I'm trying to power an AD8052 opamp with a +/- 5V power supply. The positive +5V supply for the op amp is generated by an LM317 voltage regulator (the input is a 9V battery). While, the negative -5V supply is generated by the ADM660 whose input is the LM317's +5V output. The ADM660 negative output directly powers the negative supply of the AD8052 op amp (see schematic below).
Now, when I view the ADM660 output on a scope i get a pretty stable -5V voltage. However, if I probe the +5V input voltage I get quite large switching noise (order of 150mV and frequency is approximately 44khz). If I disconnect the ADM660, the noise on the +5V line disappears. The 10uF capacitors are tantalum low ESR as recommended by the datasheet. Placing a large electrolytic cap between the +V input and GND of the ADM660 actually reduces the noise a bit but it is still significant.
1) Is this a normal phenomena with those kind of switch capacitor voltage inverters or am I doing something wrong?
2) What can be done to reduce the noise amplitude?
3) If bypass caps are the way to go, what values would you recommend (I already have three caps on the +5V line which are placed about 2cm from the ADM660) ?
If it makes any difference, the board is a 4 layer one and the +5V input to the ADM660 is routed through a via to a +5V plane.
Thank you in advance,
Whilst my area is really inductor based power supplies, I thought you may be interested in the article here: http://www.tnt-audio.com/clinica/regulators2_impedance1_e.html This looks at the impedance of the LM317 across frequency and the anecdote at the end may be a similar problem to the one you are having - the LM317 may be oscillating due to the low esr input cap on the ADM660.
There are a few things you could try to help
- Add a capacitor into the LM317 feedback loop to improve the low frequency performance
- Ensure there is a continuous load (the output impedance of this type of regulator tends to improve under load)
- Add a ferrite bead between the LM317 and the input cap of the ADM660 - it may help reduce high frequency noise
You may also want to try changing the ADM660 switching frequency - I believe you can either set 25kHz or 120kHz. One frequency may give better performance for you
This is a common question that I think we can help you with. Consider the two phases of the inverting charge pump operation:
Phase 1: C1 (the flying cap, which is C38 on your schematic) is applied in parallel to the input cap (C35 on your schematic). So C1 charges to Vin.
Phase 2: C1 polarity is reversed and C1 is applied in parallel to Cout (C31 on your schematic). So Cout charges to -Vin.
During Phase 1, C1 must be charged from the input caps (C38). And if those input caps don't have a good charge path from their source (LM317 Vout), then the load on them that C1 represents (do to the change in input capacitance-->C1||Cin) will cause a drop in the input voltage during phase 1. If Cin=1uF and C1=10uF, then the new input capacitance is basically increased by 10x during Phase 1--requiring a lot of current and time to recharge Cin during Phase 1. Does all that make sense?
There are a couple of easy fixes:
1. Change the switching frequency to 120kHz and decrease the value of C1 to 2.2uF. This will dramatically reduce the load that C1 places on Cin.
2. Add more input cap. I think the Cin is far too low (its only about 1.1uF now). Try 10uF or so. Then when C1 is placed in parallel with Cin, the change in input capacitance will be much smaller.
3. Change to ceramic capacitors. Unless your application requires the use of tantalums, I highly recommend switching to ceramics. The ceramic capacitors have a much lower ESR (then both tantalums and electrolytics) which will help tremendously in charge pump applications. I would recommend switching Cin, C1, and Cout to ceramics.
I hope that helps. Let us know if you are still having trouble or if you need anything else.
I would also add that the 10uF Jon is recommending should be placed right next to pin 8 and oin 5 of the ADM660. C35 is probably too far away from the ADM660 to really keep the source impedance low.
Thank you all for your help. I've tried some of your suggestions and here are my results:
1. What seems to have the greatest influence is the LM317's output bypass capacitors. First I changed all three of them to 10uF so the output capacitance is 30uF. This drastically reduced the noise - see photo below:
2. Next I increased the frequency of the ADM660, this further improved the LM317's noise levels - see photo
3. Next I tried to change the ADM660 capacitors to lower value ceramic ones. However, this only made things worse (maybe the ESR was high - I didn't have the data for the cearmaic caps).
4. Adding an output load resistor to the LM317 and the ADM660 seemed to have no effect at best and sometimes even increased the noise levels.
5. Also, adding a capacitor in parallel to the LM317 feedback resistor had no noticeable effect.
All in all, with the original 10uF tantalum caps and high frequency the +5V rail looks pretty good.
However, the ADM660's -5V output is acting up (it is possible that I've missed this before but I'm pretty sure that the -5V line was stable prior to all the above changes):
As you can see the output has sharp 50mV - 25ns pulses (which are heavily coupled to the op amp's output).
1) Do you have any suggestions for reducing the output noise?
2) Am I wrong to expect a stable output from an unregulated switch capacitor converter?
Looks like you've made a lot of progress so far. Here's a couple of comments in response to your results and questions:
1. Ceramic caps: I'm surprised that ceramic capacitors don't help your ripple--especially for the output cap. Ceramics have much lower ESR's then tantalum. Try using the same value (10uF) the tantalum ones and see if that doesn't help.
2. Vout ripple: The short spike in the Vout ripple is due to parasistic inductances. Probably in the output capacitor or in the trace to the output capacitor. Ceramics usually have a lower ESL than tantalums, so that may help with the parastic inductance spike. Or you can always add a small LC filter to the output.
3. Stable Output: The output of an inverting charge pump is alway "stable." Meaning it does not oscilallate because it has no feedback path (it is not sensing the voltage and adjusting any of the power delivery). So it is never unstable. However, like all switching converters, it will have output voltage ripple, which is normal.