#hcsr04

I got a HC-SR04 sonar from eBay for £3.50. The TRIG pin triggers the sonar, and a few microseconds later the ECHO pin goes high for as long as the travel time of the ultrasound. For a full description of the module please check the doc.

The formula to get the distance from the travel time is:

\[distance = speed~of~sound \times \cfrac{travel~time}{2}\]

The speed of sound is about $340~m~s^{-1}$, and we divide by two because the ultrasound goes back and forth covering twice the distance.

The easiest way to measure how long the ECHO pin stays high would be with a polling loop. However I plan to integrate the sonar into complex systems where the main loop might be slow, which would decrease the resolution of the measure. The alternative is to use interrupts. We have the choice between external interrupts and pin interrupts.

Pin interrupts require a library which wouldn't compile on my environment at first, so I used external interrupts instead. Here is teh code.

I first tried against a wall 1.5 m away. I plotted a histogram of about 300 values and compared them to a normal distribution.

The values are grouped in bands approximately 4.4 mm apart from each other. That's the resolution of the system. The sonar uses an ultrasound frequency of 40 kHz, so the theoretical resolution is 4.25 mm (at temperatures where the speed of sound is $340 m~s^{-1}$). Our experimental value is pretty close.

The histogram does not match the normal distribution too well. I have no explanation as to why, if you know better give me a shout. The standard deviation is $\sigma = 4.74$ mm, which means that (if this is a normal distribution indeed) we can expect 95% of the results to be within a $4\sigma = 19$ mm range. That's the precision of the system.

I then tried at about 3.8 m, a distance close to the maximum range of the sonar. For practical reasons the wall had to be at a slight angle. This and the increased distance lead to a few missed echoes and a couple of invalid values which I removed from the results. Again I plotted about 300 values and compared them to a normal distribution.

There are still bands approximately 4.4 mm from each other: as expected the resolution hasn't changed.

The histogram matches the standard distribution better than in the previous case. The standard deviation is $\sigma = 8.06$ mm, so 95% of the results shall be within a $4\sigma = 32.2$ mm range. The precision decreased slightly. It probably has more to do with the angle and the quality of the reflective surface than with the increased distance.