Mega Meet 2009 Atlanta GA April 17 2003
- Slides: 38
Mega. Meet 2009 Atlanta GA April 17, 2003 Mega. Squirt and Mega. Squirt Logo are trademarks of BG Soflex, LLC.
Intellitach Bowling & Grippo April 17, 2003 Mega. Squirt and Mega. Squirt Logo are trademarks of BG Soflex, LLC.
Tach Noise – It Sucks! Ø Everybody has experienced noise of some sort in the electrical signals used to determine engine position (i. e. the tach signal). Ø All ECU/ECMs need this signal on engine position to determine: Ø 4 Ø Engine RPM Ø Position of Crankshaft Ø Ignition/Injection Points. This is probably the biggest stumbling block for any installation!
Tach Noise Makes Me Cry… Ø 5 Crankshaft and Camshaft sensors (Variable Reluctance or Hall) are susceptible to the following sources of noise: Ø Spike Noise, e. g. ignition spark firing Ø Flux Noise (df/dt) – “Phantom Tooth” Ø Channel Crosstalk from adjacent position sensors. Ø The noise sources can act randomly or periodically Ø Even one pulse of noise will invalidate the position signal and cause all sorts of havoc.
Intellitach – What Is It? 6 Ø Intellitach is a signal conditioning device that removes noise artifacts from VR and Hall sensor signals. Ø Intellitach is basically a signal interceptor that constantly monitors the incoming signal source and performs signal conditioning to eliminate noise. Ø Intellitach provides a clean bipolar output signal that any ECU/EMC can use for a trigger source. Ø Realtime signal monitoring is available with Intellitach.
Intellitach ADC VR/Hall Sensors ADC OUT Digital Signal Processor ADC OUT USB CAN Patent Pending 7 ECU
8
Intellitach - Features Ø Ø 9 Intellitach employs the following features: Ø Balanced Sensor Inputs – 3 channels Ø Direct Digital Sampling (DDS) Ø Digital Signal Processing (DSP) Ø Bipolar output signal Ø USB 2. 0 output (HID mode) for signal sampling/monitoring Ø -40 to +125 deg C Temperature Range components Ø CAN connectivity OK – lets see how it works……
Conventional VR/Hall Input VR ECU Vzero 10 Ø This circuit is a zero-crossing detector with infinite gain. Ø Circuit is single-ended, little common-mode signal rejection. Ø Some circuits use automatic adjustment of hysteresis based on input amplitude.
Intellitach VR/Hall Input Instrumentation Amp VR 12 -bit ADC STM 32 11 Ø VR/Hall Signal maintains symmetry with the use of an Instrumentation Amplifier arrangement. Very high Common Mode Rejection Ø Analog signal is sampled by high-speed (1 usec) Analog-to-Digital Converter (ADC) – Direct Digital Sampling (DDS). Ø Sampled signal is now digital and can be processed using Digital Signal Processing (DSP) techniques.
12
Intellitach Processor Ø 13 Intellitach uses the latest in ARM processors, the Cortex-M 3 core (ST Microelectronics STM 32). Although this device is a general-purpose processor it possesses features that enable DSP functionality: Ø Multiply-Accumulate and Hardware Divide in 2 cycles Ø One microsecond ADC sampling with DMA Ø Interrupt chaining Ø The STM 32 operates at 72 MHz, nearly 90 MIPs Ø The processor has a USB 2. 0 port and CAN connectivity.
DSP Operations Ø Once the VR signal is digitized, all sorts of nifty signal processing operations can be performed. Ø The real beauty of maintaining the VR/Hall amplitude data correlated with time sampling is multi-fold: Ø 14 Ø Data operations can be performed in real-time with minimal phase delay effects. Ø Variances in zero-crossing points from effects like L-C phase shifts can be compensated. Ø Zero-crossing thresholding can take on many attributes depending on RPM, amplitude, etc. The following illustrates some of the signal processing techniques implemented…. .
DSP Operations – Median Filter 15 Ø Suppose the VR or Hall signal has ignition noise spikes embedded in the true signal: Ø Is there a simple way to detect the spikes and remove them, leaving the original signal?
Rescued by a Median Filter 16 Ø A Median Filter are non-linear filters that known for their ability to effectively remove impulse noise. Ø Median Filters are non-linear because they do not follow superposition: Ø Gosh – what an Yickky equation…. Luckily its much simpler than this. Ø Follow the upcoming example….
Median Filter – How to do it… Ø Suppose the ADC streaming samples values and one of them is a noise spike 2, 3, 5, 4, 6, 29, 8, 9 17 Ø For us it’s easy to see that the 29 value is a spike noise because its not the same magnitude as the surrounding values. Ø What the median filter does is take an odd number of values, sorts them numerically, and uses the middle value. Ø For this example, assume the sample size is 3 (odd number), as samples come in the sort is performed and the middle value is kept…
Median Filter – 1 st Set Ø Taking the first three values: 2, 3, 5, 4, 6, 29, 8, 9 Ø Sorting them: 2, 3, 5 Ø Use the middle (median number): 2, 3, 5 18
Median Filter – 2 nd Set Ø Taking the first three values: 2, 3, 5, 4, 6, 29, 8, 9 Ø Sorting them: 3, 4, 5 Ø Use the middle (median number): 3, 4, 5 19
Median Filter – 3 rd Set Ø Taking the first three values: 2, 3, 5, 4, 6, 29, 8, 9 Ø Sorting them: 4, 5, 6 Ø Use the middle (median number): 4, 5, 6 20
Median Filter – 4 th Set Ø Taking the first three values: 2, 3, 5, 4, 6, 29, 8, 9 Ø Sorting them: 4, 6, 29 Ø Use the middle (median number): 4, 6, 29 21
Median Filter – 5 th Set Ø Taking the first three values: 2, 3, 5, 4, 6, 29, 8, 9 Ø The 29 stays on the end… Sorting them: 6, 8, 29 Ø Use the middle (median number): 6, 8, 29 22
Median Filter – 6 th Set Ø Taking the first three values: 2, 3, 5, 4, 6, 29, 8, 9 Ø The 29 stays on the end… Sorting them: 8, 9, 29 Ø Use the middle (median number): 8, 9, 29 23
Median Filter – Its Magic!!! Ø So, the original series with the 29 noise spike: 2, 3, 5, 4, 6, 29, 8, 9 Ø Becomes the new series that is nice and clean: 3, 4, 5, 4, 6, 8, 9 Ø 24 Note the following: Ø You have to use an odd number of samples Ø There is a time-delay introduced that is ½ of the sample size – but this can be accounted for in the zero-crossing value to compensate. Ø Larger sample sizes have increased computation time because of the sorting – a sort size of 3, 5, 7, or 9 is plenty!
Median Filtering is used in Image Processing to remove sharp edges 25 Noise is a sharp edge, too.
Adaptive Thresholding - VR 26 Ø Detecting the threshold is easy – just find the zerocrossing point on the waveform. Ø However, during zero-crossing the VR output voltage is near zero volts, and it is most susceptible to outside noise effects. Ø At higher wheel speeds (resulting in high signal p-p amplitudes) the detection threshold can be expanded to help eliminate noise detection.
Thresholding - Amplitude 27 Ø Wheel teeth often are not precise – this will cause direct amplitude variations in VR signal: Ø This amplitude variation will cause all sorts of headaches for adaptive VR sensor circuits!
Thresholding - Amplitude Ø Many VR signal conditioner circuits (e. g. LM 1815) use an adaptive mode based on amplitude – the circuit re-arms when the current waveform passes 80% of the previous waveform’s amplitude: Peak 80% Doesn’t quite measure up… Ø 28 Variations in amplitude due to teeth mismatch or acceleration/decel can result in missed teeth.
Thresholding - Frequency 29 Ø The use of frequency can be a better indicator: Ø Waveform period can be used to scale threshold.
Adaptive Thresholding - Freq 30 Ø Detecting the threshold using the waveform period has a potential issue – missing or non-periodical tooth counts will result in varying periods. Ø The solution is to implement a lag filter arrangement which will effectively smooth out the tooth periodic variations, as well as cycle-to-cycle changes: Ø You pick a value of Coeff that matches the number of teeth. With this, along with an amplitude scale factor At you can arrive at a threshold function Th:
The Phantom Tooth…. . Phantom Tooth - Yikes! Missing Tooth Area 31
The Phantom Tooth 32 Ø Have you ever had the problem where you get good sync up to a certain RPM when it then goes all to hell? Ø During the missing tooth portion, any little bump, etc on the wheel can result in a new tooth signal – a Phantom Tooth. You can blame the low df/dt for this. Ø During low RPMs the PT amplitude is also low. Ø Higher RPMs the phantom tooth grows… Ø Eventually the phantom tooth crosses the threshold and becomes a real tooth – scary stuff here! Ø No fear – the frequency-based threshold adjustment will exterminate the phantom tooth!
Channel Crosstalk 33 Ø There are several distributor arrangements which include crank and cam position sensors (i. e. Toyota 5 M-GE, Honda Civic, etc). Ø For example, the Honda distributor yields a channel with 24 pulses, another with 4 pulses, and the third with one pulse per revolution. Ø Since the sensors are in close proximity the signals from one channel can couple into the other channels. Ø This crosstalk can cause issues at higher RPM and low threshold settings….
Crosstalk 34 Ø Here are the 24 and 1 pulse outputs: Ø But the channels like to interfere with each other. .
Crosstalk 35 Ø One channel’s amplitude spills into the other one: Ø However, we have a trick to fix this….
Channel Crosstalk 36 Ø Since we are digitizing each waveform we can perform our own “anti-crosstalk”. Ø We can generate a feed-forward signal based on adjacent channels and simply add it back to the channel in question. . It’s the exact same way that adaptive noise cancellation headphones work! Ø If you take the interfering channel and invert the signal (phase shift by 180 degrees) and add it back to the channel in question you effectively remove the crosstalk:
Channel Crosstalk 37 Ø The only trick with this is that the amount of amplitude of the offending channel’s signal needs to be determined in order to apply the same amount out -of-phase. This can be done by monitoring the signal while changing the feed-forward gain. Ø Multiple channels is just the sum of each channel – simply determine the inverse of each channel and add it back in! Ø Since all amplitudes for all channels increase proportionally (same distributor shaft for all, and signal is proportional to RPM) one feed forward gain per offending channel is sufficient. Ø Also note that the adaptive thresholding also can eliminate channel crosstalk…
Other Filtering Techniques 38 Ø Since we have the sampled data, we can implement other DSP algorithms to eliminate noise…. Ø One simple method is the Least-Mean-Square Adaptive Filter (LMS, a. k. a. Weiner Filter). This is basically a self-adjust Finite-Impulse Response (FIR) filter arrangement: Ø This is something we will investigate…
39
- Calendar april 2009
- 27 april 2003
- 2003 april 20
- Till we meet till we meet at jesus feet
- Megacontent
- Kilo hekto deci centi milli
- Mision imposible 6 mega
- Taylor overby mega
- Mega phylogenetics
- Mega giga tera peta
- Ecuaciones dimensionales
- Mega reto
- Mega vs giga
- Cakrawala mega indah
- Arduino alimentazione esterna
- Mega tourism
- Mikro makro mega
- Mega vesica
- Sifat mega
- Mega giga tera
- See think wonder
- How many miles is it from
- Centi
- Mega goal 4 unit 1
- Long lot survey system ap human geography
- Mega prefix
- Ottawit mega stores inc
- Mega actor
- Juan pablo laguna la mega
- People laughing
- Mega diversity countries
- Arduino mega wikipedia
- Mega 128
- Engineering notation
- Arduino interrupt timer esempio
- Nizamettin aydin
- Deci mili micro
- Remote peering in osaka
- Niveis virtua net