As there is no PSRR data for overclocked ADCs i spend some time
and measured it by myself (PSRR of AD9288 under overclocking condition).
As far i can see the PSRR is only about 25db for 50MHz signals (ADC clock 125MHz)
and only 18db for 200MHz signals (ADC clock 125MHz).
This means, with the measured 18mV PSU ripple (before my mod) the ADC error
is about 1mV with 50MHz and 2.25mV with 200MHz signal which is
in worst case about 1/2 LSB of the AD9288.
So why any mods ? Well, the original design (with overclocked AD9288)
was done by Instek, these guys calculated everything proper, no issue at all.
Tekway's design is working too - on paper at least. In real world
we have not a single ADC but at least 2 physical ADC die's, so in worst case
the error need to be multiplied by 2.
(Instek's original design have better power supply, so still no issue at all for Instek DSOs).
This means with input freq. starting from 150MHz the error is above 1/2LSB
and finally about 1 LSB with 200MHz input signal.
1LSB is maybe not much, but i don't like it, therfore i did modified the PSU a littlebit
to improve a bit 5V / 3.3V supply.
The mod is very simple:
4700/10V instead of C19 (3300/10)
1500/10V instead of C14 (1000/16)
3 x 47/6.3V SMD on the bottom side instead of C22 (330/10)
and DC/DC converter (so the PSU is cold now) instead of KA378R33.
The part numbers are:
2 x EEFCD0J470R (Farnell/Element14 9694714) - that's the 47uF SMD caps soldered on bottom side of PSU pcb.
1 x PTH04070WAH (Farnell/Element14 1470479) - dc/dc converter
2 x GRM32ER71A476KE15L (Farnell/Element14 1797018) - soldered as close as possible to PTH04070 pins
1 x EEUFM1A152L (Farnell/Element14 1855213) - 1500uF
1 x EEUFM1A472L (Farnell/Element14 1219459) - 4700uF
The Ti DC/DC sonverter is recommended for Altera FPGA designs, so it will
definitely work better for fast transient /high load as the KA378R33.
After all i did measured the ripple, it is now about 11mV (instead of 18mV) for the 3.3V
and 61mV (instead of 76mV) for the 5V.
The attached 3.3V ripple measurment is again with digital/analog filter, so you can see only 6mV
Vpp ripple, but on good analog scope it is in real 11mV.
This is maybe not that much, but now in worst case the ADC error is a little bit higher that 1/2LSB.
you can compare to what was before :
https://www.eevblog.com/forum/index.php?topic=1571.msg41294#msg41294Total price of the mode, something about 16USD.
You can of course do it with combination of TPS73701/TPS7A8001 (to produce clean 3.3V from 5V)
but my goal was to reduce the 5V ripple too and replace the unnecessary "heat-engine" (KA378R33).