Dither a power converter's operating frequency to reduce peak emissions
Add a small circuit to spread the peak RFI energy.
By Bob Bell and Grant Smith, National Semiconductor, Phoenix, AZ; Edited by Brad Thompson and Fran Granville -- EDN, October 13, 2005
Designers of dc/dc switching power converters face the challenge of controlling EMI (electromagnetic-interference) emissions produced during normal operation. If large enough, these emissions conduct through power lines or radiate to other assemblies within a system and can compromise a system's performance. Emission peaks typically occur at the converter's fundamental switching frequency and gradually reduce in amplitude at each higher order harmonic, with most of the emitted energy confining itself to the fundamental and lower order harmonics. Modulating, or dithering, the power converter's operating frequency can reduce the peak emissions by spreading EMI over a band of frequencies.
Most modern PWM controllers use an external resistor to set the operating frequency, which typically increases with decreasing resistor values. For example, the LM5020's internal oscillator delivers a regulated 2V at its programming pin (RT), and a programming resistor connected to RT sets the current that RT delivers. The oscillator also delivers a proportional current into an internal timing capacitor (Reference 1). The period of the timing capacitor's ramping voltage determines the oscillator's frequency.
The external dithering circuit in Figure 1 comprises a simple stand-alone comparator-based oscillator configured to operate at approximately 800 Hz. The output state of comparator IC2 goes high upon power-up. R1, R2, and R3 set the comparator's positive input, which initially rests at 2.9V. The voltage at capacitor C3 ramps up toward the positive threshold.
When the voltage at the comparator's negative input reaches the positive-threshold voltage, the comparator's output switches low, which also lowers the threshold at the comparator's positive input to 2.1V. The voltage at capacitor C3 then ramps down toward the new threshold, and, when it reaches the lower threshold voltage, the cycle repeats. The voltage across C3 approximates a triangular wave with a minimum voltage of 2.1V and a maximum of 2.9V.
To dither the LM5020 controller's PWM-oscillator base frequency, the triangular wave generated by IC2 modulates the current from the controller's RT pin. Resistor R5 sets the percentage of modulation dither. The right side of R5 is fixed at the RT pin's regulated potential of 2V, and the low-frequency triangle wave coupled from IC2 through capacitor C2 drives R5's left side. For R5 with a value of 64.9 kΩ, the peak-to-peak current through resistor R5 is approximately 12 µA. With the dither circuit disconnected, the steady-state current that RT sources is approximately 121 µA, and the 12-µA p-p dither current thus represents 10% total modulation.
An LM5020-controlled PWM flyback dc/dc converter, IC1, evaluates the dither circuit's effectiveness. The circuit's fundamental operating frequency is 250 kHz, which the controller's RT resistor sets. The red trace of Figure 2 shows the conducted emissions on the circuit's positive input-power line without the dither circuit in operation. The peak emissions are narrowly confined around the fundamental oscillator frequency of 250 kHz with a measured amplitude at the fundamental frequency of –24 dB.
Connecting the dither circuit to the controller's RT input produces the blue trace of Figure 2. Conducted emissions around the fundamental frequency now disperse around the fundamental frequency with maximum amplitude reduced by –34 to +10 dB.
| Reference |
Click here for more Design Ideas! |
-
Gents, what some are failing to consider is that a sine wave is akin to a delta function in the frequency domain, so all of its energy is concentrated in a very narrow frequency band.
By proper dithering, the amplitude of that very strong peak is spread to much lower amplitudes in a much wider band...imagine a delta function spreading into a large-sigma Gaussian that's very wide...so that the highest parts are still down in the noise. This is the idea.
What can you offer as an alternative? To not use PWM switching power supplies or any other kind of oscillators or digital circuits?
FP Giles - 2006-12-4 16:34:00 PDT -
There are several posted comments from which it is obvious that there is a lack of understanding of some basic properties of Fourier transforms and, hence, of how sprectrum is populated. As has been pointed out by other authors, by dithering a carrier it is certainly possible to remove power from one band (moving it to another band). This has nothing to do with any averaging performed by spectrum analyzers: it is simply math.
What remains an open question is how a pacemaker may respond to this kind of interference. Depending on the particular design, it might or might not be sensitive to the average interfering power in a broad frequency band, making dithering questionable. However, this is an issue that has to be addressed by the FCC.
Pere Palà - 2006-15-2 06:16:00 PST -
The problem with this technique is that it only reduces the peak amplitude of the EMI, and it only does that if you average over a large enough piece of the dither cycle. Most victims of this EMI, for example a pacemaker, don't respond to the peak EMI signal but to the integral over their susceptable bandwidth. Unless you can move this energy out of the victims bandwidth, which is often several orders of magnitude wide, it has no effect.
The FCC, in trying to keep things managable, only regulates the peak interference power radiated at any one frequency. They sweep a narrow receiver slowly over the band of frequencies of interest and check that the receiver never picks up a strong enough signal to cause problems.
I see two possible results of this technique. One is that the FCC may have to correct the enforcement procedures using much more complicated and expensive wide band receivers or integrating the output of the existing receivers over wider bandwidths further exacerbating the problems of getting FCC approval. The other possibility is that designers will realize that these authors have their heads so deeply wedged in the rulebook that the mixture of dioxin from the paper and solvents from the ink have addled their brains!
Douglas Butler - 2006-27-1 09:06:00 PST -
I am certainly not an RF expert, but isn't an FM signal described by the Bessel functions? When modulated, the carrier (aka clock frequency, switching frequency...) will be reduced - possibly even eliminated given the right modulation index. Of course, the total energy will be the same, but if we're trying to reduce peaks, is this not a valid approach? Measurement technique has nothing to do with it. The "carrier" will be reduced no matter how it's measured.
It seems a lot of people think this is a valid approach, as it continues to be used in such modern standards as PCI Express and SATA.
If they're all wrong, please explain why.
Michael Dunn - 2006-20-1 07:37:00 PST -
I suggest the analysis and recommendations contained in this article are completely wrong.
The authors suggest that dithering the switching frequency of an oscillator, which is basically what the power convertor is, will reduce the peak radiated power. There is absolutely no reason why this should do so, and designers of frequency-hopping transmitters would be as critical of the suggestion as I am. It is simply wrong.
Where the authors may have erred lies in figure 2, which purports to show the spectrum of the emissions. This graph obviously comes from a spectrum analyser of some sort. I suspect that the time-averaging option has been enabled, so that the AVERAGE power seems to be reduced. But the PEAK (instantaneous) power will not be reduced, and neither will any interference caused.
Further, close inspection of the blue profile shows a hint of beating at the top of the peak. This probably reflects some sort of interference or beating between the sweep frequency of the spectrum analyser and the 800 Hz dither frequency. This represents a serious lack of understanding of how to use a spectrum analyser.
Had the authors used the spectrum analyser correctly, they would have found that there was a central spike with sidelobes 800 Hz away: the sort of spectrum you would get from an FM transmitter. In effect, this is simply a form of frequency modulation.
I am appalled that this article was presented by National Semiconductor, and equally appalled that EDN published it.
Roger Caffin (Dr) - 2005-28-11 15:55:00 PST
