LPC1769 ADC in burst mode peaks every now and then

[lpcware]

Content originally posted in LPCWare by robert hulsebos on Mon Jul 14 00:49:34 MST 2014
We use the ADC of the LPC1769 in burst mode (no interrupts, no DMA). Just continuesly convert 4 stable voltages (4 channels) and read them at random periods using their indivual data registers. Not caring about OVERRUN.
Every now and then (may take minutes or even hours to reproduce) the AD0[1] conversion leads to peak value that is not plausible. (value outside range of possible values, but no 0xFFF)
VDDA and VREFP are bound to stable external ref voltage. Our oscilloscope never triggers on AD0[1] exceptional value, so no peak is there in hardware. Seems the ADC is seeing ghosts??

Anyone had similar experience?

[lpcware]

Content originally posted in LPCWare by MarcVonWindscooting on Thu Jul 31 07:27:08 MST 2014
Hey Robert,

yes, I had similar problems. I found the following article very helpful. It's for a different uC, but the findings are probably applicable for most other uCs:

http://embdev.net/topic/stm32f0-adc-input-noise-investigations#new

A few suggestions:
[list=1]
  [*]What do you mean by 'not caring about overrun' ? Maybe it's only a software/race condtion problem?
  [*]Are you checking with the scope and the ADC finds a ghost peak at the same time?
  [*]Are you sure, that the voltages are in range on all ADC inputs? In my case there was some (high impedance) noise despite 100nF in parallel to the ADC input on one channel. I was able to cancel that by enabling the internal pulldown of the pin although I'm not really sure if that is allowed ;-) I should have added a pull down in the first place....
  [*]Are your inputs signals of low impedance? Otherwise add a capacitor in parallel.
  [*]Is your reference voltage free of noise? On the other hand, in my current project (LPC1754, motor control) I use a LM5009 (switching regulator!) for generating 3.3V for the LPC and I use a ferrite and a decoupling capacitor to derive VREF and I'm really happy with the results!
[/list]