DC/DC Converter Noise Troubles

[MosFett]

I'm building a battery powered audio amplifier which uses a DC/DC converter to generate a negative rail. My problem is that the converter I designed generates way too much interference in the rest of the circuit. I've spent a few days trying to figure out how to reduce the interference but with no luck. As you can see in the screenshot below, there is a lot of high frequency (about 50MHz) noise present when the converter's inductor (L1) is being charged. I see this noise show up in the rest of my circuit and could even pick it up in an oscilloscope probe (shown in the scope screenshot). Could anyone give me some pointers on how to reduce this interference?

The scope screenshot shows the negative rail (ie. converter output) on CH3 (Pink) and EMI I picked up using a scope probe clipped to its ground cable and circling L2 in CH1 (Yellow).

[palpurul]

Something like this happened to me a while ago. I designed a low noise differential amplifier powered by switching converters (+/-15V). I was getting a lot of switching harmonics at the output. No matter how heavily I filtered the output of the switching converters I was getting the same harmonics with almost same amplitude.

I ended up designing the power board seperately and piggybacked the amplifier board and power board with their bottom ground planes facing each other. That fixed my problem.

Desinginig two boards seperately is probably overkill (depending on your requirements ofc), but it worked for me.

Is that a PCB or are you breadboarding?

[T3sl4co1l]

Is it actually ~50MHz and not just pulses -- wavelets, ringing?  If so, I'm not even mad, I'm impressed.

Tim

[thinkfat]

It looks like you're hitting some resonance at 50MHz which the switching pulse of the converter excites. How did you decouple the converter from the rest of the board? Do you have a single ground plane for the whole system? Maybe if you isolate the switching ground from the rest of the system it gets better. Just tie them together at a single point, maybe you could try to couple the grounds by an inductor.

Does the interference pattern change if you actually put load on the negative rail?

Also, looking at your schematics - you have some pretty large capacitors there, 220µF and 100µF, what type do you use there? They look quite large for ceramic caps and your schematic shows them as non-polar. But if you do actually use electrolytic caps, be reminded that they have quite a huge inductive component in their ESR and also there's self-resonance. Electrolytic capacitors are good for maybe up to 1MHz or so and for higher frequencies they're sort of acting like an inductor.

[MagicSmoker]

Personally, I'd try a light RC damper (say, 100p + 10R) across the 1N5819 Schottky or replace it with a fast conventional PN diode. This is because the '33063 is no where near fast enough to excite a 50MHz resonance, but the Schottky is.

[MosFett]

I am using a PCB I designed, not a breadboard.

The noise looks like a pretty clean 50 MHz-ish sine wave (I'll add a zoomed-in screenshot).

The closest thing to decoupling I have between the converter and the rest of the board is the LC filter I have on the converter's output. I have one common ground plane - I'll try splitting it in two in my next batch of PCBs. What types of decoupling are useful for isolating converters (I'm sorry if this basic question - maybe you could point me to something to help me read up on this)?

Adding an additional load to the rail did not affect the interference (although I did get a bit carried away and overload my converter IC, releasing the magic smoke  . Time to make some repairs...).

I'll give your damper idea a shot tomorrow once I get my board back in working order.

[T3sl4co1l]

Is there... actually no input filter cap?  How much wire length is there to the next capacitor on +12V?

This isn't anything to do with switching circuits.  This is abusing the MC33063's output transistor driver as a negative resistance oscillator.  You can pull the same tricks with the -- normally reasonably stable -- three-terminal regulators, LM317 and such, by putting them on a resonant tank and transmitting AM radio (although I don't think they go quite this fast).

Tim

[MagicSmoker]

Is there... actually no input filter cap?  How much wire length is there to the next capacitor on +12V?
...

I totally missed that... 

[T3sl4co1l]

Yeah neither did I at first... or I'd just as well assume there's a big fat cap hidden somewhere else on the schematic that we're not seeing, since it's just a screenshot.

On that note, if 1.5nF is all that's there, it would be resonant with about 6.7nH at 50MHz.  Probably more than that in strays between ground, the shunt resistor, and whatever else is nearby, so that's probably not relevant.

Also, why is 1.5nF from CS to GND?  Isn't CS measured relative to VDD?  Wouldn't you filter it there..?

Oh and another note: this is a bootstrap configuration, so the IC's ground reference is relative to the negative output.  VDD bypass is returned through C15.

Tim

[thinkfat]

Regarding the 50MHz, it may not be a resonance that is excited by the switching pulse after all. It looks like a stable oscillation. If it was ringing, shouldn't it decay quickly after the pulse. But it appears to be present during the whole ON period. So, maybe there's something coupling into the feedback?

[MosFett]

Is there... actually no input filter cap?  How much wire length is there to the next capacitor on +12V?

There is a 220uF cap between VDD and ground less than a centimeter away from the MC33063 - sorry I forgot to show that in the schematic segment.

This is abusing the MC33063's output transistor driver as a negative resistance oscillator.

I don't understand how the transistor driver is acting as a negative resistance device. It seems like a higher voltage at the base would result in a higher current. Maybe I'm looking at it the wrong way - could you elaborate on why it's behaving like a negative resistance oscillator?

Also, why is 1.5nF from CS to GND?  Isn't CS measured relative to VDD?  Wouldn't you filter it there..?

I added the 1.5nF cap from CS to GND after I saw the oscillations because I didn't see much noise on VDD but I did at CS, so I wanted to try filtering CS. It helped reduce the amplitude of the oscillations seen at CS by about a factor of 5, although it didn't affect the amplitude of oscillations noticeably at any other node.

[magic]

Dunno, try a 100nF THT ceramic bodged right onto GND/VCC pins.

Regarding the current sense cap, the point was that this node is referenced to VCC, so it should be filtered to VCC rather than to GND, if at all.

[OldEE]

How much power do you need?  Can you use a charge pump for the negative voltage?

Does the amplifier need frequency response down to DC?  If not how about a single rail supply and and capacitors on the input and output?

Larry

[MosFett]

Can you use a charge pump for the negative voltage?

I was going to use a charge pump originally, but later realized it wouldn't work; I'm powering the device w/ a 9V battery and 12V DC jack (one at a time ofc), so I need the device to work all the way down to about 5V in order to use most of the battery's energy. I built a dynamic range compressor into my amp which needs a minimum of 12 volts to operate. To get -10V out with 5V in (so, 15V Vdd-Vss) I'd need two charge pumps. The problem here is when 12V is input, Vss would be -24V, giving 36V total out which is the absolute maximum rated voltage for my op amps. A 12.1V input could fry my circuit then. The MC33063 takes care of these issues.