ECM Proportional Control

Robert Rauscher
7/9/2000, updated 12/4/00

Copyright 2000, Robert Rauscher, All Rights Reserved.

This is specific to the 747 ecm. However, the methods and madness
applies to many of the GM ecm's. (Use at your own risk, no guarantee
that any of this is correct).

Note:

        Values such as 0x22D are eprom table addresses. These you
        can easily change.

        Values such as L0004 are RAM variables. These are used by
        the ecm during it's processing.


INTRO:

The proportional and integral fuel correction logic is complex. There
is no way that I can cover all aspects of the ecm logic. This paper
is designed to give a working view of what the ecm is trying to
accomplish. I have also covered some changes that can help when
tuning hot street engine.

The proportional PID term starts with the O2 sensor reading, and ends
with a correction to the final calculated bpw. The PID proportional term
always works in the opposite direction of the immediate rich/lean O2
sensor reading.

The purpose of the PID proportional controller term is twofold. First,
it is used to attain the desired AFR quickly. Second, it is used to
maintain the switching of the O2 sensor about the desired AFR point.

The PID controller term starts with the O2 Error Term. This is the
amplitude of error (or difference), between the O2 sensor and the
desired O2. This value is used as a lookup index for the actual
proportional error term.

The combination of the proportional term and the INT term is referred
to as the closed loop correction term (CLT). When in open loop, neither
the proportional term or the INT term as used.

There are two filtered O2 sensor values kept by the ECM. One is a fast
filtered value that follows the changing O2 sensor reading closely. The
second is a slow filtered value that lags the changing O2 sensor reading.


AIRFLOW TERM:

The volume of air moving through the engine is used for many of the
PID term lookups. The airflow is defined as grams of air, per second,
per cylinder, that is moving through the engine. The term is abbreviated
as gms/sec.

This term is calculated from the current MAP and current RPM. Volumetric
efficency is not used. In the '747 code, the max value is clipped at
64 gms/sec. The calculation is gms/sec = rpm * map * constant. It is
not the most accurate term, however, it is sufficent.

It is important to realise that it takes a finite amount of time for a
change in the bpw to be reflected in the O2 sensor reading. This time
frame is referred to as the 'transport delay'. This delay varies according
to the airflow through the engine.


O2 ERROR BOUNDRY TABLES:

Two tables define the desired O2 sensor upper and lower limits. These
limits are used to decide whether the O2 sensor is either: below, above,
or within the limit window. These are the error boundries. These values
are the actual voltage output of the O2 sensor.

A third table defines the O2 sensor cross point. 

The desired O2 Sensor tables are located at: 0x494, 0x499 and 0x49E. The
tables are indexed based on airflow, in gms/second. They are five-line
tables from 0 - 64 gms/sec, every 16 gms/sec.


0x494: Mean R/L Threshold, used to determine (somehow), the direction,
(and speed?), of the O2 sensor. End result is a moving richer, or
moving leaner bit (L000E, b6). Also used to determine the occurance
and an O2 x-count.


0x499: Rich O2 Threshold, used to define upper window boundry for
the O2 Error Term.


0x49E: Lean O2 Threshold, used to define lower window boundry for
the O2 Error Term.

The above table values are converted by: O2mv = TblVal * 4.34

It is rare that any of these three tables will need to be changed.

!) The slow filtered O2 error term is the delta of the current slow
filtered O2 value and the proper window boundry, upper or lower. This
term is then used to lookup the proportional gain error from another
table.

The lower window boundry is used to calculate the error term whenever
the O2 Sensor is below that value. The upper window boundry is used
whenever the O2 Sensor is above that value.


This error value is then used to lookup various proportional values. If
the Slow Filtered O2 value is within the window, no proportional
correction takes place (prop term = 0).

When at idle, defined by L0002, b7 == 1, the above tables will have
a bias added to them. The effect is to lower the AFR for a more stable
idle. This value is located at: 0x2F7. (remember this one).

The flag at L0008, b1, (rich/lean) is set with a O2 sensor higher than
the upper window boundry. That bit is reset with an O2 sensor lower
than the lower window boundry.

If the O2 sensor is within the window, the bit is not changed.


PROPORTIONAL ERROR TERM:

Once the slow filtered error term is calculated, it is used as a
lookup index into the proportional gain table at: 0x4A8.

Another lookup is done for a proportional multiplier term, indexed
by airflow in gms/sec, at: 0x4C7.

The two above terms are multiplied together, divided by 256 and
the result is used as the adjustment to the CLT term. 

The math:  

    PropTerm = (Gain @ 0x4A8[O2err] * FlowGain @ 0x4C7[gms/sec]) / 256.


This makes the FlowGain (@ 0x4C7), dictate a percentage of the gain
table (@ 0x4A8). The end result will always be lower then the starting
gain term.

To create the CLT, the proportional term is either subtracted, or added
to the INT term. If the O2 sensor is lean, the prop term is added to the
INT. If the O2 sensor is rich, the prop term is subtracted from the INT
term. The result is the CLT term.

The CLT term is then used to adjust the BPW calculation.


For highly tuned engines (IE: hot street stuff), the proportional
gains can to too high. The effect is the engine surging under steady
state cruising. Reducing the gain and multiplier values can reduce
this problem. The values in these tables can sometimes be cut in half.


PROPORTIONAL DURATION TERM:

For the proportional duration term, lookups are done using the same
indexes as the prop-gain. The slow filtered O2 error term is used as
a lookup index into the proportional duration table at: 0x4B5.

Another lookup is done for a proportional duration offset term, indexed
by airflow in gms/second, at: 0x4C2.

The results of the lookups are added together for the proportional
duration term.


AFR:

The upper and lower R/L threshold tables can be used to move the
basic AFR. As they form the O2 Error Window, by raising the table values
toward the richer side, the overall engine AFR can be pushed toward
the rich side (lowered AFR).

This is important to understand, these three tables define the overall
AFR when in closed loop. The proportional term is always trying to
move the current O2 sensor toward these table boundries.


INTEGRATOR:

The INT seems to follow the proportional term, however, is does not. The
INT moves independent of the proportional term. The INT is updated
whenever an O2 sensor x-count does not occur within a certain timeframe.
Even with the proportional gains set to 0, the INT will still move.

Movement of the INT takes place whenever the slow filtered O2 error
term exceeds the value at: 0x2F4, which is typically 4, along with the
INT delay timer expiring without any x-counts.


STOICH:

The value labeled 'stoich' needs to match the O2 sensor AFR switch
point. On the '747, this value is located at: 0x2AA. The value is 147,
for 14.7 AFR.

Conjecture: This stoich value, plus the R/L O2 error boundries, define
the O2 sensor output levels. This ties the O2 sensor calibration into
the bpw equation.

As the R/L O2 Error Boundries define actual O2 sensor stoich switch
point in output volts. The 14.7 AFR switch point of the O2 sensor is
now 'calibrated' into the bpw calculation.

If a different fuel is used, with say a stoich value of 16.1, by
placing the value of 161 at 0x2AA, the bpw is again calibrated for
that fuel. If you now want to command the ecm to change the bpw enough
for a 15.3 AFR, by using 153 in the bpw calculation, the bpw will
be increased the proper amount to produce that AFR (in open loop).

Remember that when in closed loop, the stoich value is used in
the bpw calculation. In open loop, the desired AFR is used in the
bpw calculation (say, when it's 5 degrees out, and the engine is
first started, the desired afr might be 9.0).

Another example: Say a new O2 sensor was developed, for gasoline,
that costs $3.00. Good deal. However, the stoich switch point output
voltage is at 625mV, opposed to the current sensors switch point of 450mV.

By changing the R/L O2 error boundry table values, this new O2 sensor
can be calibrated for the bpw calculation. Instead of the tables
straddling the 450mV value, set the table values to straddle the new
625mV O2 sensor value. 


IDLE:

When in idle, the proportional gain and the proportional duration
use specific values. The prop gain is at: 0x2F9, and the prop duration
is at: 0x2F8. These replace the table lookups at: 0x4C7 and 0x4B5,
respectively.

For a rumpity-rump engine, it helps to set the idle prop term low.
This reduces/eliminates proportional action. This can have the effect
of smoothing the idle.

Also, do not forget about the O2 error bias value at: 0x2F7. This
value can be used to decrease the AFR at idle. Again, for a rumpity
engine, it can calm down the idle.


RESET's:

A proportional gain and duration reset take place whenever the INT is
reset. It is also reset whenever an O2 x-count is logged. The most common
cause of an INT reset, is the BLM 'cell in use' changing. There is an
option to have the INT reset on an accel enrichment (pump shot). An INT
reset is the setting of the integrator to it's mid-point (128). This
is where no integrator fuel correction is made.

Enable 'pump shot' reset of INT is located at: 0x006, b5.

Enable 'BLM change of cell in use' reset of INT is located at: 0x006, b4. 

(set to 1 to enable option for reset).

Other causes of an INT reset: DFCO, PE, open loop.


PID:

The PID part of all this is: 

Proportional: This is the Proportional Term described above.

Integral: This is the Integrator or INT term mentioned above.

Derivative: There is either not one used, or the resets of the INT
and Proportional Term could be considered. (note that the loop is
slow enough to not require a derivative term).


-eof-