Reduce EMI by sweeping a power supply's frequency
Edited by Bill Travis
John Betten, Texas Instruments, Dallas, TX -- EDN, May 27, 2004
Switching power supplies can be notorious noise generators. You should prevent this noise, which is conducted, radiated, or both, from returning to the input source, where it can potentially wreak havoc on other devices operating from the same input power. The goal of an EMI (electromagnetic-interference) filter is to block this noise and provide a low-impedance path back to the noise source. The larger the noise, the greater the size, expense, and difficulty of the filter design. Power supplies that operate at a fixed frequency have their largest EMI emission at this fundamental, fixed frequency. Emissions also occur at multiples of the switching frequency but at diminished amplitudes. The simple circuit in Figure 1 makes the switching converter operate over multiple frequencies rather than one, thereby reducing the time average at any one frequency. This scheme effectively lowers the peak emissions.
The circuit in Figure 1 is a self-starting oscillator with an oscillation frequency of approximately 500 Hz. When you apply power, C3 begins to charge up from 0V, and the output of the TL331 comparator is in a high-impedance state because its noninverting input sees a higher voltage than that of the inverting input. As C3 charges, its voltage crosses the voltage reference of the R1-R6 divider, and the comparator output trips to a low state. The voltage on R6 instantly drops to a lower reference level because R5 is now in parallel with R6. C3 begins to discharge toward this new reference level because R3 is simultaneously in parallel with C3. The cycle repeats after C3 discharges to the voltage on R6 when the comparator output reopens. You must carefully select the components to ensure that the two voltage-reference states of R6 are lower than the upper and lower possible charge states of C3. The circuit uses C3 to adjust the oscillator frequency; you should select C3 to have a lower value than C2. The oscillator's frequency is approximately equal to
Capacitor C2 ac-couples the ramp voltage of C3 into the UCC3813's oscillator pin. The injected signal adds to the charging current of CT during its positive portion (ac signal), thus increasing the controller's operating frequency. During the injected signal's negative portion, some of CT's charging current disappears, slowing the controller's operating frequency. Figure 2 shows the effects of the injected signal on the charging of CT. R4 controls the magnitude of the current that is injected. Reducing R4's value increases the range, or spread, of the operating frequency around its nominal fixed frequency. The injected signal's oscillation frequency, which C3 sets, controls the frequency-sweep rate.
The differential EMI-current measurement of Figure 3 (1 dBµV=1 dBµA) shows the before-and-after effects of adding the frequency-shifting oscillator. This design easily achieves a 10-dBµA reduction with a 12-kHz sweep window. A wider window further reduces EMI, but the modulator frequency may be noticeable in the converter's output ripple voltage. It is also desirable to make the injected ramp voltage as linear in shape as possible to prevent the switching converter from spending excess time at its switching-frequency limits. The nonlinearity can result in an EMI response with two distinct frequencies. You must take care not to operate the circuit below the power converter's low-frequency limits, or saturation of magnetics may occur. This circuit demonstrates a low-cost, small-area approach to reducing conducted-EMI emissions.
Check out our Best of Design Ideas section!
-
I notice that several of the other posted comments deride the use of spread spectrum. Spread Spectrum is a popular and proven technique that is most likely used in the PC you're reading this with. Not only will it reduce the signal measured on an EMI receiver, it can reduce interference to many communications devices having the same, or more narrow, IF bandwidth.
However, for conducted EMI measurements (as in the article) a 9 kHz IF filter is normally used. To obtain a 10 dB reduction in the measured signal the spectrum must be spread out over a 90 kHz width. Also, the reduction sited in the article is 10 dBuA. It is actually 10 dB. So, for typical EMI measurements, and typical communications devices, the proposed spread spectrum is not enough to improve things.
DAVID CUTHBERT - 2004-28-7 07:56:00 PDT -
The proposed improvised 'modulated spectrum' solution might perhaps help you trick an EMC qualification test but it does not really reduce actual radiated power levels.
If you want a GOOD solution:
- reduce the dV/dt 's and dI/dts
- provide magnetic shielding for the inductors
- provide adequate input / output filtering (RTFDS)
- good layout practice (small loops, controlled return currrents)
- if applicable/possible synchronize your ADCs to the PSU switching frequency
- build a prototype!!!!!
There are several nice controlled slope dc/dc controllers on the market, e.g. Linear Technology's LT1738 and LT1683.
All this may cost you somewhat increased power losses, space, prototyping effort, and/or money, so the optimal trade-off will be different for each design case.
Many analog cheers!
Marco van Steen - 2004-20-7 07:36:00 PDT -
"This scheme effectively lowers the peak emissions. "
It just makes it harder for them to track you down....
Nicholas Allen - 2004-28-5 09:25:00 PDT -
Another approach is to drive the RC pin with an external frequency source that varies (pin commonly used to sync convertors).
If this is done there is no reason why a small micro can not be used to drive the frequency on a psuedo random basis (within limits) but an RC will still be needed to allow the system to start up until the micro can take over.
Bob Moreton - 2004-28-5 00:44:00 PDT -
Yet another article on how to get round the regs. A GOOD designer would design so that his product didn't radiate !! As one who had to earn a living as well as enjoy leisure time using the radio spectrum, I get quite insensed at the the modern "dirty man" attitude to EMI. I have to contend daily with emissions from TV sets that can be received at 200 yards !! and received well enough to tell what channel they are tuned to....the audio FMs the PSU. It does NOT reduce the EMI it may spread it around, and because of the way it is normally measured with a sweeping analyser it may give a lower reading....but it is CRAP engineering !! I dont suppose any modern digital engineers know anything about RF.
Alan Melia
Carrowchip Electronics
UK
Alan Melia - 2004-27-5 15:54:00 PDT
