Correct me if I am wrong but it seems to me Knock sum is actually ignition retard amount I will try to get it in correct scaling later.

It should be possible to log Knock Sensor Voltage. Once it is sampled at high rate and coverted to wave we should be able to hear the mike.

Has anyone tried to verify it is actually Knock sensor voltage?

Has anyone experienced Knock Sum higher than 36?

Is there any DSP chip within the ECU that processes the RAW Knock sensor input and what we have as Knock sensor ADC value is preprocessed FFT?

 

36 is a actual cap in the ECU. You can change this number to whatever you want with:

05-06 - Address: 1782 , 1D, scaling = uint16

03-04 - Address: 1382, 1D, scaling = uint16

You can log raw voltage in the newer Evoscans.

As far as ignition retard amount. General rule of thumb is every 3 knock sum is -1* retard

 

The knock voltage logging is just the 0-5v value and doesnt really show any rhyme nor reason since its noise not necessarily knock. If you are good with this form of logging (its the same as the AEM) you can interpret it for what would be real knock and what would be phantom, but I have seen 3.9v and no knock, and seen 10 counts and .9v.

 

I know it is noise, but how it is handled in the ECU? I cannot find where is the Fast Fourier Transform. Is it only peak detection at any frequency? The knowledge that it is actually knock sensor voltage would help me to get rid of possibility that there is a Digital Signal Processor which pre-processes input via FFT and is connected to the pin on processor that is logged via MUT_30.

 

Maybe you should also be considering that maybe the knock voltage is not logged fast enough. We are not using a scope here, rather what the ECU is letting you see through its logging output.

Such as, you might get a 4V spike and a true knock, but in the logs only see a value around 1V. The ECU will detect it, but just not report the high voltage to you. What I have tended to notice is that sustained high amplitude knock voltage gets reported as knock sum. But as I was saying, it may be only showing this because it is not fast enough to show the spike inbetween.

 

the voltage might go through a filter respective to frequency at whatever load/RPM it's in and the outcome retards spark. The frequency for knock might be the same for all intensities, loads and rpm

 

I've been logging raw knock voltage forever w/ my ECU+ and it is logging right off the sensor's output w/o going through the ECU and I can attest to the fact that magnitude of voltage by itself has very little to do with what the ECU determines is knock. Sometimes it is downright baffling.

Heck, I thought I'd just pull open a random log real quick.

You'll see below that 2.7V is not knock, yet a few seconds later 1V is 6 counts.



(sorry for the enormity... Shrinking the PNG actually increases the filesize over the limit)

 

Time to do a search on scholar.google and find the method mitsu-heavy used. It's there somewhere

 

remember in engineering that you can determing an unknown signal by doing a convolution integral. I have a feeling the analog knock sensor voltage probably gets put into a convolution integral with a Known mitsubishi signal. And this is were the magic determines real knock from false knock.

 

I actually did use matlab to work on this a year ago but never finished the product. since matlab does convolution integrals for you.

 

 

most of the time or am i missing some of that magic at the moment.

 

well the magic lies in the FFT of the convolution integral. Taking the %power of the FFT and if it get higher than that setpoint it is real knock. This is my guess on how to do it. its up to the Dis-assembly gurus to figure out the algorithm of this convolution integral.
im telling you this for sure, just looking at the waveform of the knock voltage will do you NOTHING.

 

US patent 4508079. Mitsu's been at it since 1983. Mechanism not too different than the current, except distributor retard .

Maybe the voltage (vibration signal) goes through a rpm/load filter, a -10 - 45* expansion stroke filter for each cyl, then a frequency filter.

 

Thanx, this is exactly the information I have been searching for.

I could not get any further than US pat 5121729 and 5287837