The Power FC incorporates filtering and you can see the knock filtered or unfiltered. The unfiltered one is in the Basic Data packet and the filtered one is in the Advanced Packet. "Advanced Knock" reads lower than "Basic Knock" for this reason. (I've seen them plotted... maybe later I'll search for that post.) I know for a fact that Advanced Knock using our sensor works just fine.
At lower RPMs, you can retard the timing by just 1 degree and see knock reduced from say 8 (quite consistently... as in +/-2) to maybe 1-3 (again, quite consistently so you can definitely tell that it is reduced). At higher RPMs (let's say 5K and up) the knock numbers are higher. The minimum might be around 6-8. You can't make it go down to 1-3 at higher RPMs no matter what you do, so it isn't real knock, it is noise. It is harder to distinguish signal from noise at higher frequencies, so there's no surprise there. Better filtering might be possible, but there is a limit to the effectiveness and we are talking about knock levels that are far below audible knock] (which would be around 45 or so with this sensor). What Copilot does is retard timing until retarding it further doesn't reduce the knock. (That was Little Rocket's idea which I implemented in code... it took me about 200 hours to write all that code, including keeping track of how many degrees cells were retarded, whether or not it was done learning about that cell, displaying information about knock learning, etc.) When it is done learning, you can retard another 1, 2, 3 degrees as you see fit and then the timing is optimal for the AFR and VVT you have in each cell. So having a program like Copilot analyze the changes in knock is the answer to a somewhat noisy signal. It's a "magic bullet" that handles this signal to noise limitation.
