermarking, as it is called these days) for subsequent decoding, in the audible range, are exactly how SQ and QS encoding as well as the original Dolby Surround worked; they are an encoding method that apply certain characteristics to certain sounds in the audible range, which are then undone on the receiving side; that was my only point. Unless Neural is using some sort of out-of-band information, the "watermarks" have to be in the audible range, which would make them - umm - audible. Both QS and SQ used fixed-formula phase changes which could be undone on the decoding end; it stands to reason that these could be made variable which would produce a theoretically infinite number of channels. But if Neural has figured out a way to inaudibly manipulate audio through something other than frequency or phase angle manipulation, then I stand corrected.

In no way am I trying to upset anyone, I just like being open and forthcoming as long as money can still be made. I understand that there is a pending patent, and I have no interest in company secrets, and perhaps that is what is causing me to feel as though the methods are somewhat obfuscated. Sanjay himself said "until they're more forthcoming in their white papers, we won't know the recipe of their special sauce."

If you find me being accusatory, that is my only accusation, that the methods of watermarking seem to be obfuscated in the technical papers under math for mathematicians instead of audio-based for sound engineers. So if it's not a company secret, it would be nice to know if there are phase angle changes, which of course would be audible in the un-decoded form.

I get lost in the math in the Neural papers; I just want to know how it sounds. And yes, I'll admit, even after 35 years in Pro Audio and an Emmy Award for audio engineering and 25 years as a C-coder, I don't understand all the Neural papers; obviously the Neural technical papers come from a different mindset than mine. It sounds as though the starter of this thread feels the same way.

My background comes from spending decades smelling burning coffee and cigarettes (the cigarettes weren't burning; the coffee was) in recording studios dealing with temperamental singers and pickers who sometimes had to be hosed down in the shower. The Neural ones are math intensive, so it's a different mindset. Most of Penteo labor is involved in the physics of correcting tape damage before a simple 200-line C-code pattern-matching algorithm. I don't in any way mind sharing that; there's no secret that all the splices and tape damage are what cause stereo mixes from being able to be surround-stable. I have spent up to two hours just correcting the "burble" caused by a splice going through the machine on a famous mix. The splice always was there; in simple stereo (or on mono AM-radio!) it was inaudible.

Another Penteo nemesis is the distortion often caused by under-biased master recorders, especially when under-biased on either the left or right channel only. Right now I'm working with Janis Ian's "At Seventeen", which is under-biased on both channels, and it makes it nearly impossible to make sound clean. In the 1970s, the bias was usually set by a 2nd engineer, so its accuracy would have been dependent on his hangover. I know this, because it used to be me.

Because I worked for so many years in making master tapes, I know their shortcomings. The original mixer got to sit behind a console in the sweet spot with an acoustically tuned room; Penteo tries to replicate that experience exactly in the 3.0 version, so that the listener heard exactly what the original mix engineer would have heard, even if the mix deck wasn't perfectly calibrated. Fortunately, we can tell, and correct it. In a tuned room, it sounds exactly like a 3.0 mix, even though it's only coming from two channels, because the phantom center image is that accurate.

Rob, if you have a specific paper in mind that might go into more practical detail from a sound and waveform point of view, as opposed to the math involved, especially with audio samples of isolated components, I'd love to read and hear it. Read mine at http://www.penteosurround.com/pwp.pdf, and let me know if there's anything equivalent.

I would love to hear how Penteo material sounds with the Neural encoding/decoding; perhaps we can work that out. I'm planning on having some private meetings here in San Francisco at Polarity before and during the AES. Because Telos and Fraunhofer used Penteo exclusively for its current demo platform at WZLX, (you've probably read http://www.rwonline.com/pages/s.0054/t.6776.html) and because it was used at the past NAB show floor, it obviously works well with the Fraunhofer IIS aacPlus Surround system; I would hope that it would work equally as well with Neural. It would make a great platform/format set for XM's services.

Obviously Penteo and Neural aren't competitors, they are complimentary; Penteo is a service provided to record companies and movie studios at the mastering level, correcting past stereo mastering errors and then slicing up a stereo mix into components using a pattern matching algorithm. It assumes that the original stereo mix (especially from the pre-digital master era before 1985) is unstable (which it is) and must be time-base corrected before analysis, producing both a time-base corrected stereo master as well as a parsed 5.1 master.

Penteo,

Thanks for the reply. The very idea that the inclusion of mathematics would obfiscate the intention of being open and/or transparent pertaining to any technology is probably a problem not unique to audio.

If there is too much math, I apologize.

Now then, the concept of using phase and intensity within matrix style encoders and decoders is not at all new...energy (in our case intensity) determines the width position while "phase" (or in our case correlation) determines the depth position. Orthogonality of "energy" and "phase" is an excellent starting point for gaining control of position within the 2-D stereo "image" as they do not interact.

Taking it to the next level by dividing the spectrum into several dozen bands improves the positioning integrity as the is less chance of "collisions" in the time/frequency domain. Think of it as having a one lane highway vs. a ninety lane highway, with the same number of cars travling at different speeds, each has more paths to avoid the other.

As a person who has worked with other peoples content I'm sure you understand the the spatial pedigree of stereo mixes from different sources varies widely. With no idea of what the "phase", ITD's (interaural time delays), tape azimuth errors, etc. within the stereo aggregate the encoder can NOT simply "add additional phase" to the channel pair and produce any kind of predictable results. It's just a waste of time.

That being said, Neural's approach is to test the resultant position and spectrum of the encoded part of the downmix by comparing the original input with an internal reference upmixer over many bands of spectrum. All spectrum and positional errors are corrected, and the resultant "LwRw" 2-D stereo representation of the multichannel original is sent on it's way; with complete confidence of what's going to happen on the upmix side. This kind of "modem" makes position encoding absolute i.e. if you put an in polarity pair into the left front and right front inputs of the encoder, they image into the left front and right front, if you put an out-of-polarity pair into the left front and right front inputs of the encoder, they still image into the left front and right front. That's the difference. Matrixes don't behave that way at all. "Magic surround" (unexpected position results) happens so much that matrix equipment and software manufacturers dedicate chapters in their usage manuals on it's existance.

Neural's watermark system doesn't suffer from this...the positional encoding is absolute, error free and has a ridiculous capacity to survive just about any kind of medium: lossy compression, analog and vinyl included. Some entities have the belief that the watermark concept may only be digital in nature and that's just not true.

Regarding "phase", that's only one physical method that may be used to alter interchannel cross-correlation and hence, depth positionm. When you involve a solution over many subbands, the layman's understanding of what phase is just won't cut it.

Hey, I'm used to the burning coffee also. My designs (and my poor war-torn body) have found themselves on the road and in control rooms long enough to be referred to as "vintage"...a kind way of saying "old but useful"

Janis Ian's song, "At Seventeen" was beautiful. I can't comment on the archiving of the song, but the "sound" was haunting, intimate...I'm happy that you have an opportunity to save it.

Regarding MPEG Surround, I have read the RM0 specification and understand Fraunhofers efforts to acheive standardization, it is interesting that the standard is being sought even though the encoder is notated as "not ready"

For that matter, any comparisons made between MPEG Surround and the Neural System are based on early Neural prototypes (not so hot). Neural's state of the art down and upmxers are over five revisions from the early days.

Neural's adoption is excellent with over 70% of all new AVR's, new automotive adoption (don't ask), and the gaming industry. On the broadcast side, two sat stations, 70 terrestrial radio stations with an audience of over 20 million and dozens (I don't know, I stopped counting) of TV affiliates. Interesting note, Neural extended the reach of the Super Bowl into an additional 233 countries (!). For the game info, contact Dr. Mark Tuffy at THX.

All of this adoption is based on a single premis...no side data outside of the stereo pair. To be clear, Neural Surround (and the THX variants) is more of a modem than a codec (thanks, jj), making the task of traversing the world infrastucture not only possible, but easy.

John, Neural welcomes any opportunity to demonstrate the veracity of the technology. If you have the permission of the content owners to use their content, we would be happy to perform tests. It would be a pleasure to have another ally in the pursuit of more compelling surround content.

rob r.

If I may butt in here, Neural THX surround is more akin to MPEG Surround than it is to the usual run-of-the-mill matrix systems that have been known in one form or another since the Hafler Circuit article in (I forget which) electronics magazine way, way back when.

The difference between Neural -THX surround and MPEG surround is that MPEG surround requires external digital steering data, and Neural-THX inserts the steering data into the two channel signal, in an analog form, i.e not as bits but as coded directions via a spatial map (see the Neural papers for more info).

A missing channel or something of that sort is not a characteristic of Neural-THX, but sounds like something else gone confused. From a distance, it's hard to kjnow what.

MPEG-Surround is just like Neural-THX except that it has to rob bits from the audio coding channel in order to provide directional information, and can not reproduce an accurate spatial sensation without the extra DIGITAL bits that must accompany the signal. Both are 5-2-5 systems as they are commonly used, i.e. 5 input channels, 2 transmitted channels, and 5 output channels. What's more, the 2-channel signal from Neural-THX sounds quite good as a stereo signal by nature, something that MPEG_Surround has to be manipulated in order to do so nicely.

Sorry to resurrect an old thread but I must say that Neural THX mode sounds excellent with two-channel LaserDisc sources. Noticeably better than Pro-Logic or DTS Neo, at least to me.