Dolby Pro Logic II encoding test (90° vs 180° phase shift)

[3dsnar]

Here are two clips,
http://forum.videohelp.com/images/gu...165/180vs90.7z

And the reference (original 6 channel input) clip.
http://forum.videohelp.com/images/gu...helem_6chnl.7z

Each downmixed with:
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{180°} + 0.5 SR{180°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{0°} + 0.866 SR{0°}
and
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{90°} + 0.5 SR{-90°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{-90°} + 0.866 SR{+90°}

Please test them with your dolby certified decoders
and decide which ones sounds better (eg. A vs B version) with reference to the original sound.
The poll question is related to fifthelem_A.mp3 and fifthelem_B.mp3,
while speech_A.mp3 and speech_B.mp3 should sound identical
(to see that both downmixing methods are compatible with DPLII decoders)

[scharfis_brain]

I cannot tell for sure (tendency goes to clip A) which is the 180 or 90 degree clip because I curently have no possibility to play back the AC3 in direct 5.1 off my PC.

but anyways this is NOT a fair comparision.

fair would have been:

Each downmixed with:
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{180°} + 0.5 SR{180°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{0°} + 0.866 SR{0°}
and
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{+90°} + 0.5 SR{+90°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{-90°} + 0.866 SR{-90°}

or Rockarias style:

Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{0°} + 0.5 SR{180°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{180°} + 0.866 SR{0°}
and
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{+90°} + 0.5 SR{-90°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{-90°} + 0.866 SR{+90°}

but you are testing two types of four concurring types.
the traditional mixing with 180°
and rockarias inverse mixing with 90°

And I think this is perfectly audible when you can hear "weapons loaded" at the very beginning of the sample.

sample A lets it sound from the center.
sample B lets it sound from the surround.

But this is not due to the different phases but more due to the reason that one of the surround channels (in relation to each other) become inverted with your 90° matrix. (IMO)

(this is the thing I always tried to explain as center surround issue to Rockaria. But it seems to be a slightly different effect here)

so would you redo the samples, please?

[Rockaria]

http://forum.doom9.org/showpost.php?...1&postcount=42

Originally Posted by scharfis

IMO this matrix is useless.
Just imagine the MonoSurround-condition: Rear left and rear right are carrying the same signal.
With this matrix you'll succesfully eliminate the mono surround out of the downmix.
This means to me, that

Dolby Pro Logic Left   Right   Center Rear Left Rear Right 
Left Total        1.000 0.000  0.7071  0.866     -0.5 
Right Total      0.000 1.000  0.7071  -0.5        0.866

is a derivation from the faulty matrix above and should not be used.
You may experience wider sounding surround channels, because every middle information (mono information) is weakened in the surround downmix.

Originally Posted by scharfis

or Rockarias style:

Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{0°} + 0.5 SR{180°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{180°} + 0.866 SR{0°}
and
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{+90°} + 0.5 SR{-90°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{-90°} + 0.866 SR{+90°}

but you are testing two types of four concurring types.
the traditional mixing with 180°
and rockarias inverse mixing with 90°


(this is the thing I always tried to explain as center surround issue to Rockaria. But it seems to be a slightly different effect here)

Originally Posted by scharfis

I found, that Rockaria's personal matrix has problems with centered surround effects that are meant to be reproduced by both surround speakers (DPLII) or the surround-back speaker (DPLIIx).
Inverting the phase of one of the Surround channels like this modified matrix does can enhance the perceived width and separation of both channel because it is simply something like crosstalk reduction. But At the cost of centered surround sounds. That's why I prefer the unaltered matrix.

.....
....

@scharfis, do you see your posts are 'arbitrary itself' ?
Would you stop posting against anybody with no basis & no understanding at all? That does not help anything.

traditional mixing with 180° :: what the traditional means? your myths?, your Gods' words?
inverse mixing with 90° : youself admitted you have no clear picture of the DPL II model already?
If you continue posting like this, I will regard them as personal attacks, only to depreciate any resonable conversations.

[scharfis_brain]

I don't see a contradiction to myself here. I think that these quotes still enhance the statements of my first post in this thread.
Also it is not a personal attack as I said many times in past, too.
If you are interpreting my answers as personal attacks it is your problem. Not mine!

traditional mixing with 180° :: what the traditional means? your myths?, your Gods' words?

The wikipedia style of matrix. The Besweet style of matrix. The AC3filter style of matrix. The ffdshow style of matrix. Is this enough traditional style?

inverse mixing with 90° : youself admitted you have no clear picture of the DPL II model already?

Of course not, as I already did some manual mixes myself, build some active Dolby surround mixing ciruits etc.

If you should find irony in the last sentence, keep it for yourself. Thanks.

Actually I think you are the person that is not able or willing to do a objective conversation without personal attacks.
I never did attack you. But you do!

[Rockaria]

@scharfis, I am pretty much impressed by your convenient logic adaptible to any situations.

In my previous quotes :

The contradiction in the 1st and 3rd quoted messages from you looks to me clearly an irony. (and the 2nd one just a justification)

The 'inverse 90 degree' is actually the +-90 degree phase shifts, to be correct, based on my best reasoning & understaning which describes the DPL II model from Dolby's unclear documents.

Look at the 1st quote.
Now do you distinguish which is useless? : my useless model vs your invaluable objections

scharfis, I just want you not to play with my id at all.

@3dsnar,
The two fifth_elm clips showed a distinct difference in the first part, the _b clip sounded some wider fronts
But I have no idea which is with better seperation fidelity without the original 6ch clip(in any form such as mp4) as I mentioned in other thread.

[edit] Yeah, I see tebasuna has the same opinion.
Now with the original 6ch AC3, I see the fifth_elm_B has the better seperation fidelity. voted.

[3dsnar]

OK, thanks all of you for votes.
BTW. The original signal is provided too.

Yes, B is 180 deg phase shift (the simpler approach),
A is 90 deg. phase shift.

The sign variations are not important, because they result
in phase invertion in the decoded signal and do not affect the separation (as Tebasuna already noticed). Therefore A and B are enough to distinguish between 90 and 180 phase shifts.
I agree with Scharfis that the 90 deg phase shift should have been prepared with the same sign style, to be fully consistent with the 180 deg. shift. Maybe next time

(early) conclusions.
1) There is a noticable difference between the 90 deg phase shift and 180 deg phase shift
2) 90 deg seem to produce significantly worst results, thus probably the DPLII downmixing equation is based on 180 deg. shifts (simple sign change).

---

BTW. Please test Aud-X DSfilter DPLII decoder (or rather DPLII decoder simulation),
by sending the output as AC3 through SPDIF to your HT amps.
I am curious of your opinions and especially your thoughts regarding its quality vs certified dolby decoders quality

3d

[Rockaria]

Originally Posted by 3dsnar

OK, thanks all of you for votes.
BTW. The original signal is provided too.

But after my post in other thread, without any notice or acknowledgement.
http://forum.doom9.org/showpost.php?...9&postcount=63

You might want to add two more models to fully reflect the discussions in the related threads:
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 BL{-90°} + 0.5 BR{-90°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 BL{90°} + 0.866 BR{90°}
and
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{180°} + 0.5 SR{0°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{0°} + 0.866 SR{180°}
...
You might also want to test with original 6ch(mp4 format) vs dpl II mix(music + speaker test clip), to compare reasonably.
The FFDShow can switch between PCM/AC3 in digital out mode on the fly making it easy to compare.
As I said many times, the speaker test clip is most generous on any models(more than 95% of the seperation quality).

Also the matrices values(Ls1,Ls2,Rs1,Rs2....) might affect the seperation quality by the phase shift degree change.
Comparing the Wikipedia matrix with the current one would be reasonable(also making it reasonable using the variable reference than the values).

So far, the fifthelem_B showed noticeable better seperation, but the test is not setup to compare the seperation fidelity(original vs DPL II).

Originally Posted by 3dsnar

Yes, A is 180 deg phase shift (the simpler approach),
B is 90 deg. phase shift.

Can you provide the proof in a source format(no dll)?

Originally Posted by 3dsnar

The sign variations are not important, because they result
in phase invertion in the decoded signal and do not affect the separation (as Tebasuna already noticed). Therefore A and B are enough to distinguish between 90 and 180 phase shifts.

Again, it's just your very dangerouse assumption.
You can test it with FFDShow which can adjust the matrix value with the sign.
It shows no difference only with the simplest speaker test file.
Also prove exactly which Tebasuna's observations concure your assumptions.

Originally Posted by 3dsnar

I agree with Scharfis that the 90 deg phase shift should have been prepared with the same sign style, to be fully consistent with the 180 deg. shift. Maybe next time

also check my quote above. indeed, the polls feels like something public.

Originally Posted by 3dsnar

(early) conclusions.
1) There is a noticable difference between the 90 deg phase shift and 180 deg phase shift : agreed.
2) 90 deg seem to produce significantly worst results, thus probably the DPLII downmixing equation is based on 180 deg. shifts (simple sign change).

I agree again we are using totally differernt languages. Apparently it seems what you wanna see.
---

Originally Posted by 3dsnar

BTW. Please test Aud-X DSfilter DPLII decoder (or rather DPLII decoder simulation),
by sending the output as AC3 through SPDIF to your HT amps.
I am curious of your opinions and especially your thoughts regarding its quality vs certified dolby decoders quality

My independant Ad. :
Sorry, unfortunately, I am very much satisfied with the FFDShow with probably 1000% of proven better features and stability.
I also believe they will provide the DPL II encoding with 90 deg. phase shift very soon(with current adjustable matrix).

[Rockaria]

Was it a blind test to fool people by an anonymous person?

Originally Posted by 3dsnar original in the poll description

Here are two clips,
http://forum.videohelp.com/images/gu...165/180vs90.7z

And the reference (original 6 channel input) clip.
http://forum.videohelp.com/images/gu...helem_6chnl.7z

Each downmixed with:
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{180°} + 0.5 SR{180°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{0°} + 0.866 SR{0°}
and
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{90°} + 0.5 SR{-90°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{-90°} + 0.866 SR{+90°}

Originally Posted by 3dsnar that I captured just after his conclusion

Yes, A is 180 deg phase shift (the simpler approach),
B is 90 deg. phase shift.

Originally Posted by 3dsnar what you are seeing now

Yes, B is 180 deg phase shift (the simpler approach),
A is 90 deg. phase shift.

You definitely need to provide the encoding source of your test poll.

[3dsnar]

I have informed you of the input included (after you suggested so).
You should read my posts more carefully.
-----------
I am asking to verify my DPLII simmulation decoding algorithm,
not the entire DSfilter.
-----------
Here is the matlab code that I used for downmixing.

function [y, y2]=DPLIIdownmix(s)

FL=s(:,1);
FR=s(:,2);
C=s(:,3);
LFE=s(:,4);
SL=s(:,5);
SR=s(:,6);

y=zeros(length(FL),2)
y2=zeros(length(FL),2);

wgF=1;
wgC=sqrt(0.5);
wgA=sqrt(0.75);
wgB=sqrt(0.25);

Nrm=wgF + wgC + wgC + wgA + wgB;
wgF=wgF/Nrm;
wgC=wgC/Nrm;
wgA=wgA/Nrm;
wgB=wgB/Nrm;


%180 deg phase shifts
y(:,1) = FL*wgF + C*wgC + LFE*wgC - wgA*SL - wgB*SR;
y(:,2) = FR*wgF + C*wgC + LFE*wgC + wgB*SL + wgA*SR;


%90 deg phase shifts
y2(:,1) = FL*wgF + C*wgC + LFE*wgC + wgA*imag(hilbert(SL)) - wgB*imag(hilbert(SR));
y2(:,2) = FR*wgF + C*wgC + LFE*wgC - wgB*imag(hilbert(SL)) + wgA*imag(hilbert(SR));

[tebasuna51]

Originally Posted by 3dsnar

BTW. Please test Aud-X DSfilter DPLII decoder (or rather DPLII decoder simulation),
by sending the output as AC3 through SPDIF to your HT amps.
I am curious of your opinions and especially your thoughts regarding its quality vs certified dolby decoders quality

Sorry, I have attached my PC to a old amp with only stereo input (with DPL I capability).

To test 5.1 or DPL II I need to burn a CD and go to other room with a DVD/DivX/mp3 player attached to the SONY receiver.

I tried, to make some kind of test, use your Aud-X DSF in GraphEdit but don't accept WAVDEST or DUMP output. Only accept DirectSound output (like you say in other thread).

For my configuration your Aud-X DSF is unusable.

[Rockaria]

@3dsnar, you seem to be very confident about ...what?
Anyway, I myself won't engage in such flip-flop games any more.

http://en.wikipedia.org/wiki/MATLAB : something that costs money for what I do not owe.
http://en.wikipedia.org/wiki/Hilbert_transform : seems to be the logic for transforming a square wave form to a strange curve with 90 deg. phase shifts

Possibly we can find the corresponding function in any forms of avisynth/sox plugin..
It's gonna take a long while to prepare the verification test.... oh well, nothing to believe.

It seems it has just lengthened the route further, forcing me to continue my own approach.
At least, I believe I have proved the reference model 4 better than 2 practically, and 1/3 better than 2/4 theoretically. period.

/fooled

[tebasuna51]

Test with a SONY STR-DE495 receiver, DPL II Movie mode.

There are a clear difference between A and B in "weapons loaded" (?) and "Yes, Sir" (3 sec.).
There are presents in FL, FR, BL, BR original audio, and not at Center channel.
In A "Yes, Sir" is basically only at Center channel.
In B "Yes, Sir" is present in all five channels, then is not perfect but better than A.
Then my vote for B.

I tried also a C option (matrix 3 from original thread) with indistinguishable differences with B (at least for my ears).

[ursamtl]

If you have Plogue Bidule, why don't you just model the Dolby encoding using it? Bidule has a built-in Hilbert Transform filter (although in my experiments, it seems to produce a -90° phase shift, so for the +90° shift, you'll need to invert the signal). Another option for the Hilbert is to use Christian Budde's excellent Phasebug VST plugin (freeware). This gives you the possibility of any phase shift in a 360° circle.

If you don't have Plogue Bidule, try Audomulch.

I haven't done much with DPLII since I've never been impressed with its results, but I can tell you that in my experiments with Ambisonics, the 90° phase shift used in some encoding designs is essential. For example, in the superstereo circuits I've modelled using Plogue Bidule and Phasebug, the 90° phase shift balances the ambience nicely across the surround speakers, but moving it to 0° or 180° forces the ambience all to one surround speaker on another. Again, the application is different from DPL, but I've read a lot of surround sound documentation over the past couple of years and the Hilbert Transform is everywhere!

Finally, I know you guys are passionate about your points of view, but let's relax and enjoy this experimentation.

Regards,
Steve.

[scharfis_brain]

one question: how was the hilbert transform performed nearly 20 years ago when Dolby introduced their "Dolby Surround" ?

I know, that bandpass filters alter the phase. But the phase deviation is not constant with changes in frequency.
Also I think that 25 years ago computers weren't fast enough to do such calculations in realtime.

Do you have an idea how dolby 'could' have made the frequency independent phase shift?

[ursamtl]

The Hilbert Transform was done electronically. You can find quite a few circuits around on the net with a bit of judicious googling. It's sometimes known as a "dome filter."

[scharfis_brain]

Oh! That's nice stuff. Chaining some filters to achieve a near to constant phase shift over a defined range of frequencies.

[scharfis_brain]

I just did a small test with simple 180 degree mixing how the sign affects the surround image:

The Normal mixing:


And the inverted matrix:


samples can be downloaded here:
http://home.arcor.de/scharfis_brain/...raditional.rar

3dsnar: now compare my inverted sample to your 90 degree sample. both show the same behavior: "weapons loaded" comes from the center.
And that is why your 90 degree sample also is inverted.

I hope that it now should be clear what I meant all the time that Rockarias personal (inverted) Matrix destroys some of the surround information.

[Rockaria]

I will also have to agree that 'Rockarias personal (inverted) Matrix' destroyed scharis brain cells totally.
Sorry now brainless, I feel the responsibility. So what can I do for you?

@ursamtl, I appreciate your invaluable unbiased information.
I will look into the mentioned tools and try to find a way in my environment to apply the DPL II 90 degree phase shifts by the spec in the resonable models.

By the way, I seem to have made some annoying useless neighbors unavoidably.
My wife says she's gonna move right now although I want to give them some more opportunities to make themselves good citizons until this weekend.
The apatment is wooden, transfering the earthqakes every once in a while and the alu. shutters shap ultra sonic waves poking awaken my ears, soul and body more than 30 times a day. Their languages are nothing but noises to me, but they seems proud of keeping differnt languages in every different villlage, just 5 miles away.
My best idea atm is turning up the VOA(the foreign language to them) out from the window around the same origin(balcony) to neutralize. And it seems working now. But my wife says it's gonna go back tomorrow like yesterday saying there are such memories that don't last a day.

[3dsnar]

Originally Posted by scharfis_brain

3dsnar: now compare my inverted sample to your 90 degree sample. both show the same behavior: "weapons loaded" comes from the center.
And that is why your 90 degree sample also is inverted.

I hope that it now should be clear what I meant all the time that Rockarias personal (inverted) Matrix destroys some of the surround information.

Scharfis, I agree with you (partly).

Yes, the inverted version causes wrong position of some sounds in the surround panorama (in case of the 180 phase shifts).

The 90 deg. phase shift in general destroys much more significantly the complete sound (i.e. the inverted 180 deg shift differs from the inverted 90 deg shift. The 90 deg is even worst.).

So far it seems that the the best results are obtained with this equation:
Lt = FL{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.866 SL{180°} + 0.5 SR{180°}
Rt = FR{0°} + 0.7071 C{0°} + 0.7071 LFE{0°} + 0.5 SL{0°} + 0.866 SR{0°}

Cheers, 3d.

[Rockaria]

It's gonna be my last wasteful discussion on the unclear theory until I get the resonable fully considered test results to verify the models, unless Rockaria is pointed directly or indirectly hereafter.

Originally Posted by Rockaria's draft test plan

4. identify target DPL II models(formula) to compare
<m11>
Lt = mix(Lf.0°, C.0°, Ls1.-90°, Rs2.-90°) == mix(mix(Lf, C), mix(Ls1, Rs2).-90°)
Rt = mix(Rf.0°, C.0°, Ls2.+90°, Rs1.+90°) == mix(mix(Lf, C), mix(Ls2, Rs1).+90°)
<m12> rears(m11).-90°
Lt = mix(Lf.0°, C.0°, Ls1.-180°, Rs2.-180°) == mix(mix(Lf, C),-mix(Ls1, Rs2))
Rt = mix(Rf.0°., C.0°., Ls2.0°, Rs1.0°} == mix(Rf, C, Ls2, Rs1)
<m13> m11.-90° == m11?
Lt = mix(Lf.-90°, C.-90°, Ls1.-180°, Rs2.-180°) == mix(mix(Lf, C).-90°,-mix(Ls1, Rs2))
Rt = mix(Rf.-90°, C.-90°, Ls2.0°, Rs1.0°) == mix(mix(Rf, C).-90°, mix(Ls2, Rs1))
...
<m21>
Lt = mix(Lf.0°, C.0°, Ls1.+90°, Rs2.-90°) == mix(mix(Lf, C), Ls1.+90°, Rs2.-90°)
Rt = mix(Rf.0°, C.0°, Ls2.-90°, Rs1.+90°) == mix(mix(Rf, C), Ls2.-90°, Rs1.+90°)
<m22> rears(m21).-90°
Lt = mix(Lf.0°, C.0°, Ls1.0°, Rs2.-180°) == mix(mix(Lf, C, Ls1),-Rs2)
Rt = mix(Rf.0°, C.0°, Ls2.-180°, Rs1.0°) == mix(mix(Rf, C, Rs1),-Ls2)
...

When -180° = -, +90° = Hilbert(), -90° = -180°.+90° = -.+90°

As it explaines, the Rt in <m12>, has the image mixed with all 4 channels in the same phase region 0, making it the worst choice for the seperations theoretically.(it's a very simple math). It seems concuring my test results with the prevailing <m12> having Lf<Rf, if you read my original thread.
Now the seperation efficiency more clearly looks like m21>m22>m11>m12, which of course needs some simple listening tests.

If I were to explain why the rears have the coefs shared on different channels based on the <m21>, <m22> model is :
. there may exists some overlappings of the images spanning on the different phase regions.
. by comparing the two identical but somewhat overlapped rear coef images, it would be possible to get the closer original channel image enhanced by the servo feedback.

/OPEN minded & mutual respects to the anonymous...