Hallo Bogdan,

I understood datasheet as following:
Sampling phase is critical. Between machine cycle 8 and 9 input signal is not allowed to undergoe 'any' change. In this small time interval change of input signal must be smaller than 1LSB. That's the real important criterium. If you can guarantee, that in all cases this change is smaller than 1LSB, then you do not need any filter. But this would be an ideal situation, absolutely theoretically, almost never observed in real life. On ground connections alone you have ground shifts of up to some 100mV, only from voltage drop of fast currents accross non-avoidable inductivity. So, when 0V of analog circuitry is not identical to 0V of ADC of mcu then this criterium cannot be fullfilled without additional measure. Then you will always need some low pass filter: If 0V of ADC is shifted by ground noise, input signal is also shifted by the same amount, when filtering capacitor is connected from input of ADC to 0V of ADC (mcu). Only because of functionality of filtering capacitor! The result is, that input of ADC of mcu will not see this ground shift.

The demanding of slew rate control or minimum time constant of low pass filtering is only another pronouncing of this criterium. It's the answer of two questions:

1. If it is assumed that input signal looks like a ramp function in this small time interval, which is true for most signals, what then is maximum allowed change rate, means slew rate?

Answer: Mcu clock of 30MHz gives 400nsec for one machine cycle. 1LSB is about 5mV. So, maximum allowed slew rate is 5mV/400nsec = 12500V/sec = 12.5V/msec. Let's have some rounding and we have 10V/msec. Voilà.

2. If it is assumed that input signal makes a 5V step change, what then is minimum time constant of low pass filter that during 400nsec filtered voltage will change less than 1LSB?

Answer: About 400µsec, because 5V x (1-exp(400nsec/400µsec)) = 5mV. After some rounding we have 500µsec. Voilà.


Bye,
Kai