Fero's knock logging (Fknock1.log), kd = knock_val - noise_val and kr = knock_val/noise_val.
The same plots, but plotted by igncount (eg. cylinder):
The major problem with the logs is the absence of a thrustful indication of whether there are real knocking or just noise. What can we do about that?
- Record knock sensor output and primary trigger output in a stereo wave file, such that the knock "music" and primary trigger can be played back on the table (with correct timing wrt. engine phase).
- if we assume that the noise is symmetrical around the kv/nv=1.0 that brings us to: anything above 1.3 is real knock.
I tested a very simple method in the perl script bin/knock_analysis.pl for determining if knocking is present in the samples: (kr > k1 && kd > k2 || kd > k3), however without knowledge of when knocking actually occurs, it involves a lot of guessing.
Ex, is this a knock? (I think, and a bad one): adv=9.75 rpm=5084 map=149 knock_val=315 noise_val=156 kd=159 kr=2.019
Todo's:
- decide which knock detect method to use (and implement) in firmware
- get some logs which contains information about whether or not knocking is present in sample N
- devise some guidelines for how to setup the configuration