To recreate the problem, put the sensor in 1-5 second polling cycle. Point the sensor at an angled surface at a distance of a few metres and the radio appears to lose contact. Putting hand in front of sensor causes readings to appear again. (Maximum tested range got up to 445cm so it's not the sensor range.)
I suspect it's something to do with invalid sensor readings when the code is still waiting for a PWM pulse to produce a distance.