Hi, I'm using the LT1461 4.096V version as an ADC reference for the AD7989-5. I have the output decoupled with a 10uF X5R 50V ceramic capacitor. Input power is +5V.
For my noise test my DC input voltages are
VIN- = 3.96V
VIN+ = 0.13V
I perform a burst of 256 readings 70 times and 4ms apart. I then overlay the 70 burst readings and look at my SD of counts to determine noise levels. On the past 2 revisions of this design the noise test data was always flat with a Gaussian distribution of noise. For this rev I noticed ringing over the duration of 256 readings and it happens during every one of the 70 bursts. The graph below shows the 70 plots overlayed. Please note I am reading all 18 bits and throwing away the bottom 2 LSBs.
Some probing around with a 1:1 probe revealed the ADC reference was ringing at the start of each burst. The oscilloscope screenshots appear below. The blue trace shows the 256 conversion pulses and the purple trace is the AC component of the reference output.
Adding another 12uF directly across the ADC ref and gnd pins dropped the ringing by about 50% but I was only able to completely eliminate it by adding a 4kOhm load on the reference output. I know other references such as the LT1460 require a DC load but I wasn't aware one was needed with the LT1461. Does this sound right or am I masking a problem?
Here's the relevant section of my layout. The next layer down is solid ground plane.