Does joint-stereo destroy Dolby Surround Sound ? (rewritten)

ajung · Member Joined: 19 Dec 2002

This question is not very easy to answer, but we will try here to do so. The central question is, how well can an Mpeg-1 Layer-II audio stream, which is normally used on SVCDs and DVDs, reconstruct a complex signal like the Dolby Surround audio signals, using the method "stereo" or "joint-stereo".

We know, that the "phase information" between the two channels hide the Dolby Surround sound signals. But what does "phase information" means? To get an answer on this question, we will look at the encoding and decoding of the dolby surround signals:

Imaging you have 4 signals: L,R,C and S (Left, Right, Center and Surround) which you want to mix into a stereo signal Lt and Rt (Left-total and Right-total). Dolby Laboratories has invented back in 1987 the dolby surround singal encoding/decoding scheme, it's called a matrix surround encoding/decoding.

The encoder is actually very simple to implement, as shown in the equations in the figure above. The Left-total signal contains the left channel, the center channel reduced by 3dB (=1/sqrt(2)) and the surround signal also reduced by 3dB and shifted in phase by 90°. The same is done for the Right-total signal but in contrast to the Left-total one adds the negative of the 90° phase shifted surround signal!

Now, how do I phase shift the surround signal? Phase shifting is a modification of the signal in the frequency domain. Imaging you have transformed your signal using a fourier transformation. You obtain as fourier coefficients complex numbers. These numbers can be represented by an amplitude and a phase. Now a phase shift means, that I add a fixed value to the phases of all fourier coefficients. I our case, we add 90° respectifely pi/2 to it. This kind of phase shift is more a theoretical approach, but this is how one should do it given the specifications from Dolby. A more pratical approach is the phase shift of 0° and 180°, which is aquivalent to adding or subtracting the surround signal to the Left- and Right-total signal. But pay attention, by adding the surround signal with the 0°/180° method to the stereo signal, one gets very annoying effects: for example a fade from rear (surround channel) to the front (center channel) will not fade through the middle, but from rear to left to center. This is well known in the community, but there is a solution to the problem: Using a time delay of 7-8ms for the surround signal one gets arround the cancelation effect and the fade will go through the middle. But we are geting a bit off topic, lets get back the decoding of the surround signal.

For the decoding I have implemented a simple passive matrix decoding, that means the recontructed left channel (Lr) is Left-total, the right channel (Rr) is the Right-total, the center channel (Cr) is the sum of Left- and Right-total (-3dB) and the reconstructed surround signal is the difference between Left- and Right-total (-3dB). Now it gets clear, why the surround signal has to be mixed with 180° phase shift into the Left- and Right-total signal!

Now I had two questions: How sensible is an Mpeg-1 Layer-II (mp2) encoder on the complexity and amount of the stereo in the signal, when using the "stereo" and "joint-stereo" method and how well can it reconstruct the dolby surround signal and the four channels mixed into it?

For the first part of the question, I used two datasets:

As encoder I used the TMPGEnc Plus 2.57 from Pegasys Inc.. The encoding was done with different data rates, ranging from 64 kbit/s to 384 kbit/s, and the methods "stereo" and "joint-stereo". In the following figures, we used as a color coding: 64, 96, 128, 160, 192, 256, 384 kbit/s = red, green, blue, cyan, magenta, yellow, black. In the following figure, we show the correlation between the mp2-signals and the reference signal over the frequency range (0-22kHz, using 44.1kHz sampling). The correlation was calculated in a 300Hz window (more exactly the correlation between the fourier coefficients in this frequency window) from a 5 second sound sample:

Here we choose omega2=omega1+300Hz. As result, one obtains the figures:

It can be seen, that the "joint-stereo" method produces better results for "little stereo" then when using the "stereo" method, where the signals are separatly encoded and no information about the common signals in the left an right channels are used. But as soon as the amount of stereo signal increases, the "joint stereo" signal gets worse, but the "stereo" methods shows no changes. For "full stereo" the "stereo" method produces better results then the "joint stereo" method! When using data rates of 192 kbit/s or higher the there is actually no difference.

As a reference I gave also the correlations for the surround channel of the dolby surround signal, which I will explain in the following, how this was produced. To produce the dolby surround signal I used as sources for L,R,C and S white noise signals and mixed them into the Left- and Right-total channels as shown above. Now I compare the decoded surround channels (Lr,Rr,Cr and Sr) from the original Lt and Rt with the ones from the mp2-Encoder (see the gray arrows in the above figure). As a result one obtains the following figures:

Compared are the left-channel (which gives the same results as the right channel), the center channel and the surround channel. One sees clear differences between the different channels:

Using the method "stereo" we obtain better results for the dolby surround signals then when using the "joint-stereo" method. But when using bitrates of 192 kbit/s or higher, there is no difference between the two encoding methods and the signals are well reconstructed in the frequency range below 16kHz!

So, as a conclusion:

often worse

192 kbit/s or more

no difference

160 kbit/s

stereo

<=128 kbit/s

depends

My personal suggestion: Just use 192 kbit/s and stereo encoding!

Note: These tests were made with the encoder TMPGEnc, so I don't know, how other encoders behave!
If you are interested in the Matlab program, you can download it here: mp2test. The program is licensed under the GNU General Public License.

For more information, just have a look at my website: http://andreas.welcomes-you.com/projects/dv/index.html

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

That's some awesome work ajung and it definitely answers some questions from your last post!

Furthermore, it is definitely commensurate with the experience that people have been having on the forums -- that joint stereo encoding at 128 kbit/s has a definite adverse effect on Dolby Surround encoding.

Also, it agrees with my advice that for standard VCDs (i.e., at 224 kbit/s MP2 encoding), there is no benefit in using joint stereo over standard stereo (in terms of quality) and could theoretically be adverse effects.

As you have pointed out though, it may depend on the actual encoder as well. In joint stereo encoding mode, it can alternate between "standard stereo" and mid-side encoding depending on the actual audio. Hypothetically (on a perfect encoder), it should mean that joint stereo should never actually be worse than standard stereo. Of course, it appears not to be the case for the TMPGEnc audio encoder.

Thanks again.

Regards.
_________________
Michael Tam
vitualis' Medical Rant :: stay updated, read the rant
The Medicine Box
GNU VCDImager Authoring Guide Series

kwag · Member Joined: 12 Jun 2001

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

As above.

There is no difference in stereo separation between standard stereo and dual channel. There is the assumption that the MPEG audio encoder is working properly of course.

If anything, standard stereo should work BETTER than dual channel as it more intellegently allocates bitrate between the channels --> better audio reproduction.

As for "falsing" the signal, as Dolby Surround encoding works on the principle of the phase of the audio, you can naturally get this effect on standard stereo recordings WITHOUT specifically encoding for Dolby Surround. Dual channel not doing this could simply reflect that it doesn't reproduce the original audio as well as standard stereo.

Regards.
_________________
Michael Tam
vitualis' Medical Rant :: stay updated, read the rant
The Medicine Box
GNU VCDImager Authoring Guide Series

kwag · Member Joined: 12 Jun 2001

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

Well, since you like to drop terms out of a hat, how about you explain to us all how two channels, encoded as separate streams (i.e., BOTH stereo and dual-channel) are any different to each other in terms of stereo separation.

"Stereo" is basically dual channel where the bitrate for the L and R channels are VBR rather than CBR.

What is clear to me kwag is that you don't even know what these MPEG encoding modes even mean.

Regards.
_________________
Michael Tam
vitualis' Medical Rant :: stay updated, read the rant
The Medicine Box
GNU VCDImager Authoring Guide Series

kwag · Member Joined: 12 Jun 2001

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

ajung · Member Joined: 19 Dec 2002

Hi,

sorry, I was in holidays, so I couldn't give any comments on this discussion.

I agree with everyone how says, one should give references on the things he is writing. I had the same problems you were discussing all over. Just have a look at the mpeg standard, since there we should find the correct specifications:

http://mpeg.telecomitalialab.com/standards/mpeg-1/mpeg-1.htm

But when you want to have a look at the specifications, you have to pay 44CHF, which is 25EUR (?):

http://www.iso.org/iso/en/CatalogueDetailPage.CatalogueDetail?CSNUMBER=22412

Probably we should have a look in our libraries, if someone can find something...

Everyone will agree, that when we know the specifications of the mpeg1-standard, we can think about how the different encoding methods change the original signal.

So if someone has any new ideas, I would be happy to ready any papers or specifications or even sourcecodes about the encoding methods. If you "kwag" or you "vitualis" have anything like that, please post this here

.

I want to be absolutly clear: The tests I did are based on TMPGEnc and there I couldn't see any difference between dual channel and stereo (equal bitstreams in may tests).

KingJohn · **Banned** Joined: 22 Jun 2001 Location: UK

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

And curiously enough KingJohn, you are the only moderator here to lose your "moderator-hood"...

Regards.
_________________
Michael Tam
vitualis' Medical Rant :: stay updated, read the rant
The Medicine Box
GNU VCDImager Authoring Guide Series

KingJohn · **Banned** Joined: 22 Jun 2001 Location: UK

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

Ah KingJohn, how petty you are...

See unlike some, I like to have factually accurate information posted on the forum. kwag's interpretation of "stereo" and "dual-channel" are inaccurate.

And let us characterise it correctly KingJohn. You threw a temper tantrum in your previous incarnation when you could no longer handle simple criticism, some of it quite valid, of some of your methods. As a moderator, you were the only one who abused that power, moving posts to the "Closed" forum when someone (correctly) disagreed with some of your comments.

Regards.
_________________
Michael Tam
vitualis' Medical Rant :: stay updated, read the rant
The Medicine Box
GNU VCDImager Authoring Guide Series

kwag · Member Joined: 12 Jun 2001

ajung · Member Joined: 19 Dec 2002

ajung · Member Joined: 19 Dec 2002

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

Great info guys!

@ kwag: please note...

kwag · Member Joined: 12 Jun 2001

ajung · Member Joined: 19 Dec 2002

@kwag:
Can you specify, which encoder (version, pa-model, downmix-model etc) you used and from which source (true 5.1 ac3 sound,...) ?

BTW Dolby Pro Logic II Decoder are much more sophisticated than Pro Logic (I). They try to predict the direction of the sound. For more info, I suggest reading the specs from http://www.dolby.com/tech/ (Dolby Surround Pro Logic II Decoder Principles of Operation).

kwag · Member Joined: 12 Jun 2001

vitualis · Moderator Joined: 01 Oct 2000 Location: Australia

You are basing your judgement on 128 kbit/s MP2??

Actually, this makes a wacky kind of sense now (speculatively so anyway). Your audio is bitrate starved to begin with and since stereo can dynamically adjust the bitrate "need" between channels, it is not inconceivable that it can also do it badly (e.g., poor VBR encoding for video can actually WORSEN the video quality -- e.g., older version of TMPGEnc).

If the encoder inappropriately decreases the bitrate of one channel too much, you get the effect show by the low bitrates lines on ajung's graphs -- that is, poor fidelity with the original audio.

This may simply be a symptom that the MP2 encoder used has an algorithm that allocates bitrate poorly at lower bitrates. Paradoxically, the "dumb" dual channel method of bitrate allocation may be yielding superior audio fidelity.

Regards.
_________________
Michael Tam
vitualis' Medical Rant :: stay updated, read the rant
The Medicine Box
GNU VCDImager Authoring Guide Series

kwag · Member Joined: 12 Jun 2001

@vitualis,

Surround signals are carried on a frequency band from 100HZ to 7Khz, and that's a long way from 128Kbps ( and bit rate starvation ) so your statement is invalid.

@All,
What I can see is that this problem is only present in Dolby Prologic II and not in regular Dolby Prologic. In Dolby Prologic II, there is channel separation on the surround channel ( rear speakers ) and what I think is happening is that the stereo bit rate differences are causing phase shifts that are affecting the rear channel separation. This is why the artifacts are heard. The receivers are probably switching from Prologic (surround) to Prologic II (surround 2) in very short (millisecond) bursts and that is what's heard on the rear speakers. When the signal is dual channel, the bit streams are constant on each channel, and there's no effect from the adjacent channel as there is in stereo. That's the only thing I can think of which could be the cause for this signal "falsing". So there HAS to be a problem with the encoders. If the encoders are even slightly altering bit rates on the low frequency surround signals components, which should remain untouched from bit rate variations from the adjacent channel even if there's no sound on a stereo channel, then this is where the problem is. Maybe the encoders should follow a minimum "floor" bit rate to take this into consideration, so that even if there's no sound on one channel, the bit rate will never drop below a specified minimum and then the low frequency surround signals won't have any variations.

Some data here: http://www.extremetech.com/article2/0,3973,32655,00.asp

-kwag
_________________
KVCD.Net - Advanced Video Conversion
http://www.kvcd.net

Martyn1980 · Member Joined: 27 May 2002 Location: United Kingdom

My god ive been away for a few months, but the same arguments are here

My problem with Dual CHannel is, and i think ive asked this before to Virtualis, when i encode to Dual Channel, i lose the centre speaker in Dolby Pro-Logic 2, its fine if i use Joint Stereo or Stereo.

Is this a common problem, or have i just got a dodgy amp.

I personally prefer Dual Channel as it does seem to have better seperation that standard stereo, but i wouldn't normally use less than 224k, as 128 means theres only 64k per channel, which seems very inadequate to me.

kwag · Member Joined: 12 Jun 2001