THE Flash ECU knock sensor thread

[Greg E]

Can you explain more what the alternative strategy is? ie: What are you changing about knock detection in the alternate approach? I only ask because I can tell you that not filtering at all also does not work ideally (on an AEM 3/S car running 10.1s), yet somehow Matt M. managed to go 8's on a stock ECU and piggybacks. Just trying to understand what the proposed change is here, because the stock approach filters out the lower frequencies so there's less background noise in the signal. If that's what is being eliminated, I don't think you are improving the knock detection - you are just going to end up raising the threshold and knock is going to sneak through (and hammer the bearings). Been there, seen that.

"Hey look, zero knocksums and it is pulling like beast... Cool!"
>boom<

Just saying...

I need to be clear in that the stock code doesn't do much to protect your motor. It tries to be proactive but lets face it, excessive pressure in the cylinder flattens bearings and unfortunately we can only detect that when it becomes audible enough. In the case of detonation itself, you screwed when the "sound of horse power" is just as audible as the knock event itself.

There is no possible way to write software that uses the sensors we have available to do anything more than it does. As magical as people believe me to be, even I can't change the laws of physics.

All that aside, the issue with high band amplification is all you are doing is trading range for resolution. The hopes is that the higher resolution will reveal events slightly higher than the norm in a desperate attempt to prevent what could be a knock event from occuring before it gets severe enough to do damage.

Its like the stock MAF. Highly highly accurate in a small range. Works great on BPU setups. If you need more range, you must loose some resolution.

To answer your question about raising the threshold, the issue is the inconsistency of the signal being reported to the CPU. This is a result of the filter itself being used on the stock boards filtering 95% of what we need blocked out but letting that 5% thru in an inconsistent manner causing voltage spikes which look severe enough to be considered real knock. You can raise the threshold as far above that as you want. Problem is, what if real knock happens around that same level? How can you write code to know the difference?

If some of the bigger setups are seeing what is determined to be false knock, just raise the knock detection threshold in the areas needed. That's better than removing the filtering AND raising the threshold even higher because now you are hearing even more engine noise (and lower frequencies will be seen as more voltage at the ECU, so will raise the background noise).

If you are adding code to only look for knock on the compression stroke a bit before ignition and a short period after in an attempt to do per-cylinder knock control, then that's probably a good idea (and AEM does do this to an extent). But that's still different than removing the filter.

So back to the original question: What is different about this alternative strategy, specifically?

We have no where near the capacity to monitor events within each individual stroke of the motor and store it and compute individual responses for it. The stock code doesn't even do this (everything is a timed event assuming the previous event happened). It wasn't until they got into the CAN BUS stage where each individual cylinder got its "own ROM" so to speak.

You are familiar with how the AEM works and this alternate setup isn't much different. A simple 2D table with RPM as the axis and knock voltage as the limit value. Tuning requires you do what the pros do and run without any sort of knock detection at all. Monitor the noise levels as you tune. Make your adjustments incrementally and make a judgment call. Copy the values in that you logged. Any .02Volts over that limit = 1 knock sum. Fuel enrichment/timing retard and octane are still stock code.

[knights19]

Grab a log showing TPS, RPM, KnockADC (or knock voltage), Knock Base and knock sum. Post it here. I'll walk you thru how to determine if the knock is real and how to tune it out if it's not.

Sorry to resurrect an old thread, but this is all new learning for me when most of cars on here are perhaps no longer around or long since set-up fine
I have a 93 engine that I rebuilt, re-bore and plus-sized pistons, mounted in the rear of a small car running 4 wheel drive. I get a lot of noise as the motor is near solid mounted with the rear diff (engine transverse backwards with Ford 4x4 gearbox) mounted to the engine and a driveshaft jack-shaft also mounted to the other side of the block.
Reading through this thread has helped my understanding, but i'm uncertain to whether the ECU is thinking it is seeing a faulty sensor because it is or because of the high background noise.
Battery does get disconnected frequently so I get the learning function, but it does it persistently around the same revs. I tried backing off timing but got the same. I have plugged in another sensor (though didn't bolt it in) and had similar result but didn't get a log for that.
In the data I see a drop in knock voltage and knock base right after the RPM background switch over point at 4406 rpm, just before the 9 count incident which flat lines the timing, but only lasts for 4 seconds and returns to .2v whereas the knock count doesn't recover. To me, the above makes me think is is the noise but don't wan't to start adjusting without some moral support and first event of the year is in 3 weeks!
Are you able to walk me through how to tell if it is real or false knock please Greg?
Next step I was thinking of doing was replacing the sensor then running isolated wiring direct to the ECU.

[Unlogic]

The constant knock value of 9 for such a long period seems to indicate a faulty sensor, bad wiring or a bad knock circuit in the ECU.

[RealMcCoy]

Sorry to resurrect an old thread, but this is all new learning for me when most of cars on here are perhaps no longer around or long since set-up fine
I have a 93 engine that I rebuilt, re-bore and plus-sized pistons, mounted in the rear of a small car running 4 wheel drive. I get a lot of noise as the motor is near solid mounted with the rear diff (engine transverse backwards with Ford 4x4 gearbox) mounted to the engine and a driveshaft jack-shaft also mounted to the other side of the block.
Reading through this thread has helped my understanding, but i'm uncertain to whether the ECU is thinking it is seeing a faulty sensor because it is or because of the high background noise.
Battery does get disconnected frequently so I get the learning function, but it does it persistently around the same revs. I tried backing off timing but got the same. I have plugged in another sensor (though didn't bolt it in) and had similar result but didn't get a log for that.
In the data I see a drop in knock voltage and knock base right after the RPM background switch over point at 4406 rpm, just before the 9 count incident which flat lines the timing, but only lasts for 4 seconds and returns to .2v whereas the knock count doesn't recover. To me, the above makes me think is is the noise but don't wan't to start adjusting without some moral support and first event of the year is in 3 weeks!
Are you able to walk me through how to tell if it is real or false knock please Greg?
Next step I was thinking of doing was replacing the sensor then running isolated wiring direct to the ECU.

You have a knock sensor or wiring problem... The knock sum count you are seeing is classic failure management output, and no car continues to detonate long after you lift the throttle...

[knights19]

Even with the higher background noise?

How can I tell if there is significantly higher background noise than what would be expected?

Seemed odd to me that the trigger was the same as the RPM switch over point to triple gain.

I'll change the sensor on your advice and re-test.

[knights19]

i changed the sensor back to the original (one fitted was a new non-oem one i fitted when i rebuilt the engine), ran a new piece of shielded wire direct to the ecu and hey presto, car runs right for the first time ever! Thank you.
Log shows a huge knock spike but with loads of noise too. surely cant be seeing that much knock? No CEL onknock.power run.PNG

[RealMcCoy]

i changed the sensor back to the original (one fitted was a new non-oem one i fitted when i rebuilt the engine), ran a new piece of shielded wire direct to the ecu and hey presto, car runs right for the first time ever! Thank you.
Log shows a huge knock spike but with loads of noise too. surely cant be seeing that much knock? No CEL onknock.power run.PNG

What are the actual knock sum numbers? Graph isn't much help without the numbers.

Do you have Chrome configured to light the CEL on for a knock event?

[knights19]

Ah yes, graph much more scary than the numbers.
Run 5 5-1-19.jpg
CEL on knock is set to 5 but I guess the time it sees it is not enough to set the CEL.
Didn't quite get to 7200rpm but I ran out of road at 93mph.
Can't hear knock but car is generally too noisy for that anyway (rally car).

[anyonebutme]

There's a super easy way to test all this, pull 5 or 10 degrees of timing at that area and run it again, if the knock is still there assuming your AFR isn't wack, then it's phantom. This is one of the things that a dyno really helps out, because watching the torque curve in real time for the spike and cliff dive helps a ton. IIRC there's some knock sensor filtering things that can be changed but the stock filters are actually damn good, even with solid engine mounts and the high baseline that built engines have.

What I did notice is your timing jumps from 18 to 23 instantly, which jumps tend to set off knock even if 23 steady doesn't knock. Might be worth smoothing out the timing map where that happens.

[knights19]

Will look at that, haven't changed it from the base Chrome setting yet.
Does anyone have a log of noise for a standard motor to compare the noise too?
I wonder if the differential mounted to the side of the engine is increasing the noise.