After having read the paper, I am simply wondering what the reason is for the elaborate setup with the Audio Interface.
Is it really neccesary to send waves to pick up? Wouldn't a demultiplexer(i.e., sending a signal to just one of the rows at a time, swapping rows very quickly) do the same thing? Or is the induced latency of that technique really so bad it will become audible unpleasing?
Soundplane scans at about 1kHz. if each of the 512 junctions were scanned in sequence, that would leave hardly any time per junction to get a good sample.
In general, sampling for a longer time is better, and the continuous wave approach has the advantage of gathering data at each junction truly without any gaps.
A scanned matrix without carriers would not sense pressure. You would only have binary on/off sensing like a standard computer typewriter keyboard. Capacitive sensors require an AC signal generated on one plate, with the amplitude on the other plate representing the distance. While it's possible to calculate distance in response to an impulse (the scanned method that you suggest), it's much more workable to generate steady carrier frequencies.
By the way, I believe that the latency is the same either way. Scanned binary keyboards have a latency of (sample rate) / (number of columns), but the Soundplane has a latency proportional to (sample rate) / (number of FFT bins). It's possible to use overlapping FFT to reduce the latency, while a matrix can scan binary switches at higher sample rates to reduce latency.