ADC question

readw · Joined: 16 May 2004 Posts: 4 Location: Leichhardt

Each time I read the ADC channel I get a slightly different value. ie if I read it 16 times I will get 16 numbers that are different by +- 1. If I read the channel 16 times then divide the answer by 16 and then repeat the process I will still get a different result by +-1 from the previous set of numbers. I am using this number to position a stepper motor and unfortunately because of this I am getting a result that is causing the motor to continually step on one direction or the other.

I have turned off the interrupts.

Any suggestions?

Thanks

Warren

albertsm · Joined: 09 Apr 2004 Posts: 5916 Location: Holland

You need to eliminate 1 bit of the resolution.
You can do this by dividing by 2, or shift the result to the right.

Another solution would be to use an external a/d converter with higher resolution and better stability.

What also could be a problem is the reference voltage. Did you tried various options?

readw · Joined: 16 May 2004 Posts: 4 Location: Leichhardt

Hi Mark,

Thanks for the response.

I am using a voltage reference source 4.096volts on the avcc port of an m323 and this appears to be working ok. Being a 10bit adc I thought I would get a reasonable conversion but maybe not. I have tried different settings in the config but the one below appears to be the only one that works

Config Adc = Single , Prescaler = Auto , Reference = Internal

The documentation suggsts that this processor should be able to accept different settings. Is this correct.

If I wanted to reduce the resolution is it best to convert the number to a binary value and then use the shift command? Are you able to post an example of this?

Thanks

Warren

davematt · Joined: 25 May 2004 Posts: 6 Location: Gippsland, Australia

I have sometimes got weird ADC results with the following solution. Read the channel No. one less than the one you want, then read your desired channel. This sometimes fixed the problem, I have no idea why but would appreciate any hints.

eddy · Posted: Fri May 28, 2004 1:17 am Post subject:

I read in the datasheet (or on AVRFreaks) that you should not use the first reading. You could try reading the channel twice and see if it has the same result as reading another channel.

DyneDude · Joined: 25 Apr 2004 Posts: 5 Location: Melbourne, Australia

I haven't used the AVR ADC channels for any real precision, so the last bit "wobble" hasn't been a problem. I have however found that with many commercial ADC boards (ISA and PCI) that I get better results by reading a channel twice, but ignoring the fisrt reading. You could also try accumulating, say, 4 readings and dividing the result by 4 (shift).

Jeremy Wempren · Joined: 18 Jun 2004 Posts: 8

the source gennerated by the MCS.lib indicated that it IS reading the adc 2 times and using the second result. more than likely the best thing to do is AND the binary result and kill the last bit.

eg. result anded by
1011011 1111110
would give 1011010

esentialy canceling out the last bit

JW

albertsm · Joined: 09 Apr 2004 Posts: 5916 Location: Holland

Reading twice is only needed for the older non-mega chips.
Most new chips do not have the error and only need 1 conversion.
But to be safe and compatible, you can do 2 readings as is the default in the lib.