KV46 ADC accuracy issue

[tckwok]

We are now using Kinetis KV46 series controller and want to have all ADC results with less than < 4 counts swingings, given a very stable source.

In our setting, all of the ADC ch are enabled with ADCA & ADCB ch 0,1,2,3 configured as differential and the rest single-ended. Then,

we supplied a 3.0V to ADC VRef_H and a sample 1.2V to one of the differential ch to check how accurate we can do, after making sure the stability of the power source is as fine as 1mV level with a 6.5 digits DMM. Yet we observed the ADC reading is swing at around 8~15 counts. Therefore, we tried to follow a few steps from the article " how to increase the ADC converter accuracy in an application " provided by NXP, like section 4.2 < minimise the ADC errors related to the external environment: ( to name a few moves we did)

1. adding differential filter with R = 180ohm , C = 8.2nF under our current 50kHz sampling

2. even shut down the PWM module to avoid cross-talking

3. ensure a stable power and Vref_H source

4. those PCB layout techniques

and we repeat with test with the EV board : TWR-KV46F150M 

and got the same 8~15 counts swinging.

Q : how can I reduce those counts to < 4?

Is it caused by a wrong set-up? or what else can we do?

p.s we treat averaging technique as last resort because it wastes our sampling resource

Thank you so much for reading this and helping us!!!!!

<how to increase ADC accuracy >

https://www.nxp.com/docs/en/application-note/AN5250.pdf 

[myke_predko]

tckwok@asmpt.com‌,

Could you:

1. Explain the application in more detail. It's not that helpful saying what you "want" without explaining the application and what is being measured.  
2. Have you put an Oscilloscope on your power supply as well as VREF_H and VREF_L and the incoming line?  I'd be hesitant to depend on a DMM only to indicate that the line is clean/noise free (especially considering that the DMM connections add capacitance/inductance to the circuit which will help filter things).  When you put on the 'scope probe, please note the input and line impedance and explain how you are grounding it.  
3. Describe how you have wired your board/ADC (pictures of the schematic and PCB would be helpful).  I'd also be interested in your power management and clocking.  Your ground plane needs to be explained as does line impedance as you're implying that the sample rate is fairly high.  

Based on what you've written, a 4 count "swing" means that with a 3.3V source there is a 3mV or 0.1% accuracy which is going to be very difficult to achieve, especially in an MCU where lots of things are happening (not just the PWM).  Grounding between the MCU and the source of the signal (and oscilloscope) you're measuring is probably going to be the most critical parameter (followed by power supply stability/noise).  The PCB layout/wiring instructions in the referenced AppNote are good but probably not sufficient to get the accuracy you want.  

You're getting around 10mV accuracy which is what I would consider very good for any MCU based solution - you will be stretched to improve that to the level that you are looking for.  Most likely, you will need to hire a circuit design/PCB layout consultant with the skills to bring the noise down/accuracy up to the levels that you want - look for somebody who has RF layout experience (with embedded antennas) as a starting point.  

myke