TEM cells are best suited for eliminating highest noise components early in prototype cycle, and for testing improvements after a failed EMC test. So far I have only tested prototypes, but the setup has already proven useful at spotting mistakes early.

Limits and background noise

These tests were performed without any advanced correlation to far-field measurements. A rule-of-thumb limit for TEM cell measurements is that anything over 40 dBµV (equals -67 dBm) is significant risk of failing in EMC tests.

Open TEM cells will catch some radio signals, even though they are not as sensitive in outside reception as actual antennas would be. The background noise in my office consists mainly of FM radio in 88 MHz to 108 MHz band and a nearby TETRA basestation at 390 MHz. Above 500 MHz there are more noise sources, including GSM base stations.

To be able to take the background noise sources into account in analysis, I have added gray background to them in TinySA-App.

Case 1: High conducted noise

This device is early in prototype cycle. The PCB integrates a isolated flyback DC-DC converter for power-over-ethernet operation. It also has a microcontroller and Ethernet interface. Early testing allows easy changes in PCB layout and component choices to eliminate EMC noise.

Emissions before modifications. blue: radiated noise, red: differential conducted noise, green: common-mode conducted noise

Zooming in on the fuzzy section below 100 MHz reveals that it is full of harmonics of the 300 kHz SMPS frequency. This noise is also very visible in the conducted measurements, indicating insufficient filtering on the power supply input.

To test the effect of filtering, I added a series ferrite bead and a 100 nF capacitor before that. Together with the existing decoupling capacitors, these form a low-pass pi-filter on the input supply. This should block especially the differential component of conducted emissions.

Comparison of emissions. blue: radiated noise before modifications, green: radiated noise after modifications, red: differential conducted noise

The modification eliminated practically all conducted emissions. It also significantly reduced the broadband radiated emissions.

There remain significant emission spikes at multiples of 50 MHz frequency. These correspond to the ethernet interface RMII reference clock signal. To address this, I'm adding a series termination resistor in next PCB revision.

Case 2: Excessive GPIO slew rate

A different device is further developed and may be undergoing EMC tests in near future. To reduce risk of failing the tests, a pretest look at any emissions is done.

Radiated emissions before modifications.

Overall the spectrum looks clean, but there is a significant spike at most multiples of 25 MHz. This corresponds to the clock rate of an external communication interface. The microcontroller used has a configurable GPIO slew rate, which is currently set to highest speed for the GPIOs involved. As a test, I reduced the slew rate to next slower setting:

Radiated emissions after software change.

This simple software change eliminated most of the emission spikes. Further testing and oscilloscope measurements are needed to verify that signal integrity on the communication bus is not affected. Thanks to the advance testing in TEM cell, these software changes can now be performed at leisure before the test laboratory visit.