Design Article
Reduce EMI by varying power supply frequency
Robert Kollman, Senior Applications Manager, Texas Instruments
2/4/2009 4:00 AM EST
Have you ever tested for EMI and found that no matter what you do in the way of filtering, you are still a few dB out of specification? Here is a technique that may help you pass the EMI requirements or possibly simplify your filter design. This technique involves modulating the power supply switching frequency to introduce sideband energy and change the emission signature from narrowband noise to broadband, effectively attenuating the harmonic peaks. Note that total EMI is not reduced, only redistributed.
With sinusoidal modulation, two variables of the variables that can be controlled are modulating frequency (fm); and how far you vary the power supply switching frequency (Δf). The modulation index (Β) is the ratio of these or:
Figure 1 shows the impact of varying the modulation index with sinusoidal waveforms. With Β=0, there is no frequency shifting and only one spectral line. With Β=1, the frequency signature begins to spread and the center frequency component has fallen 20%. With Β=2, the signature has spread even further and the largest frequency component is 60% of the original case. Frequency modulation theory can be used to quantify the energy in this spectrum. Carson's rule indicates that most of the energy will be contained in a bandwidth of 2 * (Δf + fm).

Figure 1: Modulating the power supply switching frequency spreads the EMI signature.
(Click this image to view a larger, more detailed version)
Figure 2 shows even larger modulation indices and illustrates that over 12dB of peak EMI reduction is possible.

Figure 2: Greater modulation indices can further reduce peak EMI.
(Click this image to view a larger, more detailed version)
Choosing modulating frequency and frequency shift are important considerations. First, the modulating frequency should be higher than the EMI receiver bandwidth so that the receiver does not measure both sidebands at the same time. However, if you choose too high a frequency, the power supply control loop may not be able to adequately constrain the variation, resulting in output voltage variation at the same rate. Additionally, the modulation can cause an audible noise in the power supply. So a modulating frequency not too far above the receiver bandwidth, but out of the audible range, is typically chosen. Obviously, from Figure 2, a high variation in operating frequency is preferable. However, it is important to recognize that this will impact the power supply design. Namely, choose the magnetics for the lowest operating frequency. The output capacitor also needs to handle larger ripple currents due to the lower frequency operation.




Lunta123
2/6/2009 3:37 AM EST
Hi Robert,
have you really missed the LTC6902 from 2003?
This wasn't really news.
Sign in to Reply
bcarso
2/10/2009 1:16 PM EST
Is this what we really wnat to do though? We may complain about emission limits but they have been developed to make the EM environment clean enough for everyone. It is fouling the commons to play tricks to get around the instruments.
Sign in to Reply
bcarso
2/10/2009 1:19 PM EST
By the way my previous and more lengthy comment was evidently rejected by the system because I attempted to use the word bee eye tee see aich as a verb in the sense of complaining. This criterion is silly and ridiculous!
Sign in to Reply
Donscoggin
12/9/2010 12:21 PM EST
Dithering the frequency is a valid method of meeting requirements. Reducing peak emissions seems reasonable to me. For example, Quasi Peak values are often used to give a quality factor to emissions to meet standards.
Sign in to Reply