adc reading errors

How much variation are you seeing? In real life, a fast 16 bit ADC will never deliver constant codes.

What kind of amp are you using, and what's its gain?

Lowpass filtering between the amp and the adc will help, if you can stand some reduction in signal bandwidth.

What do you mean by "The Vref is connected to the Opamp output." ?

Is this one of those plastic breadboards? They're bad for this sort of thing. And separating analog and digital grounds is usually bad, too.

John

Hello! John, Thanks for your reply.

The amp is the instrument amplifier INA128. And I just put the +1.6V signal to its in+, and the in- connected to the analog V-. The Ref pin of INA128 is connected to the +2.5V after a Opamp. The Vout of the INA128 goes to +4.1V, which I measured via a multimeter to confirm it too. Gain=1. I did put 10Ohm and 1nF low pass filter for the ADC input signal.

Lots of variations I see, not just 1-3 LSB variations, even the MSB too. not sure if the uP read the code correctly or because of the circuit noise etc.

What do you mean by "The Vref is connected to the Opamp output." ?

--- The Vref of the ADC need a low impedance.

They are plastic breadboards. What should I do? Make a PCB instead of working on the breadboards?

Would the noise make such big difference on every bits?

Cheers,

B-B claims 3 lsb's p-p noise, which is pretty impressive. The INA128 will add seome of its own, but nothing radical. If you're seeing codes all over the place, something is bad wrong.

Try grounding the ADC inputs and see what you get. You may have a clocking problem or something. How are you getting the codes out of the adc?

I usually lay out a board to do anything complex, but this case is simple enough that you could still do it by hand. I like to get a sheet of bare copperclad FR4 and build breadboards directly on the (single!) ground plane, with axial passives, wire, and little commercial adapter boards to break out surface-mount parts. That can do very low noise stuff.

John

You could be reading the serial interface incorrectly, for example reading each bit on the wrong edge of the clock so that the data is changing at that instant. SPI interfaces have a "clock polarity" and a "clock phase" setting; both these must match the ADC.

Look at the serial output on a scope and see if you still see the MSB transitions then. I bet you don't. But if you do, remove the amplifier and try shorting the ADC inputs to ground. Check you have the correct supplies and reference voltage (you could connect the reference to the positive supply if unsure).

I would use the "dead bug" prototyping method on plain copper clad PCB material. (Or make a real PCB). You may also be able to buy an evaluation board for the chip from the manufacturer, use it as a starting point.

No.

Thanks, John. I tried a ground plane on a sheet metal, which is actually the base of one of the breadboard I have. Now what uP got from ADC were almost all zeros, as I had before. I got all the ADC input zero when I use the C compiler build in SPI function, and I thought it might be something wrong on the code, I then changed the code to something like the following, not using any built-in function. Afterwards I was able to read in different wrong ADC input every time. Then I seperate the analog and digital ground, not much help. Now I added in a metal plane as a ground plane, and the readings are almost all zero.

Each instruction takes about 0.25usec, i.e., the ADC sampling rate is about 4M, below its max. 6Mhz.

output_low(ADC_CS); //Enable ADC for (i=0; i

OK, but what's the electrical setup? Is this a PC or a pic or what? What sort of electrical port are you using, and how is it physically wired to the adc? A little clock ringing or such could be bad. Have you scoped the digital lines?

John

Hi! John, It is a PIC18F452, and the PIN_C3 and C4 (SPI/SCLK/Din) are used for sclk and data input. PIN-B7 is used for CS pin. Flying leads are used to physically connect them. I haven't got a scope on hand now.