Why do SMPS sound rolled off in the high frequencies?

[stretchneck]

I have used a number of SMPS to power my music server over the years, but none of them has got close to my Paul Hynes SR7 in the high frequencies.

Some computer power supplies, such as high end corsair, and the AX1600i I had have very good voltage regulation and ripple suppression - the AX1600i sounded to me as identical in the lower and mid range, but rolled off / unsatisfactory upper band.

I use Hypex SMPS in my amplifiers and these are completely perfect accross the frequency spectum to my ears.

Has anyone else experience this, and any thoughts reasoning why these computer smps sound rolled off in the high notes, yet the Hypex power supply doesn't seem to do this.

Thanks

 

[Mark Tillotson]

Well human hearing is heavily rolled off in the high notes too, so I think you'd need to do some measurements to establish the truth of the matter.

 

[stretchneck]

Well human hearing is heavily rolled off in the high notes too, so I think you'd need to do some measurements to establish the truth of the matter.

I know for a fact that my hearing is rolled off, I can't hear anything above 16kHz. But why does my NUC (19v input) have much better high end with a Paul Hynes SR7 linear supply, whereas a high spec'd ATX power supply (AX1600i) sounds very similar, but really rolled off and no prescence in the high octaves. No other factors were changed in the comparision, including my ears

Unfortunately I don't have any measurement equipment... I'm just looking for potential technical reasoning.

I know that not all SMPS exhibit this. I recently tried a Melco streamer, that uses SMPS, but sounded perfect across the entire listening frequency.

 

[TNT]

Do some measurement to document the difference. I suggest in the electrical domain. Report here and we can discuss further.

//

 

[abraxalito]

@Mark Tillotson, @TNT - do you have specific measurements in mind? Do share.

Changing power supplies has no effect on FR so the 'rolled-off' effect probably is as the result of HF noise injection into your signal. To understand further I'd need to see a sketch of how your various systems components are interconnected but the most likely mechanism for noise injection is common ground impedance coupling over single-ended audio interconnects. The Hypex in my understanding uses balanced ICs for its inputs which is likely the reason its own SMPSU isn't injecting HF noise.

For those who are interested in some technical nitty-gritty, I recommend this article : https://www.eetimes.com/power-tip-47-tame-conducted-common-mode-emissions-in-isolated-switchers-part-1/

 

[TNT]

Yes, do a fr sweep measurement on the output of the amp or a pink noise spectral analysis using the two different PSs to see if there indeed is a difference that correlate with the listening impression.

//

 

[abraxalito]

What's your proposed mechanism for how a PSU swap can affect frequency response?

 

[stretchneck]

@Mark Tillotson, @TNT - do you have specific measurements in mind? Do share.

Changing power supplies has no effect on FR so the 'rolled-off' effect probably is as the result of HF noise injection into your signal. To understand further I'd need to see a sketch of how your various systems components are interconnected but the most likely mechanism for noise injection is common ground impedance coupling over single-ended audio interconnects. The Hypex in my understanding uses balanced ICs for its inputs which is likely the reason its own SMPSU isn't injecting HF noise.

For those who are interested in some technical nitty-gritty, I recommend this article : https://www.eetimes.com/power-tip-47-tame-conducted-common-mode-emissions-in-isolated-switchers-part-1/

Thanks for the feedback - this is helpful.

I use hypex and melco as example, and I am talking about the use of Computer ATX SMPS used in a music server. Looking at Melco, they use TDK IEC filters, Innuos also use them on their computer server even though it's a Sean Jacobs linear power supply - Innuos use a Schaffner FN 9255 E dual stage filter by the looks of it. SGM sound galleries use a dual choke in their power supply.

John Swenson provides some insight here are as well:

"The traditional cap only filter (transformer, diode bridge, big cap) produces raw DC with a sawtooth riding on top. That sawtooth produces lots of high frequency components that the regulator has to deal with. Traditional regulators do very well at low frequencies, but have lousy characteristics at high frequencies which means a fair amount of those high frequency components from the cap-only filter get through to the regulator. Fancy discrete regulators do well at blocking the high frequency components, but add cost and complexity to a PS. Our approach is to use a properly designed choke-based supply whose ripple is a perfect sine wave, no high frequency components, thus a traditional regulator works very well. The discrete regulator is not needed to deal with the high frequency components, since there aren't any."

JS-2 Linear Power Supply – UpTone Audio

I'm wondering if the effect I'm heating is mains borne or RFI

 

[abraxalito]

John Swenson there is talking about normal-mode noise rather than common-mode noise.


The TDK IEC filter you cite almost certainly is going to let a lot of CM noise through. Its notoriously difficult to filter (mentioned in my linked article) - a very low capacitance CM choke combined with high inductance is called for. Those two aspects (low C, high L) are mutually exclusive. In general a mains filter should be used as close to its rated current as possible to get the best CM filtering - a 10A one as cited seems to have too high a current rating for a computer. The graphs in the DS you linked to for the Schaffner filter illustrate this principle well - as the current rating rises, so the CM rejection falls.