I might suggest however that you take a series of periodic readings and run them through a running average filter so that you can avoid detecting a glitch. Certainly a 1/2 second or so filter delay would not cause an issue would it?

I think in this case your suggestion is great as always Mike.
I was thinking of solving this problem by different approach, I take 4 or 5 speed samples and only consider them valid when they all indicating over speed exactly as if I was debouncing a switch.
What do you think?
Looking at the waveform of the sensor it was perfectly clean and I think there will be no noise problem especially when I'm using opto-coupler to interface it to my 8052 but designing for rare events makes my circuit reliable especially when i'm going into mass production.

Regards
mahmood