If you're referring to what I think you are (the AI Lowpass Filter section of the M Series User Manual), the 750 kHz cutoff refers to the 628x devices. For the 625x, the specifications actually state that the -3 dB cutoff point is 1.7 MHz. While this might not seem desirable given the card's max sample rate is 1.25 MHz, having the higher bandwidth reduces the settling time (very useful considering the channels are multiplexed into a single ADC).
So you clearly can't avoid aliasing if you have higher frequency content on the input terminal. If there are high frequency components to your signal that are giving problems when sampled, I would suggest using an external lowpass filter.
Somebody has suggested incorporating configurable anti-aliasing filters in the DAQ idea exchange, but for the reasons mentioned in that thread I don't see it happening:
1. Having a fixed filter chosen for the maximum sampling rate of the device only works when the device is operated in single channel mode and at the highest rate.
2. How should NI decide what filter characteristics to use? Filter type, number of poles, ... The worst case filter would probably add a considerable amount of expense but would be overkill for a large majority of applications of the device. And it would not be appropriate for use at lower sampling rates.
3. Users who do not understand aliasing (and there seem to be quite a few of them), might be fooled into thinking that they did not need to worry about this if a built-in filter existed.
4. Adaptable, non-sampled data filters are not easy to design and build, especially if they need to be tuned over a wide range as would be required to implement this idea. Sampled data filter, which can be quite adaptable, have the same aliasing issues as the A/D converters. They cannot be used as anti-aliasing filters.
5. Filter settling time would be added to the multiplexer switching and settling time. This might significantly reduce the maximum achievable sampling rates for some multiplexed systems.
6. For some applications intentional undersampling is used. Any device with built-in filters would need to have an option for disabling the filters for such applications.
Best Regards,
John Passiak
