# repeated data conversions

• posted

I was wondering just how bad data is corrupted through repeated conversions. I have this idea to test it and was wondering how hard it would be to set up.

Essentially you have a ADC, DAC, buffer, and control system.

you initially setup the buffer with the required digital data.

Then the buffer sends the data into the DAC and the DAC feeds the ADC which then goes back into the buffer. The control system handles all the necessary stuff and counts each time a full conversion of the buffer has happened.

diagrammatically,

DAC -------> ADC ^ / \\ / \\ / \\ / \\ / \\ V Buffer

After the nth conversion(of the buffer) you can then compare the buffer with the original data and see just how different it really is.

I'm wondering if such a system could be designed to handle virtually and DAC and ADC for testing purposes?

Any ideas?

Thanks, Jon

Loading thread data ...
• posted

"Information theory" says that every conversion gives you error of the least significant unit of measurement. No arguments on this! How many repeats of it will give you recognisable error from original depends on your setup. One way is to define the ending error and let the system repeat those cycles, counting them, till it gets there.

Have fun

Stanislaw.

• posted

Well, I don't know if thats information theory but the error is obviously in the least significant bits because that is how the devices work. The issue is how the error propagates. Also there is a difference between the mathematical representation of the error and what actually happens. i.e., just cause you assume it is guassian distributed doesn't mean it actually is and I'm interested in what really happens rather than what happens mathematically(not that it can't be described mathematically but its much more complex that I care to deal with).

Well, I want to actually see what happens... ideally I would take a snapshot after each step and I could then have a time series of how the error propagates.

The reason I'm interested in this is because I have some things I'd like to do but they require going in and out of the digital domain many times and I'm not sure the signal will ultimately be affected to due the conversion process. For example, if the signal is dithered then how will that affect the outcome? with errors from the conversion process creep into significant bits?

Jon

• posted

Your guess is as good as mine, the converter cannot close one eye and/or tilt its head to "better" read the displayed value. It will simple guess depending of the size of the known to it smallest digit. For this accuracy the converter IS built, you want better accuracy use more smaller units in conversion or use statistics. Nothing is absolute. "Significant" means what percentage of the whole? On how many digital digits on what accuracy of the analog conversion against what reference. When looking for creeping errors it is good to have _very_ reliably performing setup.

Good luck

Stanislaw.

• posted

If you wanted to model the basic parameters (# bits, AWGN or other noise) of the system, you could probably model it in Simulink pretty easily. If you actually want to test physical parts, you could still use Matlab/Simulink to provide the original data, and to recirculate the data in digital form using the data acquisition toolbox, putting the current digital state into successive dimensions of a variable. Then you could plot the whole thing pretty easily. You'd just need to attach a supported digital I/O to the computer. Try a new ADC/DAC pair? Just hook them up and press run again.

-- John

• posted

On Fri, 06 Jul 2007 20:19:16 +0000, Jon Slaughter wrote: ...

When I was a kid, we used to do this with a tape loop. We'd say something, and it would echo, and echo, and echo, getting noisier, and noisier, and noisier until it was like listening to ocean waves.

As far as your scheme, the only errors will be in the analog part.

Have Fun! Rich

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.