Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[kevinpwhite]

I'd appreciate thoughts on the following, please !

Configuration
-   I have JRiver MC28 on Win 10 Pro 20H2 playing out to a Creative X7 5.1 capable external DAC pre/Amp and thence Audioengine active front and rear speakers and sub-woofer. 
-   I have connected a Clarity M peak loudness hardware meter to the system which is reading the output from JRiver MC.

Rear Channel Phase Correlation
-   The Correlation Meter for the L / R Rear Channels indicates that channels are out of phase.
-   The negative reading can be reversed by changing the polarity of either (not both !) of the Rear Channels in DSP Studio’s Parametric Equaliser section.
-   This is distinctly audible when playing any of several mono test tones (pink noise, sine or square waves).
-   I have set up separate Reverse Polarity Filters for (SL) and (SR); toggling between them demonstrates the effect very clearly.
-   The screenshots from the meter (below) show the ‘Standard’ MC28 condition (OOP) and with (either of) Rear L / R reversed relative to the other. The Correlation Meters are at bottom right.

Meter Function Comments
-  In 5.1 Mode, the Correlation Meter compares Left/Right (Top) as well as Left Surround/ Right surround with 2 bar graphs.
-  Phase issues can result in audible loss of low frequencies, unintended alterations to a mix, among many other problems.
-  Readings toward the positive end of the meter indicate proper phase alignment (indicated in blue), and phase problems are shown in orange to the left of the meter.
-  If the two inputs are identical (a dual-mono signal) they are ‘fully correlated’ and the meter indicates +1 (Cosine of an angle of Zero degrees phase difference between the two channels)
-  If the two signals are completely different (eg. a widely spaced stereo input), they are ‘fully decorrelated’; the phase difference = 90 degrees and the cosine is Zero.
-  For a dual-mono signal where one channel is polarity-inverted, the phase difference is 180 degrees; Cosine = -1 so the meter indicates at the negative extreme.
-  Around + 0.3 positive correlation is typical for a stereo signal with a decent mono component. The rear channels typically have a higher mono component for music compared to eg. a true surround mix from an action film soundtrack.

Comments
-   The normal source music is from commercial recordings (44kHz 16 Bit CDs or 192/96kHz 24 Bit Hi Res downloads , both in FLAC format) which should be phased correctly !
-   It is possible, though unlikely, that there is a coding issue in MC28 - although I would have thought one of the many users worldwide would have picked up on that by now.
-   There is however very little discussion concerning phase correlation / polarity that I’ve been able to locate on Interact.

Request
-   If anyone has any insight into this phenomena I’d much appreciate if you can advise

Many thanks in anticipation,

Best regards,

Kevin

[mattkhan]

Your dsp studio configuration is required info here I think, can you share it?

Perhaps also show audio path while playing

[kevinpwhite]

Hello and thanks for the reply...

Here's the DSP Audio Path and basic settings.

Please forgive ignorance; is there a simpler way to extract the complete DSP Settings to show/send you ?

Regards,

Kevin

[mwillems]

Your dsp studio configuration is required info here I think, can you share it?

Perhaps also show audio path while playing

I'm not familiar with the loudness meter you're using, but if it's measuring sound in the room (rather than in line output), have you checked the physical wires/cables connected to the speakers?  If one or the other is plugged backward (amp "+" connected to speaker "-", or RCA cables reversed, etc.), that would also create a polarity inversion for that speaker.  I'm sure you've thought of that, but it might be worth verifying that both speakers are plugged in with the same polarity. 

Please disregard if the loudness meter is an in-line meter (which would cut the speakers out of the equation).

[kevinpwhite]

....if the loudness meter is an in-line meter (which would cut the speakers out of the equation).

It is using a VST PLugin loaded into DSP Studio, so completely in the digital domain.

[mwillems]

It is using a VST PLugin loaded into DSP Studio, so completely in the digital domain.

Got it, sorry for the static!  It might also be helpful to include a screen capture of your DSP studio screen, including the output format tab settings.

[kevinpwhite]

Here you go....

[mattkhan]

Hello and thanks for the reply...

Here's the DSP Audio Path and basic settings.

Please forgive ignorance; is there a simpler way to extract the complete DSP Settings to show/send you ?

Regards,

Kevin

You can use the save DSP to file option in DSP studio

I think your audio path says you are using jrss to upmix to 5.1 right? Ok confirmed by the pic. I don't think I have ever seen a public description of jrss so probably it depends on how this is extracting surrounds from the stereo input.

[mwillems]

I think your audio path says you are using jrss to upmix to 5.1 right?

I noticed that too.  I wonder if the upmix is driving it?

Kevin, if you have any native 5.1 test clips, it would be worth trying one of those to see if you see the same phase issues.

[kevinpwhite]

Yes, per the screen, it is set with JRSS mixing (recommended).
I've saved the DSP Preset file and attached as a .zip - does that contain all the DSP Settings ?

I shall have to find a suitable 5.1 native track to test with - Apollo 13 Launch should do the trick.....   I will find that and try it out shortly.

K

[mattkhan]

https://en.m.wikipedia.org/wiki/Matrix_decoder gives an example of the way some matrix upmixers work and you can see there is a phase shift to extract the surrounds so I would guess that jrss is implemented like this. I imagine it could be tested using some test signals to verify if you wanted

[Hendrik]

Roughly, the way the upmixer works, is to take the front L/R channel, and remove the commonality between Left and Right so that "center" audio does not get repeated in the surround, as that would feel odd and destroy the stereo image (and it applies a bit of delay)

The formulas used for that look fine to me and I don't see how one channel would necessarily get out of phase from the other in normal content where both channel are equally active.

[mwillems]

Roughly, the way the upmixer works, is to take the front L/R channel, and remove the commonality between Left and Right so that "center" audio does not get repeated in the surround, as that would feel odd and destroy the stereo image (and it applies a bit of delay)

Presumably, the delay applied is the same for both rear channels?  That would be an obvious place that phase shift could be introduced if it weren't the identical on both sides.  But I doubt that's the issue (it would be obvious in other ways I expect).

Alternatively, Kevin mentions using test tones (Pink noise, sine waves, etc.), where both pre-mix channels might be the same or very similar.  I wonder if the observed decorrelation is just an expected consequence of removing the summed content from two signals that are basically the same?

[kevinpwhite]

OK....  that was interesting !

I found a Fraunhoffer 5.1 Test file ( https://www2.iis.fraunhofer.de/AAC/ChID-BLITS-EBU-Narration441-16b.wav ) instead and have checked with that.

DSP Output Format - Channels set to 5.1 with No up- or down-mixing

-  With the native 5.1 source, I can confirm that the Rear L/R Correlation meter is indicating correctly. It also sounds right to the ear 
-  But with any 2 Ch source and JRSS mixing selected, the Rear phase reversal occurs.

I have had a trawl to see if I could find anything about JRSS mixing and perhaps a suitable test file to help explore further, but so far no joy. Are you aware of naything which may help illuminate, please ?

In the meantime, it would appear that the takeway nugget from this is:

"If using JRSS upmixing to 6 Channel, a Phase reversal occurs between Rear Left and Right; Setting up a Reverse Polarity (SL or SR)  filter in DSP provides a workaround."

[kevinpwhite]

First, thank you for the several other angles - I hadn't refreshed/seen those before my last post.

...... Kevin mentions using test tones (Pink noise, sine waves, etc.), where both pre-mix channels might be the same or very similar.  I wonder if the observed decorrelation is just an expected consequence of removing the summed content from two signals that are basically the same?

The test tones I mentioned were explicitly mono signals. So I think it's a little more complicated than 'simply' removing the common components from L and R channels and then feeding the residue to the rear. There must be some other signal manipulation taking place or, with a mono input and therefore L and R identical, you'd expect there to be zero Rear output ?

[Hendrik]

Its not entirely just that part, the final simplified formula comes out to this:

Ls = 0.625 L - 0.375 R
Rs = 0.625 R - 0.375 L

If L and R are identical, it should just result in a much quieter version of L or R, and not influence the phase.
Now, if one channel would be entirely silent, one would be a moderately quieter version of it, and the other one an even quieter phase inverted version of it - but that effect is intentional to dampen the surround a bit.

[kevinpwhite]

Its not entirely just that part, the final simplified formula comes out to this:

Ls = 0.625 L - 0.375 R
Rs = 0.625 R - 0.375 L

If L and R are identical, it should just result in a much quieter version of L or R, and not influence the phase.
Now, if one channel would be entirely silent, one would be a moderately quieter version of it, and the other one an even quieter phase inverted version of it - but that effect is intentional to dampen the surround a bit.

Thank you for the additional clarification.  Although the scenario doesn't quite seem to fit the present situation !


I have been searching to try and find some leads on JRSS^TM mixing - so far without any success. It seems strange that it doesn't seem to appear anywhere in the Googlesphere at all !  Do any of you have any clues to offer that might help ?

JRSS only seems to relate to things like the Royal Statistical Society here in the UK and a US Gov Security organisation...


The 'JR' bit isn't by any chance related to JRiver ?

[Hendrik]

JRSS is our own audio mixer, you won't find it anywhere else.

[kevinpwhite]

JRSS is our own audio mixer, you won't find it anywhere else.

Ah, so I was on the right track !  Presumably, therefore, the JRiver Developers will understand the inner workings and what the mixing algorithms are intended to deliver. I'd appreciate if we could get a definitive statement about the characteristics that I've highlighted today.


Is this one for Matt Ashland or is someone else better placed to respond, please ?


Best regards,


Kevin

[Matt]

Hendrik is a JRiver developer.

[kevinpwhite]

Hello Matt ! Thank you..... then it looks like all the thinking power needed to resolve the connundrum is already on the case


K

[Hendrik]

As I've hinted at above, depending on how coherent the left and right channel are (or not are), a certain amount of phase shift in the upmixed channels is expected and intended as part of the upmix effect.
This will entirely depend on the signal, and a fully coherent signal will not phase shift, an incoherent one will - but neither extreme should be realistically rather common in a stereo source, but more often happen in test material.

As far as I am concerned, the upmix is working as expected. Phase shifting one channel would destroy some of the characteristics of the mix. As mattkhan pointed out above as well, phase changes are a typical tool for audio mixing into a different number of channel.

[mattkhan]

In the meantime, it would appear that the takeway nugget from this is:
"If using JRSS upmixing to 6 Channel, a Phase reversal occurs between Rear Left and Right; Setting up a Reverse Polarity (SL or SR)  filter in DSP provides a workaround."

that looks like very much the wrong takeaway, probably the real takeaway is "if you don't like upmixing, don't use it"

you can directly verify the extreme cases quite easily using the approach described in https://wiki.jriver.com/index.php/Verifying_DSP_Studio

basically first play a sweep to both L & R and take measurements into REW of L then SL then SR, you'll see the output is identical on each channel except for a gain difference (i.e. the reported impulse response is the same, surrounds are about ~12dB down). This is the fully coherent case Hendrik mentions

repeat the above but playing the sweep to the L (or R) channel only, now you'll see a gain difference between the SL and SR channels (of ~4.4dB as given by the formula Hendrik mentions above) and reversed polarity (impulses offset in the attached pic so you can see what I mean)

real content will be constantly varying between these extremes

[kevinpwhite]

Good evening mattkhan and Hendrik,

Please forgive the interlude prior to replying to your last couple of posts. The additional information you’ve provided is all very helpful in better understanding what’s going on ! The light slowly dawns….. 

I have now tried a direct A/B comparison between a 5.1 source and its 2Ch counterpart (both native commercial recordings of the same music), toggling the Polarity Filter in a similar manner. This demonstrates the following characteristics:

5.1
•   Ls/Rs Correlation Meter typically between +0.3 and +0.5; an occasional blip to around -0.2
•   Rear Sound is clear and distinct (as you would expect)
•   If a Reverse Polarity Filter is applied, the meter obviously goes negative;  the sound is noticeably ‘thinner’ and with less bass

2Ch with JRSS Upmix to 5.1
•   Ls/Rs Correlation Meter typically between -0.5 and -0.8
•   Rear Sound is also clear and distinct…..
•   If a Reverse Polarity Filter is applied, the meter goes positive BUT the sound is noticeably affected. I’d almost describe as a balance / bass shift toward the Rear Right speaker

I have also observed that following both Correlation Meters together gives a neat visual demonstration of the effects of Hendrick’s formula and tallies with the results shown in mattkhan’s REW graph.
•   The movements of the L/R and Ls/Rs Meters clearly change in harmony
•   The higher the positive correlation on L/R, the less the phase shift below on Ls/Rs
•   The lower the correlation on L/R, the greater the phase shift on Ls/Rs

The phase behaviours seem also to vary depending on music type, including some instances (eg. electronic music, such as parts of the Interstellar soundtrack) where the L/R meter is hard over on +1 and the Ls/Rs meter also goes significantly positive for quite lengthy stretches …. This seems to tally with marked differences between Ls and Rs Peak Level values, while when the Ls/Rs Phase is in the (more normally seen) negative region, the Ls/Rs Peak values appear to be very similar. Some predominently spoken voice recordings show strong positive L/R values accompanied by low LS and Rs readings, with an occasional dip onto the negative side (again moving in harmony).

So, a rather long way round to the conclusion that the effects originally described (mostly with mono test signals) are not representative of ‘real world’ listening. And that the behaviour of the upmixed signal shouldn’t be directly compared with discretely encoded 5.1 source material. 

My apologies for appearing to cast aspersions about the effectiveness of the JRSS encoder –  and I agree the sentiment in mattkhan’s last post

"if you don't like upmixing, don't use it" 

Fair comment !  For the record, I will be continuing to use it and am pleased to have gained a much better insight that at the start of this exploration.

Thank you for taking the time to discuss and clarify !

Best regards,

Kevin