Bipolar SMPS with Manufacturer Recommended Circuit

[Potato Cakes]

But presumably you have an audio interface that can go up to 20kHz or 30kHz, should give some idea of what is going on at 220Hz.

The audio interface being used is a Sound Devices Mix Pre3-II, which according to the spec sheet goes to 22kHz.

 

[JohnRoberts]

Attached is a generic capacitance multiplier schematic that I found. I did see some others that recommended another resistor that is parallel with the capacitor which forms a voltage divider. I do not know which values to use yet.

Thanks!

Paul

The second resistor biases the base of the multiplier at a lower voltage so ripple on the unregulated does not cause the pass trasistor to saturate.

JR

 

[dogears]

I think you need to figure out why the supplies that switch at 100kHz are producing 220Hz before worrying about filtering it out.

 

[Potato Cakes]

I think you need to figure out why the supplies that switch at 100kHz are producing 220Hz before worrying about filtering it out.

Right now the 220Hz is inconsequential to me as I know through filtering (the 2200uF cap mentioned earlier) I achieve the desired result regarding noise. It's just that have to do filtering via something else than a large cap as discussed earlier. If I can't figure out a solution soon I will have to hire some help.

Thanks!

Paul

 

[dogears]

And by using the really big hammer approach you put it into a fault mode. There may well be something simple going on.

 

[ccaudle]

The audio interface being used is a Sound Devices Mix Pre3-II, which according to the spec sheet goes to 22kHz.

Actually the spec sheet says it goes quite a bit higher:
10 Hz to 80 kHz +/- 0.5 dB re 1 kHz @ 192 kHz sample rate

You already used that to check the spectrum of the power supply output. Through some large capacitors, I assume? You didn't actually show how you connected the power supply to the audio interface.

But the point is, just use the same setup for time domain.
Search sound card oscilloscope

 

[ccaudle]

The noise level is relative

Of course, which is exactly why I asked what the reference level is, i.e. relative to what reference voltage.
Sound Devices makes it more difficult than necessary to figure out the reference level, but the spec sheet for your interface says that the maximum input level for line input is 28 dBu, and the minimum line input gain is -20 dB, implying that for 0dB gain 0 dB FS is 8 dBu. I think, it is not entirely clear how the gain staging is setup, whether the max input spec is for 0 dB gain or minimum gain.
Assuming the max input is specified at minimum gain, for mic input the maximum input is 14 dBu, but minimum gain is 6 dB, so again 0dB would be 8 dBu. That is consistent, so I'll go with that.

8 dBu is 1.95V RMS. In your original spectrogram the high frequency noise floor was around -100 dB, and the 220 Hz noise peaked at -80 dB. It is not clear from your screenshot what FFT parameters were used, so hard to compare to normalized per Hz values, but just using the raw values, that -80 dB peak would be around 0.2mV. That is pretty low for a basic switcher. It would be good to know where it is coming from, but considering that the manufacturer noise plus ripple spec is 120mV peak-to-peak, you probably won't get a lot of help from Cui trying to get even lower than that.

The -80dB show in actual practice is too high and clearly audible.

Clearly audible where? At the output of the audio circuit being tested? Are you trying to power a really simple transistor circuit of some kind? Any kind of op-amp based audio circuit is going to have quite high power supply rejection at 220Hz, at least 80dB, which would put that -80 dBu peak on the raw power supply at around -160 dBu at the output of the audio circuit.
Of course a single ended transistor circuit with no feedback is going to have essentially 0 PSRR, so it matters what kind of circuit you are powering.

It also is important to verify that you are actually measuring what you think you are, so checking your setup with a pair of batteries, and other known high quality power supplies would be useful, as well as the switchers you have under varying load conditions (i.e. put some different sized resistors on the output so you can see if the noise varies with the amount of current supplied).

 

[Potato Cakes]

Actually the spec sheet says it goes quite a bit higher:
10 Hz to 80 kHz +/- 0.5 dB re 1 kHz @ 192 kHz sample rate

You already used that to check the spectrum of the power supply output. Through some large capacitors, I assume? You didn't actually show how you connected the power supply to the audio interface.

But the point is, just use the same setup for time domain.
Search sound card oscilloscope

My mistake. I was looking at the THD specs just below.

 

[Potato Cakes]

Of course, which is exactly why I asked what the reference level is, i.e. relative to what reference voltage.
Sound Devices makes it more difficult than necessary to figure out the reference level, but the spec sheet for your interface says that the maximum input level for line input is 28 dBu, and the minimum line input gain is -20 dB, implying that for 0dB gain 0 dB FS is 8 dBu. I think, it is not entirely clear how the gain staging is setup, whether the max input spec is for 0 dB gain or minimum gain.
Assuming the max input is specified at minimum gain, for mic input the maximum input is 14 dBu, but minimum gain is 6 dB, so again 0dB would be 8 dBu. That is consistent, so I'll go with that.

8 dBu is 1.95V RMS. In your original spectrogram the high frequency noise floor was around -100 dB, and the 220 Hz noise peaked at -80 dB. It is not clear from your screenshot what FFT parameters were used, so hard to compare to normalized per Hz values, but just using the raw values, that -80 dB peak would be around 0.2mV. That is pretty low for a basic switcher. It would be good to know where it is coming from, but considering that the manufacturer noise plus ripple spec is 120mV peak-to-peak, you probably won't get a lot of help from Cui trying to get even lower than that.



Clearly audible where? At the output of the audio circuit being tested? Are you trying to power a really simple transistor circuit of some kind? Any kind of op-amp based audio circuit is going to have quite high power supply rejection at 220Hz, at least 80dB, which would put that -80 dBu peak on the raw power supply at around -160 dBu at the output of the audio circuit.
Of course a single ended transistor circuit with no feedback is going to have essentially 0 PSRR, so it matters what kind of circuit you are powering.

It also is important to verify that you are actually measuring what you think you are, so checking your setup with a pair of batteries, and other known high quality power supplies would be useful, as well as the switchers you have under varying load conditions (i.e. put some different sized resistors on the output so you can see if the noise varies with the amount of current supplied).

It is clearly audible when monitoring back through any audio interface or when monitored back after making a recording. The screen shot is from an analyzer plugin that is useful for doing a variety of audio tests. As for the Sound Devices Mix Pre, it is very precise on setting levels, which I did so at unity gain (0dB).

I am indeed measuring what I think I'm measuring. I first I thought I was getting an issue because of an impedance mismatch but that is not the case. I had this same issue with another interface when I was home doing these same tests.

Again, my conclusion is that there just needs to be proper filtering on the output of the PSUs. I will point again to the results I got from using an oversized cap across the +V of PSU 1 and -V of PSU 2 (two PSUs that for the bipolar PSU as described above). But the problem is that an actual 2200uF cap is way too much for the PSU itself, even though in a different circuit it did managed to power on after cycling several times. Hence, I am now looking to use a capacitance multiplier scheme so I can replicate the noise results of using a large capacitor without causing the PSU to freak out. My problem is I don't know where to start with selecting component values per the capacitance multiplier schematics posted earlier.


Thanks!

Paul

 

[dogears]

It isn’t clear why a SMPS switching at 100k would produce 220 Hz. It’s possibly a beat frequency from the two supplies, or some kind of low load hiccup. Either way it shouldn’t be there.

Test it with only one supply connected, then see if a minimum load affects the frequency.

If you can eliminate the cause of the 220 Hz you can avoid using the huge cap, and avoid solving the new problem that creates.

 

[ChrisMadog]

Woah! Woah! The square things in the schemo are not bridge rectifiers. They are AC/DC isolated converters.

If I understand well, you are suggesting half-wave rectification for each rail. That would result in output voltages of about +/-160V (for 115V mains) and double for 230V mains. I don't think it's what the OP wants.

Aha - yes,, I understand now.
My bad.

Disregard everything I said.

ChrisM

 

[ccaudle]

It is clearly audible when monitoring back through any audio interface

And "It" is the noise on the 15V rail coupled directly to an audio interface? Or "It" is the effect of the power supply noise on the noise at the output of an audio circuit powered by those regulators? That was the question, what is the connection used for measuring and listening to this noise you are describing.

I am indeed measuring what I think I'm measuring.

My point with that comment was to verify that the noise is actually on the output of the regulators, and is not for example being picked up in the wiring to the audio interface through magnetic induction. That is probably not what is happening, just because the spectrum shows about a spread of several tens of Hz, magnetic interference from power line related frequencies would usually be pretty narrow, but when you see something that doesn't have an obvious explanation it is good to verify your tools setup just to get a baseline for comparison.

I first I thought I was getting an issue because of an impedance mismatch

That doesn't really make sense, but it does bring up a question which been asked a couple of times before, which is what load do you have on the power supply when measuring? Some power supplies either lower the switching frequency, or skip switching cycles entirely, at low load. See post #30 above.

 

[ach]

This is not from the same power supply, but it is a tech doc I received from a similar SMPS manufacturer. The *strong* recommendation for this circuit is to include additional diodes as shown below, specifically to help prevent start-up and switching issues. Perhaps not a panacea for noise, but could help with stability.

 

[Potato Cakes]

Condensed summary of findings posted here:

https://groupdiy.com/threads/smps-class-a-microphone-preamp-discoveries.82639/