I am working on a switcher with ADC sampling at the same frequency as the main switcher PWM frequency, they are already synchronized to reduce noise when sampling the ADC. There are also several small flybacks (10watts at 12Volts output) on the board, I am thinking of syncing their PWM as well, but I am wondering if syncing a flyback is the same as syncing a forward converter for timing? Is it worthwhile to do this or is the noise reduction going to effectively be minimal since there is noise all the time in SMPS due to the fast current changes in the inductors, or is switching noise really something that is worth taking into account even for small switchers?
If multiple converter circuits are constant frequency, its a normal precaution to sync them, whether A-D conversion is being attempted or not, simply to avoid beat frequency effects in regulation.
Depending on the number if bits, syncing A-D conversion can reduce jitter, but it won't reduce noise-induced error. If it is timed to a certain part of the noise period, repeatable error can be reduced.
Even your humble dual slope A-D converter multimeter uses a conversion rep rate that is loosely related to the local mains for reduced LSB dither.
On may flyback converters, you are operating them with "current mode" controller. It is bad if one converter switches on or off just at the time that another is deciding to switch off. It is extra bad if two are switching off at exactly the same time. You can end up with an small oscillation from the interaction. When you sync them, you want to be sure you don't put the converters into that situation.
In the past, I have solved this issue by using the switch off of one converter as the sync pulse for syncing the turn on of the other. This puts an unwanted cross talk between the servos of the two converters but it does avoid the oscillation issue because it removes the feedback path in one direction between the converters.
If you don't sync the converters on purpose, it can be a good idea to make sure that they run at very different frequencies. This pushes the beat frequency up high enough that the output filter can reduce it.
Others have mentioned the desirability of not having switchers operating on different but nearly the same frequency. I can attest to that being a very good idea. I just got bit by it. ;-) Two supplies running from the same input bus, physically separated by a few inches but "talking" to each other on the input bus were the problem. Apparently the control loop in them is noisy (marginally stable?) at around 4-5kHz; there's a broad peak in the spectral noise out at their outputs around that freq. It's low enough frequency to be impractical to L-C filter. But when we got a board with those two switchers running at frequencies different by about 5kHz, the 3.3V-output one (which supplied the ADCs through some filtering that was quite good at the switching freq) had a couple millivolts of 5kHz on its output, and that caused sideband spurs at 5kHz on the digitized samples.
Something not mentioned by other posters, at least as far as I could see, is that you can make things much better if you not only synchronize them but run them at phase offsets to interleave the pulses drawn from the input side. This is a really big advantage. Some switcher controller manufacturers such as Linear Technology offer parts that do this for you.
If the current drawn from each supply is essentially constant so that the switcher runs with each cycle like every other cycle, then if the ADC sampling is synced to the switching frequency, effects from the switchers alias to DC, and to the degree they are constant, they can be calibrated out (subtracted out--one nice thing about aliasing to DC is that you have no phase to worry about!). Another trick is to put the (aliased) spur signal in a frequency band you aren't interested in.
But I can also offer you some hope: the board I've been working on has several switchers on it, to supply 1.0V, 1.2V, 1.8V and 3.3V, for some fairly heavy-duty digital processing. There are two analog -->
ADC channels that run at about 100Ms/s. I can see spurs on one channel at the switching frequencies and harmonics, with max amplitude around -120dBm for one of the fundamentals, rapidly trailing down to much lower levels. Almost all that is picked up by amplifiers and filters in front of the ADC. On the other channel, I see three power supply fundamental spurs at -135dBm or less, and practically everything else is lost in the noise floor at -155dBm (this with a
10Hz resolution bandwidth). I believe almost all of what I see is being coupled in magnetically at this point. We did kill the 3.3V switcher noise by going to a linear regulator down from 5V for the critical analog parts; the very slight overall lowering of efficiency was well worth the performance gain.
I am measuring the current through an inductor using a shunt resistor feeding opamp and ADC, the inductor current is up to 55amps with 15amp max ripple at 100kHz. The ADC sampling is synchronized to always occur right before switching. Is it best to run long traces from the shunt to the opamp/ADC, or is it best to put the opamp and ADC as close to the shunt kelvin terminals as possible (and thus next to the large planar inductor).
Also I am measuring the output voltage of the inductor synchronized with the switching at 100kHz and filtered by a capacitor, using a resistor divider to measure the voltage and an opamp/ADC. Is it best to have the resistor divider close to the inductor output voltage in this case, or also move it away and have a longer trace for the inductor output voltage?
Ah, I had not realised the ADC was to measure the SMPS itself.
My experience has been mainly with smaller buck switchers. But quite low noise circuits, and the same principles should apply.
The magnetic field from the inductor will impose a voltage on any loops created by the PCB tracks, according to the area of magnetic field they see (and the geometry). And the magnetic field drops off very quickly as you move away from the inductor.
There will also be capacitive coupling from the big voltage change on the switching node. But the shunt is low impedance (yes?) so we only need to consider the magnetic (low impedance) coupling.
You need to minimise the length of the low-level signal tracking, which means putting the opamp right next to the shunt.
If you are filtering out the ripple, have a passive filter at the ADC which can then be far away.
Hey, I bet you could measure that inductor current using the magnetic field picked up by a "probe" track. You wouldn't need the shunt at all!
I made a little H-probe out of a 10mm loop of wire soldered across the end of a piece of coax. Connect to a sensitive scope input set to 50 ohms. If you have a prototype board you might want to make one and see what you find - it gives you a nice feel for the magnitude of the effects and how the field behaves.
In general you want to minimise the length of the low-level signal wires. So I would move the divider away. Although if it is high voltage there might be safety reasons why you would do the division locally instead.
Ya I thought that sounded kind of like a current transformer. I have to measure 60Hz current as well (too low frequency for a probe track I think, thanks for the cool idea though, I might put one under the ferrite inductor just to see what the scope shows :) (probably noise as usual on my cheap USB one)