Note on the pdf document showing the thermal MAF sensor output.

 The sensor output voltage vs actual mass air flow is nonlinear and not a
 simple function.  It as a complex polynomial and can only be reasonably
 dealt with by a lookup table using a uC as the controller.

 This effectively makes a simple analog-555 approach impossible.
 Craig Taylor


 Unfortunately I have to agree but we have come up with a custom $5 555 that
 solves the problem. It consist of an A to D followed with a look up table
 followed by a PWM generator all for $2.50 less programming and shipping:)

 Paul Lamar

 Craig replies;

 I think that is the best approach.

 Craig Taylor



Is the mass flow vs. voltage chart real data (taken on an actual sensor)?

I plotted it and then did a curve fit (two ways).  The curve fits
reasonably well to an exponential regression..  I would think it
possible to come up with an analog circuit that follows the curve
reasonably well.  I believe the gain curve of many transistors (around
Vt) is somewhat exponential..

Then if needed, add some piece-wise comparator based tweaking to the
pulse width - above a certain flow voltage, adjust the fuel up (or
down) by a certain amount.

I'm just saying, don't get scared into digital by the non-linear
response of the MAF.  This is non-linear, but not in terrible way..
If you go to digital (a lookup table), do so for other reasons.


Matt-


I don't know but it matters not now as we can adjust the curve
based on dyno engine test using the look up table. It is much
simpler than it seems with perhaps one page of code.
Read the MAF voltage. Use the number to go into the look up table. Read the value in the table and send it to the PW generator. Start over.

The program code won't change from chip to chip. Only the values in the table.

With a ten bit A to D you have one part in 256 part resolution or
.02 volts. If the mixture range is 10:1 to 20:1 the resolution on mixture setting is .04. In other words 15:1 or 15.04 :1.


Paul Lamar


Matt, Paul

Matt you make a good point.  The results of a simple analog "linearizing" circuit won't be exact but could be within 5% or so without too much strain, I think.

 One suggestion is to use a rate multiplier to produce an X**2 (X squared) result.  That could be squished around to be an "X to the 2.xx" by changing the gain of the input to one of the multiplier ports.
I just might try to simulate that approach.  Of course, you can use a CMOS 555 as a rate multiplier.  Handy little chip :-)

Craig Taylor


Craig returns;
,

Rather than build a multiplier, it seemed more reasonable (thanks for the suggestion, Matt :-) to just use a simple piecewise non-linear shaper circuit.  The one attached has five different slopes and 4 break points.  The parts cost is maybe 50 cents for the opamp, resistors and diodes.  Resistors and diodes are no more than a few cents each in 100s.

 I used a 0-50V input for this demonstration because the larger voltage made the diode break points more apparent on the graph.  For 5V maximum input, the diode breaks visually disappear on the output curve.  That's because the diodes turn on over a 100 to 200mV range and blend the straight lines together.  Looks nice!

 The design can be done with a sheet of graph paper and a ruler and a few hours.  Just plot the MAF calibration table for your favorite MAF and then figure out how many segments you want to use to get "close enough".  I would think more than 10 would be overkill.

 The 40V supplied voltage divider sets the input voltage where each break occurs. The diode series resistors determine the slope of the segment between the breaks.  As each break point is passed, the diode associated with that voltage turns on and adds it's resistor in parallel with those already on.  Lower resistance=higher gain =steeper output slope.
 Use the opamp supply for the voltage divider bias supply.  Hey, it's already there!

To test out the design, replace the MAF output with a variable voltage (use a 10k - 50k pot).  Use a decent DVM to measure the output voltage of the opamp as you change the input with the pot.  Easy.

Craig Taylor

Craig, how long ago did you send this. I think your clock is off in your
computer. That moves you way up in the inbox and I might miss your messages.


You might have to set it each time you turn it on.


Paul Lamar

Paul, I sent the graph and circuit about 08 Oct, 2009.

My computer clock seems OK.  I am using the Yahoo mail system ;-)

The graph posted above was derived from the curve of the Ford MAF sensor that was posted here about mid September of 2009.  I have not tested either a MAF sensor or the piece-wise linearizing circuit.  It is all modeling done on a reliable SPICE simulator I have used for years.

 After lots of messing around with multiple analog techniques, I still think the look up table is the way to go.  The flexibility is just too powerful.  For instance, you can use it to "dial out" the dead time of the pulse fuel injectors.  At least to a first order.  That should greatly improve idle.

Craig Taylor


I have no idea why it showed up in my inbox as unread:)
Do you have any experience with the MOS Chip  I/O?
I am having a heck of time sorting that out.
I want to use it in the PS/2 bidi byte  mode but I can't
seem to get it configured properly.
The nomenclature does not seem to be standardized.

ELP!!! :)

Paul Lamar