Add filtering to SMPS power supply

[Bo Deadly]

Before using it I decided to connect it to my scope to see if there were any noise... To my surprise there's major ripple at 1.5kHz. Close to 600mV.

You have to load it. It's throttling.

If you want to do a proper measurement, load each voltage separately with resistors. But they'll have to be biggish. Like 100R 5W for +-16V and maybe 680R 5W for the 48V. If you only run it for a few seconds, you might get away with paralleling smaller wattage together.

Or two modules might be enough to pull the SMPS IC out of throttling mode.

 

[elskardio]

You have to load it. It's throttling.

If you want to do a proper measurement, load each voltage separately with resistors. But they'll have to be biggish. Like 100R 5W for +-16V and maybe 680R 5W for the 48V. If you only run it for a few seconds, you might get away with paralleling smaller wattage together.

Or two modules might be enough to pull the SMPS IC out of throttling mode.

Thanks squarewave

I will run more test tomorrow with proper loading.

 

[Tubetec]

I wonder what REWs spectrum/fft with 192khz sampling would reveal in your common or garden walwart output , probably spikey as a porcupine .

Would it make sense to add a filter at the psu end ,so the spikes never travel down the cable  , or would  a  common mode inductor be better served up at the far end so it  stands a chance of balancing out  anything picked up in  the cable from external sources as well ?

Ive definately seen the interactions Abbey talks about between different switcher supplies in guitar pedal rigs where nowadays its common to find multiple Smp's  with different pedals ,sometimes with different voltages . Ive noticed quite a high failure rate in stomp boxes operated in this way , and also chances of interference are bigger , typically you get big bunch of 50hz splattered with buzz saw harmonics .  I did take a look at someones pedal tray  one time who had a particularly bad noise problem, was three supplies , one 18volt smp,  reg'd down to 9 volts rails to run Boss style pedals ,then another specialist pedal switcher 20 volts rail , individually each pedal worked cleanly , but by the time you patched all the pedals up and formed small ground loops all over the place S/N had degraded to about 20db , and the guy actually played like that for years ,he still had his sound ,noise was like a foghorn in the backround though .

 

[ruffrecords]

Thanks squarewave

I will run more test tomorrow with proper loading.

To meet current energy efficiency regulations, most SMPS go into a power saving mode under no load conditions so it is anyone's guess what the spectrum looks like at that time.

Cheers

ian

 

[ruffrecords]

I just found this snippet from Rod Elliot's web site on the topic of his dual rail SMPS project:

At light loading (i.e. well below the maximum current allowed for), the UC3845A chip will operate in 'hiccup' (aka 'skip cycle') mode, with a burst of oscillation followed by a period where nothing happens at all.  This can make the switching frequency appear to be much lower than it really is, so increasing the output ripple.  This is just one of the reasons that the output requires additional filtering.  Adding a 'brutal' filter (having very high capacitance) will delay the output from reaching full voltage, especially if current limiting is implemented.

When I first made measurements on the Meanwell ELG-150-C which is a 300V SMPS I only had it lightly loaded and I got ripple that was very difficult to trigger on the scope. It had 2.8V peak to peak ripple at 50Hz with what looked like 400mV spikes of 13.8KHz. With 4 x 33K 5W resistors strapped across it which would draw about 37mA, the 50Hz disappeared and the ripple was about320mV at 28KHz. So clearly these SMPS do need a decent load before they get anywhere near their 'normal' operating regime. The prototype Mark 3 mixer I am building right now currently has 8 modules fitted which should draw about 100mA so I might try connecting that up to the Meanwell and see what the ripple looks like.

Cheers

Ian

 

[elskardio]

I just found this snippet from Rod Elliot's web site on the topic of his dual rail SMPS project:

Thanks Ian for the good info!

I tested the power supply with a 100R load on the V+ rail and I get the same result... I will test it this week running 2-3 modules in a rack to see how it looks on the scope.

 

[dogears]

I wanted to resurrect this thread with a question.

With a feedback SMPS like a Meanwell IRM series, is there any risk to using an LC filter with as low a frequency as possible? Is it possible to adversely affect the feedback regulation of the PSU itself?

I found some designs that appear to use an LC filter after these with a corner frequency of around 7kHz. With similar components you could get down to ~300 Hz. Should you go as low as possible?

 

[Bo Deadly]

Not sure about the LC filter causing problems (that's not in the "feedback" path anyway) but I don't think it's necessary to us an LC filter that low. If the SMPS goes into "hiccup" mode then you could get LF noise. But I personally would never be satisfied with using an SMPS in hiccup mode in the first place. Otherwise, all of the noise should be HF which a moderate sized LC is great for (100uH would be fine). If you must have the next level of quiet, add a capacitance multiplier. Or, if you need to get to a specific voltage and you're ok with more than a diode drop, use proper regulators (LM317 / LM337).

 

[dogears]

Thanks. The only reason I said "that low" is because the inductor series goes from 0.47 uH to 220 uH at the acceptable current limit, with DCR ranging from .002 ohms to 0.37 ohms in the same range. With a 1000uF cap that's 7kHz to 300 ohms resonant frequency. The unit switches at 65kHz.

As for "low enough", it's a first order filter that range covers -21 dB to -74 dB at 65kHz. Kinda not sure where to draw the line.

100 uH with a 1000 uF cap is -68 dB at 65kHz. Seems reasonable to me... but is there any reason to not go lower?

 

[Bo Deadly]

You really don't need to filter that low. SMPS don't like a lot of capacitance on the output anyways. Sometimes the datasheet lists a limit. 100uF would probably be fine.

 

[abbey road d enfer]

100 uH with a 1000 uF cap is -68 dB at 65kHz. Seems reasonable to me... but is there any reason to not go lower?

I don't know if it's reasonable, I don't know if going lower has drawbacks, however that's exactly what I did on my 51X lunchbox, 100uH and 2x470uF. It works for about 3 years now... I didn't spend much time experimenting.

 

[john12ax7]

It might be a case of read the datasheet and then try it and see.  Max capacitance spec will limit C.  Also some SMPS will want to see a minimum current at turn on which could limit L.

 

[dogears]

Unfortunately datasheet is sparse, no mention of max capacitance or minimum current at turn-on. I built it with a CRC, 47uF 0.47uH 1000uF and it's fine. Very, very low ripple on the scope.

I'm an (not-electrical) engineer though and I don't work like this in my day job.. try it and see isn't really the usual approach. So I assume there's some analytical optimization method.

I read somewhere or other a good starting point is to go to 1/10th the switching frequency as your -3dB. For this unit, that would be 6.5kHz. There are lots of ways to skin that cat with inductors and cap combinations.  Would it be worthwhile to try to "design" a CLC filter for a nice shape based on the impedance? Using something like this?
http://www.wa4dsy.net/filter/hp_lp_filter.html


I gather that the inductor size along with the current draw affects ripple current which combined with the ESR of your capacitor changes the ripple voltage. Is it just a balancing act between response time and lowering ripple?

 

[dogears]

Well, for those following along at home I found a really nice article about it.

http://www.interpoint.com/product_documents/DC_DC_Converters_Output_Noise.pdf

An excerpt:

An external LC filter can affect load regulation. Using the smallest L value possible and keeping the inductor ESR small will help. The LC filter also acts as a series resonant circuit across the power converter output terminals and can affect the stability of the converter if the filter is under damped and the filter’s resonant frequency is inside or close to the control loop bandwidth. Setting the filter’s resonant frequency above 30 kHz should suffice for any Interpoint power converter. If the filter’s resonant frequency is within or close to the converter’s bandwidth, keeping the Q below 2 should be adequate for most Interpoint converters.

 

[Bo Deadly]

I believe what that is mostly saying is that if you use a larger inductor and smaller capacitor your LC filter will be more of a resonant circuit and you could get a huge peak in the filter response which of course would be not good.

But if you use 100uH or even 1m you should be ok if you don't use a strangely small cap. There is a combination of inductor and capacitor that will yield a Q factor of 1 (considered "critically dampened"?). For 100uH, I think that would be 100uF to get critical damping. Meaning it will be perfectly flat until you get to the corner frequency and then it's a perfect 12dB/octave slope down. I'm not so good at the math end but there are online calculators for that sort of thing (I just model everything in LTSpice first).

Not sure why higher ESR would be a problem. I would think that would help with damping.

But again, I would not get too carried away with this. The ripple at the switching frequency is like 200mVpp. So with the 100uH / 100uF example again, that gives you ~60dB of attenuation at the switching frequency which means your 200mVpp becomes 0.2mVpp. I think that's good enough!

 

[dogears]

We do this to learn, no? Otherwise I could just buy something that already works. I know I'm nuking it, but I want to actually understand best practices.

I think one potential issue is that if the control loop crossover is above/around/at the resonant frequency of your filter - AND your LC filter is underdamped - you could introduce instability. And possibly instability that is load dependent.

A 100uH inductor with a DC resistance of 0.16 ohms and a 100 uF cap with no load has a Q of 6.1 and a 13 dB peak at the corner frequency (1591). This guide suggests keeping the Q down to below 2, which would be 100uH with a 1000uF cap - frequency of 503 Hz and Q of 1.9. Abbey is spot-on, as usual.

But that is without a source or load impedance or considering the ESR of the cap, which seems to potentially help with a lot of these problems.

I suppose the application matters as well - if you have something that can draw a LOT of energy rapidly, like a power amp, a very large inductor may limit the transient response more than caps can respond (though this seems unlikely to me).

It's also worth noting that downstream filtering may make a lot of this moot. A later RC or diode-capacitor filter will eat up quite a bit of remaining ripple.

 

[Ale Dalto]

As john12ax7 points out, SMPS actually may not like caps directly on the output. It's not like you're filtering rectified AC. You're trying to get HF hash. And since the inductor will effectively be isolating the SMPS from the cap that follows, I would use a large inductor and small cap instead of the other way around like you're schem shows. I would use the largest inductor you can find with the right current requirements and then just pick a cap that gets down below 20Khz. For example, you should be able to get at least 1mH which with a 10uF cap gets you down to 1.5kHz. You might also use a common mode choke but I have heard people say it doesn't matter. Personally I always liked the idea of using a common mode choke to completely isolate both outputs of the SMPS and then use the point right after the choke as 0V and star ground point. Same applies if you're stacking two to make bipolar. You'll just need two chokes. Of course SMPS must be isolated type.

What do you think of placing a 0.47uf on the output of smps? Can cause problems?
In order to get a clc output filter like this

 

[abbey road d enfer]

What do you think of placing a 0.47uf on the output of smps?

I don't think it would change anything, since there's already a capacitor at the output in the smps.

 

[Ale Dalto]

I don't think it would change anything, since there's already a capacitor at the output in the smps.

Dumb a question.
This smps has a single or dual voltage output? I read single output but in the pin out I see +Vo and -Vo.
I need a +/- 5v power supply on my board.

 

[abbey road d enfer]

Dumb a question.
This smps has a single or dual voltage output? I read single output but in the pin out I see +Vo and -Vo.

-Vo is teh negative pole, +Vo is the positive pole of a single source.
A dual supply may use two single sources coupled in series by connecting teh positive pole of one to the negative pole of the other. I hope it's not too confusing.

I need a +/- 5v power supply on my board.

Can you be more specific? Do you really need dual rails?