Questions about pipeline carlibration with LMS algorithm ADF

[windfire]

lms algorithm

I'm working pipeline carlibration algorithm with LMS ADF. now, my sigmadelta adc & pipeline adc work normally in matlab ,
but my carlibration algorithm's realization met some question,
some one can give me good suggestions?:?:
Thanks!

 

[n1cm0c]

carlibration

What is ADF ?
What is your implementation of calibration ?

 

[windfire]

Re: Questions about pipeline carlibration with LMS algorithm

least Mean Square (LMS) Adaptive Digital Filter calibration

 

[n1cm0c]

Re: Questions about pipeline carlibration with LMS algorithm

What is the error model you are calibrating against then ? Removing PN dithering from the decorrelated output ?

 

[windfire]

Re: Questions about pipeline carlibration with LMS algorithm

my errors contain comparator offset. opamp gain error,non-linear error ...
now,my adf realization met some question in matlab , do you have research about it?

 

[n1cm0c]

Re: Questions about pipeline carlibration with LMS algorithm

I did study the problem in the past.
I looked mostly at background calibration using PN sequences to dither the input signal and decorrelate the PN sequence out at the output. The LMS error model there is the suppression of the PN sequence, by adjustment of gain and dac error.
You are trying to correct also for gain error, like Murmann perhaps ?
Or you have a slow but accurate ADC in parallel with your fast pipelined ? This has been done in the past, and has been recently picked up again.

 

[windfire]

Re: Questions about pipeline carlibration with LMS algorithm

yes,i use a low speed but high accuracy sigma delta adc
I have some questions:
1.sigma delta adc's fs=f/n,pipeline adc 's fs=f,when sigma delta's osr=m.
the sigmadelta adc's bandwidth is fs/2/n/osr,but the pipeline adc's bw=fs/2,
when the data from sigmadelta according FIR, the signal (>fs/2/n/osr) filtered
do you think its a problem;
2.when pipeline adc's first stage output 2bit data, how do we realize it in adf?
i sent message to you,please check it!
THX

 

[n1cm0c]

Re: Questions about pipeline carlibration with LMS algorithm

I don't know if I understand your explanation correctly.
The sigma-delta ADC supplies the "error-free" conversion result. The LMS algorithm will use the difference between this value and the pipelined ADC one.
The aux ADC could have been sigma-delta or anything else, accurate but slow. In the case of the sigma-delta it needs to repeatedly sample the input (over sample it) to push down the noise, so in the implementations I saw I beileve that there was a sample-and-hold to store the input voltage until the aux adc finished the conversion.

If you don't have a sample and hold and your input signal BW is beyond the nyquist BW of the sigma-delta you are using, then it won't work.

Are you doing background calibration ?

 

[windfire]

Re: Questions about pipeline carlibration with LMS algorithm

yes, when input signal is sampled by pipeline adc (fs), the signal also is sampled by sigma delta with THA(fs), sigma delta's sample rate is fs/N . but i simulate model in matlab,it has no influence adding the THA. the signal (>fs/2/N/OSR) is always filtered by FIR.