Upmixing - where are we at? Have people compared upmixers?

[KMO]

I just noticed that Dolby Surround puts a lot of signal to the center channel, more so than Auro3D. In order to level things out and make it sound even better, I reduced the center channel level by -6 and increased the overall master volume to compensate. I've noted that in my settings for anyone wishing to replicate:

Unless they have a fundamental speaker misbalancing error that needs correcting, that sort of recalibration is going to disturb the balance - you're effectively reducing the relative level of all mono content, by reducing a speaker that's dedicated to mono. It's not just a speaker weighting change. To actually "even" it you'd need to send more mono content to L+R, not just reduce C.

I would be interested to know what the balance of L/C/R mono content looks like in Dolby Surround. I'm used to Dolby Pro Logic II, and I just spent a few minutes testing what it looks like.

Test signal was L+R pink periodic speaker cal (500-2000Hz) noise from REW. Measurements all done at (or close to) main listening position, isolating particular speakers.

(CW=0 is what Dolby Pro Logic II Movie does, and CW=3 is the default for Music mode, so L+R total level equals C level. That signal was at 78.8dB SPL on L+R without the Dolby Pro Logic II enabled, so we can see that CW=7 doesn't interfere with L+R at all - it totally mutes C. Whereas with CW=0 L+R still get a bit of leakage 30dB down).

In that test, it does look like C needs a bit of a boost - possibly 2dB. A separate test suggests that's half down to my calibration or mic position - C is 1dB low compared to L without the DPLII. The other 1dB could just be general uncertainty about how signals acoustically sum.

I had assumed the DSU "Centre spread" was something like Centre width 3. If you really needed to drop it 6dB to get in line, that would mean something more like Centre width 1. But then having done that, all your mono would be far too quiet. Unless they'd screwed up and it was something like L+R=75dB and C=81dB, so the total mono level was too high to start with?

 

[Chromatischism]

Unless they have a fundamental speaker misbalancing error that needs correcting, that sort of recalibration is going to disturb the balance - you're effectively reducing the relative level of all mono content, by reducing a speaker that's dedicated to mono. It's not just a speaker weighting change. To actually "even" it you'd need to send more mono content to L+R, not just reduce C.

I am using Center Spread as noted in my post. It is close to the Auro3D and Stereo balance now.

 

[KMO]

I am using Center Spread as noted in my post. It is close to the Auro3D and Stereo balance now.

I know - I was writing on that basis. Whether it's on or not, reducing the centre speaker level will reduce the relative level of anything panned to centre position in the mix compared to stuff panned left or right or rear.

Can you provide any measurements to illustrate the effect of this tweak? That would reveal whether centre-panned stuff was too loud before, and if it's now too quiet.

(I guess this tweak works if you're playing totally mono stuff though, so there's no issue of balance with stuff panned to other positions...)

 

[Chromatischism]

I know - I was writing on that basis. Whether it's on or not, reducing the centre speaker level will reduce the relative level of anything panned to centre position in the mix compared to stuff panned left or right or rear.

Can you provide any measurements to illustrate the effect of this tweak? That would reveal whether centre-panned stuff was too loud before, and if it's now too quiet.

(I guess this tweak works if you're playing totally mono stuff though, so there's no issue of balance with stuff panned to other positions...)

I don't think there is any center panning in upmixed content. I also don't listen to anything that's mono.

 

[Dj7675]

I am using Center Spread as noted in my post. It is close to the Auro3D and Stereo balance now.

I agree DSU with center spread is really good. Without it a lot of the L/R is dumped into the center.

 

[Chromatischism]

I agree DSU with center spread is really good. Without it a lot of the L/R is dumped into the center.

It actually took me a while to figure out what was bothering me about it. It wasn't until I was sitting up by the AVR, which is also next to the center speaker. I then noticed how loud it was compared to Auro3D, so I simply wanted to make them closer to equal, so the center now has a channel level of -6 on my DSU quick select. Then with Center Spread on, it's pretty good. I like it a lot more now and go between it and A3D. If you make the changes I did, you can switch between them to understand where they work best without all this extra nonsense like boosted bass or center levels.

The decision on which one to use is basically: do I feel it's better for this track to mostly stay up front (A3D) or is it conducive to bringing the surrounds into play more (DSU)? The answer really depends on the content and my mood. And if it's a movie, it's between DSU and Neural: X. Now that I have DSU tweaked, I should test it in more movies instead of defaulting to Neural: X every time.

 

[KMO]

I don't think there is any center panning in upmixed content. I also don't listen to anything that's mono.

Maybe I'm not explaining myself very well.

In stereo content there will be stuff that is panned by the original mixer of the material to left, centre or right in the stereo mix. They will be crossfading to achieve the levels they want. They will be expecting a certain balance to be maintained. In particular, something coming sent to both left and right at -3dB should roughly match something sent to just left or right alone at 0dB.

The upmixer gets a lot of flexibility in where it places stuff throughout the speaker arrays, but it needs to try to preserve the basic levels of the stereo mix.

If centre spread is disabled (or width=0), something pure mono coming in both L+R in the 2.0 mix at -3dB will come out of just centre, at 0dB, to adjust for that difference between one and two speakers.

If centre spread is enabled, it will be sending signal to all of L+C+R in some proportion, aiming again to achieve the same overall 0dB level. In the absence of measurements from you, I'm going to guess that DSU's centre spread is the same as DPLII's width=3, which seems to be roughly centre -2dB and left and right -7.5dB each.

Now, maybe DSU is like width=3, and you don't like the centre still being 5dB louder than either L or R for centre-panned signal. It's too "narrow" for you. You'd prefer to have width=4 or more, so the centre is not louder. That's fair enough - the weighting of C versus L+R is a matter of taste.

But you can't correct that width well by turning down centre alone. Something in the stereo mix with L+R both at -3dB, that should end up summing to 0dB is now coming out of 3 speakers at -7.5 to -8dB, which is too low. Stuff panned to centre by the original mixer is now too quiet compared to the rest of the mix, by about 3dB. You may have corrected the "width" to your taste, but you've upset the original mix by de-emphasising stuff the original mixer put in the centre, compared to stuff they put left, right or surround.

Illustrating in numbers:

Width	Left level = Right level (dB)	Centre level (dB)	Total level (dB)
0	-inf	0	0
1	-11.5	-0.7	0
2	-9.3	-1.2	0
3	-7.4	-2.0	0
Chromaticism's (width 3, C lowered)	-7.4	-8.0	-2.8
4	-4.8	-4.8	0
5	-4.0	-6.7	0
6	-3.5	-9.1	0
7	-3.0	-inf	0

All this is predicated on my belief that Dolby Surround's spread mode will have a correctly-summing set of L/C/R, like all the widths of Dolby Pro Logic II.

If you can show that Dolby Surround is failing to correctly level centre-panned content, then "-6dB to centre" could be valid advice. Otherwise, it's as objectively wrong as generally advising "+6dB to left" or "+10dB to 3kHz and up". There may be a taste reason for it - trying to simulate width=4 - but doing it is going to have measurable adverse consequences.

 

[Lbstyling]

For what it's worth, I ran logic 7 on a lexicon processor up to 2 years ago when I a/b'd it against auro 3D and made the change over.

General perception was Neural x seams to have rolled treble when I compare to Atmos, and less dynamic range. Atmos pulls the stage to the front more than Auro. Auro makes the concert feel bigger and more 'fun'.

When I played it for a friend, he cracked a smile as soon as we went to auro every time.

This was is a previous house, so I'm starting again. This time it's straight to Auro for me.

 

[Chromatischism]

Maybe I'm not explaining myself very well.

In stereo content there will be stuff that is panned by the original mixer of the material to left, centre or right in the stereo mix. They will be crossfading to achieve the levels they want. They will be expecting a certain balance to be maintained. In particular, something coming sent to both left and right at -3dB should roughly match something sent to just left or right alone at 0dB.

The upmixer gets a lot of flexibility in where it places stuff throughout the speaker arrays, but it needs to try to preserve the basic levels of the stereo mix.

If centre spread is disabled (or width=0), something pure mono coming in both L+R in the 2.0 mix at -3dB will come out of just centre, at 0dB, to adjust for that difference between one and two speakers.

If centre spread is enabled, it will be sending signal to all of L+C+R in some proportion, aiming again to achieve the same overall 0dB level. In the absence of measurements from you, I'm going to guess that DSU's centre spread is the same as DPLII's width=3, which seems to be roughly centre -2dB and left and right -7.5dB each.

Now, maybe DSU is like width=3, and you don't like the centre still being 5dB louder than either L or R for centre-panned signal. It's too "narrow" for you. You'd prefer to have width=4 or more, so the centre is not louder. That's fair enough - the weighting of C versus L+R is a matter of taste.

But you can't correct that width well by turning down centre alone. Something in the stereo mix with L+R both at -3dB, that should end up summing to 0dB is now coming out of 3 speakers at -7.5 to -8dB, which is too low. Stuff panned to centre by the original mixer is now too quiet compared to the rest of the mix, by about 3dB. You may have corrected the "width" to your taste, but you've upset the original mix by de-emphasising stuff the original mixer put in the centre, compared to stuff they put left, right or surround.

Illustrating in numbers:

Width	Left level = Right level (dB)	Centre level (dB)	Total level (dB)
0	-inf	0	0
1	-11.5	-0.7	0
2	-9.3	-1.2	0
3	-7.4	-2.0	0
Chromaticism's (width 3, C lowered)	-7.4	-8.0	-2.8
4	-4.8	-4.8	0
5	-4.0	-6.7	0
6	-3.5	-9.1	0
7	-3.0	-inf	0

All this is predicated on my belief that Dolby Surround's spread mode will have a correctly-summing set of L/C/R, like all the widths of Dolby Pro Logic II.

If you can show that Dolby Surround is failing to correctly level centre-panned content, then "-6dB to centre" could be valid advice. Otherwise, it's as objectively wrong as generally advising "+6dB to left" or "+10dB to 3kHz and up". There may be a taste reason for it - trying to simulate width=4 - but doing it is going to have measurable adverse consequences.

Interesting, but a lot of this is based on guesses.

All I can say is that the reason I made the change was the center was too loud and too dominant. I started at -5 and it wasn't enough, so went to -6. It is much closer to Stereo and A3D now.

 

[KMO]

Interesting, but a lot of this is based on guesses.

Indeed, which was why I was asking if you could provide measurements of your system. I guess not?

If anyone else has REW and Dolby Surround, and wants to help out by running level checks akin to what I did for DPLII, please do!

I've not managed to find any concrete answer online. Would also be good to check Auro-Matic.

From my understanding Dolby and Auro-Matic are quite fundamentally different in that Dolby Surround like DPLII before it is trying to specifically steer stuff based on matrix encoding, with the primary use being 2.0 encodes of film/TV. The "centre spread" and "centre width" controls effectively partially undo the steering to centre, sending some of it back to L+R, making it less objectionable for 2.0 music.

Whereas Auro-Matic doesn't do matrix-based steering at all. It's more akin to a "DSP" spatialising mode, so inherently is L+R focussed for 2.0 material, so will naturally be very wide, and more stereo-like.

 

[Chromatischism]

So what kinds of measurements would we need to look at?

 

[KMO]

Okay, well a detailed set of tests that would give us all the info necessary. This isn't exactly what I did - it's a bit more thorough.

1. Put mic in main listening position, leave it there.
2. Make sure your levels are calibrated, so L/C/R are all giving the same SPL with the speaker calibration signal we'll be using below. Best done using multichannel HDMI input. If not same SPL, record the SPLs so we can adjust.
3. Sort out 2-channel input, not multichannel. (I had to switch to Java drivers to do this - ASIO seemed multichannel only?)
4. Send in a mono speaker calibration signal - test signal simultaneously to both L+R.
5. Check overall level, in each mode (DSU spread off, DSU spread on, Auro-Matic, and may as well do stereo, just to crosscheck with original calibration)
6. Repeat step 2 with a L-only signal.
7. Isolate speakers by disconnecting speaker or power-amp cables. (I only messed with L/C/R - output from surrounds was negligible for DPLII, so left them).
8. Repeat step 2 and 4 tests with only L speaker connected, then with only C speaker connected.

If we have all that, we can figure out what the general L/C/R weighting is of the modes, and how they compare to DPLII widths measured above. Although given that the new algorithms are more fancy, it's possible that such simple measurements won't tell the whole story.

 

[Chromatischism]

Send in a mono speaker calibration signal - test signal simultaneously to both L+R.

Is it ok to use HDMI Channel 3 with the center disabled for this? The AVR will split to L and R.

If we have all that, we can figure out what the general L/C/R weighting is of the modes, and how they compare to DPLII widths measured above. Although given that the new algorithms are more fancy, it's possible that such simple measurements won't tell the whole story.

Right...DSU moves sounds based on frequency and who knows what else.

 

[KMO]

Is it ok to use HDMI Channel 3 with the center disabled for this? The AVR will split to L and R.

No, because we're specifically trying to test the upmixer handling of 2.0 source material. Not how the AVR downmixes multichannel. It's crucial that the test signal be in a 2.0 container.

(Oh, and it's also crucial that the AVR doesn't know you're disconnecting speakers - we're trying to see how it balances things when they're all connected).

But for step 2, sure, just do level checks on HDMI channels 1,2,3.

My only concern was a potential level mismatch between Java and ASIO drivers, but they seemed to match, as long as I set Windows mixer to max for Java. The stereo cross-check in step 5 was to cover that.

 

[Chromatischism]

I don't know how to test L&R at the same time without using HDMI Channel 3?

 

[KMO]

I don't know how to test L&R at the same time without using HDMI Channel 3?

If using REW, use the "Generator" window. If set to 2-channel, you just select "L+R" as Output at the bottom. (If it was multichannel, it would offer primary and secondary channel).

I was using pink periodic Speaker Cal (500 to 2000 Hz).

For the level readout, I was using the SPL Meter window (SPL C Slow), but more specifically the logger window to try to pin down the level to 0.1dB.

 

[database]

This is my current stereo music upmix setup:

- 3 channels only, LCR. No satellite channels
- Dolby Surround with center spread enabled
- 2 ms additional delay on center channel after calibration

This is the only upmixing setup I've heard that doesn't sound worse than stereo in some way (tested Auro, DPL II, Neural X, and more software upmixers, which all had major issues). Not only that, it actually sounds better than stereo, and I say that having been a stereo purist for years. It's extremely faithful to the original stereo sound, while making it better by making fewer compromises: with stereo I've struggled between setting up LR for a wide soundstage vs having a tight and focused center image. The 3ch upmixing setup lets me have both: a perfectly focused and clear center image from the center channel (without the compromises of a phantom center as documented in Toole's book) and LR spaced to provide a spacious soundstage.

Unfortunately I couldn't find any use for the satellites with this or any upmixing setup. They all sound gimmicky and destroy the balance of the music as heard in stereo. For dolby surround in particular, sounds are sometimes extracted out of the mains L and R and placed partially in the mains and surrounds. The solution was to turn off all surrounds, and then the upmixer no longer extracts any sound out of the mains.

I've been listening to music this way for about a year and haven't looked back.

 

[dasdoing]

extracting stereo to sides never made sense to me. the information just isn't there. the only thing that might make any sense is to use a reverb VST with first reflection simulation and send that signal there

 

[JoachimStrobel]

There is an old photo from the 60’s around showing Frank Sinatra with a three speaker (LCR) setup. Adding a third speaker to fill the stereo gap makes a lot of sense and can niceley be extracted from the stereo signal. Originally Dolby added the surround channel in mono as Dipol speaker only for ambient humming background sound. Simulating ambience with 2 satellite speakers has always been a joke. With 4 heights, side and rears it might work, but hardly from existing stereo recordings.

 

[Tim Link]

I haven't heard some of the high end up-mixers out there but for the last few weeks I've been listening to a simple 3 channel up mix that produces produces inter-aural crosstalk reduction when the speakers are placed at about 1 foot center to center spacing. It's been blowing my mind every night. I don't see myself ever going back to a 2 speaker setup, and I'm preferring this hands down to any of the Dolby or DTS upmixes I've tried, or even Atmos discrete multi-channel recordings. This is just rocking my audio mind! It's just freakin' beautiful. I'm sitting here in wonder and amazement right now listening to It's a Beautiful Day - Marrying Maiden
I posted about it here: https://www.audiosciencereview.com/...k-elimination-reduction-par-excellence.39157/

and here: https://www.diyaudio.com/community/...tom-center-image-problem.393540/#post-7216172

I had my doubts at first but with the speaker spacing dialed in for my listening distance it's just melting my soul every night! I've never heard a system sound like this before except in experiments where I used a physical divider right up to my nose to separate left and right channels. I consider this a legitimate substitute for the awesome sounding but ergonomically unlivable center divider wall between your legs and in your face.