Dear all,

until recently I was confident in my filtering practices, however now I would like to put my rigour to the test and see what the community thinks.

I have attached an image of RFI filtering I added onto some differential signals. I shall in point form outline my line of thinking and aim to discuss each point.

Let us focus on SIN+ and SIN- differential signals coming IN to the circuit from pins 14 and 7 at the left of the image.

1. C204 and C205 are X type 0603 package capacitors. The intent is to redirect RFI noise to the metal chassis/heatsink to avoid reflecting incoming noise back out and to avoid transmitting internally generated noise out.
2. FBP26 (Ferrite bead pack) is a common mode choke so will present a high impedance to common mode noise/signals and the aim there is to try and get the differential/common mode noise through the caps instead. The caps would also help to get rid of the differential noise which might pass through the common mode chokes easily.

Questions from me:

1. The 0603 package leave ~0.6mm of clearance between the signal pad and RF GND pad (i.e. chassis which is bonded to earth). Typically I maintain a design rule to keep 1mm of clearance from signals to earth tracks, but in these small components lies an exception. Is there a risk of a well bonded chassis leaking noise back in through capacitors of this nature not only through the capacitance, but also through this smaller clearance? I am assuming a well bonded earth with lots of surrounding heavy duty metal.

2. Whether the ferrites are common mode chokes or just ferrites, my idea was to try and force the noise out through the caps. Now I wonder if it isn't better to have the ferrites in front of the caps to try and eat up the noise and dissipate it as heat before it gets round to the caps. This however would reverse the situation for the internally generated noise but I assume in most cases that long cable runs result in much more RF type noise than internal switching circuits @ ~100MHz? This concept also applies to replacing the ferrite with a small resistor and forming an RC network where the R is intended to help dissipate some of the noise as heat rather than blasting it all through to the chassis.

Please keep in mind that this is only signal filtering so minimal components is desirable (hence me limiting myself to just one cap and one ferrite).

Please also keep in mind that the philosophy described in this post is already in my design so I am very hesitant to change it unless you all seem to think it is an absolutely god awful way to go about filtering.

I look forward to your responses of your own best practices and what you have/have not had good experiences with.

Kind Regards,

James

 

It's not clear to me from the circuit which way the signals are flowing (I think right to left), but starting from signal source and ending in Rx input I'd arrange things in this order:

120 ohm termination resistors
choke
common mode caps

Mismatch between the two common mode caps will reduce their effectiveness. A third differential mode cap slung between the CM caps will help bring this back into line. As a rough rule of thumb, the differential cap can be about 10X the value of the CM caps. I know you don't want extra bits, but such caps are small and cheap and the performance enhancements can be large.

I'd be inclined to take the common connection of the CM caps to the signal ground, not to the chassis. Chassis grounding works well with feedthrough caps and pi filters at the boundary to the outside world, but I'd hesitate to use it in the position shown. And I'm not sure how it would be possible to make a good low-inductance connection between the 0603 caps on the PCB and the chassis.

Resistors would be cheaper than a choke for sure - it would be worth giving them a try if you've got access to EMC testing facilities to get some quantifiable results. It's very satisfying to get huge improvements from a well designed filter.

The filtering rolloff should be at least 10X the signal frequency for good signal integrity. If you're using resistors and both CM and DM caps then the common mode rolloff is given by fCM=1/2∏RCCM, and the differential mode rolloff is given by fDM=1/2∏(2CDM+CCM).

The cap clearance should only affect the maximum working voltage, there's not enough stray C to leak significant noise.

All that said, it's a pretty good circuit you've got and it should work well as is, these notes are just tweaks regurgitated from my tinkerings in this field.

 

Thanks so much Darren,

This is my first really big design so I appreciate the detailed reply. The prime motivation for a minimal component count isn't cost but space. The amount of circuitry I am having to hammer into this small space is currently beyond my comprehension but I am also trying not to subsidize my signal integrity due to what I have to pack in (BGAs on both sides!).

The circuit I posted is just one of umpteen interfaces.

What I have decided to do is this.

1. In some locations I have moved the chokes out front and the caps behind. In others I have kept it as per previous. This I will use as a bit of an experiment to see if there is any noticeable difference. This experiment also extends into ESD. I like the caps out front to give a low impedance path to ground for ESD strikes while the chokes show a high impedance path in essence 'forcing' the ESD zap down the caps. However there is a concern here in that caps will probably tend to die from ESD (keep in mind I also have ensured the back end op-amp circuitry is ESD protected, so it is not a huge risk if the caps do die but an interesting way to test).
2. I have kept the termination resistors at the back end as I want to smash them as close to the receivers as possible and the receivers will be located across the PCB whereas caps/chokes I place right next to connectors. In the future however, wherever possible, I agree, they should probably be right up front.
3. I have changed any signals to use my GND instead of EARTH as the grounding point of the caps. After your comments and reconsidering how noisy our chassis ground will be (this is a high power variable speed servo drive application, so hecks of switching noise will be picked up on the chassis), I think you are correct that it is wiser to shunt it to GND. My primary motivation for doing it to RF_GND (i.e. EARTH) was for ESD, but in reality I'm not too concerned about shunting ESD onto my big chunky ground plane.

Many thanks for your advice and acknowledgement of my concerns. I look at this as a learning process and your comments will forever be embedded in my mind for future interface design.