MS3 PRO Knock Detection Thread
 

Ok so I've searched and read a lot of threads with the word "knock" in the title. But I have a few questions regarding how to setup and get knock detection working with a MS3. I have a MS3 PRO if it makes a difference. I know it has windowing and a lot of settings, it looks much better than the older stuff.

1. What knock sensor is best? Use the stock sensor? GM sensor? Something else? I have a BP that's bored to 84mm, so not sure if that really makes much difference for the stock sensor.

2. If I use the stock sensor, can I just tap into the existing wiring, or should I run a new shielded wire all the way to the sensor? Would it be better to run a new shielded wire?

3. What about settings for the knock control? Does anybody know what "works" well?

I want to try to do something using a knock sensor, and would prefer to not drop 600 on a J&S Safeguard if the MS can do the same thing just as well.

I know all about det cans, I use them.

I want this thread to be about using a knock sensor and software in the best way possible to have protection if something causes the engine to detonate.

I've played with mine a bunch but not been real successful yet. I can't get it to reliably detect knock and only knock -- either it misses knock or it falses on other noise.

--Ian

Thanks for the feedback Ian. How is your setup regarding my 3 questions? Stock sensor? Stock wiring? Stock sensor location? What settings are you running? Are you on standard 83mm bore or oversize?

Your results are what I've had a long time ago using a KnockSense MS, but supposedly the MS3 has a better system.

Yes, yes, yes to stock sensor, wiring, and location.

Settings in my build thread here:
https://www.miataturbo.net/build-thr...3/#post1215621

84mm Supertechs, Carillo A-beams, GTX2863R.

(edit -- updated knock sensor settings are here:)

http://www.codrus.com/miata/fm2r/May30/knock-may-1.png
https://www.miataturbo.net/attachmen...ine=1433056043

--Ian

Here are my settings:
1) Stock Sensor and wiring.
2) Initial settings per Reverant's Base map. (windowing and coolant scaling in particular)
3) Frequency filter confirmed by other info based on piston diameter.
4) Thresholds set by free reving logged values plus 15%
5) Gains based on makings the individual cylinder's values from free reving match one another. They make logical sense based on the sensor location.
6) I don't have Det Cans, yet, so confirmation of these settings is based upon places of MAP and RPM where the Knock Detection pulled timing. Then, when I reduced advance at those places, Knock Detection no longer detected. This demonstrates to me that the detection was based on a too-much advance phenomenon, such as knock.
7) I only get false trigger now when start kick-back occurred (during start tuning) or reving to redline and letting off quickly. I think this is likely gear noise or something like that.

https://www.miataturbo.net/attachmen...1&d=1433106208

https://www.miataturbo.net/attachmen...1&d=1433106208

https://www.miataturbo.net/attachmen...1&d=1433106208

The knock window settings are very very retarded. If you have peak pressure after 40 degrees ATDC you will misfire. I wouldn't start later than 20 degrees and wouldn't have longer than 20 degree window.

I don't know the code but it's most likely just looking at peak noise rather than integrating over a crank angle window. Therefore the knock window needs to be around where peak pressure would occur.

The top graph is BTDC. This is not a defense, just information.

So basically the window ends at 20 ATDC across the RPM range, and the beginning and window width are what changes.

So what would you recommend I set them at? Say listen from 4* ATDC to 24* ATDC? Something else?

I'm guessing that once you hit peak pressure, pressure is going down, and thus so is temp, so the odds of it detonating AFTER peak pressure is very very low? Or is this wrong? If true then windowing should stop after peak pressure of course.

I could have sworn I read in the help file that units were atdc for start of window, negative number is btdc.

Anyway you wouldn't want to start any earlier than about 8-10 atdc. At low speed you might have a longer duration but it's hard to say because there's no load component to the map. Low speed always wants to knock more, but it makes less boost.

High speed knocks less, but it makes more noise, Exponentially more.

The knock window settings were meant to start listening for knock a few degrees after the ignition event. I generally took the ignition timing at 75kPa or more absolute pressure, as pinging is - in my experience - unlikely below 75kpa.

For example, most of my basemaps have about 24* of timing at 3000rpm/75kPa. So I set the knock window at 3000rpm to start at 24*BTDC, then set a knock window duration of 44* to stop listening at 20*ATDC.

I hope this makes sense.

Was hoping you would come in. That's what I thought you were thinking on those windows.

Apology Accepted.

Got a question for you guys.

I got the knock sensor hooked up and software setup, but I'm showing 0.0% knock all the time. I actually cranked it this morning, and it showed nothing again, but then at one point I cranked it, and suddenly it was working! I tried tapping on the block and sure enough it registered knock. I thought crap, no idea but now it's working. Key cycled and it kept working. Key cycled again and it died, hasn't worked since.

I tested the sensor, it's good. I tested the wiring between the sensor, it's good.

I redid the wiring, still don't work. Swapped sensor for anther that also test good, still nothing.

Any ideas? I'm starting to think the knock module in the MS is bad...

One good way to test the module is to disconnect the sensor and connect your PC sound card to it. With the engine running, play a sine sweep file and see if the sensor responds.

Knock doesn't occur right after the ignition event. It has to go through the burn delay period first, which can be 10-20 degrees at full load, then bulk burn period-knock is much closer to area of 50-75 percent burn.

Translation : listening before TDC will only pick up noise.

Makes sense. Better windowing should result in better detection / rejection of false signals.

What I hear you saying is that the window should / can be a fairly uniform, Start at TDC, end at 25 ATDC, regardless of RPM. Is that a good interpretation?

What is your definition of "full load"?

*EDIT: Anyone have input regarding CLT scaling?

As a follow up, I did the test Matt described above and got no response, so I shipped the ecu back. Will report back once I get an update from DIY Autotune.

In the meantime, bump for this discussion to continue. :)

I have question for getting no-knock reference. It seems a lot of people set this by reving the enging in nuetral and using those values as a baseline.

Would it be better to pull say, 5 degrees from the timing table (guarantee a no-knock condition) and make pull from idle to redline, and use that as your baseline? Reason I ask, maybe the noise profile will be different under load?

Bump for input!

Also I sent my ECU back, DIY confirmed the knock module chip was bad and they fixed it under warranty! Awesome to see they support and stand behind their products.

It should be arriving today so I'm hoping to get the knock settings all setup and see how it works. I currently have my windowing from 8*ATDC to 28* ATDC.

For this I entered -8 for Knock Window Start, and 20* for duration. Does this sound correct?

MS3 PRO Knock Detection Thread
 

I would start with those settings. Then you can experiment with the window at certain rpms. I would keep the length at 20 degrees and move from -8 to something later (maybe move in 5 degree increments) to see if the detected values are noticeably different.

Ok so I got my ECU back from DIY Autotune. They said the knock chip was bad, and they replaced it.

Anyways, I got the ecu back , hooked it up, and I'm still not getting any knock input! I went through all the troubelshooting again, ending with me playing a Sine Wave Sweep into the MS3 PRO and still got nothing.

I emailed them but it's the weekend so nothing back yet.

Can anyone take a look at my settings and make sure there's nothing obviously wrong?

Pics:

https://www.miataturbo.net/attachmen...ine=1435521608

https://www.miataturbo.net/attachmen...ine=1435521608

https://www.miataturbo.net/attachmen...ine=1435521608

https://www.miataturbo.net/attachmen...ine=1435521608

MS3 PRO Knock Detection Thread
 

What happens if you crank the gain to some absurd level?

I tried that a while back, it made no difference. I turned the gain as high as it would allow and still got nothing.

I ended up sending the ECU in to be repaired again, they fixed it today! Even sent me a log showing it working. So now just waiting to get it back and into the car.

So, my questions for anyone that will take the time to answer:

1. Are the settings I posted good? The only thing I plan to change vs what I posted screenshots of is to change the duration from 50* to 20*. I used 50 for testing. That and I'll have to tune the knock threshold of course.

2. Does the MS3 detect knock all the way to redline? Or does it get overwhelmed with noise at some RPM?

MS3 PRO Knock Detection Thread
 

I've done knock sensor development in a lab and I can tell you that noise increases exponentially with rpm and it's entirely possible that the sensor is blind to 1 or more cylinders at certain speeds and loads. As in, the engine can be knocking but the sensor won't pick it up. You don't hear about that when people tune spark on stock ecu knock systems for other cars (subaru etc) because somebody at the OEM set the gains and frequency and window and thresholds in order to mask/work around the blindness.

That's all by measuring knock with cylinder pressure sensors and comparing to lab grade knock sensors/accelerometers.

Knock control and knock limit is very much judgment calls and opinions.

Ok thanks for the info.

But hear me out. I don't have a lab, dyno, or a million dollars to do this to the OEM level. So what can I actually do to make the most of this? That's the question I need answered.

For example:

1. To set knock threshold, I've read people do this by reving in Neutral. This seems odd. I think pull say 5* and do a pull to redline, and use that as a baseline of no-knock. Agreed?

2. Once I know what the baseline threshold is, how much "higher" should I set the real knock threshold? I would think some % higher than the baseline. Say, 10%? Or should this vary with RPM? Perhaps more sensitive at low RPMs (because less noise) but a higher threshold at higher RPMs (since more noise)?

3. How much authority should I assign to the knock sensor? I've read some setups only pull up to 4*, but I believe Subaru does a lot more than this. I think the norm is something higher than 4*, likely more like 8*?

Answers to these questions would help me a lot.

The issue you are having with 0% is not a hardware issue, it's a firmware issue. DIY should have known this.

<p>Are these settings different car to car? How does an OEM get away with a single setup for all motors. Could we potentially crowd fund some sort of lab grade testing and apply it to all setups?</p>

Stock sensors work a lot different though right, as in actively filtering and listening to multiple frequencies from the same sensor/speaker.

All of the megasquirt stuff I've seen just seems to generate a duty cycle off of signal to noise ratio or something. It's largely unfixable, at least like OEM.

So we'd need a signal amplifier on the knock sensor, crossovers and filters stolen from a stock ecu somehow, maybe by sweeping a signal against stock ecu jimstim to watch and plot what input frequencies and levels pulls timing. Then make a knockbox or somehing to send PWM or TTL signal to the megasquirt.

That's all way over my head. I like my detcans and moderation tuning.

How so? I'm running there latest firmware.

MS3 PRO Knock Detection Thread
 

Well let me ask a question. Are you willing to advance spark and induce knock in order to adjust these settings?

Yes I'll do it, but not at full boost. But at light-moderate loads I'd do it. I'd prefer to wear some type of det cans while I can hear it in my ear, and see the response from the knock sensor.

MS3 PRO Knock Detection Thread
 

Your options are limited without a loading dyno, or any dyno really. When you start banging through gears on the street the drive train movement can affect the noise curve, especially if you don't have a nice even road surface.

You can hold it in some higher gear, give it part throttle, and take some logs as it runs up in engine speed. Then try to repeat it but it with more spark and see if you notice something.

What I would personally do is try to induce knock at low speed lugging where the noise is naturally lower, the engine likes to knock, and cylinder pressure is lower so the risk of damage isn't so high. Floor it at 1000 rpm in 5th, or maybe 1200 or 1500 at most.

You should hear more of a faint chirping sound when it knocks, especially at 1000. The background noise from combustion is very low at such low speeds, but there's still potential for other noise.

Ok thanks for the info. I can do that. I can run about 10 PSI boost a 1,000 RPMs. Should I stay out of boost for this test or let it boost?

Also what specifically would I be looking for? I'm guessing to determine what real knock looks like so I can set my thresholds accordingly?

MS3 PRO Knock Detection Thread
 

Ok im coming back to this now.

1. There's no easy answer to this without a lot of risky testing. Pull 5 degrees method is closer to what I'd do, but you've got transient and drive train effects from doing pulls on the street.

Running in neutral unfortunately doesn't load the engine down so much, but we don't know the sensitivity of background noise to load. That's the kind of thing you study when you configure a knock sensor in a lab.

You could do both and see if any of the numbers make any sense.

2. If you're not willing to beat the shit out of the engine to find out (who could blame you), 10% is as good as any. It's kind of arbitrary as nobody can seem to agree on where the line is first crossed and how much knock is too much.

3. Ok so if you had a modern model based ecu you would have a minimum spark curve map that's tuned in the combustion portion of the engine torque model. This tells the ecu how much spark it can pull without misfiring. It's used for cat lightoff, knock retard, and torque reduction requests on gear shift. That's done by sweeping spark and looking at combustion and torque fluctuation.

I don't have any kind of cylinder pressure data or misfire data. So we go off rules of thumb, past experience, educated guesses. I'd go with 4 because I'm concerned that more than that could overheat exhaust and melt something or cause a misfire. But you can always increase it later.

MS3 PRO Knock Detection Thread
 

Can you really make 10 psi boost at 1000rpm? Is that with vvt maxed out?

MS3 PRO Knock Detection Thread
 

Stick it in 5th at 1000rpm. Floor it up to 2000rpm. Save log. Advance 1 or 2 degrees, do it again. Compare knock sensor activity. Keep advancing until you hear audible knock. Shouldn't be too loud. See if you have some kind of trend of spark vs knock sensor activity. Be mindful of air intake temps changing knock limit.

There's a small risk of preignition but I think it probably won't be an issue. That will make a louder sound and be more unpredictable than spark knock.

If you can't get any knock sensor activity but you can hear it audibly, or its just way too low, I would wonder about the gain and the frequency settings.

Thanks for the reply sir!

Here is my current VVT map:
https://www.miataturbo.net/attachmen...ine=1435872618

1.- Alright I think I'll do the pull 5 degrees approach. I am 99.9% sure my current map does not knock, as I already pulled 3* from it when I know it knocked on that map at some point. I believe it was knocking in the 2000-2500 RPM, 150+kpa range with very high AITs, but at the time I just pulled from all the boosted cells to be safe.

So I think I'll do a baseline the way it is now (minus 3) and another baseline at minus 5, and another at minus 7, etc. Just to "see" what changes. For my knowledge, more data will be useful and help me see what's normal.

2. If I break this engine, I'm tempted to build a cheaper one and use it for testing. There's a lot of things I want to test, and knock control is for sure one of them. But for the question 2, I'm thinking once I have the data from the test in point 1, it should be easier to establish a knock threshold.




I'll do that once I get everything back together and on the road again. I've tuned my fuel at low speeds doing 500-2500 RPM pulls in 5th, and I noticed then that I needed 11.0:1 AFRs at low RPMs to prevent detonation in boost. It was audible no doubt.

I'm doing a pulley upgrade too so new boost level will be around 30 PSI at 7,700, it was 23 PSI at 8,500 before. 30% more pulley now.

Hi to all!

How to tune integrator time constant?

The MS3PRO module uses the TPIC8101. The datasheet and application notes do a pretty good job of speaking to the integrator time constant. I ~think~ I remember the integration time being determined by the knock window, but it has been a long time since I looked at this stuff.

changing the knock window integrator time constant changing itself?
you manually set the integrator time constant under the bandpass filter. Initial setting 150 ms

If you read and understand the datasheet I linked you will see there are 2 parameters, integration_time and time_constant. The parameter time_constant is a static value set in the knock sensor parameters window, integration_time is configurable as a function of engine speed via the knock window settings.

https://cimg1.ibsrv.net/gimg/www.mia...919a87f623.jpg
I ask about this integrator time constant. Why 240?
How to choose this value?

Yes, and I pointed you to a resource that tells you what that value is and how it plays into the knock sensor reading as reported in MS.

Score: eggplant

think of that setting as a global gain, except smaller numbers give more gain, larger numbers give less. I believe this is also described in the MS manual, only they refer to it as a filter parameter.