Feature Request - knock sensor, additional Points

[AbeFM]

As alluded to in my other post for help, I think taking the graph from 10 to 20 points would solve a lot of problems. I've been looking at this data for a long time now (months+), and 16-20 points would cover me.



You can see my frustration, here's a couple of different approaches, either leaves quite a bit on the table. With ~20 points I could do a pretty good job of hugging the curves. The red and heavy blue are 10 point curves, the light blue is 30 points, but 20 would do a pretty good job.

With all the 16x16 tables, just adding 10x1 hopefully wouldn't be that big an impact.

[AbeFM]

Furthermore, I have a question for people at large:
Would you rather have a map where you draw the background noise of the engine, then a single number for the % margin you give that (i.e. ignore noise 20% over graph), or would you rather draw the limit? I tend to lean towards a description of the baseline plus a margin, but perhaps people like the other way as it lets you give more/less room in different places?

[tpsretard2]

would rather a map

[AbeFM]

Do you have a feeling for how many points you need?

You can clearly see points coming in and out of resonance on my motor, I could likely use as many points as I could get... BUT, 18+ seems like it would let me draw a line within a factor of 2 in most places.

[tpsretard2]

like you i have very similar plots. I has a number of 20 in my head before. however 16-18 would be doable.

[AbeFM]

Yeah. 32 is probably an upper limit on what would be useful, and 18 probably a lower one, from my curve.

More would be awesome - if there was a good way to develop the chart, a sure candidate for auto-tune - but ten is actually too few to do a good job, once I look at the data.

[vw_chuck]

What this knock control needs is something called adaptive gain. Most OEM knock control strategies use this so that it takes all the background noise out and you basically have the same voltage level for all RPM's. Then you end up with a single threshold calibration that if it goes over that level it considers it a knock event.

[AbeFM]

Chuck, at first blush, I really like the idea. But the limits would have to be fairly wide - if you look at my engine's numbers, you can see there's a ton of spots where the gain would have to change at very fast rates over short spans - and setting the gain that high that fast would mean missing it in other places?

Certainly it is a possibility. It is the kind of feature I would want to play with, but doubling the number of points would instantly solve the majority of the problems with the system.

It'd be great if we could make an impression but this is talking to a wall here.

[muythaibxr]

What this knock control needs is something called adaptive gain. Most OEM knock control strategies use this so that it takes all the background noise out and you basically have the same voltage level for all RPM's. Then you end up with a single threshold calibration that if it goes over that level it considers it a knock event.

More detail on how this works please...

[Matt Cramer]

The version I've heard from an OEM engineer worked sort of like this:

The ECU keeps a moving average of the knock level heard in the past X engine cycles. X probably should be user-specified. This moving average could be done across the board or per-cylinder.

If knock level > Y * moving average, apply knock retard.

[AbeFM]

Sounds reasonable, but you'd have to pick y appropriately. Or perhaps scale it with where you are with a chart, much like what we have. Otherwise, in places with steep slopes, your threshold will be artificially lower, which in turn will make you raise Y which in turn makes you see sensitive in the places where the curve is flatter.

I could review some logs in detail to get an idea for what numbers would work and what wouldn't and how tight they would have to be - but allocating a few more points wouldn't require much in the way of coding - assuming there's memory, etc, for them. I get knock pretty often now and don't know how much to trust it. I've got a series of low resolution "best guesses" trying to cover the spectrum, so I can load a different table for a high RPM day from a low RPM day.

[CSXRT4]

In the oem subaru code, it has a moving average for each cylinder and then a curve that defines a limit above that average.

Also, its hard to tell for sure because ive only been able to see the code in assembly and its fairly complex, but I think it compares the knock level noise outside the knock window to noise inside the window.

[AbeFM]

That seems reasonable. I hope the moving average doesn't include previously detected knock. I would think it would HAVE to be per-cylinder to be useful at all - I'd be curious to see what it looks like as well.