Guess the paper’s author put all his devices on the same desk and set TX power to maximum level. The problem is that LoRa modem sometime detects LoRa preamble even in strong enough noise 
and surely sometime is able to find a corellation between strong enough LoRa signal and ideal preamble waveform. As far as I can recall, Semtech’s name for false positive is “ghost decode”.
I have a personal knoledge of the guys who had observed the same behaviour and then invested a lot of money into developing “LoRa CSMA” w/out doing enough nmr of the field tests. Their activity ended up in epic fail…