We recommend eitherADA4899 or ADA4932 as a single-ended-to-differential driver and ADA4899 as a differential-to-differential driver (shown on page 18 & 19 in the AD7626 datasheet).
I am curious as to why the newly released AD8475 is not mentioned as a suitable driver for the AD7626.
I assume that for sequential step inputs into the ADC the settling time would be on the order 1 LSB or ~15ppm.
The AD4899 is specified for a settling time of 50nsec to 0.1% of 2V and does not seem to settle to a level required for 16 bit accuracy even from the graphs.
The AD4932 specifies a settling time of 9nsec to 0.1% of 2V and Figure 49 indicates it is considerable better at 20 nsec, but is hard to say if it is good enough for 16 bit settling, It does have better harmonic distortion numbers than the AD4899 ot AD8745 and lower noise density.
The AD8745 settles to 10 ppm in 50ns of the same 2 volt swing and would seem to be a better choice. Full power bandwidth is less but 15MHz would seem sufficient.
The specific application that this will be used for requires a sampling rate of 6.6MSPS and we can tolerate a 14 bit level of performance at that rate. We are not doing equivalent time sampling or anything that requires we exceed nyquist on the input.
The slew rate of the AD8475 is it's biggest limitation in driving the higher sample rate parts. Also see the bandwidth graph, figure 15, where for the full scale input range of the AD7626 (8.192 Vp-p), you're limited to ~ 2.5 MHz of signal bandwidth. In your application this device might be adequate.
Thanks. I looked at slew rate and was thinking single ended, so that clears things up.