Measurements and Limits of Conducted EMI: Quasi-peak, Average, and Peak Measurements

We have not yet explained what the rationale is behind the two types of limits - average and quasi-peak.

Historically, quasi-peak was meant to simulate human responses to noise. Humans have a slowly increasing level of aggravation or annoyance to a persistent disturbance. Therefore, to simulate this (subjective) response, there are built-in attack and release rates in quasi-peak detection. The signal level is effectively weighted according to the repetition frequency of the spectral components constituting the signal. So, the result of a quasi-peak measurement will always be dependent on the repetition rate. The higher this frequency, the higher the measured quasi-peak level. Further, because of the finite charge and discharge time constants involved in quasi-peak detection, the spectrum analyzer must sweep considerably slower in this setting. Therefore, peak detection can be carried out, and this turns out to be much faster.

Further, with peak detection, we will always get the highest reading (usually followed by quasi-peak and then by the average, see Fgr. 7).

So, to perform EMI scans quickly (at least in the initial stages of design), most engineers prefer peak detection. The results of this are then compared against the published quasi-peak limits, and compliance is sought. If it’s achieved, in effect, some additional headroom (safety margin) has been gained, because if a quasi-peak reading had been carried out, it would have certainly given a reading less than the peak reading. The headroom is "nice to have," because it will help account for various uncontrolled parasitics that may show up at a later stage. So, quasi-peak detection really needs to be performed only when we are marginally failing the peak detection test (with quasi-peak limits applied).

Note: Usually, if we can meet the quasi-peak limits, we automatically meet the average limits too (using average detection and average limits). But there are stray cases where CISPR 22 quasi-peak limits may be complied with, but the average limit test is failed. That is a bad sign, possibly indicating that a major re-design of the PCB layout and/or transformer is required.

It’s often colloquially stated that 'at frequencies below approximately 5 MHz the noise currents tend to be predominantly differential mode, whereas at frequencies above 5 MHz the noise currents tend to be predominantly common mode.' But this may or may not be true always. Certainly at frequencies above 20 MHz, any conducted noise is most likely attributable to inductive pickup, For example from radiation spilling out from the cables.

And we know that that is inherently common mode in nature. However, since radiative pickup is not necessarily the main (or only) source of common mode noise in a switching converter, we should be prepared for surprises. For example, we recall that unequal line impedances can convert a (high-frequency) common mode noise into a (high-frequency) differential mode noise.

Finally, we observe that standard conducted EMI emission limits are typically only up to 30 MHz. We can ask - why weren't the limits set even higher? The reason is that by 30 MHz, any conducted noise is expected to automatically suffer severe attenuation in the mains wiring, and therefore won't really be able to travel far enough to cause interference much down the road. However, since the cables can certainly still radiate (locally), typical EMI radiation limits cover the range from 30 MHz to 1 GHz.