Should phantom power be supplied by a linear supply?

[Bo Deadly]

You can't do a proper null test if you recorded your samples in different time. You need to record simultaneously and use the same converter.

If you send the same recording through both units with the same settings and record with the same rig, you should be able to do a meaningful digital diff. You will need to time align at the sample level but if there is a distinct identifiable waveform to key on, it shouldn't be too hard. Once the two tracks are sample aligned, just invert one and mix. The residual is the difference.

But in this particular case, I would find it difficult to accept any difference in audio attributed to the power supply. If I were doing this, I would perform the null test procedure with the same unit as a control that shows the null test procedure is in fact working. And, if possible, the same unit should be used with only the power supply swapped out to account for any variance in the other parts of the unit.

If there is an actual difference in the audio attributed to the supply, that would simply be a bad supply. This would not be shocking as there are examples of pro-audio gear using SMPS incorrectly (my MOTU Traveler mk3 phantom has a 2kHz whistle in the noise floor which presumably is from the DC converter). But I would be shocked if there were a difference that could possible make one unit "sound a bit better".

 

[ccaudle]

why don't you do a mult and then a null test?

Sounds like a good test for the new test script from @user 37518 .

If there is an actual difference in the audio attributed to the supply, that would simply be a bad supply.

Yes, but not necessarily out of spec, could just be bad for the application. I'm not familiar with the compressor being used, but information I could find indicates that it is a class A FET compressor with transformer coupled input and output running from 48V. If that is single ended circuits with limited feedback, it could have fairly poor PSRR compared to e.g. a modern op-amp. It is within possibility that high frequency ripple from a switching supply could cause modulation of the audio through a simple single ended circuit. Not saying that is the case, just that the possibility is worth investigating.

I have not been able to do the null-test comparison on those two files myself yet, maybe later today and then I can make a more informed guess what is going on in this particular case.

 

[abbey road d enfer]

If you send the same recording through both units with the same settings and record with the same rig, you should be able to do a meaningful digital diff. You will need to time align at the sample level but if there is a distinct identifiable waveform to key on, it shouldn't be too hard. Once the two tracks are sample aligned, just invert one and mix. The residual is the difference.

But in this particular case, I would find it difficult to accept any difference in audio attributed to the power supply. If I were doing this, I would perform the null test procedure with the same unit as a control that shows the null test procedure is in fact working. And, if possible, the same unit should be used with only the power supply swapped out to account for any variance in the other parts of the unit.

That wouldn't be a valid test, since the clock differences, both in phase and frequency would introduce differences in the reconstituted audio.
When recording at different times, the actual original timing of the samples can vary over one period of sample rate. At 96k SR, it represents 1/100th of a 1kHz sinewave. It means the signal at a particular time is defined with a precision of about 6%. Enough to produce significant difference in a null test. AFAIK there are no DAWs that allow moving sample by a fraction of SR period.
A proper comparison would need to have two identical units recorded at the same time, with the only difference being the PS, but there would be differences due to component variations, probably more significant.
Admittedly, the differences are just emerging from noise.

If there is an actual difference in the audio attributed to the supply, that would simply be a bad supply. This would not be shocking as there are examples of pro-audio gear using SMPS incorrectly (my MOTU Traveler mk3 phantom has a 2kHz whistle in the noise floor which presumably is from the DC converter). But I would be shocked if there were a difference that could possible make one unit "sound a bit better".

+1

 

[JohnRoberts]

I like the idea of trying to null the same PS to another sample of the same PS. If you can get a deep null this way, it adds credibility to your test. If "same" to "same" does not null deep, we have a problem with the null test.

JR

 

[ccaudle]

It will be difficult to time synchronize samples processed at different times.

There are some software tools available to help:
DeltaWave audio null comparator
Audio DiffMaker (older than DeltaWave)

Note I have not used those myself yet, I just found them but won't be able to install and use until later tonight at least (after 23:00UTC).

AFAIK there are no DAWs that allow moving sample by a fraction of SR period

Most do not by default, but you can sometimes add the capability.
This Eventide sub-sample delay, for example:
Eventide Precision Time Align

I will be interested to see the code @user 37518 is using for his spectrum based diff tool. Using cross-correlation you should be able to find the delay to fairly high precision, but the "easy" way is to just move to the closes sample position.

I suspect it is something like this example:
signal fitting with sub-sample resolution

 

[user 37518]

There are some software tools available to help:
DeltaWave audio null comparator
Audio DiffMaker (older than DeltaWave)

Note I have not used those myself yet, I just found them but won't be able to install and use until later tonight at least (after 23:00UTC).



Most do not by default, but you can sometimes add the capability.
This Eventide sub-sample delay, for example:
Eventide Precision Time Align

I will be interested to see the code @user 37518 is using for his spectrum based diff tool. Using cross-correlation you should be able to find the delay to fairly high precision, but the "easy" way is to just move to the closes sample position.

I suspect it is something like this example:
signal fitting with sub-sample resolution

Yeap, I am using cross-correlation to match the audios, because visually matching the sound files is not a very reliable way to do it, unless they are almost identical. I will be posting the script soon.

 

[rjb5191]

You can't do a proper null test if you recorded your samples in different time. You need to record simultaneously and use the same converter.

I used the same converter, but I understand the issue. There's no way to do this when the point is to switch between the two different power supplies on the same unit.

 

[rjb5191]

I like the idea of trying to null the same PS to another sample of the same PS. If you can get a deep null this way, it adds credibility to your test. If "same" to "same" does not null deep, we have a problem with the null test.

JR

Yes, this makes sense. I will do this also.

 

[Bo Deadly]

That wouldn't be a valid test, since the clock differences, both in phase and frequency would introduce differences in the reconstituted audio.

Popycock.

Effects of clock phase will have minimal effect when looking for differences that would be audible. I've done diffs this way and it can yield very informative results. Its really the only practical way to do it.

I used WavePad which, last I checked, is free to some extent to create various stimulus with short breaks of 0's at the beginning to make it clear how the samples need to be aligned using Audacity.

But yes, you might get marginally better results if you split to both units and capture simultaneously.

 

[abbey road d enfer]

Popycock.

Thank you.

Effects of clock phase will have minimal effect when looking for differences that would be audible.

You may not have properly read my point. I'm not talking about clock phase variations. I'm talking about making the two files perfectly sync'ed with more precision than the +/- one half period that you can get by aligning them in the timeline.

 

[dogears]

Quick way to start I suppose would be to use a scope to look at the business end of the +48V rail. And check to see that the voltages are the same...

 

[Bo Deadly]

You may not have properly read my point. I'm not talking about clock phase variations. I'm talking about making the two files perfectly sync'ed with more precision than the +/- one half period that you can get by aligning them in the timeline.

I think I understand what you mean. But for basic null testing, that level of precision should not be required and I know that because there was a time when I got super pedantic about audio analysis using Octave and seeing how much I could null-away and it worked surprisingly well even without properly synchronized recordings. Converters average together many individual conversions before emitting a sample so, even at HF, the values should be similar enough that you can do a pretty good digital diff with separate sample aligned recordings.

Although if I ever have an infinite amount of time and resources, I would be interested in studying the exact differences between:

a) digital null of separate sample aligned recordings
b) digital null of properly synchronized recordings
c) analog null

 

[abbey road d enfer]

But for basic null testing, that level of precision should not be required

According to rjb "there's an audible difference above the noise floor." For me it means the difference is just emerging from the noise floor, so it's in the LSB's. I suggest it could just be the result of dithering, whethere it's intentional dithering or just noise dithering.

I would be interested in studying the exact differences between:

a) digital null of separate sample aligned recordings
b) digital null of properly synchronized recordings

It seems one of the softwares that ccaudle and user 37518 mentioned would answer that.

c) analog null

As you know, the effects of component variations due to temperature, humidity, planet alignment largely exceed the noise floor. A 1/100th of a dB variation in frequency response results in a null error that's well above the noise floor of many systems.

 

[rjb5191]

According to rjb "there's an audible difference above the noise floor." For me it means the difference is just emerging from the noise floor, so it's in the LSB's. I suggest it could just be the result of dithering, whethere it's intentional dithering or just noise dithering.

It seems one of the softwares that ccaudle and user 37518 mentioned would answer that.

As you know, the effects of component variations due to temperature, humidity, planet alignment largely exceed the noise floor. A 1/100th of a dB variation in frequency response results in a null error that's well above the noise floor of many systems.

There was no dithering applied to the files. Anyone with more knowledge than me is very much welcome to properly align the files and make a proper null test if mine didn't meet a certain standard, which I agree probably didn't. I'm not sure how to align the files more exactly. I'll take some basic measurements from the two supplies when I have the chance.

 

[abbey road d enfer]

There was no dithering applied to the files.

the moment the signal goes through an analog circuit, some kind of dithering happens, due to noise.

 

[ccaudle]

There are some software tools available to help:

Played with those a little bit last night. First thing that threw me off was that the original files are 32bit float format dumped from a DAW. The older software (Audio DiffMaker) didn't recognize that at all, DeltaWave did, but didn't play nicely with my audio interface (seemed to want to send 32 bit audio to the interface, even though it should be advertising it is a 24 bit device).
Eventually got them both working well enough, and DeltaWave seems to do a better job time aligning. Even after time and amplitude aligning, the best null achieved was about -55-ish dB.

Listening to the amplified difference there did not seem to be noticeable distortion, frequency response changes, etc., so my preliminary guess was that the playback was started at slightly different points in the original source, that some amount of compression was still being applied, and that the interaction between attack and release time of the compressor meant that there were very slight differences in gain at each point in time in the resulting files.
Of course if rjb5191 says that the compressor was off, and the unit was only being used as a buffer amp, then that hypothesis would be right out the window.

I did listen to the files, of course, but I could not hear the differences rjb5191 described.
If anyone wants to do some blind checks to see if you can identify the files without knowing what you are listening to this ABX playback and testing tool is handy:
Lacinato ABX software
tutorial on using Lacinato ABX

 

[a soBer Newt]

Analog Devices Inc. Manufacturer Product Number LT8330ES6#TRMPBF is an amazing part for P48 supplies IMO.

 

[My3gger]

Isn't linear phantom supply so cheap to make from existing transformer it makes no sense buying separate switcher? PSRR is more often than not high, allowing simple circuit.

 

[Newmarket]

You'll need a good amount of filtering on either kind of supply. A linear supply will have 120Hz (or 100, depending on where you live) ripple, while a switching supply will have higher frequency noise. But almost any piece of pro gear you buy these days will have all switching supplies.

A properly designed regulated linear supply will not have any significant mains frequency on its voltage rails. It's what linear regulators do. Just need enough capacitance to give those regulators enough to work with. Of course it's different if you are running a power amp stage from unregulated "rough DC".
There is an issue of radiated mains frequency noise but that can be dealt with by design / spacing. Also applies to SMPS as you still have mains frequency at the input.

 

[user 37518]

Isn't linear phantom supply so cheap to make from existing transformer it makes no sense buying separate switcher? PSRR is more often than not high, allowing simple circuit.

It depends, not for all applications, you might need an extra tapping on the trafo that might not be available or use a rectifier to make a voltage doubler or tripler, some big filter capacitors and a high voltage regulator are needed, the LM317HV costs around $2 USD and is a good candidate but not the only option, the price of big reservoir caps can be quite high as well. For small power consumption and a small voltage drop from input to output, you don't need a proper heatsink for the regulator, otherwise, you need a good chunk of aluminum. All of the above can be replaced with a cool DC-DC converter IC, a small inductor and a few small value ceramic capacitors.