Related: some measurements with Audio interface capture, Audacity's Spectrum Analysis, and the iFi iDSD Nano DAC (capable of DSD 256 with free firmware update). I replaced the default Computer USB bus-powered DAC box by external power, some charger-types, a battery pack, and then my DIY low-noise Linear Regulated PSU. You will see graphs of the DAC output spectra there for each power solution. There's also some filtering involved along the USB routes.
Some context:
I worked on a custom USB connector/isolator - purely analogue things, but additional SQ gains also required AC-side filtering to achieve the best SQ.
PSUs will interact. SMPS in the chain will generally ruin it, if not the circuit being powered, it will add noise to another piece of gear.
These measurements are what they are - no claim is made that newer single-ended DACs will react the same (I expect better isolation nowadays), no claim is made that truly balanced DAC will react the same (YMWV- Your Mileage Will Vary).
Some context:
- The DAC had been released maybe a couple of years before 2017, measurements done in Winter 2017. Note the date typos, they should be a year earlier since I posted this on Dec 2017: "Dec 2017 - Jan 2018 holidays (so last year)" should really be "Dec 2016 - Jan
2017holidays (so last year)" for the custom USB connector/isolator. - It's a single-ended DAC
- It's a budget DAC
- There was no USB Isolator chip at the time for the kind of USB speed it needed to do a full isolation
I worked on a custom USB connector/isolator - purely analogue things, but additional SQ gains also required AC-side filtering to achieve the best SQ.
PSUs will interact. SMPS in the chain will generally ruin it, if not the circuit being powered, it will add noise to another piece of gear.
These measurements are what they are - no claim is made that newer single-ended DACs will react the same (I expect better isolation nowadays), no claim is made that truly balanced DAC will react the same (YMWV- Your Mileage Will Vary).
Last edited:
16 bit. Don't forget that I took this measurement at the end of a 1m long cable, since that's form that the 12V USB trigger is in. The noise floor is also affected by FFT size.
I have, as in the thread I linked to. It does work. I've done it on other power supplies too - sometimes something as simple as clamping a ferrite bead onto some wires. If you can't do that, then an alternative is to earth it right at the secondary of the transformer, to provide the smallest possible loop for the noise current to flow around.
That is what I found as well.... and it is not the subject of the analysis in this thread. Maybe it should also be examined..?
I spent a couple of months trying a few things with SMPS but, apart from not being happy with the result, one crucial element of surprise was how detrimental the SMPS was if plugged into the same source of mains, that powered my amplifier (Aleph J) and DAC (Holo May). I had to lay a separate power cord and plug it into a separate mains outlet just for the SMPS I used.
Anyway, I settled on not using SMPS and not considering them for my needs. Ended up building a CLC front-end for my liner power supply with two TO3 paralleled transitions as a series pass devices, to reach my needs of 6A constant current draw. The perceived difference in sound was just... like opening the gate to another, unknown and unexplored dimension.
That one I missed - went straight to the screenshot showing the measurements.
But I had a look/read ... Nice results and a solid improvement. Thanks for sharing.
Here are the Meanwell photos:
1 = dirty ground
2 = semi-dirty (after the choke)
3 = "clean" (output) ground, after 2 x 0.01 ohm SMD resistors (notice: there's provision for 4 of these reisitors, but only 2 were used)
The top view... I replaced one of the filter capacitors with the same-value Philips cap that I like. Notice 2 x 270ohm resistors that are loading the output when the power supply is idle => 110mA of the constant current draw... which in this case might be too much due to the minuscule size of that heatsink that runs too hot even with no load connected:
AU$76 at Mouser... I looked a bit more... AU$46 at the local shop in Sydney. So, yeah, it will power up something that needs 15V... but it sure is a noise generator with appalling ground uniformity... it also needs the fan cooling to keep the whole thing under control temp-wise even when it's idling. That is all despite the very latest and greatest ultrafast-switching diodes used (D92-02).
Just a friendly reminder for all who are planning to play with these open-frame types... the big black 400-odd volts capacitor LIKES to keep the charge for around 5-10min. Either discharge it with a (what I use) 10ohm/50W resisitor, or let it sit on the bench for a while.
1 = dirty ground
2 = semi-dirty (after the choke)
3 = "clean" (output) ground, after 2 x 0.01 ohm SMD resistors (notice: there's provision for 4 of these reisitors, but only 2 were used)
The top view... I replaced one of the filter capacitors with the same-value Philips cap that I like. Notice 2 x 270ohm resistors that are loading the output when the power supply is idle => 110mA of the constant current draw... which in this case might be too much due to the minuscule size of that heatsink that runs too hot even with no load connected:
AU$76 at Mouser... I looked a bit more... AU$46 at the local shop in Sydney. So, yeah, it will power up something that needs 15V... but it sure is a noise generator with appalling ground uniformity... it also needs the fan cooling to keep the whole thing under control temp-wise even when it's idling. That is all despite the very latest and greatest ultrafast-switching diodes used (D92-02).
Just a friendly reminder for all who are planning to play with these open-frame types... the big black 400-odd volts capacitor LIKES to keep the charge for around 5-10min. Either discharge it with a (what I use) 10ohm/50W resisitor, or let it sit on the bench for a while.
A couple of mine were with 1M cables and 24Ω term. One terminated in 6Ω is not much of antenna...
Yes, but it's relatively small difference, and noise goes down with increased samples.
If that measurement is real, that's one noisy supply.
I sweep them with an ultrasonic receiver/amplifier I have here. The computer ones have a nasty 50hz blare, while the one in some late model Marantz just screams at a higher frequency.
Not my cup of tea.
Not my cup of tea.
Now let's see what difference a common-mode choke on the output makes. Most CMC's are designed for mains input filtering and are high-inductance (1mH and up), not suitable for a ps output. Something in the 2- to 5uH range is suitable, as it maintains low output impedance over most of the audio band.
The CMC I used is 3.1uH per side, with 2mΩ DCR. For a 65kHz switcher (the MeanWell RPS-30-15), the impedance of 3.1uH at 65kHz is about 1.3Ω. With no capacitor following the CMC, noise was the same as without it. So all by itself, the CMC isn't giving us anything.
These noise spectra are with 100uF and then 1000uF after the CMC, 1A load on the RPS.
The 65kHz spike is only 6dB lower with the CMC compared to without it. The capacitor is what makes the biggest difference.
Used as a pre-regulator, the lowest noise is using a buildout resistor (such as 470mΩ) followed by a large cap.
The CMC I used is 3.1uH per side, with 2mΩ DCR. For a 65kHz switcher (the MeanWell RPS-30-15), the impedance of 3.1uH at 65kHz is about 1.3Ω. With no capacitor following the CMC, noise was the same as without it. So all by itself, the CMC isn't giving us anything.
These noise spectra are with 100uF and then 1000uF after the CMC, 1A load on the RPS.
The 65kHz spike is only 6dB lower with the CMC compared to without it. The capacitor is what makes the biggest difference.
Used as a pre-regulator, the lowest noise is using a buildout resistor (such as 470mΩ) followed by a large cap.
I compared the spec sheets for the one you tested (RPS-30-15) and the one I posted the photos of (LPP-150-15). I noticed the same noise figures (100mV pp); however, the RPS has only 0.1W max of idle power consumption (7mA) and therefore - does not need forced cooling when idle, whereas LPP, despite the claim of 10A max current capability, will need the forced cooling as soon as it is placed in the chassis (enclosure), due to idle current draw I mentioned earlier (110mA !!!)
Of course, the spec sheet for LPP does not state anywhere the idle current / power consumption. The fan is shown in the mechanical drawings section....
I read somewhere on these forums that the only thing they are good for is to toss them really high and hit them with a baseball bat really hard.
Of course, the spec sheet for LPP does not state anywhere the idle current / power consumption. The fan is shown in the mechanical drawings section....
I read somewhere on these forums that the only thing they are good for is to toss them really high and hit them with a baseball bat really hard.
I think you measured a Mean Well low end module(RPS-30-15). I wouldn't use it.
If you could measure Mean Well GSM GST module, and use a LCR filter for HF and Plus a CRC traditional filter for 60HZ, you probably would get a different result.
If you could measure Mean Well GSM GST module, and use a LCR filter for HF and Plus a CRC traditional filter for 60HZ, you probably would get a different result.
larger chokes might have mH of common-mode inductance, but much less differential-mode inductance - µH or so. Just use a bigger capacitor afterwards if you need low output impedance though.
That's the one sold here in the store, with ACA kits? Like this one: GST120A24-P1M (24V 5A)... maybe? It would be interesting to see the measurements...
However, based on the spec sheets, the GST models have higher ripple&noise vis-a-vis, compared to RPS and LPP... The external filtering works to a certain degree (and I will leave it to you to decide if that is good enough for your audio applications). However, I do not see how the noisier power supply would yield better noise performance even if the abovementioned filters (both) are used. One thing is also certain - I do not think anyone tried both filters, one after the other, to tackle the broad noise spectrum. But then again... the CRC filter didn't do much...
Just to finish... would anyone buy any of these SMPS modules, If they are sold with two separate filters to combat high noise levels? Even if they are combined into one PCB... it would be a hard sell and a pain to find the additional real-estate area to fit them.
However, based on the spec sheets, the GST models have higher ripple&noise vis-a-vis, compared to RPS and LPP... The external filtering works to a certain degree (and I will leave it to you to decide if that is good enough for your audio applications). However, I do not see how the noisier power supply would yield better noise performance even if the abovementioned filters (both) are used. One thing is also certain - I do not think anyone tried both filters, one after the other, to tackle the broad noise spectrum. But then again... the CRC filter didn't do much...
Just to finish... would anyone buy any of these SMPS modules, If they are sold with two separate filters to combat high noise levels? Even if they are combined into one PCB... it would be a hard sell and a pain to find the additional real-estate area to fit them.
The RPS series has hospital-use approvals and a lower noise spec than the GST series. That does not inspire me to order one. You're welcome to post the noise spectra of one, though.
Capacitors over 1000uF yield no reduction in the 65kHz peak.
Also note in the plot that the noise was worse (about 8dB max) from ~ 500Hz to 1.8kHz with the 1000uF cap vs the 100uF one. That is real, and is caused by the resonant interaction of the L and C chosen. Larger caps push the resonance even lower. The two-stage filters have this as well. These are not "maximally flat" networks...
- and remember that experimenting with adding a small, judicious, calculated amount of damping of both series & and parallel LC resonances might well clear things up mightily.
That's an aspect that almost never ever appears in the many 'new-hotness, genius diy mains ac filter' and similar recipes. & so why you should not beleive everything you read in such threads.
- It is also, exactly, no small part of why Mark Johnson's inline-smps adaptor filter thingy works.
@martin clark , I disagree. I modeled Mark's filter yesterday and the small added milliΩ's are inconsequential to the resonances, which are peaking below the filter cutoff. If you want a 4th order filter without resonances, just make it a 4th order Butterworth. And doubling or tripling up on the filter stages does nothing to stop the switching frequency from being pushed onto circuit ground.
It's possible that a higher inductance CMC would improve on my results, but the output impedance would be very messy. Of course, one can create all sorts of "midrange magic" with impedance peaks in the audio passband...
It's possible that a higher inductance CMC would improve on my results, but the output impedance would be very messy. Of course, one can create all sorts of "midrange magic" with impedance peaks in the audio passband...
They are kOhm, not mOhm, values.
But the real driver for my comment above - is that (and this is v esp fo mains ac filters) the output load side is not a nice, passive /resistive load, absolutely nothing like: if a rectifier is involved, or discontinuous current, you have a wideband trains of artifacts reflected into the output of the filter, and that... is a singular mess particular to each use case.
But the real driver for my comment above - is that (and this is v esp fo mains ac filters) the output load side is not a nice, passive /resistive load, absolutely nothing like: if a rectifier is involved, or discontinuous current, you have a wideband trains of artifacts reflected into the output of the filter, and that... is a singular mess particular to each use case.
They are... but they are not used in the correct combination with other components across L, to make them efficient.
According to Mark's remarks... where he goes on and on about how he spent a lot of time, used high-end measurement gear and completed the PhD on the subject... did he EVER post in his tread the measurements?
According to Mark's remarks... where he goes on and on about how he spent a lot of time, used high-end measurement gear and completed the PhD on the subject... did he EVER post in his tread the measurements?
Indeed, because at 65kHz the attenuation from larger capacitors is limited by ESR and ESL. You need a bigger choke to get more attenuation at 65kHz. Larger capacitors will give you the lower output impedance at audio frequencies that you wanted though.
There are both. The 1k-ish Ohms across the 2.2uH L is only meaningful in the multi-MHz range.
But this whole issue isn't about the load - it's about the source.