Hallo Mahmood,

I have drawn a schematic for you, implementing an analog idea, if you want to proceed in this way. It's similar to that, what Charles proposed.



From picture you can see, that sensors of all four wheels are used. First RC filter suppresses noise. Schmitt-trigger input inverter follows and produces fast and stable trigger signal for TLC556, which is a CMOS version of dual precision timer 556. TLC556 is connected for producing very stable high pulse, means TLC556 works as monoflop. High pulse duration is Tx = ln2 x Rx x Cx. If now speed increases, duty cycle changes, but Tx remains constant. So, low pass filtering gives a speed proportional voltage: 5V x Tx x f. f is the number of revolutions of wheel per time.

The signal of both wheels of each axes is simply averaged by adding both voltages, providing a first order correction of signal when driving curves: If left wheel is running faster and right wheel is running slowlier, because car drives a curve, signal at input of comparator CMOS TLC352 remains constant.
For achieving an increase of redundance, there is a second signal conditioning for the other axes. As TLC352 has open drain output, the two comparators can be connected as 'wired-and'. So, you would need only one input line of mcu.
Output of circuit gives negative logic signal, means, when at least one axes shows too high speed, output of circuit goes low.

Low pass filter is 2-pole type, which highly improves ripple rejection. Second stage consists of parts, which are about ten times higher in R and ten times smaller in C. So, first RC-filter is not disturbed by a too low ohmic succeeding circuitry. The one resistor '40R' is neccessary for not allowing the signal to leave common mode input voltage range of 0V...Vcc-1.5V.

Comparator TLC352 has some hysteresis, achieved by 1M resistor. Of course, you can also use TLC393 for comparator.

Make RC >= 10 x RxCx and ripple will be less than 0.5%.

The benefit of an analog solution is always questionable, of course, when mcu is fast and skilled enough to do the job without additional analog signal conditioning. This point, and the question whether there's indeed a need for a curve correction and increase of redundance, can only be answered by yourself, as long as we do not have any more details.

Bye,
Kai