The 44Khz sampling is probably from the youtube encoding, it was most likely recorded at a higher bitrate than that then compressed down, so more extreme ultrasonic variation may be happening. The ultrasonic itself isn't audible to your ear so it isn't necessarily a problem - but the combination of the ultrasonic waves need to average out to recreate the wave, but the noise present in this wave is well within human hearing. Viewing the original 100Khz waves is not useful, as those waves are not audible to the human ear. The point is of the system is that your ear kind of "averages out" the frequencies that it can't hear, like a low pass filter. The digitisation process has effectively applied a low pass filter at 44Khz, whereas the stated objective of the system is to create a sound wave that when passed through a 20Khz low pass filter it averages out to the original wave. I'll be kind and go with a 15Hkz low pass filter, as that's the upper limit of the majority of human ears.
I've applied a 15Khz low pass filter to better recreate the human ear, shown below.
This is 10Khz (well within human hearing)
and this 5Khz
Getting better now, how about 2Khz
ok so a 2Khz low pass filter would mostly filter out the noise. The wave is still a poor reproduction of the source signal however, as you can see it's not a symmetrical sine wave, it's skewed, but this gives you an idea of what they're going for - they want to see the above wave (well, a symmetrical version of the above wave) when you run a 20Khz low pass filter on the output.
A 10Khz signal wave will look good awful, the noise will destroy the wave and will sound way off if you compared it to a true 10Khz sound. I suspect that's why we only saw 500hz to 20hz
(20min delay)