fluctuating reference voltage of ADC

[saurabh17g]

I am using ADC - ADS805 with a sampling rate of 5MHz. The response of the supply is stable and the ripple is in millivolts. But when the circuit with clock (5MHz), RAM , counters , ADC is used, the vcc fluctuates with a ripple of 0.5V. Due to this , the internal reference pin voltage(vcc/2) of ADS805 fluctuates. This results in spordic output of ADC. we have connected a standard tectronix powersupply whose current rating is upto 10A.
The power supply output fluctuates only at the edges of the clock.The fluctuations are high at the rising edge which is used by both adc and the counters and low at falling edge which is used only by the adc.
Plz suggest me

 

[karesz]

Hi,
it seems me to be a bypassing& /or GND problem...
Did you applayed the "FIGURE 5. Recommended Reference Bypassing Scheme" & "FIGURE 12. Recommended Bypassing for Analog Supply Pins"?
K.

 

[fcfusion]

I agree with karesz. It seems like a Ground problem to me. That sort of thing happens when, for some reason, the ground is not the same for every component.

What is the shape and frequency of ripple?

 

[dick_freebird]

I would begin by using the lowest-ESL bypass caps possible on
all the digital "stuff". Get the current loops to not involve the
larger ground network.

Digital outputs driving off-board can be a big slug of charge.
Series resistors before the connector might help that a lot,
so long as the remote end can tolerate degraded edge rate.
You might be able to "source terminate" and call it nearly
a freebie.

Running a polyphase clock scheme might also help you - that
is, say, run a 4-phase clock and divide down such that the
ADC rising edge is phase 1, AC falling is phase 3; logic
clock rising is phase 2 and logic falling is phase 4. By this
skew, you would pick up a quarter cycle worth of supply
glitch settling window. If the perturbation is really impulses
then maybe this helps significantly. Or even lagging out the
digital clock so its edges hit just after sampling of the analog
quantity.

But I think cutting the primary "kicks" is where to begin.

 

[karesz]

Hi,
I found (randomly) these script yet a half hour ago ...
K.