Enclosed SMPS versus open frame regarding noise

[thor.zmt]

I was being generous with my comment about a maxi of 1 millivolt at mains frequency from a linear supply. The Coutant and Kingshill supplies I used to work with specify around 200 microvolts peak to peak (referred to as low frequency noise as I assume they weren't really looking a lot beyond 20KHz .

Thats nearly 40dB noisier than my SMPS... Just saying.

Thor

 

[Matt Syson]

2 microvolts peak to peak over a DC to say 30 kHz bandwidth would be impressive however the point is that in the work I have been doing for 45 years it is the whole 'package' that is important and all mixing desks have a fair amount of rail to rail capacitance designed in because of course signal current flows in localised loops so with usual resistance capacitance isolation of individual or blocks of stages the whole system can remain 'quiet' as long as noise is not introduced into the grounding system. A mix amp stage has around 30dB of 'gain' (when referring to ground noise and to obtain better than around minus 70dB total output noise means that the whole system has to be resistant to interference either 'radiated' (received) or conducted. The struggles of the past to get low noise, distortion free audio circuitry are being compromised by cheap, inadequately filtered power supplies, then some complain that clean, low noise amplification sounds 'sterile'. Some people are never happy!

 

[Bo Deadly]

You say "mine does -120dB" without stating what is 0dB. That means your statement cannot be evaluated.

When I say "my USB interface can do 120 dB", I mean typical specs associated with USB interfaces like dynamic range or EIN. I'm not sure what my old MOTU Traveler mk3 specs are but these days EIN of 127-ish is not uncommon.

If it is for example an Audio Interface with 0dBFS at +22dBu or +20dBV, to make things comparable, the top of my chart would be at -60dB and the bottom at -180dB.

If I'm not mistaken, the 100Hz peak in your plot is at -117dBV which is well above the EIN of a typical audio interface. And your noise floor looks like -151dBV?

Notice that in my second plot, that noise floor is being amplified +60dB using a THAT 1510. I don't recall the specifics of what I did when I took that measurement but my guess would be that I attenuated the input just enough for the output to reach 0dBFS which means the noise floor represented in that plot is quite a bit lower than the -120dB shown. SO I think our plots are quite comparable.

Having said that, I'm not suggesting that there's anything "wrong" with your SMPS. I acknowledge that those peaks are very very small. I was just trying to make the point that a lowly MeanWell APC-16-350, which is $10 on Mouser, can be very quiet and that for DIY at least, I don't recommend that people roll their own.

 

[ccaudle]

This is a APC-16-350

How did you connect to the mic amp? Use the amps built in caps and just connect the two signal inputs to power out and local ground?
Calibration to dBV should be relatively straight forward, have some device output a sine wave in a range that your DMM can measure accurately, adjust for 1V output, and then connect that to the input and adjust until it is 0 dB FS.

 

[ruffrecords]

If I'm not mistaken, the 100Hz peak in your plot is at -117dBV which is well above the EIN of a typical audio interface. And your noise floor looks like -151dBV?

Don't forget Thor's FFT plot is almost certainly per root Hz. So, you need to integrate from 20Hz to 20KHz to obtain a noise floor figure that we audio designers would recognise. Assuming approximately Gaussian noise this means you need to add about 43dB to the 'apparent' -151dBV noise floor which makes it about 108dBV in my book. Most FFT software will show this automatically (REW does for instance). Not good enough for the EIN of a mic pre but plenty good enough for a power supply.

Cheers

Ian

 

[ruffrecords]

Having said that, I'm not suggesting that there's anything "wrong" with your SMPS. I acknowledge that those peaks are very very small. I was just trying to make the point that a lowly MeanWell APC-16-350, which is $10 on Mouser, can be very quiet and that for DIY at least, I don't recommend that people roll their own.

LED supplies are possibly an unexplored seam of gold for DIY PSUs. I have a Meanwell ELG-15--C500A which will output 300Vdc at up to 500mA. My plan was to use it as the HT supply for a big tube mixer with a chunky linear regulator to drop and regulate the volts down to 275V. Maybe all it needs is the right kind of CLCLC filer on the output?

Cheers

Ian

 

[Bo Deadly]

LED supplies are possibly an unexplored seam of gold for DIY PSUs. I have a Meanwell ELG-15--C500A which will output 300Vdc at up to 500mA. My plan was to use it as the HT supply for a big tube mixer with a chunky linear regulator to drop and regulate the volts down to 275V. Maybe all it needs is the right kind of CLCLC filer on the output?

That's a hefty supply. As you know it's really important to match an SMPS to the load.

Just brainstorming here but Constant Current supplies are designed to run at the rated current so I could see a capacitance multiplier with an RC that also has a shunt regulator to servo current. That would allow you to dial in any output voltage from 300-150. Maybe using some kind of circuit with a BD140 and TL431 for example. Of course the circuit would require a lot of fiddling. Specifically it should only shunt some maximum and then stop. If the load were momentarily disconnected for example, you don't want the shunt to eat all 150W and burn your house down. Even 50W would be troubling. So again, the SMPS would have to match the load fairly well to go this route.

The alternative would be to simply not operate it in the CC region. That's ok. I like the CC ones because they're half the size of the CV ones. The noise is twice as much but it's still nothing compared to the rectified garbage of a linear supply and, because of the high switching frequency, it's easy to filter with relatively small parts.

So I'm guessing that in this case a simple two transistor CM (or one mosfet) would be the way to go. Then do as tube circuits do and just use RCRCRC as necessary to make whatever voltages you want.

But again, I'm just thinking out load here. That's definitely something I would spend a lot of time fiddling with in LTSpice. And then probably three iterations of PCBs.

But I definitely like that supply. I could see that being worth it for sure.

 

[thor.zmt]

When I say "my USB interface can do 120 dB", I mean typical specs associated with USB interfaces like dynamic range or EIN. I'm not sure what my old MOTU Traveler mk3 specs are but these days EIN of 127-ish is not uncommon.

Ein for line and microphone input's varies.

Let's keep it simple. My measurements used a calibrated Audio Precision system and tested power supply loaded at a substantial percentage of its rated power. My measurements show absolute levels. At the output of the power supply, capacitively coupled with < 1Hz LF cutoff.

1uV = -120dBV & -140dB re +22dBu for context.

Without knowing reference levels and methodology it is hard to be sure what your measurements are actually showing, while mine are absolute and can be compared to any other measurement taken with relevant reference levels.

Having said that, I'm not suggesting that there's anything "wrong" with your SMPS. I acknowledge that those peaks are very very small. I was just trying to make the point that a lowly MeanWell APC-16-350, which is $10 on Mouser, can be very quiet and that for DIY at least, I don't recommend that people roll their own.

The specific SMPS you mention in the specific way you applied it (you do not state current draw or if there is anything else in the circuit) is quiet in measurements you do not provide sufficient detail to be able to compare them for me.

If you need a LED power supply with 350mA constant current supply in your audio project, your suggestion may be very usable.

I have no specific application for it, sadly.

Thor

 

[thor.zmt]

LED supplies are possibly an unexplored seam of gold for DIY PSUs. I have a Meanwell ELG-15--C500A which will output 300Vdc at up to 500mA. My plan was to use it as the HT supply for a big tube mixer with a chunky linear regulator to drop and regulate the volts down to 275V. Maybe all it needs is the right kind of CLCLC filer on the output?

When using commercial SMPS there are a few "gotcha's".

First, internal EMC measures may couple noise into the system. Especially 2-Wire types often have high coupling capacitance between 1/2 Mains voltage and ground. Alternative 3-Wire supplies can cause earth loops. It is important to recognise the issue and to take action during system integration to mitigate or eliminate these issues without compromising electrical safety.

Second, SMPS often feature "Hickup" protection. This can cause problems during startup of an Audio Device full of big capacitors. It is best to include active filtering (capacitor multiplier?) that is also set up as soft-start.

Third, as discussed, if startup and indirect noise coupling are addresses, we may still experience a lot of feed-trough of switching noise. Using LC + active filtering can eliminate this.

An alternative approach would be to use a "shunt type" noise sink (essentially an AC only Shunt regulator) that uses the DCR & Inductance of the LC filtering as "ballast". They can also provide enough "pre-load" to reliably keep the SMPS out of burst mode which commonly causes significant audio band noise.

I have implemented this type of "de-noise" both inside SMPS (commercial), as stand alone post SMPS filter "adaptor" and inside equipment externally powered by SMPS. Using NJM5534 & external transistor I get at least 40dB low audio band (< 1kHz) noise reduction with a few mA standing current and a few 10's mOhm of inductor DCR. As the frequency rises the inductor impedance rises and noise suppression improves with 20dB/decade up to a few 100kHz.

Below you can see the electronic module that does this denoise in a mass produced PSU (the small green/silver board) that was redesigned from the OEM's standard offering with better filtering and optimised feedback loop, but is otherwise commodity parts based and a very basic quasi-resonant design, very reliable and cheap. PSU is nominal 60VA.





The advantage of this "shunt de-noise' is that losses are minimal (this is nowadays legislated) and the commonly good regulation etc. of the SMPS are retained and also cost is basically "nominal".

Thor

 

[Bo Deadly]

I have implemented this type of "de-noise" both inside SMPS (commercial), as stand alone post SMPS filter "adaptor" and inside equipment externally powered by SMPS. Using NJM5534 & external transistor I get at least 40dB low audio band (< 1kHz) noise reduction with a few mA standing current and a few 10's mOhm of inductor DCR. As the frequency rises the inductor impedance rises and noise suppression improves with 20dB/decade up to a few 100kHz.

That sounds interesting. Do you want to share a schematic?

 

[thor.zmt]

That sounds interesting. Do you want to share a schematic?

My own (given how bitchy it was to get the results and keep a stable circuit) I will keep as "trade secret".

But the basic principles are well documented:

Finesse Voltage Regulator Noise!

While Wenzel focuses on feed forward circuits (hint, it's a good idea in a known setup, better than feedback,but requires basement level design in) the same principles work with feedback based circuitry.

My design started like wenzels final design, but switched to feedback from feedforward to make it universal.

Under ideal conditions feedforward is superior. Adjusted for real world feedback wins as "design, fit, forget" in mass production designs.

Thor

 

[Admiral-dk]

Very interesting approch Thor

I understand that you keep some info for yourself - but any further hint on the series part ?
I mean - do you still only use the series resistor (and the Inductor is elsewhere) or is it a series inductor instead of the resistor .... or even both a series RL before the shunt ?

Best wishes

Per

 

[Matt Syson]

Thanks Thor for that interesting diversion towards another 'solution' to noise however it is fortunately not the direction I am taking for my applications so no conflict of interest.
Ideas to persuade a switcher to come up 'cleanly' to give a 5 volt supply (with remote sensing to compensate for cable loss) which is capable of allowing booting of microcontrollers (several that need around 3 Amps in total because a load of other stuff is also hanging off the 5 Volt rails) is one of my 'goals', where currently a LM338 linear regulator ' just does the job and boots cleanly every time

 

[ccaudle]

do you still only use the series resistor (and the Inductor is elsewhere)

I guess Thor can answer for himself, but I think he already stated it pretty plainly:

a "shunt type" noise sink...that uses the DCR & Inductance of the LC filtering as "ballast"

Any switching power supply will have an inductor on the output, his description of the circuit is that it uses the parasitic resistance and intrinsic inductance of that component as the series element before the shunt device.

 

[Admiral-dk]

Well I must admit that it isn't totally clear to me ....
And I should probably now admit to have constructed a few SMPS systems over the years and at least one of them made it into production ....
This does not by any means make me an expert on the subject - and that is the reason I ask -> to learn. Don't get me wrong - I know about Shunt supplies - but have only seen them in Tube Gear (Measure Equipment, like Scopes etc).
So when I see a smart new way (to me), to use a simple and Elegant circuit ,to solve an annoying problem, with out much Loss / High Efficiency - I ask to clarify ...!!!!
Remember that to assume - is to make an Ass- (out of) U- (and) Me ..... and I've been there too often because I didn't ask

Per

 

[thor.zmt]

Ideas to persuade a switcher to come up 'cleanly' to give a 5 volt supply (with remote sensing to compensate for cable loss) which is capable of allowing booting of microcontrollers (several that need around 3 Amps in total because a load of other stuff is also hanging off the 5 Volt rails) is one of my 'goals', where currently a LM338 linear regulator ' just does the job and boots cleanly every time

I suspect you are running into the protection system. Depending on the precise circuit there may be modifications that can be performed to delay the onset of the foldback or "hic-up" protection, but its non-trivial.

You could add a substantial capacitor-bank with a delayed application of power to the CPU board only once this bank is "full".

Thor

 

[thor.zmt]

Any switching power supply will have an inductor on the output,

Correction, every SMPS should have a second LC filter. It's surprisingly uncommon and mainly because few "SMPS Designers" actually understand how the SMPS feedback loop really works.

Extra LC, especially small and cheap has DCR and affects DC Regulation. Trying to take the feedback after the LC filter causes stability issues if the filter actually does something.

There is a simple solution we know from Audio Amplifiers, namely Cherry's "NDFL" adjusted to operate across a "slow" optocoupler. But that needs more than copy/pasta.

his description of the circuit is that it uses the parasitic resistance and intrinsic inductance of that component as the series element before the shunt device.

Precisely. Wire DCR, trace DCR all also count.

Say we have 10mV audio band noise, mostly 100Hz.

We have 50mOhm DCR in the system (mainly inductor DCR).

How much current must be drawn on the far side of the 50mOhm to make 10mV appear across the 50mOhm resistor?

0.2A.

So we would have a chunky heatsink and 0.3A quiescent current allowing for 0.2A RMS, maybe 0.4A for "Arbitrary Waveform".

Or switching to an AB output (implies capacitive or inductive coupling) may be preferable if we have SO MUCH noise to kill with so little DCR.

If the current frequency, phase and amplitude in our shunt "current sink" is correct the totality of the noise will appear across this resistance and the output will be noiseless.

In reality we never get 100% but 60dB below 10kHz are quite easy without any adjustments and mass production reliable, using feedback.

Wenzel shows Feed forward, however the conversion to feedback should be obvious. Feedback allows very large audio band noise reduction with basically parasitic resistances as ballast.

The 0.3A quiescent current also illustrates the limitations. It is better to start with a fairly low noise PSU if we want to "finesse" noise.

BUT it works without any DC voltage drop (except a few 10mV), to it is the ultimate "LDO".

And it means we can keep Agency compliance, EMC etc. to people who and live and breathe it and often can certify in house even FCC, meaning it costs essentially nothing in China.

With a Constant Current LED Supply applying a DC shunt regulator and decent LC pre filtering could be a really good choice, if we are prepared if needed to burn up the entire CC current.

Again, 5534 plus shunt transistor can work great. I like to hang a PNP or LM337 off pin 5 for that, dumping an extra few mA into pin 5 to wake up the third stage npn transistor a little.

There are also tricks with the 5534 to up the 2nd stage current (and we can add our own degenerated but still lower noise external input differential) that work just as well here as they do in audio applications.

Thor

 

[thor.zmt]

Ideas to persuade a switcher to come up 'cleanly' to give a 5 volt supply (with remote sensing to compensate for cable loss) which is capable of allowing booting of microcontrollers (several that need around 3 Amps in total because a load of other stuff is also hanging off the 5 Volt rails) is one of my 'goals', where currently a LM338 linear regulator ' just does the job and boots cleanly every time

Seems you run into protection and the kind of "runaway" situation where low input voltage causes local switchers on the CPU module to draw extra current, which either triggers a hard shutdown and "hic-up" protection or may trigger the fold back current limiting and the system becomes stable at a very low current limit and never so to speak, "gets it up".

The simple solution is to use a 10A rated switcher to drive the 3A load.

Or a little less, whatever allows a clean start-up. For DIY this is probably the way to go.

Now if it's commercial gig and if you intend to make a few 1,000 or more, where the "use a fat supply" doesn't work, tell your supervisor to put bread on my table and I help out.

Thor

 

[RogerAF]

Ideas to persuade a switcher to come up 'cleanly' to give a 5 volt supply (with remote sensing to compensate for cable loss) which is capable of allowing booting of microcontrollers (several that need around 3 Amps in total because a load of other stuff is also hanging off the 5 Volt rails)

Isn't there a way to hold each microcontroller in Reset until power stabilizes? And/or what about sequencing the power to each device?

 

[Matt Syson]

Thanks for the suggestions but it is a very specific application for a product that was sold in the thousands around 30 years ago and needs to be a simple 'Plug and play' solution for those who are seemingly incapable of just repairing the old, simple linear supplies. Replacing a few dried out electrolytics is SO tough!