Question on Spread Spectrum

Hi,

I have a question about implementing spread spectrum in a Fly-back converter IC. The converter input is rectified AC line voltage. The converter is operating in DCM mode with 'constant' Ton and constant frequency. The load is a string of white LED.

My question is about how best to modulate frequency of the converter (+/-3 % required). If I modulate frequency in a sine wave shape, I am concerned about visible flicker in the LED light due to low frequency inter-modulation products between rectified line frequency of 120 Hz and the sine wave which may have a frequency close to 120 Hz but not exactly 120 Hz.

I am wondering, if I will have better results if I modulate with a pseudorandom-bit-sequence-generator (PRBS) of cycle repetition frequency close to 120 Hz. My thinking is that the spectral density will not be concentrated at the fundamental frequency and so instead of getting a high amplitude, low frequency intermodulation product (as in the case of the sine wave), I will get a low amplitude intermodulation product between the PRBS fundamental (~120Hz) and the AC line. This will hopefully help to reduce flicker of LED output.

My question is: does anyone here have any experience about this. Is my argument correct? Any other ideas on how to avoid flicker?

Thanks!

powerelec said:

The converter is operating in DCM mode with 'constant' Ton and constant frequency. The load is a string of white LED.

Click to expand...

First you'll have to clarify this, since the converter can't be both constant Ton and constant frequency...

My question is about how best to modulate frequency of the converter (+/-3 % required). If I modulate frequency in a sine wave shape, I am concerned about visible flicker in the LED light due to low frequency inter-modulation products between rectified line frequency of 120 Hz and the sine wave which may have a frequency close to 120 Hz but not exactly 120 Hz.

Click to expand...

It's possible that it will show up on your output. The normal function of frequency dithering is to just spread the spectrum of the switching interference, which should work in the same way an FM modulator works. Ideally this only affects the output frequency content near the switching frequency.

But in a DCM converter, modulating frequency can also modulate the gain of the converter (depending on the control type), and it's not unthinkable that this could modulate the output at the dithering frequency.

What it comes down to is whether the bandwidth of the control loop is capable of rejecting the disturbances in switching frequency caused by the dithering. For most SMPS, rejecting frequencies up to 120Hz should be easy. If you do see low frequency ripple on the output, I would blame it on too slow of a control loop, rather than the dithering itself.

I am wondering, if I will have better results if I modulate with a pseudorandom-bit-sequence-generator (PRBS) of cycle repetition frequency close to 120 Hz. My thinking is that the spectral density will not be concentrated at the fundamental frequency and so instead of getting a high amplitude, low frequency intermodulation product (as in the case of the sine wave), I will get a low amplitude intermodulation product between the PRBS fundamental (~120Hz) and the AC line. This will hopefully help to reduce flicker of LED output.

Click to expand...

Using a pseudorandom profile would make the dithering more effective in terms of EMI, but for output ripple it still comes down to the bandwidth of the dithering vs the bandwidth of the controller.

mtwieg said:

First you'll have to clarify this, since the converter can't be both constant Ton and constant frequency...

Click to expand...

Apologies for not being clear! Frequency is constant. Ton is defined as 'constant' because the circuit is not expected to sustain any large load changes. Also the 120 Hz ripple on the input voltage is ignored by making the feedback loop very low bandwidth (< 12Hz) by placing a large capacitor at the output of the error amplifier. Also the input voltage is unfiltered, so that it is varying from 0V to Vac peak and back to 0V every 120 Hz half cycle. This architecture is used because a high power factor is desired and because in this way, our average primary current will be sinusoidal and in phase with the input voltage. There is feedback from a resistor in series with the LED load, but this feedback is only used to compensate for slow changes such as temperature variation or slight changes in Vac_rms. A large output capacitor is used because output power supplied is varying over one half cycle but our load is constant power so we need to store the excess somewhere and keep ripple in output voltage and current low.

It's possible that it will show up on your output. The normal function of frequency dithering is to just spread the spectrum of the switching interference, which should work in the same way an FM modulator works. Ideally this only affects the output frequency content near the switching frequency.

But in a DCM converter, modulating frequency can also modulate the gain of the converter (depending on the control type), and it's not unthinkable that this could modulate the output at the dithering frequency.

What it comes down to is whether the bandwidth of the control loop is capable of rejecting the disturbances in switching frequency caused by the dithering. For most SMPS, rejecting frequencies up to 120Hz should be easy. If you do see low frequency ripple on the output, I would blame it on too slow of a control loop, rather than the dithering itself.
Using a pseudorandom profile would make the dithering more effective in terms of EMI, but for output ripple it still comes down to the bandwidth of the dithering vs the bandwidth of the controller.

Click to expand...

As described above my converter loop is ultra low bandwidth. I do not want to modulate the duty cycle to compensate for 120 Hz noise on the input. I will filter that noise using a (huge) output capacitor. My LED current will also have 120 Hz ripple but that is OK because the ripple frequency is higher than the persistence of vision frequency of the human eye. I can also modulate spread spectrum at ~120 Hz either sinusoidal, triangle or pseudo-random. What I am truly concerned about is inter-modulation difference products between my spread spectrum and AC line, showing up as flicker in the LED light, because 1-2 Hz is perceived by the human eye as below the persistence of vision threshold and hence flicker

So ultimately my question is: will the perceived flicker in the LED current be less if I use pseudo-random or if I use sine/triangle of similar amplitudes to modulate.

Thanks for taking the time to read my post mtweig!

Okay, I see your concern then.

If your dithering rate is the same as or near the line frequency, then it may modulate the input current waveform, causing frequency spurs at very low frequency and at intervals above the line frequency. This could cause poor power factor, as well as increased low frequency ripple on the output. Simulation is probably the best way to tell.

Personally I'd suggest using a transition mode converter, where variable frequency is built into the converter itself. The frequency dithering is perfectly synchronous with the input voltage, so no spurs are produced.

Alternatively you can use a dither rate much greater than the line frequency and converter bandwidth. If you do that then it shouldn't matter what the actual dithering profile is (so long as its bandwidth is limited properly). Is there any reason why you want to to use a low dither rate?

By transition mode, I am guessing you mean where the subsequent clock cycle begins at the moment when the current in the secondary for the previous cycle decays to 0? Yes that would have in-built spread spectrum. But I am the chip designer, not the end user so I have to design this chip in the DCM constant freq mode as required in the specification.

Regarding using a dither rate higher than 120 Hz, one concern is that inter-modulation products may still appear at the output. For eg: with a 241 Hz dither I could get inter-modulation products of 241-2*120 = 1 Hz. I don't know if the magnitude of these higher order products is larger or smaller than the fundamental (f_spread-f_line). Another concern is that if we put the dither frequency to much in the audible range then we may get a loud humming sound from the inductor due to magnetostriction.

Well if you're designing the IC then you have more latitude to deal with the problem.

Are you implementing any kind of input current shaping feedback loop at all? Average current mode control would probably deal with this issue entirely.

Another possibility might be to also modulate on time along with frequency, so that the gain of the converter becomes more frequency independent.