I'm using an AD7984 to capture the output of a dual slope integrator circuit. I'm using it in /CS mode, 3-wire, with no busy indicator. Sampling at 500 KHz. Vref = 3.3V. SCLK runs at 12 MHz. It's a 16 channel system, but I'm using 16 devices, so they all operate in parallel. We are dropping 2 LSB's to make it a 16 bit system (I know - what a waste of good bits!). We're using it in single-ended mode (long story), and driving it with a MAX4452, with an RC filter at the input using 18 ohms/1 nF.
The portion of the waveform I need to capture is a short period of constant voltage roughly 320 nS wide, which comes at the end of a slope down from 2.2V. I'm having trouble getting the correct output codes on a number of channels, though not all. I can apply DC voltages of about 2.2V at the input and get the correct output codes. Also, shorting IN+ and IN- gives output codes around 1090, which I guess makes sense and together these would seem to prove the ADC is functioning correctly for DC signals.
But when the dual slope integrator is running, five of the channels produce output codes in the range of 15,280, with a few bits of noise, even though I can find no evidence of signals at that level. I've put the scope on Vref, VDD and VIO and they're all clean, especially Vref.
Here's my question: The data sheet says the acquisition time is 250 nS. But what exactly does that mean? My CNV pulse is 500 nS wide, so it stays low (acquisition mode) for 1.5 uS. But does the acquisition time spec mean that the analog input has to stay constant for 250 nS before the rising edge of CNV? If that's the case I'm probably violating that spec, but we've tried stretching that constant voltage area, with no resolution of the problem.
The scope traces I've attached show the signal on top, CNV in the middle, and SDO on the bottom, for the 'running' case, and for the case where the input is shorted to ground.
Any help you can provide would be appreciated!