RP2350 ADC INL/DNL

[greiman]

I am looking for other users experience with the Pico 2 ADC.

I found lots of noise when using USB power. I followed a suggestion and put a LM4040C30 3.0 volt reference from the Vref pin to ground. That improved the noise problem.

The DNL/INL problem is better but still not great.

I used a 16-bit AD5693 DAC to generate ADC input. I stepped from zero to 3.0 V in one millivolt steps. I get this error chart for the ADC at 12-bit resolution. I subtracted a large zero offset and adjusted for the 3.0 V reference.

The three big jumps are at these multiples of 512: 1536, 2048, 2560.

This is better than the RP2040. Here is an early example for a Pico 1. It has noise since I didn't use a voltage reference chip. It goes to 3.3v since 3V3 is the reference.

I checked the DAC performance with a 6 1/2 digit DMM. Here is the result:

[earlephilhower]

Very interesting results. There's a minimum of 9 ENOB per the dataset, but that looks more 7.5 if my back of the envelope math is correct.

Have you opened up a discussion on the RPI forums? The actual designers are active there and might be able to offer assistance.

[greiman]

Have you opened up a discussion on the RPI forums?

I see the claim of 9 ENOB and I looked at the forums but found nothing.

The latest datasheet still has this so I am waiting for more info.

12.4.5. INL and DNL Details to follow.

I really like the RP2350 but often need an ADC for my projects but it looks like strike two for the ADC.

[greiman]

Here Is a SAMD21 ADC with Vref the internal 3V3.

Better accuracy between 0.3 and 3.0 V.

[earlephilhower]

Microchip has had a few more years than the RPI folks getting the analog portion right. Analog ASIC design is always a dark art practiced by wizards. :laughing:

Just spitballing, but is it possible to design a transfer function to get you closer to what you need? If multiple chips are repeatably off in the same way, is there a 4K entry lookup table that can map from ADC readout to actual physical value? I know you can't add information but maybe you can shift the errors around?

[greiman]

Just spitballing, but is it possible to design a transfer function to get you closer to what you need?

Probably a quadratic fit would be pretty good for the RP2350. I don't like the DNL jumps that many CPU chips have.

I have lots of experience with ADC chips so if I need more accuracy I add a chip. I have written drivers for a number chips.

For high accuracy I like the MCP342X family of 18 bit ADCs. They have a builtin reference and INL: 10 ppm of Full Scale Range.

There are lots of good SAR 12 bit chips. For many years I have used the MCP320X family.

[greiman]

Here is an example of a 12-bit MCP3202 with a Pico 2. I just put the ADC and DAC chips on a bread board with no caps so there is a lot of noise.

Most reading are within a millivolt. Since the least significant bit is 3000/4095 = 0.73 millivolt this is about as good as you can expect.

I use three parts, MCP3202, 3.0 V reference, and a 220 ohm resistor.

[greiman]

I have been trying to reduce the noise on Pico 2 3V3. The schematic has a note that I can run the RT6150B Buck/Boost converter in the default PFM mode or the alternate PWM mode.

I loaded this program:

void setup() {
  pinMode(23, OUTPUT);
  // digitalWrite(23, HIGH);  // PWM Mode
  digitalWrite(23, LOW);  // PFM Mode
}
void loop() {
}

And I get this trace on 3V3 note the time scale is 20 μs/division.

I get this in PWM mode at 500 ns/division.

Still not great.

I tried an Adafruit Feather RP2350 with this program:

void setup() {}
void loop() {}

And get this on 3V3:

Looks like I should use an Adafruit board for analog apps.

Anyone checked other new RP2350 boards?

Edit: I am using a scope probe with with a typical ground lead so it picks up some noise since I have lots of computers and other equipment in my lab.

Here is a trace with the probe tip touching the ground clip. I am impressed with the quality of the products Limor Fried creates at Adafruit.