# Why does radiated emission of a PCB decrease as the frequency of the signal increases?

I was just reading Henry Ott's book "Electromagnetic Compatibility Engineering" and chapter 16.3.3 where he discusses how changing reference planes is detrimental to EMI.

Quoting from the book:

When a signal trace changes from one layer to another, the return current path is interrupted because the return current must also change reference planes

which he explains increases loop area and radiated emissions.

Then I then came across this statement:

At 247 MHz (diamond marker in Fig. 16-9B), the emission is almost 30 dB greater for the case where the signal transitions from the top to bottom layer, versus the case where the signal is routed on a single layer.

That is in perfect accordance with the previous statement, but:

"Above about 2 GHz, the interplane capacitance is sufficient to reduce the impedance of the return path, and hence, the radiation in both cases are about equal"

And indeed by observing the graphs, above 2 GHz emissions in Figures 16-9 A,B are approximately the same.

So my question is:

Does designing for frequencies above 2 GHz have less restrictions, in that you don't have to take such issues as changing reference planes into consideration? That would be counterintuitive to what I've considered so far about higher frequency signals and how they are more susceptible to EMI.

Thanks

• No, life gets ahrder as the frequency increases. However, there's such a mess of absorption, relfection, changing impedance etc etc going on, any particular result is pretty meaningless to draw general conclusions for other boards from. If your board needs low EMI, then you need to test the final configuration. The only general things that can be said are treat all HF signals as transmission lines, avoid large changes in impedance, stripline doesn't radiate like microstrip can. – Neil_UK Mar 31 '18 at 15:43
• @Neil: to say nothing of specific PCB characteristics across frequency. – Peter Smith Mar 31 '18 at 16:23
• The surface losses will rise at higher frequencies. These losses will be your friend. – analogsystemsrf Apr 1 '18 at 1:57

It is not easier to design for higher frequencies and the confusion you are experiencing is caused by the misconception of thinking that the two solutions compared in the book would be as good at 2 GHz when they are actually as bad.

In other words, traditional signal and power traces are great antennas anyway at those frequencies, but usually the signals spectrum has already faded so low that it is not a problem anymore. This can be easily understood if you think for example the envelope of the spectrum of a trapezoid which is quite common signal waveform. The amplitude decreases 20 db/dec after the frequency of $\frac{1}{\pi \cdot T}$, where T is the signal on-time and 40 db/dec after the frequency of $\frac{1} {\pi \cdot t_r}$, where $t_r$ is the rise time of the signal. Therefore, for example the spectrum of a signal with risetime of 10 ns has a second knee at 32 MHz whichs is almost two decades under the frequency range of interest. Therefore, the signal amplitude has faded to about 1/10000 because of this knee alone at the frequency of 2 GHz.

If we were studying discontinuties in structures designed for signals having considerable spectral content at 2 GHz, it would be clear that the circuit with discontinuties would behave worse.

Radiation factor here is the shunt effects increase with frequency and the series inductive loop is large enough for the aperture of 247 MHz to be sufficently large > 10% of wavelength to leak enough signal to make 30dB difference including its signal and return path inductance and loop aperture.

Loops and slots at 1/4 wavlength are efficient radiators. Feedthrus contribute a bit of inductance depending on length to diameter ratio, which overall with the signal loop can cause sourcesignal to be reflected and radiated more in the entire loop from source impedance mismatch.

Although Friis path Losses increase with frequency, they also leak easier thru loop, slots and unintended radiators ( sharp edges) easier with smaller wavelengths. In this case the dielectric shunt capacitance of the circuit board served to attenuate more with rising frequency due to L/C ratio which affects impedance. (Z^2)

In fig B the dominant unintended radiators are visible at 1f,2f, 3f and 5f although the 1f is closer to 270 Mhz than (the authors report of ) 247Mhz with 290MHz (2.9GHz/10) per division and 2f at ~ 540Mhz (just under 2nd division at 580 MHz. But that is just my interpretation from interpretting the display.

Bravo (+1) for reading the Ott Book. It is a must read for all budding EE's, which I read in 1980 and ought to be taught in school.