Dolby Surround 2

[trg100]

I've been experimenting with using BeSweet (1.4b11) to convert 5.1-channel AC3 files to DS2 encoded 2-channel versions. I have used WinDVD 4 and PowerDVD 4 in 4-channel mode with DPL2 enabled to play back the DS2 tracks on my 4-way SBLive and speakers.

Screen shots can be seen here (http://members.aol.com/trg100uk/ds2_tests.html).

I've made myself a test 5.1 AC3 file with SoftEncode. It's just a mono CD track switched around the channels (See 1st pic). I've also checked the results are simliar with a DVD soundtrack.

I've found many odd things and it would take ages to explain it all. Basically though, it works :) I get reasonable separation on the rear speakers. However,

1. The DPL2 and 4-speaker combination in both DVD players only works properly if I take the 2-channel DS2 encoded wav file and rencode it to 2-channel AC3 with SoftEncode. Playback of the wav or an mp3 gives various problems with channel mapping, relative volumes, and even crackling popping noises :(

2. The volume in the portion of the DS2 file representing the centre channel is much higher than the other sections. Compare the results of downmixing with -s surround (2nd pic) and -s surround2 (3rd pic). The dialogue is disproportionately loud in the movie track too.

3. There is some slight leakage of sound to the wrong channels - mostly muffled high frequencies. But I guess it's not gonna be perfect.

Can anyone offer an explanation for points 1 and/or 2? What kind of experience have others had with DS2 and DPL2? Any other comments?

[trg100]

The screenshots (http://members.aol.com/trg100uk/ds2_tests.html) page now has a 4th image. This shows the result of headac3he's DS2 downmix algorithm. Here the relative volumes seem more sensible. The centre information is slightly quieter in both channels than the left/right only sections are in their respective channels. This presumably accounts for the fact that the former is played through 2 speakers.

So good so far. All is not great though, look closely at the first 1/3 of the left channel section. This quieter section (check original in SoftEncoder shot at top of pics page) is bizarely over amplified?!

Apart from this small problem it's possible to reduce a 5.1 AC3 to a 2-channel file, compress it and save loads of space!!! That's cool, no? Now if only the DPL2 decoders in PowerDVD/WinDVD would work properly with mp3 files. At the moment I have to re-encode to AC3 (see previous post) which is nonsense.

Also, I tried to compensate for the very loud centre output in BeSweet using -C -12db but this does not seem to have any effect when using -s surround2. A bug? Wouldn't like to say:) Pity DSPguru is away.

[kxy]

Originally posted by trg100

1. The DPL2 and 4-speaker combination in both DVD players only works properly if I take the 2-channel DS2 encoded wav file and rencode it to 2-channel AC3 with SoftEncode. Playback of the wav or an mp3 gives various problems with channel mapping, relative volumes, and even crackling popping noises :(


How do you decode your mp3, I am assuming you are using lame, but what about the command lines/preset?

Have you tried vorbis? If not give it a try. Try -q 4.99 and -q >= 5 and see what kind of result you will get. For me -q 4.99 produce warbling effect on my rear speakers and -q produce less artifacts.

[trg100]

How do you decode your mp3, I am assuming you are using lame, but what about the command lines/preset?
I'm confused about whether you're talking about encoding or decoding. I am using PowerDVD and WinDVD to decode all file types because I am not aware of anything else that has Dolby Pro Logic 2 routines. The mp3s that I tried to play back were encoded using lame with --alt-preset insane (didn't care about size). I don't think this is important though since the (poor) results are the same using a wav file or an mp3, the only reason I tried mp3 was because WinDVD won't play a wav file. The screenshots (http://members.aol.com/trg100uk/ds2_tests.html) show the wav files in CoolEdit.

Have you tried vorbis? If not give it a try. Try -q 4.99 and -q >= 5 and see what kind of result you will get. For me -q 4.99 produce warbling effect on my rear speakers and -q produce less artifacts.
You're talking about some king of multichannel .ogg file I guess, but I'm not exactly sure what. If you know of a Pro Logic 2 capable vorbis decoder I would be very interested! I never tried .ogg because neither DVD players recognised them.

[kxy]

okay, it is I who got confused in regards to your test. I have a external amp decoder for my computer along with 5.1 speakers. Therefore I did use my dvd-player to test out the results.

[MaTTeR]

Didn't PP(Peter) write the DS2 plugin and implement it in WinAmp long ago? You might check the preference dialog in WA. I know he wrote the code in a couple of hours, so I think it should be useable now.

[trg100]

Thanks Matter but the only DS2 plug-in I know of is the one written by DSPguru. I don't think this works with winamp. I can't find anything about a DS2 plug-in for winamp, there's nothing on Peter's homepage (http://www.blorp.com/~peter/). I looked at the options for all the winamp plug-ins I have but can't see anything about DS2.

Have I got the wrong end of the stick?

[frank]

2. The volume in the portion of the DS2 file representing the centre channel is much higher than the other sections. Compare the results of downmixing with -s surround (2nd pic) and -s surround2 (3rd pic). The dialogue is disproportionately loud in the movie track too.That's why I started some tests for DS2 (Dolby Surround Pro Logic II) encoding. Results:
1. The centre channel sounds too loudly. (+16 dB!)
2. The Surround channels are 180 degree wrong encoded to the stereo waves. Lt should have the negative surround component, not Rt.

I encoded a DS2 wav with HeadAC3he, and all seems to be ok.
But who knows the used downmix matrix??

Thanks for the pictures trg100.
I'll make some suggestions in a new thread.

It's a pity that DSPGuru is gone away for some weeks.

[trg100]

New pictures here (http://members.aol.com/trg100uk/ds_tests_2.html).
--------

Hey frank it's good to hear that someone else is interested in this issue and that you've spotted the centre channel volume problem in BeSweet too. I look forward to your suggestions :)

I don't really follow your second point but only because I don't know enough about how the surround information is encoded. Would the problem you describe affect sound placement?

Are you using a Prologic 2 or equivalent compatible external amp to decode dolby surround? If not, what software do you use for playback?

The new pictures page (http://members.aol.com/trg100uk/ds_tests_2.html) has screen shots of some new tests I ran using a constant test signal. It's easier to see what's going on. This time normalisation was used in both BeSweet and HeadAC3he.

There are three main issues:

1. The aforementioned high volume on the centre channel in BeSweet's DS2 downmix (screenshot 2).
2. HeadAC3 seems to have a normalisation problem. The very start of the wave is strangely loud, see screenshot 6.
3. The lack of Dolby Pro Logic 2 enabled software to play back DS2 files. PowerDVD 4 and WinDVD 4 work well but only when wav/mp3 files are transcoded to 2.0 AC3s which is weird and annoying!

[frank]

My article was ready and then crashed my computer :(
Hey your last pictures show very good my theory of operation. Look at my thread about downmix matrix.

1) The surround signals are encoded with 180 degrees phase shift into the stereo tracks. Lt must have the negative part. The basic downmix equations you can read in the ATSC A52 document to AC-3 standard.
DVD2AVI sources include downmix matrix too.

2) I only own a Pioneer DS receiver (but with MPEG-2 mc audio decoder :) ). But I have seen the first DS2 receivers from Denon, Yamaha. And I intend to buy... later.
I create SVCDs with best sound. And from now I want to encode in DS2, because it is fully compliant and needs not more space (192 kbps).
I see DS2 upcoming. It doesn't violate the SVCD standard, because of analog stereo tracks.

3) I haven't seen the normalize peek because I hate normalizing. It's not professional. AC-3 has defined levels. I use normal compression and static gain of 10dB to compensate the matrix attenuation -> ever the same dialog level!

4) The DS2 system is very new. Software players and drivers have bugs. We must wait. -> PowerDVD recognizes my SPDIF only on AC3 DVD movies!!

[kxy]

Originally posted by trg100
[B]New pictures here (http://members.aol.com/trg100uk/ds_tests_2.html).
--------

The new pictures page (http://members.aol.com/trg100uk/ds_tests_2.html) has screen shots of some new tests I ran using a constant test signal. It's easier to see what's going on. This time normalisation was used in both BeSweet and HeadAC3he.

There are three main issues:

2. HeadAC3 seems to have a normalisation problem. The very start of the wave is strangely loud, see screenshot 6.



Since DA has being very busy lately, I took the liberty of quoting him.

Here is the a quote
"oh wait, I did my own test file, and I think what he describes as "normalizing" error is in fact an "encoding error". At least I got the same "error" when I decoded the ac3 with sfse and normalized afterwards. I'll check the graphs again on that page, but I guess it is not HAC3's fault."

Attached the pic with SFSE decoded ac3. (I guess SFSE can be seen as reference decoder...) If one knows how ac3 works, one wouldn't be too surprised about this effect...

About your file, the SFSE produces the same "problem", so it is *not* a decoding/normalizing problem of HAC3. (see pic)
Maybe it has something to do with DRC.

ssfe pic (http://icdweb.cc.purdue.edu/~yaok/sfse.png)


About the lowpass problem that canadian_fbi posted, in the other thread. It has *nothing* to do with HAC3 and BeSweet. BeSweet is using shibata's eq for lowpassing. Hac3 just tell vorbis to lowpass since HAC3 has *no* lowpasser.

[trg100]

Interesting kxy. I was careful not to cry "bug!" since I wasn't confident that there wasn't a sound reason for this behaviour. At the time though, it was unexpected. I can't check now and my shots don't quite have the definition to prove it, but I am almost certain this peak did not occur with BeSweet. I wonder why that is.