john2020
Full Member level 5
ADC averaging techniques
hi all,
Can anyone tell me where there are some good resources on averaging
integrate-and-dump' oversampling techniques for ADC's?i got some information,which am sharing with you here..
ADC circuits are not hard if you don't need much accuracy and don't push to
get the max from your converters. But getting the last bits of performance
from any ADC takes a lot of work and attention to detail.
What you show here is not an N sample average. This is an IIR filter. If
you analyze the response to an impulse input you will see that the value
never goes away completely until the resolution of the calculation allows
it to fade into the noise.
This is a perfectly good filter for the right application, but I don't
think a scale is necessarily the right app. I would suggest that an output
rate be selected R along with an oversample rate, N. The ADC should be
sampled at a rate of N * R. N samples are averaged giving one output at a
rate of R. This will still have some "noise" in the output, but it should
eliminate most if not all of the ADC noise. It will not correct for ADC
errors such as INL.
//Alpha is value between 0-1
y=newvalue();
x=x*Alpha + (1.0-Alpha)*y;
//x is now your new value...
Imagine that you read a new value 10 times a second. Then if Alpha is .1 you
would be averaging 10 samples.
if anyone got more ideas here,pls do share with me
hi all,
Can anyone tell me where there are some good resources on averaging
integrate-and-dump' oversampling techniques for ADC's?i got some information,which am sharing with you here..
ADC circuits are not hard if you don't need much accuracy and don't push to
get the max from your converters. But getting the last bits of performance
from any ADC takes a lot of work and attention to detail.
What you show here is not an N sample average. This is an IIR filter. If
you analyze the response to an impulse input you will see that the value
never goes away completely until the resolution of the calculation allows
it to fade into the noise.
This is a perfectly good filter for the right application, but I don't
think a scale is necessarily the right app. I would suggest that an output
rate be selected R along with an oversample rate, N. The ADC should be
sampled at a rate of N * R. N samples are averaged giving one output at a
rate of R. This will still have some "noise" in the output, but it should
eliminate most if not all of the ADC noise. It will not correct for ADC
errors such as INL.
//Alpha is value between 0-1
y=newvalue();
x=x*Alpha + (1.0-Alpha)*y;
//x is now your new value...
Imagine that you read a new value 10 times a second. Then if Alpha is .1 you
would be averaging 10 samples.
if anyone got more ideas here,pls do share with me
