Knock activity - Real? Or should I turn down the gain?

[Hodgdon Extreme]

Turbocharged 2JZGE. Was up and running for 15 years on (ancient) Haltech E6k. Boost was 17-20psi, 470-500whp @ 16º timing, passed thru 5 owners before coming to me. Never broke.

Now I'm up and running on G4x and starting some WOT tuning and knock setup. I have dual Bosch flat-response "donut" knock sensors paralleled into the single knock channel of the ECU. I have the overall gain set to 5, which is too high, but I'm seeing cylinders 1 & 2 start to show a lot more activity than the rest up around 5000rpm.

This is with only ~10psi boost. I had the timing set to ~14º for my first pull. For the second datalog (the longer one), I pulled 2º out throughout the whole boosted section - but didn't seem to make much difference.

Fuel is 93 pump gas.

I don't see how this longblock could have lasted so long if this is real knock. But it's concerning to me that cyl's 1 & 2 are bundled in with the rest of the cylinders until ~5000rpm, and then go wild.

My fueling is plenty safe - so I can remove WBO2 and put some leaded 110 gasoline in the tank and re-test...

...or should I turn the gain down on cylinders 1 & 2 to bring them inline with the other cylinders, set the overall gain and move on with life?

File links:

Calibration file: https://drive.google.com/file/d/1lDzC2umjPEDHINTK422wqlKYku26fiDk/view?usp=sharing

Shorter log file with one WOT pull: https://drive.google.com/file/d/1XcsKyabG8allEhu3ItAOVHGLgC1UCIEb/view?usp=drive_link

Longer log file with several WOT pulls: https://drive.google.com/file/d/1MYnrH_DAMQkL2P_KsDBmd9955ftUuJGu/view?usp=drive_link

(while out driving, there was a BMW with no front bumper to cover a giant intercooler - I couldn't help but play with him a little. dumb, I know - especially when I'm concerned about knock)

 

[Rossobianconero]

Is 93 us? so like 98 ron?

Probably Adam can help us with the 2 knock sensor into one knock signal, normally you would run one knock sensor one signal input. So I don't know if that could cause that problem. It should not be even close to knock on 97-98ron on that lambda and timing. Are you using any sound confirmation of the knock? are you using your own knock detection system? That would confirm it either way. My guess is that or the frequency is not the right one, or you cant run two knock sensors on 1 signal input. You will probably end up with a gain per cylinder that is not all at 1.0 when the knock detection is setup. 

[Adamw]

I dont know enough about audio circuits to know if two sensors paralleled into a single input is acceptable practice or not.  

The gain is effectively just a volume knob on your amplifier so it wont affect the trend you are seeing.  Most likely you are seeing more mechanical noise due to incorrect frequency.   I would also increase your window start to 20deg and length to 50 get some further noise out.   Oh, and as Rosso says, the cyl gains would usually need adjustment to balance the cyl levels also.  

[Hodgdon Extreme]

Thanks for the thoughts, gentlemen.

Fuel is USA 93 = (RON+MON)/2 . I also don't think 10psi on an 8:1 engine with 0.75 Lambda and 10.5* spark timing would be knocking.

Jose @ Link live chat was the one that told me paralleling dual knock sensors into one input would be just fine. I could try simply unplugging one of them at a time and retest to see if there is some kind of interference going on - but I doubt it...

Probably the easiest next step is to simply put some fuel with a big MON into the tank and retest.

Regarding narrowing the knock detection window: will doing this increase chances of real knock going undetected? If so, how much? What are the reasonable/practical limits to the knock detection start/end window?

Let's assume, hypothetically, that racegas and narrowing knock window yields same results I'm seeing now:

...I'd need to turn the gain way down for cylinders 1 & 2... But wouldn't that leave the knock activity for those two cylinders way lower than the rest of the cylinders under ~5000rpm? AFAIK, the knock channel gains cannot be calibrated as a function of rpm; they are global scalars...

[Vaughan]

13 minutes ago, Hodgdon Extreme said:

Jose @ Link live chat was the one that told me paralleling dual knock sensors into one input would be just fine. I could try simply unplugging one of them at a time and retest to see if there is some kind of interference going on - but I doubt it...

I was under the impression that when trying to run two knock sensors into one input you need to use 10kohm or similar resistors in series in each signal wire before you splice them together but I would have to confirm this with a hardware person.

[Adamw]

12 hours ago, Adamw said:

due to incorrect frequency

Frequency is your main issue.  Try 6, 7 & 8 KHz to find one that gives a less erratic noise trend then make it knock to make sure it is detected.  

 

8 hours ago, Hodgdon Extreme said:

Regarding narrowing the knock detection window: will doing this increase chances of real knock going undetected? If so, how much? What are the reasonable/practical limits to the knock detection start/end window?

Of course if you went as far as reducing all the way to 0 knock window length you would never capture anything, but I said to set it to start 20after ign with a 50 length which is right where knock is going to occur.  Right now with your ignition timing at 10deg BTDC and your knock window start at 10deg after Ign it means you are listening right on TDC which is where the piston changes direction and slaps across to the opposite cyl wall.   

 

9 hours ago, Hodgdon Extreme said:

I could try simply unplugging one of them at a time and retest to see if there is some kind of interference going on - but I doubt it...

Microphone circuits are very sensitive to impedance and capacitance.  Just unplugging one sensor only removes the impedance, you still have the cable connected which is effectively a big capacitor and with nothing connected at its end probably a nice noise pick-up antenna too.  2 in parallel may work fine, I have never tried, but personally I would stick to the proven single sensor connected to a single input rather than trying to reinvent the wheel and introducing unknowns. 

[Rossobianconero]

Also, use your own knock detection system like the knockblock, to know for sure if is knock and when. Only when I know my knock detection setup is perfect I feels comfortable without it (when tuning)

[k4nnon]

On 6/14/2023 at 8:25 PM, Adamw said:

I dont know enough about audio circuits to know if two sensors paralleled into a single input is acceptable practice or not.  

The gain is effectively just a volume knob on your amplifier so it wont affect the trend you are seeing.  Most likely you are seeing more mechanical noise due to incorrect frequency.   I would also increase your window start to 20deg and length to 50 get some further noise out.   Oh, and as Rosso says, the cyl gains would usually need adjustment to balance the cyl levels also.  

Hey Adam, would these numbers be universal between engine platforms? Just reading through the post, Im not currently having any knock issues, but noticed on my built evo 9 my knock settings are set to start 10deg after and window length is 60. Would there be a more accurate/smoother knock detection setting to 20deg after and a 50 knock window? Or are these numbers based off WOT timing numbers? Let me know please and thanks..

[Hodgdon Extreme]

To follow up on this:

I enabled Modeled-Multi Fuel mode and configured things so my Lambda targets and ignition angle would be same for E85 and gasoline. I had to set the flowrate of my injectors about 10% less for E85 to hit my Lambda target, but that's a different story...

Anyway, with ethanol content registering about 60%, I did some WOT pulls and found the knock activity unchanged compared to what it looked like on gasoline. Same Lambda, same (low) ignition timing, same ECT, same ambient temp, same IAT...

So at this point, I'm confident I'm not seeing actual knock. Next up I'll test out changes to the knock detection window, and using only a single knock sensor.

More to follow