ELECTROSTATIC NOISE

·         Shielding can be used to keep noise inside the shield.  In this case, it is important that the shield have  a nice LoZ impedance to ground.  Ground is any single reference point (Mother Of All Grounds - MOAG).  The shield and the signals should share this point.  If “earth ground” is available, it should share this point also.  This will shunt RF currents (generated inside the shield) to ground.  Where will this be?  Is the negative side of the 12V supply earth ground?  Can we count on that (think of the battery case).

·         Shielding can also be used to keep noise sensitive circuits in.  In this case also, it is important that the shield have  a nice LoZ impedance to ground.  The point here is also to shunt currents (this time generated outside the shield) to ground.

·         A good RF ground path is short, wide and has a lot of surface area.  Braids are good.  Copper “tape” is good.  30 gauge wire wrap wire is NO GOOD.

·         Decoupling capacitors are REQUIRED at each IC.  Make sure that the highest harmonic of the currents in the IC is below the self resonance of the capacitor.

·         Purchased components (i.e. SBCs) should be tested by themselves before we commit to a device.   This will set a benchmark that the design can be no better than.

·         Multilayer boards with their uniform ground plane impedance are the only way to load the deck in the favor of first pass compliance.

·         Separate analog and digital ground planes if they must coexist on the same board.  These should both be bound to the reference point for the instrument (the Mother Of All Grounds).

·         If multiple ground planes are used (i.e. an “noisy” digital one and a “quiet” analog one), do not let wires traverse the boundaries.  You might as well have but one ground plane if you do that.

·         Identify your biggest noise sources.  Big currents or voltages with fast risetimes (or high frequency CW signals) will be the most obvious culprits.  My current list is the heater (If driven by a PWM signal), SMPS, stepper motors, backlight circuit, and all digital circuits.

·         Identify your most susceptible circuits.   I have that to be the thermistor circuits (current drivenºHiZ), photodiodes (ditto) and the electro-chemistry circuits. 

·         Separate the two items above as much as possible.  Put (well grounded in the RF sense) metal between them. 

·         When sensitive sensors (there’s an oxymoron for you) are far from their circuits, use GOOD shielded cables.  Look out for triboelectric effects in cheap cables.  Gore makes a good Teflon dielectric cable.

·         Each signal should have a single shield ground point to start.  With some of the RF interference signals, this rule is not so clear cut, but it is a good place to start.  The single shield return point should be at the signal reference point.  Ideally this is also at MOAG potential, but more often than not, it will not be.

·         NO CURRENT should flow in a shield.  This means that all shielded wires should have no less than two current carrying wires inside the cable.

·         Know where you EXPECT noise currents to flow.  This does not mean that physics will comply.  But, know what you think is happening.

ELECTROMAGNETIC NOISE

·         Identify all of the loops in the machine.  Start at the block diagram.  Any loop that current can traverse without encountering a HiZ circuit component will be capable of emitting or receiving B fields.

·         Magnetic shielding is expensive, possible and very difficult to apply (bend it and it is no longer Mu metal). 

·         The easiest way to quiet an electromagnetic source is to reduce loop size.  On a board, one can reduce the loop to a figure eight with two loops of equal size (far field net B field is zero).  Cables should be twisted.

·         The currents in the two wires that are twisted together in a twisted pair should always be equal and opposite.

·         It is possible to have noise sources that are both noisy in an electrostatic and electromagnetic sense.  This case calls for twisted pairs inside a shield.

·         Distance is your friend.  Keep those talkers and listeners far from each other. 

Back