Tuner@Swift

iTrader: (1)

Can the frequency output from the knock sensor be streamed into a replayable audio format? I'd like to be able to hear what the knock sensor is hearing, sort of like electronic det cans.

honki24

iTrader: (23)

The sensor outputs a voltage... I suppose if the maps were found and understood perhaps we could backtrack to frequencies... but since noone seems to begin to grasp the knock tables as it is... I don't see your dream coming true in the near future.

So... when is someone here going to take a Japanese Mitsu ECU engineer out for a few drinks and steal all their c0dez? lol.

Tuner@Swift

iTrader: (1)

Actually, the knock sensor is a piezoelectric transducer, so it outputs a frequency within its scale. The ECU employs a frequency band pass to mainly look at frequencies between 5kHz and 21kHz with focus at 12kHz. The sensor itself is basically a microphone, the ECU has to actually determine what is knock based off a background noise level and actual ignition events.

I could tap the KS lead and run it into the mic input of my laptop and use some audio spectral analysis software to see how noisy the output of the sensor is before ECU processing.

jcsbanks

I've used a Bosch knock sensor plugged into a mic input on a laptop before, works fine. Don't use the ECU's knock sensor though as the impedance of the mic input may not be sufficient to avoid altering the ECU's knock control.

An easy mounting point is on the inlet manifold.

Tuner@Swift

iTrader: (1)

I don't want to run like that, just to test some things. Ideally, I would like the initial knock sensor input to the ECU (pre-processed signal) be read from a logger and output to a software based audio processor. The problem I see is that the logging rate would not be nearly fast enough to keep up with it.

Really all I want to do is try to get as much data on how our knock detection and correction works as possible. I have a theory that it is very similar to the DSM knock chip, but processed internally by the ECU.

jcsbanks

At 62500 baud you could get 6250 samples per second on a continuous DMA dump of a single variable. I can't get a reliable connection above this baud rate.

Trouble is, 6250 samples per second will have a low pass filter at 3125Hz (Nyquist) - useless for knock processing.

It is likely the ECU uses analog signal processing (rectification, buffering/amplification, bandpass filter) before the knock sensor input goes into the chip's ADC, then the ADC is read on a timer.

Bez had a thread on it somewhere on aktivematrix... it seemed to be looking at a moving average or difference from previous reading.

Tuner@Swift

iTrader: (1)

I am going to take my ECU apart and see if it has a similar separate knock processor on it like the DSM ECU's do.

jcsbanks

All I could see on mine were a load of surface mount passives and unidentifed ICs.

I couldn't see TMS320C certainly, and gave up on trying to route anything from the plug through the multilayer board.

cwtt

On my HKS Knock Amp their is an ear phone plug where you can plug in a set of headphones and listen to the stock knock sensors. Pretty cool little function. But it is $800

cpoevo

iTrader: (1)

I have a PLX Devices R-500 and it has a function where you can plug in headphones and listen to the knock sensor. It really works too, you can hear the knock. This is not a new concept though. Years ago before ecu tuning became so easily available the old school guys tuned with headphones to listen for knock. I found a link a long time ago how to build a little thing to log knock to a tape cassete and then analyze it on the computer.