Switcher Noise Cleanup Filter

[gnuarm]

A control board with sensitive analog electronics is being fed by a switcher on another board at 7.5V.  The switcher is fed from a ~12V source that also supplies a brushed DC motor.  On the control board the 7.5V supply is regulated down with linear regulators to the various voltages required on this board.  The total draw on the 7.5V supply is around 300 mA, most of that the LED back light for a pair of LCD displays.  The digital logic on the control board is an MCU and a small FPGA.

When the 7.5V supply comes onto the controller board it will be filtered through a pi CLC filter.  I'm looking to pick an inductor to filter the noise from the switcher as well as the noise from the motor.  I know wound inductors have capacitance between the turns of the coil which acts to bypass the coil.  I'm not sure how important this is in a pi filter.  Even if there is some small equivalent capacitance, it would act to limit the floor of the filter, but since the capacitance to ground on either side of the inductor is such a smaller impedance, I would expect this to still filter well. 

Any recommendations?

I also need to filter the 12V power source, but at a much lower current, <50 mA.  There are a handful of logic and analog signals that need to be filtered as well. 

[Vovk_Z]

I usually start from an inductor with such a rated inductance so it has several ohm impedance at the switcher main frequency. This rule gives quite effective filters. Usually it is 1..10 uH inductor for an 'average' switcher.

[penfold]

The inter-winding capacitance of the inductor probably only becomes significant for the >1MHz harmonics where it can form a high-pass filter with the ESL of the caps, probably only noticeable if any resonances crop up, otherwise as you say it'll just give you a "floor" to the filter. Both L and C parasitics should be derivable from datasheet specs to make a vague prediction about the freq and Q, but of course highly dependent on layout.

Lossy ferrite helps keep "unexpected" resonances damped as does a little "real" resistance (if you have some overhead to tolerate voltage drop).
4 terminal SMT feedthrough caps have very nice HF performance so avoid a lot of the ESL issues to begin. But all that depends whether or not your circuit is actually sensitive to the HF stuff.

There are some nice "all in one" power supply filters with the L's and C's combined, but usually geared toward the >10MHz stuff, I guess they're more about complying to standards than targeting specific problems... again depends on your frequencies.

[jonpaul]

Bonjour  gnuarm

At such low powers EMI is not a big issue.

Modern switchers use soft switching and high frequencies, thus the conducted noise is easily adress with a bypass capacitor at the lod, eg 10 uF plus 10 nF.

Commercially available filters have the CM and DM chokes and capacitors combined, see Murata, Panasonic, CoilCraft, etc.

Finally a DC side filter must be rated for theDC as the core will suffer a bias.

Kind Regards,

Jon

[gnuarm]

The inter-winding capacitance of the inductor probably only becomes significant for the >1MHz harmonics where it can form a high-pass filter with the ESL of the caps, probably only noticeable if any resonances crop up, otherwise as you say it'll just give you a "floor" to the filter. Both L and C parasitics should be derivable from datasheet specs to make a vague prediction about the freq and Q, but of course highly dependent on layout.

Yes, layout is important.  We have a separate power supply board that is providing a switcher regulated voltage, 7.5V and the linear regs are local to the control board.  The control board has a number of ADC that sample slowly, but some signals come in at rather low levels.  So I want to over do the noise mitigation rather than have to figure out where the noise is coming from later.  The inductor data sheet measures parameters at 2.5 MHz, Q of 22, SRF 12 MHz.  The switching circuit is nominally 0.5 MHz.  Bourns SRU1048-220Y

Lossy ferrite helps keep "unexpected" resonances damped as does a little "real" resistance (if you have some overhead to tolerate voltage drop).
4 terminal SMT feedthrough caps have very nice HF performance so avoid a lot of the ESL issues to begin. But all that depends whether or not your circuit is actually sensitive to the HF stuff.

I'm not certain the HF noise will be important.  We also are pumping amps into a brushed DC motor with PWM around 20 kHz.  It has honkin' caps on the power rail and with the higher switching frequency we only see around 1 amp of current change during the PWM cycle.  But it is still going to make a bunch of EMI. 

There are some nice "all in one" power supply filters with the L's and C's combined, but usually geared toward the >10MHz stuff, I guess they're more about complying to standards than targeting specific problems... again depends on your frequencies.

I simed a filter with a ferrite and the 22 uH inductor with several caps.  It has resonances but nothing that is a problem.  They largely went away when I added a bit of series resistance in the AC source along with a bit of capacitance since it's coming to the board over a cable.  The guy designing the switcher seems to have used less input capacitance than the data sheet suggests... STRONGLY!  I'll have to say something to him about that. 

If you are interested here is my simulation file.  The power board uses the 22 uH coil in the switcher, so I started with that.  47 uF is about as high as reasonable in a ceramic cap without the price going up and being harder to get.  I could go with a higher inductance coil easily enough.  The DC current will be under 500 mA.  This family of coils offers up to 220 uH handling that current. 

 Pwr_Filter.asc (2.29 kB - downloaded 32 times.)

[penfold]

I'd image it would be pretty descent then all that considered.

My only thoughts (not suggesting anything's important so much, just pointing out my thought process), 5MHz is the 5th harmonic of 500k so probably well attenuated by a well designed switcher's filter, whereas 10MHz is probably more likely to creep through even a well designed one, which is where I would suspect any resonances between trace inductance, cap ESL and inductor parasitic C in your filter... ferrite bead kills those before they get past the main gate so OK. However... the output impedance of the filter will have peaks in Z at the same frequencies as the gain will around the 10MHz mark.. so not a problem for LF analogue, but that digital stuff may be a source of occasional large noise spikes should those currents get past local decoupling... so its a good case for a 10nF or two at the filter output perhaps.. I'm not sure TBH.

[nctnico]

A control board with sensitive analog electronics is being fed by a switcher on another board at 7.5V.  The switcher is fed from a ~12V source that also supplies a brushed DC motor.  On the control board the 7.5V supply is regulated down with linear regulators to the various voltages required on this board.  The total draw on the 7.5V supply is around 300 mA, most of that the LED back light for a pair of LCD displays.  The digital logic on the control board is an MCU and a small FPGA.

Any recommendations?

How about using an RC filter in front of the regulators that need to supply a sensitive circuit? Usually I start with 10 Ohm / 10uf but size according to current requirements and voltage drop. The added bonus is that less heat is developed in the regulators. LC filters are nice from hundreds of kHz where an RC filter can be dimensioned to suppress much lower frequencies.

[T3sl4co1l]

Elephant in the room is common mode noise, coming in across all those wires between boards.  Size the filter for the known differential noise, leave options open to increase it, or add another LC stage, or add CMCs.

Tim

[KT88]

Just one quick tip: Use manufactures models for the caps of the filter(s). The tiny (parasitic) inductancies often make a big difference...

Cheers

Andreas

[gnuarm]

A control board with sensitive analog electronics is being fed by a switcher on another board at 7.5V.  The switcher is fed from a ~12V source that also supplies a brushed DC motor.  On the control board the 7.5V supply is regulated down with linear regulators to the various voltages required on this board.  The total draw on the 7.5V supply is around 300 mA, most of that the LED back light for a pair of LCD displays.  The digital logic on the control board is an MCU and a small FPGA.

Any recommendations?

How about using an RC filter in front of the regulators that need to supply a sensitive circuit? Usually I start with 10 Ohm / 10uf but size according to current requirements and voltage drop. The added bonus is that less heat is developed in the regulators. LC filters are nice from hundreds of kHz where an RC filter can be dimensioned to suppress much lower frequencies.

I would have been happy with that, but there are some devices that use an unfair share of the power and the guy designing the power board wants tons of head room for the linear supplies.   "Dead weight" includes over 200 mA of current in the LCD display back lights.  Do LEDs really need highly regulated power?  I don't think so.  Likewise the speaker can suck up to 150 mA and you won't hear any noise above 20 kHz.  So the DC has to be sized for this while the noise sensitive circuits don't use a total of 50 mA.  I suppose I could provide separate filtering for the sensitive parts.  Still, 50 mA limits the resistor value to maybe 10 ohms dropping 0.5V.  The choke is 69 ohms at the switcher frequency and nearly 3 ohms at the PWM frequency, so nearly as good at the PWM rate and much better at the switcher.  If needed, the choke can be bumped up 10x. 

I guess I could ask the power board guy to run the intermediate voltage up a bit. 

[flowib]

AN 101 from LT by Jim Williams may be a worthwhile read.  In short you can try some 1206 ferrite beads, and a couple of 0805 ceramics to get rid of the Megahurtz range harmonics.