line level processing deriving a center channel from stereo channels

ebeowulf17

Joined Aug 12, 2014
3,307
Taken at the output of each of the two sections of logic addition followed by a low-pass filter, the level of the summation of the middle signal in the two channels is +6 dB relative to the level of the summation of the left and right signals. Voltage gains and losses of the upmixer are configured so that the level of the middle signal at the output of the upmixer is +3 dB with respect to input level. Thus the level of the summation of the left and right signals becomes -3 dB. From this it follows that the level of the left signal in the middle channel is -6 dB with respect to the level of the left signal in the left channel. The same applies to the right signal.

Cross-feeding of the processor attenuates the middle signal in the left and right channels each by -3 dB. Thus the level of the middle signal in the generated middle channel is +6 dB relative to its level in the left or right channel. This is the same level difference that results when deriving a middle channel by addition of the left and right channels.

The above occurs over the frequency range of 400 Hz to 1600 Hz (two octaves). So the main advantage of the upmixer versus the method of simply adding the left and right channels is the attenuation of the left and right signals in the middle channel. Adding left and right channels results in no attenuation of the left and right signals in the derived middle channel.

Granted that even -6 dB is not a lot of attenuation but it seems to work.

-Pete
I understand the psychoacoustic effect you're going for by emphasizing mids in the center channel. I can see where this might work well for home theater, but I imagine it would sound weird listening to music on such a system. Although I supposed that would just be personal opinion, per listener.

Anyway, you keep saying that you're getting greater attenuation of left and right channels in the center channel, and I don't think it's mathematically possible to achieve what you're describing. I'm almost positive you're fooling yourself on this one. When starting with a regular stereo recording, there simply isn't enough information in the recording to really separate things.

Maybe, if you assume that what you consider the "real" left and right channel information is mostly high and low frequencies, and that what you consider the "real" center channel information is mostly mid frequencies, then your claim would make a little more sense, but I don't think those are valid assumptions.

Don't get me wrong - if this system achieves a desired effect and you're happy with the results, then that's great! I'm happy for you. But I think it involves a lot of assumptions and compromises, and I think the way you're describing the results is probably a little misleading.
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
Anyway, you keep saying that you're getting greater attenuation of left and right channels in the center channel, and I don't think it's mathematically possible to achieve what you're describing. I'm almost positive you're fooling yourself on this one. When starting with a regular stereo recording, there simply isn't enough information in the recording to really separate things.
An assumption that I'm making is that the center signal found in the stereo left and right channels is identical in both channels. That is, the center signal is the signal that has exactly the same waveform in the two channels. On the other hand, the left and right signals are uncorrelated, or saying it another way, if the left and right signals at some point have the same amplitude and slope, that is purely a chance occurrence and the probablility of this occurring is very low. In stereo recordings, I think that this is almost always at least very nearly the case. There might be some slight differences between the center signal in the left channel and that signal in the right channel.

In the simulation that I'm attaching (LTspice), if you make the phase of the two input sine waves of equal amplitude and frequency identical (both 0 phase angle), then this simulates the presence of the center signal at both inputs to the processing circuit. If you change the phase of one of the input sine waves to 90 deg. , this results in a decrease of the output RMS voltage of the processing circuit by one-half.

The phase relationship equal to 90 deg. is equivalent to what on average is the phase relationship between two uncorrelated signals. So this is simulating the (uncorrelated) left and right signals at input to processing.

If you don't have LTspice, you could look at Fig. 1 attached to my first post in this thread. Fig. 1 shows the basic operating principle of the upmixer.

Regards,
Pete
 

Attachments

ebeowulf17

Joined Aug 12, 2014
3,307
An assumption that I'm making is that the center signal found in the stereo left and right channels is identical in both channels. That is, the center signal is the signal that has exactly the same waveform in the two channels. On the other hand, the left and right signals are uncorrelated, or saying it another way, if the left and right signals at some point have the same amplitude and slope, that is purely a chance occurrence and the probablility of this occurring is very low. In stereo recordings, I think that this is almost always at least very nearly the case. There might be some slight differences between the center signal in the left channel and that signal in the right channel.

In the simulation that I'm attaching (LTspice), if you make the phase of the two input sine waves of equal amplitude and frequency identical (both 0 phase angle), then this simulates the presence of the center signal at both inputs to the processing circuit. If you change the phase of one of the input sine waves to 90 deg. , this results in a decrease of the output RMS voltage of the processing circuit by one-half.

The phase relationship equal to 90 deg. is equivalent to what on average is the phase relationship between two uncorrelated signals. So this is simulating the (uncorrelated) left and right signals at input to processing.

If you don't have LTspice, you could look at Fig. 1 attached to my first post in this thread. Fig. 1 shows the basic operating principle of the upmixer.

Regards,
Pete
Well, I'll admit that there's more going on here than I initially was imagining, and that I don't fully understand it all just yet.

Although I'm still skeptical, I certainly can't debunk something I don't understand, so I apologize if my earlier comments were a bit harsh. I'm playing around with your simulated circuit, trying correlated and uncorrelated noise sources, and trying to get a better understanding of the behavior. I'll let you know if I find anything that might be useful.

For now, it's my bedtime, so I'm done for tonight. Cheers!
 

ebeowulf17

Joined Aug 12, 2014
3,307
I think I'm starting to grasp the concept here, and I understand the boosting of correlated vs. uncorrelated signals now, but unfortunately I realized at the same time that it must introduce significant distortion. Sure enough, there are inharmonic beat frequencies generated by any difference between the left and right signal. As you can see in these FFT plots of a 500Hz and 568Hz signal, passive mixing delivers a clean mix of the two sine waves, but the new circuit introduces many distortion components.two-sines-diff-freq_FFT-notes.png
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
I think I'm starting to grasp the concept here, and I understand the boosting of correlated vs. uncorrelated signals now, but unfortunately I realized at the same time that it must introduce significant distortion. Sure enough, there are inharmonic beat frequencies generated by any difference between the left and right signal. As you can see in these FFT plots of a 500Hz and 568Hz signal, passive mixing delivers a clean mix of the two sine waves, but the new circuit introduces many distortion components.
In the working upmixer, a 2nd order LR high-pass filter, Fc = 400 Hz, is inserted between the output of each of the two voltage sources and input to logic addition of that simulation, Fig. 1R-179.1.See Fig. 3 attached to my first post in this thread.

The preceding high-pass filter would reduce I presume some of the distortion generated by the custom circuit, but that lower FFT certainly looks pretty ugly. Would all of that be very audible?

Given that the custom circuit is unusual, is it possible that the FFT analysis is in error?

For testing operation of the upmixer, would you say that pink noise would be similar enough to a music recording for measuring input/ output of the upmixer?

Regards,
Pete
 
The FFT plot clearly shows that some of the distortion products are a result of intermodulation (heterodyning)...
i.e: 636 is 68 Hz away from 568. Likewise 432. And both 364 and 704 are at 2X 68Hz apart.

The higher frequency clusters may be composite second order intermodulation.
 

ebeowulf17

Joined Aug 12, 2014
3,307
First off, I should say that I have a strong background in audio production (primarily location recording of live music) but not in real audio *engineering* per se. I dabble in electronics, but I'm no expert, not by a long shot! Having said that, here are my opinions.
but that lower FFT certainly looks pretty ugly. Would all of that be very audible?
Two of the peaks are only 10-15dB below the signals, so I'd expect that to be pretty audible. A few more peaks are about 30dB down, potentially still noticable. As a point of reference, reverb decays are described in terms of their 60dB down point, implying that 60dB down is where you'll stop hearing it. I don't know how true that is, but it's something to consider.

I'm not sure how much distortion can be masked in a busy mix. Starting with pure tones makes it much more obvious what's going on. A little crunch on a voice or some background street sounds in a movie might go unnoticed. I'm just not sure how much.
Given that the custom circuit is unusual, is it possible that the FFT analysis is in error?
I doubt it. Your circuit is busy enough that I can't immediately wrap my head around all of it at once, but I don't think LTspice should have any trouble with it. (Spice is a LOT smarter than I am when it comes to circuit analysis!)
For testing operation of the upmixer, would you say that pink noise would be similar enough to a music recording for measuring input/ output of the upmixer?
Pink noise would be useful for certain tests, but doesn't tell the whole story. I did some noise tests on your circuit, and they seem to confirm the correlated-sound level boost that you described, but they don't reveal the distortion in an obvious way.

I would think an extreme edge-case real world test might be helpful. If the distortion behaves the way I'm imagining it will, the following test will make it clear: Start with two separate, very clean tracks as left and right, with no mid signal present. For example if you had a singer/song writer with the acoustic guitar panned hard left and the vocal panned hard right. Feed that signal through your circuit and record the output from the custom mid channel. A traditional passive mid mix would just deliver a neutral mix of the two sounds. A theoretically perfect, magical mid extractor would be silent, since there isn't any mid signal. I think that, with your circuit, you'll hear a lot of distorted artifacts of both tracks mixed into the middle, and that in this extreme test case the distortion will be fairly noticeable.

But I might be wrong - I'm having to stretch my imagination a little to guess how exactly this plays out in real world scenarios, so I'm not really sure. This is a little bit theory and a little bit gut-feeling. It's certainly worth testing, but some version of the test I just described is the best I can come up with, so I can't test it here (I've heard that you can load waveforms into LTspice, so maybe it is possible for me to sim this, but I don't know how just yet.)

If you can't find or generate a stereo recording with two very clean, totally discrete, unrelated tracks for the left and right, let me know. I should be able to pull something from one of my recordings and make a hard left-right mix for testing purposes. I can send you a file if you need one.
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
First off, I should say that I have a strong background in audio production (primarily location recording of live music) but not in real audio *engineering* per se. I dabble in electronics, but I'm no expert, not by a long shot! Having said that, here are my opinions.

Two of the peaks are only 10-15dB below the signals, so I'd expect that to be pretty audible. A few more peaks are about 30dB down, potentially still noticable. As a point of reference, reverb decays are described in terms of their 60dB down point, implying that 60dB down is where you'll stop hearing it. I don't know how true that is, but it's something to consider.

I'm not sure how much distortion can be masked in a busy mix. Starting with pure tones makes it much more obvious what's going on. A little crunch on a voice or some background street sounds in a movie might go unnoticed. I'm just not sure how much.

I doubt it. Your circuit is busy enough that I can't immediately wrap my head around all of it at once, but I don't think LTspice should have any trouble with it. (Spice is a LOT smarter than I am when it comes to circuit analysis!)

Pink noise would be useful for certain tests, but doesn't tell the whole story. I did some noise tests on your circuit, and they seem to confirm the correlated-sound level boost that you described, but they don't reveal the distortion in an obvious way.

I would think an extreme edge-case real world test might be helpful. If the distortion behaves the way I'm imagining it will, the following test will make it clear: Start with two separate, very clean tracks as left and right, with no mid signal present. For example if you had a singer/song writer with the acoustic guitar panned hard left and the vocal panned hard right. Feed that signal through your circuit and record the output from the custom mid channel. A traditional passive mid mix would just deliver a neutral mix of the two sounds. A theoretically perfect, magical mid extractor would be silent, since there isn't any mid signal. I think that, with your circuit, you'll hear a lot of distorted artifacts of both tracks mixed into the middle, and that in this extreme test case the distortion will be fairly noticeable.

But I might be wrong - I'm having to stretch my imagination a little to guess how exactly this plays out in real world scenarios, so I'm not really sure. This is a little bit theory and a little bit gut-feeling. It's certainly worth testing, but some version of the test I just described is the best I can come up with, so I can't test it here (I've heard that you can load waveforms into LTspice, so maybe it is possible for me to sim this, but I don't know how just yet.)

If you can't find or generate a stereo recording with two very clean, totally discrete, unrelated tracks for the left and right, let me know. I should be able to pull something from one of my recordings and make a hard left-right mix for testing purposes. I can send you a file if you need one.
Thanks for your testing with the simulation and your comments. The listening test that you propose sounds like a good idea, and I will have to see if I can come up with the type of stereo recording that you suggest.

Listening to the output of the upmixer alone, the reproduction perhaps sounds a little raspy, but not tremendously distorted to the point of being very objectionable to listen to. With all three channels running, any raspy reproduction is not apparent.

-Pete
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
[QUOTE="ebeowulf17, post: 1479625, member: 245586
If you can't find or generate a stereo recording with two very clean, totally discrete, unrelated tracks for the left and right, let me know. I should be able to pull something from one of my recordings and make a hard left-right mix for testing purposes. I can send you a file if you need one.
[/QUOTE]

That would be great if you would be willing to put together such a recording for me. No doubt there are people in my area that could do such a thing, but I can't say whether or not they would be willing. The best that I could do myself would be to record two unrelated selections of music, one on each track; that would sound very weird and might distract from hearing distortion.

Regards,
Pete
 

ebeowulf17

Joined Aug 12, 2014
3,307
That would be great if you would be willing to put together such a recording for me. No doubt there are people in my area that could do such a thing, but I can't say whether or not they would be willing. The best that I could do myself would be to record two unrelated selections of music, one on each track; that would sound very weird and might distract from hearing distortion.

Regards,
Pete
Cool, I'll put something together for you. Hopefully I'll get it done this weekend, but I can't make any guarantees. Cheers!
 

ebeowulf17

Joined Aug 12, 2014
3,307
Actually managed to squeeze it in right after work. This Google drive link should take you to a folder with 4 files.

https://drive.google.com/folderview?id=13qZgvOT083Kd0cVlkAyGFU1mZa2p6NPS

I made a left/right isolated mix for you with the reverb and delay turned off so it would be easier to hear any distortion. I also included the "real" mix with normal panning, stereo guitar, and effects on vocals, just in case you're interested in the finished version of the song.

Both mixes are uploaded in both 16/44.1 .wav and 320k .mp3 formats. There are definitely better recordings out there that would make a better test, but this is the only stuff I can readily share. Hope it helps.
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
Actually managed to squeeze it in right after work. This Google drive link should take you to a folder with 4 files.

https://drive.google.com/folderview?id=13qZgvOT083Kd0cVlkAyGFU1mZa2p6NPS

I made a left/right isolated mix for you with the reverb and delay turned off so it would be easier to hear any distortion. I also included the "real" mix with normal panning, stereo guitar, and effects on vocals, just in case you're interested in the finished version of the song.

Both mixes are uploaded in both 16/44.1 .wav and 320k .mp3 formats. There are definitely better recordings out there that would make a better test, but this is the only stuff I can readily share. Hope it helps.
Thanks very much. I burned the file "These Days"-AAC-mix.mp3 to a CD and then played it through my stereo system.

Listening to the center channel reproduction alone, I did not hear any distortion. However, the level of the the left and right channels in that speaker connected to the center channel output was fairly loud. However, this is what should occur as the output of the "logic adder", which is where the distortion would derive, is mixed with the sum of the two channels at unity gain, So I'm sure that the distortion is there, but being masked by the linear addition of the left and right channel signals.

With all three channels operating, reproduction of your mix sounded very good.

Regards,
Pete
 
Last edited:

ebeowulf17

Joined Aug 12, 2014
3,307
Thanks very much. I burned the file "These Days"-AAC-mix.mp3 to a CD and then played it through my stereo system.

Listening to the center channel reproduction alone, I did not hear any distortion. However, the level of the the left and right channels in that speaker connected to the center channel output was fairly loud. However, this is what should occur as the output of the "logic adder", which is where the distortion would derive, is mixed with the sum of the two channels at unity gain, So I'm sure that the distortion is there, but being masked by the linear addition of the left and right channel signals.

With all three channels operating, reproduction of your mix sounded very good.

Regards,
Pete
I'm glad to hear that it's working well. As you said, I have to assume some amount of distortion is there, but if it's not audible, then it doesn't really matter.

Out of curiosity, when you did the three channel listen, was that also with the special test mix, or with the regular stereo mix? The acoustic guitar was original recorded in stereo with an MS mic pair and I'm curious how the perception of stereo width was affected by the addition of the center channel and processing circuit. It was never a super-wide sound to begin with, but I like having a bit of "space" in the guitar, for lack of better words.
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
I'm glad to hear that it's working well. As you said, I have to assume some amount of distortion is there, but if it's not audible, then it doesn't really matter.

Out of curiosity, when you did the three channel listen, was that also with the special test mix, or with the regular stereo mix? The acoustic guitar was original recorded in stereo with an MS mic pair and I'm curious how the perception of stereo width was affected by the addition of the center channel and processing circuit. It was never a super-wide sound to begin with, but I like having a bit of "space" in the guitar, for lack of better words.
When I listened with all three channels operating, that was when playing the special test mix. So far I have not listened to the regular stereo mix through the system; I will try comparing 3 channel reproduction of the two different recordings tomorrow.

My suspicion is that the special test mix would have better separation and thus sound better than the regular mix given that there is a fair amount of "leakage" to the center channel with my processing.

-Pete
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
I'm glad to hear that it's working well. As you said, I have to assume some amount of distortion is there, but if it's not audible, then it doesn't really matter.

Out of curiosity, when you did the three channel listen, was that also with the special test mix, or with the regular stereo mix? The acoustic guitar was original recorded in stereo with an MS mic pair and I'm curious how the perception of stereo width was affected by the addition of the center channel and processing circuit. It was never a super-wide sound to begin with, but I like having a bit of "space" in the guitar, for lack of better words.
Listening to the regular stereo mix, it seemed as if the guitar was coming from the left and right speakers while the vocals were in the middle. When I turned up the volume of the middle speaker (with a L-pad), the singing localized in the middle became more prominent.

Listening to the special test mix, even with no attenuation of the reproduction by the center speaker, the sound of the guitar and vocals are clearly localized to respectively the left and right. Going from no reproduction from the center speaker to full volume, I preferred the change in the quality of reproduction which I would say had a somewhat improved presence.

So actually for listening with the processing I would prefer the special test mix as then there is good left/ right separation. Listening to the regular stereo mix with or without reproduction by the center speaker, I was uncertain as to where the sound of the guitar was coming from.

Take care,
Pete
 

ebeowulf17

Joined Aug 12, 2014
3,307
Listening to the regular stereo mix, it seemed as if the guitar was coming from the left and right speakers while the vocals were in the middle. When I turned up the volume of the middle speaker (with a L-pad), the singing localized in the middle became more prominent.

Listening to the special test mix, even with no attenuation of the reproduction by the center speaker, the sound of the guitar and vocals are clearly localized to respectively the left and right. Going from no reproduction from the center speaker to full volume, I preferred the change in the quality of reproduction which I would say had a somewhat improved presence.

So actually for listening with the processing I would prefer the special test mix as then there is good left/ right separation. Listening to the regular stereo mix with or without reproduction by the center speaker, I was uncertain as to where the sound of the guitar was coming from.

Take care,
Pete
Cool, thanks for the feedback! It sounds like the stereo width in the original mix holds up.

The guitar is meant to sound sort of big and spacious. Ideally it would sound like it was coming from everywhere, not just from the outsides, and still be centered, just with some width to it. Of course these are just personal preferences and my personal bias. There's no right or wrong when it comes to aesthetic choices like these.

But yeah, for testing your system, the extreme left/right panning is undoubtedly the better mix.
 

Thread Starter

PeteHL

Joined Dec 17, 2014
473
Please see the file attached to my post today. This shows some improvements that I have made to my processing circuits shown in the attachments to my post #1 of this thread. The purpose of the processing is to reproduce some of the center signal of standard music recordings by a loudspeaker positioned center front of the listener. Please let me know if you spot any mistakes that I have made in the new circuits.

The changes that I have made are the following:

1. The cross-feeding shown at FrtCntrFig5.pdf of the original design (see my post #1, this thread) is eliminated. This is made possible by reduced crosstalk in the center-front channel output of the improved upmixer.

2. The circuit of FIGURE UPMXR-1 in the PDF attached to this post is very similar to that of of FrntCntrFig2.pdf (see my post #1, this thread), except that the addition of the sum of the left and right channel input signals to the output of the logic adder section is eliminated. At Fig. 2 (original design) remove the adder including op amp U9, and then the two circuits are essentially the same. The other change that I have made shown at FIGURE UPMXR-1 is including two signal polarity separators each replacing two half-wave rectifiers.

3. I'm now referring to the circuit of FIGURE UPMXR-1 as a Lesser Amplitude Follower (LAF) as this describes it better than logic adder.

4. Q of the high and low pass filters of FrtCntrFig3.PDF of the original design (attached to my post #1 in this thread) equals 0.5. At FIGURE UPMXR-2 in the PDF attached to this post, Q of these filters has been increased to 1.

5. As is not the case for the high and low pass filters of FrtCntrFig3.PDF of the original design, the high and low pass filters of each of the two processing sections (upper and lower octave) of the revised upmixer of FIGURE UPMXR-2 have a cut-off frequency equal to one half octave above the low cut-off frequency of the pass-band of the section.

6. Overall voltage gain of the new upmixer of FIGURE UPMXR-2 corresponds to integrating the upmixer into a stereo system where all of the power amplifiers and loudspeaker systems of the setup have respectively equal fixed voltage gain, and impedance and sensitivity. This is shown at FIGURE-UPMXR-3 included in UPMIXER2021.pdf attached to this post. Differences of amplifier gain, and/ or loudspeaker impedance and sensitivity, could be accommodated for by adjusting the resistance of resistor R18 of the circuit of FIGURE UPMXR-2.
 

Attachments

Top