Exactly right. This is all behaving just like any other passive filter circuit. The attenuation difference at 65kHz between the 100uF and 1000uF in the plot is the difference between their ESR. The only way to get more attenuation is a larger-value CMC.
What if I want to use
And if you want to use the SMPS to power up a digital audio playback device, like a streamer that wants to see a uniformly low output impedance into 100ths of MHz range AND low ground noise levels? Even if a DAC has a galvanically isolated USB module, the sound benefits from paying strict attention to ground noise are extremely important.
And if you want to use the SMPS to power up a digital audio playback device, like a streamer that wants to see a uniformly low output impedance into 100ths of MHz range AND low ground noise levels? Even if a DAC has a galvanically isolated USB module, the sound benefits from paying strict attention to ground noise are extremely important.
I don't want to make this thread about MJ's filter, but for interest's sake, here's how it models on the computer. The math assumes a zero-Ohm source and perfect ground. Neither of which exists, but only the latter matters much in this case.
Here's the freq response / attenuation curve. Peaking prior to cutoff is minor, inconsequential. A well-behaved curve.
And here's the impedance and impedance phase, again assuming a zero-Ohm source and perfect ground. Yellow is Z, blue is Zphase.
You can decide if you like this impedance curve supplying power to your amplifier. I'd prefer something flatter, more linear.
If it weren't for the switching currents polluting the circuit ground, this would be a very useful noise filter.
Here's the freq response / attenuation curve. Peaking prior to cutoff is minor, inconsequential. A well-behaved curve.
And here's the impedance and impedance phase, again assuming a zero-Ohm source and perfect ground. Yellow is Z, blue is Zphase.
You can decide if you like this impedance curve supplying power to your amplifier. I'd prefer something flatter, more linear.
If it weren't for the switching currents polluting the circuit ground, this would be a very useful noise filter.
Last edited:
There's nothing preventing Mark to provide real-life measurements on his thread; ones that will include the hook-up wiring & mounting arrangements -> required to place his filter in a real-life application. Then, we can talk about how effective or not this filter really is.
Don't forget it's only rated for 3 amperes maximum, and is applied to wall-wart SMPS's that are powering line level audio gear. Preamps, DACs, Korg NuTube preamps, "H2" effects boxes, headphone amps, etc. These wimpy-current-consumption pieces of gear use SMPS's which operate at switching frequencies between 50 kHz and 300 kHz, depending on how much/little you spend. So I recommend extending the simulation window to 10 MHz instead of 0.1 MHz because that's the environment they are expected to operate in.
i do not care about measurement, i listen to the tweeters close range, if the smps supply is no good i will hear it...
otherwise, i will just listen to the music and enjow my smps fed amps...
otherwise, i will just listen to the music and enjow my smps fed amps...
It doesn't show anything of consequence, since the source's characteristics, including an unknown amount of wire from the power brick, are not part of the model. I added reasonable ESR for the L's and C's but nothing beyond that. (You can tell it's 25mΩ for the C's because the curve levels out there...) Most devices of the type you mentioned will have a cap across the power input.
There are plenty of situations that will draw consequential current. Mine is a Class A headphone amp, drawing 1 Amp.
Last edited:
The inductors are self-resonant and, in my opinion, it's not a waste of time to attempt to model that in simulation. Which is why the datasheet S.R.F. value is copied right on the schematic diagram itself.
The electrolytic capacitors are self-resonant too, and again I think it's not a waste of time to attempt to model that in simulation. More difficult, but IMO still worthwhile.
The electrolytic capacitors are self-resonant too, and again I think it's not a waste of time to attempt to model that in simulation. More difficult, but IMO still worthwhile.
1. I used the values in your circuit diagram except for adding the ESR, which wasn't there. No ESL is there, either.
2. The self-resonances you decribe are piddle compared to the many unknowns that would totally swamp them. How long are the leads to the load? What filtering if any is there? Not to mention the switching ground currents passing through unaltered.
There's no way to model that and have it be universally applicable or meaningful.
If you have an impedance analyzer that measures out to several MHz, sweep your 470uF cap and see. They're not crazy resonances. They're broad and lossy. The ones I've measured have been.
Back to the CMC approach. Here's another POV on the CMC filter. Say we want to get the switching peak, currently sitting at -50dBV, down to -100dBV. Our filter will use a large-ish capacitor with 25mΩ ESR. In theory, the CMC would have to have an impedance of 7.9Ω at 65kHz to give us -50dB attenuation. That's about 20uH.
That would give us a "16-bit" power supply and ground.
2. The self-resonances you decribe are piddle compared to the many unknowns that would totally swamp them. How long are the leads to the load? What filtering if any is there? Not to mention the switching ground currents passing through unaltered.
There's no way to model that and have it be universally applicable or meaningful.
If you have an impedance analyzer that measures out to several MHz, sweep your 470uF cap and see. They're not crazy resonances. They're broad and lossy. The ones I've measured have been.
Back to the CMC approach. Here's another POV on the CMC filter. Say we want to get the switching peak, currently sitting at -50dBV, down to -100dBV. Our filter will use a large-ish capacitor with 25mΩ ESR. In theory, the CMC would have to have an impedance of 7.9Ω at 65kHz to give us -50dB attenuation. That's about 20uH.
That would give us a "16-bit" power supply and ground.
Some reputable brands like Meanwell publish test reports on their website. These show measurement of AC input conducted noise. Look out for a Class B level pass.
Long ago I had problems with some TDK Lamda "professional industrial" 12V smps generating high levels of common mode noise on the outputs. The noise 12V to 0V was tiny, but the 0V was very noisy to the external Gnd
Ok, I managed to find a few compliance certificates.... attached.
The EMC one requires that you download the relevant IEC standard to figure out what the compliance regulations and requirements are. Many are (standards) downloadable for a fee.
The EC (level B compliance) again, states the relevant standards but does provide some figures. I looked at those (figures) in disbelief - they look too high to me... scary stuff.
... but then again, I did have to provide a separate power cord, plugged into a separate mains outlet, for the SMPSs I played with.... because when plugged into the same mains outlet as my amp and DAC, the sound deteriorated drastically... easily noticeable by me and my kids.
The EMC one requires that you download the relevant IEC standard to figure out what the compliance regulations and requirements are. Many are (standards) downloadable for a fee.
The EC (level B compliance) again, states the relevant standards but does provide some figures. I looked at those (figures) in disbelief - they look too high to me... scary stuff.
... but then again, I did have to provide a separate power cord, plugged into a separate mains outlet, for the SMPSs I played with.... because when plugged into the same mains outlet as my amp and DAC, the sound deteriorated drastically... easily noticeable by me and my kids.
To illustrate my point that the resonances of large-ish electrolytic capacitors are not crazy, spikey things but pretty tame, here's an impedance measurement from 10kHz to 10MHz of three caps: 100nF 250V film, 1uF 63V film, and 1,000uF 50V 'lytic. The latter was an Elna 1000uF/50V with about 31mΩ ESR.
You can see that the film caps have distinct resonances, but the 'lytic is smooth and broad. ESR dominates from 50kHz to 200kHz and then the ESL takes over. It is about 5nH, which is miniscule and will be totally swamped by everything that follows it, especially in the wiring to the load.
Also, this is an older 'lytic from the 90's, larger than its modern equivalents. A modern 470uF 50V cap would be about 1/4 its size, resulting in proportionally less ESL and the ESR region being broader, extending into higher frequencies.
Page 26 of this report tells you the limits on the AC power port
https://www.vecow.com/dispUploadBox/PJ-VECOW/Files/7849.pdf
Remember that 60dBuV is just 1 mV and that is for the quasi-peak limit 5 MHz to 30 MHz.
Some smps when unloaded go into a very dirty pulse skipping mode, They like to be loaded at least 25%. This is a problem for a smps powering a Class AB amplifier
https://www.vecow.com/dispUploadBox/PJ-VECOW/Files/7849.pdf
Remember that 60dBuV is just 1 mV and that is for the quasi-peak limit 5 MHz to 30 MHz.
Some smps when unloaded go into a very dirty pulse skipping mode, They like to be loaded at least 25%. This is a problem for a smps powering a Class AB amplifier
Electrolytic capacitor ESL does begin to degrade filter performance in the MHz region. See middle and bottom traces in #74 figure, top. That's one reason why simulating out to 10 MHz is occasionally illuminating.
This is the noise spectrum from the raw AC mains. How is this line input affecting the performance of the SMPS? What is the output of SMPS using a clean sine wave as input?
Put a couple of motor--->flywheel--->generators in series between that mains AC waveform and your audio gear. Problem solved!
Continuing on the CMC train, here's the RPS-30-15 with a 60uH CMC followed by 1,000uF. Setup is identical to that used in prior measurements.
Unfortunately the theory is not borne out. A larger CMC does not necessarily result in increased attenuation at high freqs. The 65kHz switching peak is 10dB HIGHER than it is with the 3.1uH CMC.
Another one bites the dust.
(Edit: The graph label says 50uH but that was a typo - it's 60uH)
Unfortunately the theory is not borne out. A larger CMC does not necessarily result in increased attenuation at high freqs. The 65kHz switching peak is 10dB HIGHER than it is with the 3.1uH CMC.
Another one bites the dust.
(Edit: The graph label says 50uH but that was a typo - it's 60uH)