There are no low frequency systems any more!

11 Aug 2020

Author: Keith Armstrong

There are no low frequency systems any more

Following on from my Blog “Ground / power bounce cause noise emissions from all IC pins”….

When we measure excessive emissions from our products at even-order multiples of a clock’s frequency, we should suspect ground/power bounce noise as being the cause. This is a useful diagnostic tool, and points us towards methods of fixing the problem at source (which is always the best approach, if you can do it).

The very wideband CM noises caused by IC ground/power bounce are emitted from every pin on a clocked IC, whether Vdd, Vss (0V, GND), signal/data input, signal/data output, reference voltage input or output, etc.  When causing problems in meeting emissions test limits, it could be being emitted from:

a) The PCB’s power distribution systems, including its external mains or DC power leads;

b) Heatsinks fitted to clocked ICs, power converters or regulators (switched or linear);

c) Any/all conductors (traces, wires, cables, etc.) connected to an IC
 – even if they are low-frequency analogue I/Os;
‘static’ digital inputs (e.g. connected to mechanical switches),
or ‘static’ digital outputs (e.g. connected to indicator LEDs).

d) Any/all conductors (traces, wires, cables, etc.) connected to a module, product, or item of equipment that uses clocked ICs (such as microprocessors, codecs, dynamic memories, FPGAs, etc.). 

It is important to be aware of the fact that because ground/power bounce noise is ‘fired out’ of all IC pins as CM noise, this means that no low-frequency analogue signal (audio, instrumentation, etc.) that connects to a microprocessor, etc., is ever free of RF noise that extends well into GHz frequencies.

This is true even if the analogue signals connect to microprocessors, etc., via active components such as operational amplifiers and linear voltage regulators, or passive components such as capacitors, transformers or baluns. The very high frequency CM noises caused by ground/power bounce will simply pass straight through them (via the stray capacitances inside the devices and/or their associated PCB traces or wires), especially at frequencies above 100MHz.

Yes, this type of noise will even pass straight into the output pin of an opamp, and come out of its input pins with little/no attenuation!  (At these very high frequencies, extending to GHz, the output impedance of an opamp is about 100 Ohms – because all the open loop gain has vanished and no feedback loops can operate to reduce the output impedance. So RF noises on traces and other conductors couples very nicely indeed straight into opamps via their output pins.)

Not only that, but switching power converters and regulators also couple significant amounts of DM and CM RF noise onto all power rails and signals, including analogue signals that connect to opamps, whether they are inputs or outputs.

Unless all DM and CM RF noise has been removed from low frequency analogue signals by good-quality filtering (and filtering at > 1GHz is not at all easy), they are going to need to use screened (shielded) cables that make good ‘360°’ RF terminations at both ends.

This termination at both ends is as recommended for high-frequency systems by Henry Ott in Chapter 3 of his important textbook: “Noise Reduction in Electronic Systems”, Second Edition, Wiley-Interscience 1988, ISBN: 0-471-85068-3.

These days, the only shielded/screened conductors that should ever be routinely permitted to terminate their shields/screens at one end only, are wireless antennas!

A (roughly) worked example of stray capacitance coupling:

A stray capacitance as little as 0.1pF has an impedance of about 8kohms at 200MHz, and a typical external cable has a characteristic CM impedance of about 150ohms.

So, if a 3V peak-peak digital clock signal (say, 1Vrms) at 200MHz has just 0.1pF of stray capacitance to a cable that exits a product without shielding or filtering,  it will cause about 12microamps rms of CM current to flow in that cable.

But it only takes 2.5µA of CM current in a long cable on an Open Area Test Site (OATS) to fail a Class B limit at 200MHz, to CISPR 32 or EN 55032 (same as for CISPR 22 or EN 55022 that these new standards replaced).

In fact, the 0.1pF of stray capacitance from the 200MHz clock would likely cause the tested emissions to be about 13dB above the limit.

At the 3rd harmonic of the 200MHz squarewave clock, 600 MHz, the amplitude has reduced to 1V pk-pk (say, 0.3V rms) but the impedance of the stray 0.1pF has also reduced, to about 2.7kohms, coupling about 11microamps rms into the external cable. Because the Class B limits are 10dB higher above 230MHz, this should only fail by about 3dB. Ditto for the 5th harmonic at 1GHz.

But what does it take to suffer 0.1pF of stray capacitance, really?

This calculator tells us that two uninsulated 1mm diameter wires spaced 30mm (1.5 inches) apart and running parallel in air for a length of 12.5mm (half and inch) have a stray capacitance of 0.1pF between them.

If both wires had a typical thickness of PVC insulation, the stray capacitance would be very slightly higher.

If instead the two conductors were 0.1mm (4 mil) wide traces in an FR4 PCB, spaced apart by 12.5mm (half an inch) and running parallel for just 6mm (a quarter inch), they would also have a stray capacitance of 0.11pF.

So it is very easy indeed to get at least 0.1pF of stray capacitance in a modern PCB assembly!

In fact, EMC engineers are quite used to the fact that, in typical modern electronic products, any CM RF noise at frequencies above 100MHz spreads very widely throughout all the traces in PCBs, and throughout all of the other conductors (wires, cables, pneumatic pipes, etc.) in the product too. 

The types of signals or powers that those PCB traces or other conductors are carrying does not matter at all. The inevitable stray capacitances between them all spread the RF noise around without caring what names the circuit designers gave each net.

« Back to Blog