Switching PSU filtering

[jasonallenh]

The commonly used Meanwell supplies already contain EMI filters at their inputs, according to the datasheets. Is there any point in installing additional filter caps (or even inductors) at the IEC inlets? I've got more than one device utilizing several of these little caged Meanwell switchers...

I cannot imagine caring about anything but the output of the switcher. Whatever is in my walls is going to likely be different than someone else's walls anyway

 

[thor.zmt]

The commonly used Meanwell supplies already contain EMI filters at their inputs, according to the datasheets.

Most SMPS that actually pass EMC do. However that filter is there to pass specific EMC legistlation ONLY. It is important to understand what that means.

EMC is commonly tested 150kHz - 30MHz and 30MHz up to whatever limit is set.

Leakage of the main switching frequency is not covered and only harmonics are filtered. Lower frequencies allow higher levels of EMI.



Multiple SMPS on the same mains circuit can create beat tones in the audible range.

Is there any point in installing additional filter caps (or even inductors) at the IEC inlets?

Interesting question. Answer is "it depends". On the precise design of switcher and filter.

In many cases using a "balanced" power transformer can reduce leakage of common mode noise into the output.

Adding capacitors both X & Y is usually constrained by both regulations and the need of keeping ELCB's from tripping and has limited benefit.

Adding a much larger value (common mode) inductor to lower switching noise and harmonics and making sure "Y" capacitors to earth are fitted on the "far side" of this inductor is valid.

Using a mains balancing transformer with high leakage inductance (say EI or LL wound with separate chambers for primary and secondary winding on a split bobbin can also act as common mode inductor.

Of course, it seems counter intuitive to add a big hunk of iron and copper transformer to a small SMPS.

Thor

 

[Matt Syson]

Another minor excitement with mulmtiple switchers is the possibility: liklihood of serious inrush currents which can be around 40 Amps per (moderate size) module so burning mains switches becomes a real possibility at switch on.
When you are talking RF many subtle small differences can lead to problems as your mans cable becomes an aerial (transmitter) and when neatly cabled in a rack with audio gear can be fun. See also my frequent 'rants' about star earthing which was and is 'good' if you are only considering powerline frequency problems but a few yards/Metres of cabling becomes a handy aerial so what may be a 'solid' ground at 50/60 Hz cqn be effectively open circuit if you are unlucky. GOOD design works on understood and managed principles NOT luck!

 

[Cqwet Dbdfte]

Shielding, grounding, filtering, and bypassing becomes more of an issue with SMPS'.
Shielded power cables are standard and low cost from Amazon etc. Gotham sells bulk double shielded power cables, if you want to roll your own.
Right, emission levels are regulatory, maybe not enough for a very low level audio environment.
Some SMPS' have freq sync input where your clock can be introduced, possibly to add a spread spectrum clock. There are a few low cost chips that will do this. This would spread the "joy" a few %.

The regulatory tests have very specific setups, not necessarily equal to every use case. If you are designing equipment to use SMPS more time and effort is needed, than 50/60Hz supplies. Getting rid of transformer fields may suggest using shielded toroidal transformers. Transformer to transformer coupling in tube amps is an issue. Shielding strong magnetic 60Hz fields in neither cheap nor easy. Some resort to remoting the mess to another enclosure. Distance helps.
Consider how good is acceptable. -174dBm/Hz is not always needed.
Sometimes low level random noise is added to A/D inputs to mask monotonic noise. Low levels may be OK.

 

[thor.zmt]

Shielding, grounding, filtering, and bypassing becomes more of an issue with SMPS'.
Shielded power cables are standard and low cost from Amazon etc. Gotham sells bulk double shielded power cables, if you want to roll your own.

There are some power cables from Eupen with build in LC filters and ferrite powder mixed into the filler of the cable that surrounds the wires.

Right, emission levels are regulatory, maybe not enough for a very low level audio environment.

A few 100mV peak-peak at 67khz or 132khz can cause headaches.

Some SMPS' have freq sync input where your clock can be introduced, possibly to add a spread spectrum clock. There are a few low cost chips that will do this. This would spread the "joy" a few %.

Off- mains (AC/DC) converting supplies come in a few subtypes, but most are variable frequency, with frequency falling as power goes up (same as self oscillating Class D Amplifiers). So synchronisation is usually not feasible.

The regulatory tests have very specific setups, not necessarily equal to every use case. If you are designing equipment to use SMPS more time and effort is needed, than 50/60Hz supplies.

Yup, just because a SMPS is certified, doesn't make it "good for audio". It is not too hard to make an SMPS very quiet:

Getting rid of transformer fields may suggest using shielded toroidal transformers. Transformer to transformer coupling in tube amps is an issue. Shielding strong magnetic 60Hz fields in neither cheap nor easy.

I had to shield an MC input stage against mains transformer noise. There was already a mild steel case for the transformers and around 20cm distance. Adding layers of mu-metal and copper inside a mild steel case got it down to below -146dBu.

Thor

 

[Matt Syson]

The main problem is people expecting the combination of CHEAP and EASY which simply doesn't happen. Identifying and correctly managing the various problems is vital.

 

[thor.zmt]

The main problem is people expecting the combination of CHEAP and EASY which simply doesn't happen.

Cheap, Easy and High Performance! And fun fact, it is possible.

Identifying and correctly managing the various problems is vital.

Yes, but they are complex and often counterintuitive.

Thor

 

[living sounds]

I've had another look at my Oberheim synth. It had a custom switchmode PSU installed, which consists of 3 Meanwell PSUs, all followed by LC filters. The +/- 15V DC and -5V DC get additional treatment by quite a lot of active eletronics and this results in very stable output voltages. However, the +5V is lacking that, and as a consequence the voltage sags when keys are pressed (music is played) and LEDs are switched on, and there is more AC voltage present on the power line than I would consider acceptable. There is some noise present in the output that might be caused by the PSU.

Now I'm thinking about replacing that +5V SMPS with a +7.5 SMPS, keep the filter, maybe add some capacitance and then follow it with a (sufficiently sized) 2A +5V regulator. Any reason not to do it this way?


The synth originally had nothing but a diode bridge following the transformer, a big filtering cap and a linear regulator followed by some capacitance.

 

[thor.zmt]

I've had another look at my Oberheim synth. It had a custom switchmode PSU installed, which consists of 3 Meanwell PSUs, all followed by LC filters. The +/- 15V DC and -5V DC get additional treatment by quite a lot of active eletronics and this results in very stable output voltages. However, the +5V is lacking that, and as a consequence the voltage sags when keys are pressed (music is played) and LEDs are switched on, and there is more AC voltage present on the power line than I would consider acceptable. There is some noise present in the output that might be caused by the PSU.

Now I'm thinking about replacing that +5V SMPS with a +7.5 SMPS, keep the filter, maybe add some capacitance and then follow it with a (sufficiently sized) 2A +5V regulator. Any reason not to do it this way?


The synth originally had nothing but a diode bridge following the transformer, a big filtering cap and a linear regulator followed by some capacitance.

With so many voltages, I would make a "universal DC input" Board with sufficient power, use a buck converter running at > 1MHz to create a "clean"(ish) low voltage DC bus at a fixed voltage and then use individual DC-DC converters again running at > 1MHz to make all the required voltages.

A modern DC-DC chip has roughly the same audio band noise as an LM317/337, at > 1MHz switching frequency a 2nd order LPF with 1KHz turnover will theoretically kill the > 1 MHz switching ripple by 120dB.

Chips which can (say) work with a 9V bus and can at 9V input handle +/-18V tracking at 0.75A output less than a square inch are very common. To make +5V and -5V at (say) > 4A again takes pretty widely available IC's.

To make 9V at high current we can use again commodity IC's. A commodity 12V/10A "Brick" is pretty cheap, now we have up to > 100VA various DC.

All these small IC's can be externally synchronised, so if there is (say) a 12MHz Oscillator, we can synchronise all switchers to 1.2MHz.

Thor

 

[Cqwet Dbdfte]

https://www.analog.com/en/products/lt8625s.html
Looks like a good part, availability scarce.
Similar parts available.

 

[Matt Syson]

I like Thor's solution of a 'master' supply running into a bunch of other converters as necessary BUT, the big BUT, they have to be effectively filtered and screened to a good standard afterwards. Yes I have measured the audio outputs of (expensive) audio gear (€35,000) which uses a master supply at 12 Volts then regulated down and up to get other rails and granted the noise when measured 20 - 20KHz with standardaudio industry measurement (Audio Precision) does show a reasonable spec BUT open it up to 300KHz and actually LISTEN to the noise it is riddled woth little birdy noises and 40dB MORE noise. Maybe decent screening and more filtering woulmd help but it was a commectial product so I am not about to investigate myself.

 

[thor.zmt]

I like Thor's solution of a 'master' supply running into a bunch of other converters as necessary BUT, the big BUT, they have to be effectively filtered and screened to a good standard afterwards.

Not so. They need to work at a sufficiently high frequency AND synchronised.

These days 1MHz+ is pretty much standard for most up to date parts. Beat-notes result from non-synchronised switchers.

Another point is to filter the noise and place the "noisy" supply onto it's own copper island. Non of the switching frequency should be allowed to escape.

This needs to balanced somewhat with the need to drain EMI into earth/mains. Ideally we have a mains earth on the chassis, in which case we can "soft earthing" to drain EMI from both the audio circuit and the switchers into earth.

In that case having chokes in the negative input line and in the supply grounds is possible, so our "noisy island" is referenced to earth but otherwise isolated for RF.

In cases without earth, it usually needs a ferrite bead in the negative supply line (which is referenced either to mains earthy or coupled to mains as common mode) and a ferrite bead in one (or all) ground connection to the audio ground, to allow a path for RF for EMC testing.

open it up to 300KHz and actually LISTEN to the noise it is riddled woth little birdy noises and 40dB MORE noise.

This can have reasons other than leakage from power supplies, one example, noise shaping in AD or DA Conversion or in Class D Amplifiers. Many DS modulators when running "idle" tends to create birdies (low level idle tones) as well, they go away with sufficient signal, like a bit of pink noise dither a few dB above the modulator noise floor.

Also, self oscillating Class D Amplifiers (which are pretty much the majority) must be not only synchronised to avoid beat-notes, but due to their relatively low switching frequency, classic LC filters with (say) 40kHz -3dB point are woefully inadequate to suppress a switching frequency at (say) 400kHz which further dynamically drops under load, as they only manage 40dB/decade suppression.

So if we have a Class Amplifier with 50V rails and 400kHz switching frequency and a 40kHz output filter, we will see around 500mV @ 400kHz at the output.

Further, at the top of a sinewave test signal the switching frequency will drop by 25...50%, so with a 50% we actually have 200kHz, so our RF level at the output varies dynamically from 500mV at 400kHz to 2,000mV at 200kHz WITH THE MUSIC SIGNAL.

It will pass EMC testing, but what will happen to the sound? Anyone know the key mechanism for electrical distortion in transducers? Eddy current distortion. And it rises with both signal and frequency in a cubic function.

Obviously a correct output filter for a Class D Amplifier should suppress the switching frequency (or carrier as I like to call it, betraying time spent in RF) by 100dB+ under all conditions, for a "High Quality" system.

Thor

 

[Matt Syson]

Hi Thor, I think we are looking at similar but slightly different aspects to largely gert to thye same vconclusions. It can be done but it needs carefull thought and is probably expensive if it is not planned at an early stage. I have spent 45 years stopping clicky noises (and digital control etc) getting into audio mixers and gahave even had to rip copper 'ground plane' traces off some boards because it had digital ground on one face and audio summing ground on the reverse of the board, which was not appreciated early on because the digital parts had not been completed when the gear was built.

 

[ccaudle]

digital ground on one face and audio summing ground on the reverse of the board

What was the result of that? Capacitive coupling causing common mode noise in the audio circuit?