an analog to digital converter problem

[Power boy]

I have a linear ccd signal input, the ccd analog output shift out
pixel's voltage at max frequency of 1MHz. According Nyquist the raw
signal's bw is 500kHz.
check page 6 in below:
http://www.zygo.btinternet.co.uk/tcd1304ap.pdf

I need to build an afe + adc circuits.

I want to increase the dynamic range. The spec says noise is 2mV. Who
knows the measurement bandwidth? whatever, just say, 2mV.

I picked an adc, super good, with bw of 16MHz. I want to do an AC
coupled, bandpass afe. I calculated the variable gain should be 0-60dB
or less. I want to limit my afe's bw, or limited harmonic for the ccd
analog inputs. Remove the inband white noise as much as possible. but
I want to consider the in band ripple will generate the uncertainty. I
want to control the in band analog uncertainty within 0.1%. How many
harmonics should I choose? To do a reverse Fourier transform might
sovle this problem but I am far from the textbooks.

I guess I need 500kHz * 15th = 8MHz. how about the ripple introduced
by the 1st to 15th harmonics for square wave?

 

[Power boy]

Congratulations.  You have made an erroneous "According to Nyquist" claim
that I hadn't thought to refute:http://www.wescottdesign.com/articles/Sampling/sampling.pdf.

Nyquist's limit says absolutely nothing about a signal's bandwidth going
into a sampler: it only says what the bandwidth must be for perfect
reconstruction.

In fact, the ultimate input signal to your CCD is spatial, and is as
bandlimited (or not) as the focus is blurry (or sharp).  In pixel-starved
applications folks usually want to have a blur spot that's smaller than a
pixel, to lose as little information as possible, even if it means some
aliasing.




Each pixel, taken as an individual thing, is 1us long and has 2mV or less
of noise.  The 1us sets some pretty firm bounds on the measurement
bandwidth.



What the heck do you mean by AFE?

If you want to sample the output of your CCD, you want to make the ADC
synchronous with the pixel clock.  You want to either sample at the one
best point in the pixel or you want to sample a burst around that point
and average.  What you don't want to do is to munge up the CCD signal --
for that you don't need to worry about harmonics, you just need to worry
about your amplifier/buffer's step response, to make sure that you're
getting plenty of the current pixel, and not much at all of the previous
one.

For 16 bits and 1us, I get about a 1.8MHz bandwidth, assuming one
dominant pole (11 time constants = 1us, if someone wants to check my
math).

thanks but you still didn't point out the bandwidth even this is
spatial. if this is spatial, I treated as sqare wave impulse, the bw
for this is still 1/(2*delta_T). To sample this ccd with big dc
offset, pink noise and white noise I want to remove the noise , dc
offset. However, a HPF introduces too much distortion due to the in
band phase delay. I can only use an AC coupled lpf. the pink noise is
still there. I couldn't just directly send it to adc which includes
too much noise. the white noise's spectrum density is flat that's why
I narrow the input bandwidth. I tested a little bit if I use ac
coupled, bessel or butterworth with cutoff at 16MHz it just introduced
rising edge delay. if I use HPF the signal is totally distorted,
within 1us the voltage level is tilting due to the phase delay. I know
the pink noise is still there.

my signal is about 5 uv with the dc offset 3V. noise is about 2mV. I
want to do 100 times averaging which improve 10 times snr. and I will
do other kind of filtering after digitized the signal.

 

[power boy]

thanks but you still didn't point out the bandwidth even this is

spatial. if this is spatial, I treated as sqare wave impulse, the bw

for this is still 1/(2*delta_T). To sample this ccd with big dc

offset, pink noise and white noise I want to remove the noise , dc

offset. However, a HPF introduces too much distortion due to the in

band phase delay. I can only use an AC coupled lpf. the pink noise is

still there. I couldn't just directly send it to adc which includes

too much noise. the white noise's spectrum density is flat that's why

I narrow the input bandwidth. I tested a little bit if I use ac

coupled, bessel or butterworth with cutoff at 16MHz it just introduced

rising edge delay. if I use HPF the signal is totally distorted,

within 1us the voltage level is tilting due to the phase delay. I know

the pink noise is still there.



my signal is about 5 uv with the dc offset 3V. noise is about 2mV. I

want to do 100 times averaging which improve 10 times snr. and I will

do other kind of filtering after digitized the signal.

I just changed detector and design approach after research a few days.

My current solution will be dual slope integration plus correlated double sampling technique. I figured out the noise source (readout, photon, dark) and some methods to remove them.
thank you all.