Cancellation of reconstructed digital signal

Hello,

I have an instrumentation amplifier with a gain of 100. 
The positive input (AO) is connected to a DAC. The negative input is taken
from a previous analog stage and also connected to a ADC --> recorded
signal (AI1).

I put a replicate of the recorded signal (AI1) at AO. However, I got an
amplified signal of quantization, noise, etc.

I want to ask which is the best way to reconstruct the analog signal. Is
wavelett algorithm to remove noise is good enough? Any other method
available (ICA, etc)?

Thanks a lot.

alexia wrote:
> Hello,
> 
> I have an instrumentation amplifier with a gain of 100. 
> The positive input (AO) is connected to a DAC. The negative input is taken
> from a previous analog stage and also connected to a ADC --> recorded
> signal (AI1).
> 
> I put a replicate of the recorded signal (AI1) at AO. However, I got an
> amplified signal of quantization, noise, etc.
> 
> I want to ask which is the best way to reconstruct the analog signal. Is
> wavelett algorithm to remove noise is good enough? Any other method
> available (ICA, etc)?

This makes no sense -- if you already have the signal recorded, why do 
you need to measure it?

It sounds like your 'real' problem is that you are successfully 
amplifying quantization noise, and now you want to get rid of it.  The 
problem that you are going to run into is this: for a filter to work 
there has to be something that distinguishes the two signals that you 
are trying to separate.

"Straight" quantization noise is not all that stationary -- sometimes it 
has all it's content at DC, sometimes it's pretty well spread over the 
spectrum.  Because of this any linear filter (and to my knowledge 
wavelet filters are linear) is going to have a hard time consistently 
distinguishing quantization noise from signal.

It's like pouring beer through a towel, hoping that water will come out 
the other side: the ingredients all 'look' the same to the weave of the 
towel, so it can't work as a filter.

All you can do is to change your quantization noise.  _If_ the recorded 
version of your signal is not where the quantization noise is then you 
are stuck with improving your DAC.  This may mean that you have to 
revise your hardware with a higher precision DAC.  If your sampling rate 
out to the DAC is generous enough, you may be able to dither the signal 
going to it -- see http://www.wescottdesign.com/articles/sigmadelta.html 
for details, or some other reference on sigma-delta modulators.

-- 
Tim Wescott
Control system and signal processing consulting
www.wescottdesign.com

alexia wrote:

> Hello,
> 
> I have an instrumentation amplifier with a gain of 100. 
> The positive input (AO) is connected to a DAC. The negative input is taken
> from a previous analog stage and also connected to a ADC --> recorded
> signal (AI1).
> 
> I put a replicate of the recorded signal (AI1) at AO. However, I got an
> amplified signal of quantization, noise, etc.
> 
> I want to ask which is the best way to reconstruct the analog signal. Is
> wavelett algorithm to remove noise is good enough? Any other method
> available (ICA, etc)?

No matter what magic spells do you cast (ICA, wavelets, fuzzy logic, 
genetic algotithms, turbo codes...) it is impossible to get cancellation 
in this setup because ADC-DAC pair introduces delay.

If you want to entertain your professor, put a FIR filter between ADC 
and DAC and update it by LMS stochastic gradient.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com