Status: because of lack of time and not clear documentation about knock chanells in 1.1.27 car runs on 1.0.73 using knock control with 9.5kHz center frequency. 9.5kHz looks reasonable because of less background noise (see graphics below). Both mentioned values works as expected - when knock occurs knock value is significant higher than noise.
I have problems to configure knock control. Engine is 20XE Opel with little too high CR for RON98 and stock cams.
Engine like knocking under some circumstances. We lower engine temperature from 95-100C° to 80-85C°. It helps a lot.
But anyway there is danger. So I spend a day for knock sensing. Find frequencies by analyzing wav files, finding what means variables across wiki, Robs and Andrey forums. My frequency is 5.9kHz and 12kHz as harmonic. Try both with similar results.
But there is big problem that I can't solve: when knock occurs (very easy to hear) spikes come in both knockvalue and knocknoise. No matter how to configure window and then integrator time constants. Due this KnockDiff not changes significant. KnockDiff only becomes more noisy - not unusual at normal combustion too. But not give clear rising.
MembersPage/PhatBob : There have been some discoveries that you may find useful regarding knock_chan
http://vems.hu/www.vems.co.uk/forum/index.php?topic=461.msg4739#msg4739
solution is simple: \n
uint8_t knock_noise_scale; // knock, noise scale factor (0x10= *1)
knock_noise_scale=00
will scale down noise and let the difference be K-0, so knock signal can act
For better analyze I even modify vemsV3.ini by adding Knocnoise and Knocvalue for datalogging.
I use Specrtogram for analyzing wav, where real knock recorded. Center of spots are my mentioned frequencies.
If frequency is chosen totally wrong, KnockNoise and KnockValue must not rising when knock occurs. But in real life there are big spikes during knocking on both. As I understand spikes is output buffered with given frequency only.
Currently 1.1.27 firmware.
Board #1437
Engine - slightly overbored 20xe. 86mm is standard bore.
Msq
http://www.vems.hu/files/GintsK/Config.txt
http://www.vems.hu/files/GintsK/tables.txt
Record of real knock:
http://www.vems.hu/files/GintsK/KNOCK4.wav
Not very usable wav of clicking wrench at diferent areas of block:
http://www.vems.hu/files/GintsK/wrench01.wav
---
M: Something is wrong here. It seems to be noise, not knock-sound. The spectrum does not have a peak at the knock frequency. To sense knock, the knockchip must listen to the specific knock frequency, not a wide-spectrum noise. The wrench method always works if the sensor is mounted and operates properly
G: Agree that something is wrong. Because of it this description is here.
G: Recorded sound is knocking definitely! Acquired with over advanced timing at partial load in 1000-2000Rpm range on higher gears. Knocking is very clearly audible through unpacked firewall too. But I am not so sure about recording and analyzing methods.
G: What kind of spectrum visualization you recommend to use? I see peaks at 6 , 9.5 and 12kHz. Did you mean my peaks did not match with calculated frequency?
G: I believe this kind of bosch sensor is wideband. So wide band noise comes to knock chip too. Bad thing about this engine - there is mech cam followers. So noise is more than usually.
G: Engine bay is tight. And I haven't chance for firmly wrench hits on back side of engine where sensor is. But how wrench test practically works? This is a other source of sound than detonation explosion inside combustion chamber. No? equally as mechanical followers or freely clearanced pistons rattling.
knock control investigations. Any comments welcomed!
- is the knock-sensor bolt on properly ?
- yes ~25Nm on factory place.
- wired properly ?
- between
- EC18 pin 2-3 (recommended)
- Wired this way. OE harness is used.
- or using EC36 pin 26 as ground ? (could work, if otherwise the power grounds are wired properly)
- or other means ? (through endplate is possible. 2nd knock channel can be tried this way, although a 100nF and a 10-22k resistor is needed)
- EC18 pin 2-3 (recommended)
- maybe there was also a problem with the soundcard mixer setting that was used for recording ?
- recorded using external USB souncard and microphone input. This input is not very good on this card - noisy. During recording equalizer is used with all frequencies below 4kHz (center) pulled low. But sample frequency was 44000 and so it stays.
One more observation: if chose knock window start at 0° and length 60° there is no Knockdiff at all - 1.00 all the time.
During experiments I found knock settings in LADA OE calibrations. Generaly window starts @ 6-18° and ends @30-42° at low rpm and @54-60° at high rpms (not window length, but end angle BTDC).
18/3-2008
Hi. You can use this software to analyse the frq. http://www.relisoft.com/ The Frequency Analyzer
Peter
Knock3
Thank You. I somewhere found Spectogram old freeware version, witch allows some adjustments for better visualization. Here is knock3.wav with magnified my 3 frequencies:
Gints
Here is additional file from same season:
http://www.vems.hu/files/GintsK/KNOCK3.wav
Just make some additional spectrum observations in this file:
Spectrum for all file:
Spectum from knock only:
ahd from only operation without knock:
