Years ago, I worked for a small company that designed and manufactured custom embedded systems. One of our projects involved the re-design of traffic-light power boards for a local traffic-light management company. To detect burned-out lights, circuits measured current through the lamps. Our customer always used a classic Hall-effect current sensor, with all its shortcomings and its large volume, weight, and the need to calibrate sensor amplifiers one by one.
At that time, around 2005, I knew Allegro Micro had a series of integrated Hall-effect current-sensor ICs, and I wanted to try them and include them in a project. This redesign offered the perfect occasion, so I suggested the use of the IC and the designers accepted it. We chose a model similar to the ACS712, but without the internal filter-configuration pin.
To remove unwanted noise, we connected a simple RC filter with a cut-off frequency of about 1 kHz to the sensor output. We deemed 1 kHz sufficient bandwidth for a 50-Hz sinusoidal (well, almost sinusoidal) current waveform.
After we received assembled boards, testing began. Everything worked well, but the current sensors failed intermittently. That's the worst kind of failure one could expect.
Testing the current sensors involved turning on simulated traffic lights in sequence, logging the current output (around 200 samples per second), comparing the waveforms to the expected results, and then checking wave period, current amplitude, as well as phase, crosstalk, and so on.
When failures occurred, they appeared as oscillation, at "nonsense" frequencies and amplitudes -- nothing similar to 50 Hz, or 220 Vrms. This result left us puzzled because we had validated the current-sensor IC before designing the complete PCB.
After three or four days, I jumped in and asked for a copy of the schematic and focused on the current-sensor circuit. Then I read the sensor's datasheet, page by page, line by line. After a few minutes of reading I found the sensor's "Output Capacitance Load" parameter; a maximum of 10 nF. But the output low-pass filter used a 47-nF capacitor with a 3300-ohm resistor. Clearly, the filter circuit exceeded this capacitance limit so we replaced the 47-nF cap with a 10-nF ceramic cap, and used a 15-kohm resistor to preserve a 1-kHz cutoff for the RC filter. The circuit now performed properly and we never again saw the oscillation.
Half an hour spent reading the data sheet would have saved three to four days of frustration, debugging, and replacing parts.
Do you feel reading a datasheet from end to end is an investment rather than a waste of time?