What do we know about knock control?
 

After doing a bunch of research, messing with forged motors and cars without balance shafts, I still dont have any real clear idea about what the "knock" maps we can see in ECUflash.

Balance shaft deletes can cause odd harmonics from 4,000 and down it seems. I saw 15 counts at 72%load and 33* timing cruising today.

Forged motors can (dont always) throw 8-10 counts in when accelerating and randomly through the rev range.

I am curious where we are at in the disassembly of this pretty sophisticated control system? I mean does it look probable that we will be able to have knock control like in DSMlink eventually or is it significantly more sophisticated than that?

Knock control parameters that would be cool:

Control of degrees per count

RPM range (like the subaru has, rpm correction ranges both low and high)

Load based control like the subaru as well?

Ignore knock voltage below a certain level

Ideas? Input, criticism, praise, etc.?

I'm hoping this will come about soon as well. With the completion of my stroker sans balance shafts, I'm worried about how noisy the engine will be and if it will come up as false knock.

I've worked out how to disable the knock control completely, but it isn't a popular option (rightly so). I had another look recently and the routines are very convoluted indeed. Bez posted up some filter information on aktivematrix previously, but I don't think it gave him success with his forged motor.

Turning it off and monitoring a filtered variable made from the knock sensor ADC channel might reveal a simple threshold system.

This would be a great option for many myself included. Having a "built" motor Seeing 10-12 counts from 7500 rpm and up, with less timing than stock sucks! lol

A *good* option do deal with part throttle fake knock due to slappy slappy de slappy loosey goosey forged pistons would be to ignore the knock sensor below 100% load. I've been doing this in my DSM for a couple years and 10-20K miles. Works great and is NOT dangerous.

I've been studying the knock sensor voltage when logging. It really is an interesting little bugger. See the AEM cars set a knock sensor voltage limit, above which timing will be pulled (user defined how much per voltage level). What I've noticed is that the car is relatively loud when driving in vacuum. When it goes into boost it gets quiet again. When it knocks it gets noisy again.

It should also only listen for knock during a certian timing window. For instance 20 degrees before and after top dead center.

I think understanding knock control is a good idea, BUT it may take a long time.

some people have asked me to implement a load threshold for knock, ie knock only matters over 140 load... I havn't had time to investigate...

I don't think any real progress will be made until Mitsu starts using pressure sensors for knock detection instead of the piezoelectric sensors that are used now.

The problem is with this kind of sensor once your change the cylinder bore or any other geometries of the engine, the resonant frequencies change and the knock sensor that was used for stock is pretty much useless. The sensor is chosen to match certain resonant frequencies of that specific engine and then filtered accordingly, based on crank angle, etc.

There was real good paper that I read on knock control with piezoelectric sensors, but I don't think I saved it. Basically, there are just too many variables that go into monitoring knock using these sensors. That's why the knock routines in the ECU are so complex.

If I had my way, I would love to completely ditch the entire knock routine and knock sensor and install some of those spark plug pressure sensors that I have read about. Knock would show up as a spike in pressure and wouldn't trigger due to engine clearances and loose pistons and engine harmonics, rattling downpipes, etc.

I don't know why Mitsu hasn't changed already. I think Toyota uses pressure sensors for knock monitoring, but I'm not positive.

As far as our ECU and the knock routines, I think Bez has posted the most information on this on AktiveMatrix, but even he had no luck altering things, as jcsbanks mentioned. It would be great to be able to 'retune' the sensor, but I don't know if it's possible. I think the sensor is chosen specifically for the harmonics and resonant frequencies of the engine and then filtered from there. I forgot most of what I have learned, though. If we can alter the tune of the sensor, that would be great, but I think it would be a ton of work, if possible.

BTW, all DSMLink has (the last time I used it), is the ability to turn off knock control below certain thresholds, like TPS and load. That would be a good starting point for our ECUs to get rid of phantom knock.


Eric

The piezoelectric sensor is NOT tuned to a specific frequency. It is true that it works best over a certian frequency range. However many (read: all) AEM 4g63 cars utilize the stock knock sensor (and in the case of 1g DSM's they use a 2g sensor since the AEM maps are tuned to this more sensitive sensor). This includes bored motors, stroked motors, destroked motors, motors with solid lifters etc. You CAN tune the sensor to work for all 4g63 motor types. I know this because AEM cars do. When setting up an AEM car you do a pull and record knock sensor voltage vs RPM with NO detonation. You know if the noise is higher than this that it is abnormal engine noise and timing needs to be pulled. I've noticed that on average less than 2v on the knock sensor is no detonation. When above 2v then I am detonating. Just turn the logger on and watch it, its interesting stuff. :)

I don't want to get into a big debate, but monitoring knock sensor voltage is not monitoring detonation. The AEM simply monitors the voltage and you set a noise threshold based on that voltage.

The ECU uses a ton of routines to filter and check with running averages, etc, etc, etc, to determine what is and what isn't detonation. Even then, it's not an exact science.

But, monitoring voltage all by itself is pretty useless. I know you are going to disagree, but I don't have the energy to argue about it in this thread. :)


Eric

OT but was just wondering if that would was my car today:)

edit; off topic

the physical sensor's are not tuned to a specific frequency, thats the job of the "filter" tables. I'm assuming these are coefficients for some type of FIR or IIR filter. In general - knock noise only occurs at a specific frequencies (like 6KHz), so theres usually a bandpass (5-10KHz), and then a threshold on the noise within the filtered band.

We would need an expert on DSP to figure out the algorithm used.

You can turn the knock control off with periphery changes but in so doing it will no longer show any knock counts which isnt helpful at all really. AEM and stock voltage coincide but its so random appearing looking at voltage spikes and then looking at whether or not it was knock (in the stock ECU). I use averaged voltage to knock count from a stock ECU when comparing AEM logs. Consistent voltage spikes in an AEM usually indicates knock. That or its noisy :)

Travis, no worries...it wasnt your car...though yours did come up with false knock when the spacer burnt through. Done and ready to go.

I have seen 3.45v spikes on the stock ECU and not be knock.

i think the ECU does a good job of discarding one off voltage spikes. its only when when the spikes are repeatable, or there is consistently more "noise" that the ecu starts considering it knock.

Despite what people are saying i think the AEM still has good knock indication. When the AEMs knock voltage starts rising, you can be pretty sure something is wrong.



edit: since knock has a specific frequency profile, does the ECU dynamically take into consideration the noise floor? or is the noise floor a parameter that is calculated and set from the factory?

From my understanding and what I have read from people that have disassembled the knock routines in the stock ECU, it is dynamic. Also, from what I understood, there really isn't a noise floor. It's more of a comparison to past events, the delta between events and time, the specific crank angle, etc, etc. It's way more complex than I can explain here.

That's the only reason why I stated that merely looking at knock voltage is useless. You can obviously pick out very bad knock, but there can be large spikes in a knock voltage log with 0 knock and there can be realtively smooth voltages with a lot of knock. It's just not a direct correlation between the two.

I can probably dig up DSMLink logs which show the 4 rawknock signals from the 4 cyinder timing events on a DSM. In that same log I can show you what the ECU considered to be knock and what it didn't. In most cases, you wouldn't have a clue where the knock is and, more importantly, the severity (the knock counts in ECU terms).


Eric