Turns out my pi-filter simulation from the previous post does not tell the
full story, as with a power filter it's best to first look at the step
response before looking at the frequency response. With that said, here are
the updated simulations which now include two decoupling capacitors at the
load (both ceramic, one 100nF and other 10μF)
Chosen Pi-Filter Configuration
Here is the step and frequency response of the chosen pi-filter
configuration. If you want to know how I came to this conclusion then see
the following sections
Comparing no filter to pi-filter, note how there is no drastic phase shift
and the attenuation is grater, with the -3dB point starting at 70kHz instead
of 170kHz
Pi-Filter Step Response
For the step response I am using a current load configured with a
pulse function, with the step/pulse looking something like this:
To help with comparison, below is what the step response looks like without
any filter present
NOTE: This plot is a bit deceptive as it looks like the circuit is
able to respond well to transients, HOWEVER if you also look at the frequency
response (see next section) you will see that we can do better
Here is the initial pi-filter configuration I have chosen, a 10μF electrolytic
capacitor, a 33Ω (@100MHz) ferrite bead, and a 10μF ceramic capacitor
NOTE: This is not my final configuration, if you have a look at
the last section you will see that lowering the ferrite bead impedance
drastically helps with step response
And here are some configurations that did not did not make the cut. Again,
it's a bit harder to see why if you just look at the step response; if you
also see the frequency response (see next section) you will note that some
configurations have sharp phase shifts that lead to a "peaky" attenuation
NOTE: I realise that ELEC/CER & CER/CER configurations have a
similar step response (with CER/CER actually having a better frequency
response). However I am fairly certain that in the real world ELEC/CER will
give better performance due to the relatively large ESR of the electrolytic
capacitor, as with low ESR capacitors it's much easier to introduce
oscillations/ringing with fast transients
Pi-Filter Frequency Response
Same order as before, first we have no filter (to help with comparison).
Note how we have a tolerable attenuation up to 10GHz where he see a drastic
phase shift and the peak that comes with it (though I suspect we don't
really care what happens after the 1GHz mark)
Here is how things look if we implement the chosen filter from the previous
section. Note how there is no drastic phase shift and the attenuation is
grater, with the -3dB point starting at 90kHz instead of 170kHz
And for completion, here are the frequency responses of the configurations
that did not make the cut. The -3dB point for these (in order) is 230kHz,
170kHz, & 65kHz
Pi-Filter Step Response vs Ferrite Bead Impedance
From here on out I am trying to see what impact the ferrite bead impedance
has on the filter performance, the results will shock you ;^)
As before, to make comparison easier here is the step response without any
filter present
And here are our candidates. Note how drastic the difference is between
the 8Ω (@100MHz) & 120Ω (@100MHz) ferrite beads. As you might guess,
am leaning towards the 8Ω ferrite bead for the pi-filter
Pi-Filter Frequency Response vs Ferrite Bead Impedance
Again, to make comparison easier here is the frequency response without
any filter present
And here are our candidates, the -3dB point for these (in order) is 70kHz,
90kHz, & 70kHz. Note how the 120Ω ferrite gives +2dB amplification (not
attenuation) around 40kHz
Tips For Power Filter Simulation
Here are some things to keep in mind when simulating a power filter:
-
Make sure to include parasitic elements in your simulation. This would
be the parasitic resistance/inductance/capacitance of cables,
connectors, tracks, planes, decoupling capacitors, relays...
-
Include a series resistance in your AC/noise source when doing a
frequency response. With my simulation is set it at 10mΩ. If you forget
this element (or set it to 0Ω) then your simulation would show
amplification before you see the roll-off, while in the real world you
would not see a response like this as the source would have a finite
series resistance (which I guess is a parasitic resistance)
-
Know what a typical current transient you are bound to run into looks
like for your circuit. For example, if you are switching a relay and you
know the could current is 10A then you can better understand what sort
of filter response you need
-
In addition to above, see if you can figure out the constant current
consumption of your circuit. Again knowing this will give a more
authentic simulation
-
It's useful to know the input/output impedance of the filter, as with
this information you can figure out what filter topology will be best.
For example, a pi-filter is designed to "match" a high input impedance
with a high output impedance, whereas a T-filter is the opposite, a CL
filter is high/low, and a CL filter is low/high