more musings on EFI controllers

[Paul E. Campbell]

> The reasons for using digital logic and microprocessors: they're
> "neat", well defined, reproducible.  They appeal to our egos because
> we can actually "control" something.  Analog tends to be "fuzzy",
> irritating and often is the controller rather than the controllee.
> But, it has the advantage of being faster to develop, easier to
> implement (depends on the complexity of the system), cheaper,
> and easier to "twiddle".

Not anymore. It takes a while to get used to some of the "tricks" that one
does with digital circuits. Also, you can "twiddle" digital just as easily
via computer. I always use battery backed SRAM's or EEPROM's in all my
projects because it takes seconds to reprogram them.

Since Lattice came out with their ISP stuff, I can just plug in a $20 chip
and infinitely reprogram it using a PC and $100 for the starter's development
package.

In short, digital stuff is now just as easy to implement and just as easy
to twiddle. Other than if you've got a 4 op-amp circuit, I don't think
analog is all that much cheaper, either.

However, the "way to do things" is not necessarily obvious. All-hardware
PLL's, tricks with counters, and such are not necessarily obvious at
first. If you can't handle that sort of thing, you'll probably go the route
that all the regular manufacturers do; do it 99% in software. This costs
more than hardware assisted software or all hardware unless your application
is trivial enough to run on a PIC microcontroller or the complexity is pretty
steep.

> If you'll read my latest posting, you'll see that I use
> a Holley aftermarket EFI throttle body system that is analog (well,
> the timing for the injection pulse, etc. is probably a digital
> function) and uses the minimum of sensors: namely TPS and temperature.
> This system works very well (it costs about $600 for everything) when
> you consider that their next more sophisticated (read digital)
> system costs double that.

Okay..assuming you can get a throttle body that works, the next issue
is which sensors to use and such. Fuel injectors will cost you say $30
(using a J.C. Whitney catalog and rounding up a little). The little
driver chip is around $5-$10; just a high current driver transistor
with some extra smarts.

As for the digital/analog end of controlling it, for analog, you'd use
a ramp generator and a comparator. In digital, you'd use a countdown
clock with a programmable start point. Programming the "curves" can be
as simple as using a ROM or a GAL. Either way, this is also low cost.

The temperature sensors are thermistors. The TPS is just a potentiometer.
Now, here the TPS might be a bigger problem. Unless you switched the TPS
for an optical encoder of some sort, you're going to have to deal some
sort of ADC. Analog just uses the potentiometer directly. Course analog
circuitry has to deal with the nonlinearity problem, too. Same is true
with the thermistor. Recommended route here since neither of these are
tremendously high speed is a slope detector ADC or something else equally
cheap.

Temperature sensors are around $5. TPS sensors are $5 and usually come
with the throttle body.

I could scrape a system identical to the Holley together for less than
$200, excluding the throttle body itself. I have no idea who makes custom
throttle bodies or where to get one, so I'd head to the local junk yard.

Now on the "processing" end, practically everything in your "simple"
system is a matter of programming the fuel pulse map. The easiest way to
do this is to simply generate the fuel map in a ROM and use the ROM for
your pulse width mapping. There is no "processor" as such. EPROMs are
as cheap as the op-amp circuit and they will trivially handle
nonlinearities of any conceivable sort.

So the simplest solution on the electronics end of things is a voltage
regulator, an EEPROM, a 2 channel ADC, a buffer (for the ADC
outputs), a programmable count down timer, a clock, an injector driver,
and some glue logic. Although the EEPROM may be separate, the buffer,
clock, counter/timer, and glue logic could probably be fit into a GAL
or other programmable logic chip. How's that for simple?

> Now, the advantage of this system is that you could put potentiometers
> allowing adjustment of how much each of these factors contributed and
> they could be placed where you could easily access them so real-time
> adjustments could be made -- and wouldn't require a computer in your
> lap, much less re-programming and assembling or compiling.

Ahh! You got me there. I've still got a reprogramming step. But a computer
in your lap isn't really that bad. Everybody around here is an expert at
balancing a beer can in one hand, shifting with the other, and steering
with their knees. You'd have to put down the beer can and fool with the
computer. On the other hand, you could route the serial cable back into
the cab trivially.

> The complexity increases exponentially (same with a digital system)
> as more control is attempted (read more sensors and therefore
> more calculations).  Adding EGO feedback would be done in the same
> manner as before (remember, this is my speculation only, not gospel).
> The signal would be processed and summed into the ramp generator, but
> now there's a serious problem -- the EGO output is extremely non-
> linear, and that becomes a B-I-G problem with analog.  (With micro-
> controller, one can store values in a lookup table for curve
> correction.)

It doesn't become exponentially more complex. Once you've gotten to the
level that you're going to have to use a DSP instead of the trivial stuff
you and I mentioned, then there's no inherent added complexity to the
hardware end. Simply add more sensors and/or actuators. On the software
side, things can get as bad as you want them to.

In what I've learned so far, it appears that if you want to do anything
very exotic and still be cost effective, you might as well give in to the
idea that you HAVE to have a DSP, not an 803x, 805x, 68xx, 80xx, 80x86, or
68xxx microcontroller. Analog Devices makes it a trivial choice with their
ADSP-21xx family that has a $100 starter's kit and $15 processors at the
bottom end that are hardly "bottom end". I'm sure TI, Acorn, and others
will follow suit. Motorola will keep catering to the high end of course.

> Well, it seems to me that the point of diminishing returns is rapidly
> found.  But, if the objective is to go fast, it is much less complicated.

Yeah, get a carburator. Screw fuel economy. Just run rich.

My objective is to upgrade my trail machine (Bronco with 36" swampers). It's
a carburated straight 6. No such thing as "aftermarket EFI" unless I buy a
funky intake manifold that bolts to 4 barrel carbs and Holley's 4 barrel
carb EFI replacement unit or move up to a V8. So my choice is going to be
limited to whose throttle body will fit on the carburator's inlet and can
I afford 6 separate fuel injectors unless I try to get one of those GM
single injector things working.