What I am looking for here is a way to settle down the closed loop response.  I don't mean just slowing it down because I do want it to acquire the closed loop condition quickly and to respond smartly to changes in the motor's load.  My feeling is that the loop is over controlling the engine, making changes to the mixture greater than are needed.  An example of this is during cruise, say around 55-60 mph.  I would see the O2 voltage swing back and forth, as it should, but I could also feel the affect on the motor as the closed loop fuel modifier changes the injector's pulse width  back and forth chasing the O2 voltage.  You could here the tone of the exhaust change pitch along with the smoothness of the engine and a slight surging as the closed loop fuel modifier would adjust the mixture to the motor way too lean then way too rich.  True, the average would be around the target stochiometric AFR, but the loop was way over controlling to keep it there.  The surging may be slightly aggravated in my situation because my 350 and cam combination does not like excessively lean mixtures.  Also the open exhaust and the fact that most of the time the rear window on the coupe is down you are really aware of what the motor is doing.
    If you haven't studied the code yet maybe a little background of the key players might be in order here.  There are two major loops running in the code.  The first one, with the highest priority, is the Ignition Trigger handler.  Obviously every time an ignition pulse comes along from the distributor this routine takes over what ever else is going on at the moment and does its thing.  The ignition handler is responsible for reading the O2 sensor ADC and determining what direction to adjust the closed loop fuel modifier (L0079) needs to be moved (of coarse this assumes that all closed loop rules have been met).  If the O2 voltage is below 447 mv then the value of L0079 is adjusted to increase the mixture, otherwise it will move in a leaner direction.  The value stashed in L0079 is a signed integer that can move a maximum of +/-16,000 counts.  If this value is positive it is subtracted from the calculated pulse width reducing the AFR, if negative it will make the mixture richer.  A value of 16,000 represents approximately a 20% change in fuel.  The amount that this accumulator (L0079) is changed by is in  L0075.  So L0079 is how far the mixture is adjusted and L0075 is the rate it is adjusted by.  The ignition trigger handler itself doesn't actually apply the value in L0079 to the injector's pulse width.  The is done by the second major routine.  This loop waits till the MAP ADC is read by the Ignition Trigger handler (every ignition event) then starts processing all the sensor inputs and computes what injector pulse width and spark angle that the Ignition Trigger routine should use and apply to the engine.  It is this loop that actually determines if it is ok to go into closed loop, determines the rate (L0075) to modify the accumulator, and apply what is in the closed loop accumulator (L0079) to the pulse width that it has just calculated.  So in a nut shell the Ignition Trigger routine determines the direction and applies the rate modifier (L0075) to the closed loop fuel accumulator (L0079) while the main calibration loop calculates pulse width, applies the accumulator, and determines the rate of change.
    At first I approached this by attempting to slow down the accumulator when the O2 error voltage was off its rail, or at least a somewhat arbitrary point close to the rail, by reducing the rate value in L0075.  I tried several variations of reducing the rate value as the O2 error voltage approached the nominal center voltage of 447 mv.  It helped slightly but wasn't very effective.  The big trouble here is the the O2 sensor's response is very slow and always lagged well behind what the fuel accumulator was doing.  Even if I completely zeroed the rate value in L0075 when the O2 voltage was off the rail (say between 800mv and 100mv) it made hardly any difference at all to how far the fuel accumulator would go.  The accumulator value would already be considerably too far the wrong direction before the O2 error voltage would even begin to change.  So this approach alone wasn't going to do the trick.  Another thing that aggravated the situation was in the way they calculated the rate of change value in L0075.  The magnitude calculated was a direct function of RPM.  The faster the motor turned the larger this value was.  Two things caused this.  One was the lookup tables (at LF554 and LF544) both increased with RPM and the other effect was that as the motor turned faster the MAP ADC was read more often releasing the main calibration loop and applying the rate value in L0075 to the accumulator (L0075) more often.  It looks like the O2 sensor's response time is pretty much fixed (at least at a given operating temperature) so by increasing the rate of change value, and how often it was applied to the accumulator, made the situation of over compensating worse the faster the motor turned.  I needed a way to normalize the rate value so that no matter how fast the motor turned it would modify the accumulator at a fixed rate (time) also.  The rate varied from about 3.3%/sec @ 1500 rpm to 18%/sec @ 4000 RPM.  What I did was come up with a set of values that would normalize the rate value to RPM.  There were originally two lookup tables, one at LF554 that was used when the O2 error voltage is below 447 mv (lean) and the other at LF544 for when the error voltage was above 447 mv (rich).  What I did was split the difference between the two and pick a point at 2500 RPM to create my new set of values.  This works out to 6000 counts/second, or 7.5%/sec.  Seemed like a reasonable starting point.  The below plots show what originally was calculated for L0075 vs RPM.   The second plot is of the results with my new set of values.  The third plot shows the O2 error voltage break points.  This is where the rate value is further modified as the error voltage gets closed to 447 mv.

    The first one here shows results of the two original tables.  You can see that rate for when the mixture was lean is twice as much as for the one if the mixture was rich.  The blue line represents how many times the rate value (L0075) is being applied to the accumulator value (L0079).  The faster the motor turns the more the ECU would over compensate driving the accumulator much farther than needed.
 
 

    This second plots shows  what is now being calculated for the rate value in L0075.  The blue line represents what the output of the calculations should be.  The lookup table only has eight entries so I can't get an exact match, but you can see by the red line that I got pretty close.  The green line is the effective results.  I was trying for a count of 6000/sec (7.5%/sec) so you can see that it very close to target.
 
 

    This last plot shows the breakpoints where the rate modifier in L0075 is further modified when the O2 error voltage starts moving off a rail.  The 'Y' unit is a numerator in the scaling of this value: L0075 = L0075 * Y/16. Currently any time the O2 error voltage is above 92 mv and below 828 mv this further scaling is applied to the rate modifier.  This wasn't very effective for off idle operation but it really helped the closed loop idle stability.

    Well at this point even with the rate modifier being calculated based on time rather than RPM and the little help that the above O2 error voltage vs modification rate table gives me the loop was still over compensating.  I didn't want to just slow every thing down because I still wanted the loop to acquire lock quickly.  What I finally came up with is a scheme that so far has been working out well.  What I do is when the error voltage crosses 447 mv I immediately move the accumulator value in L0079 half way back to the value it had at the last 447 mv crossing of the O2 sensor and then hold off any more modifications to it for a period of time (currently around 300-400 ms).  What this does is give the O2 sensor time to catch up with what the motor is actually doing and still allows the loop to acquire lock quickly.

    The below plots show some before and after operations.  They represent cruise operation around 2500 RPM.  Check out the Scaling Page for units explanations.

    This is one of my early test runs.  The O2 loop routines are in their stock form at this point in time .  You can see that the fuel is swinging back and forth most of the time around 10%-12%,  and at times closer to 14% to 16%.  You could feel and hear the motor leaning out at the bottom of those dips.  Also notice that the O2 error voltage is not really symmetrical.  There is more time spent on the rich side than the lean.  As steep as the error voltage transitions are you can see way that just reducing the rate of change during the transition period had little affect.
 
 

    Here are two runs that were taken after I applied all my modifications.  You can clearly see that the fuel modification value is much more stable without all those wild swings.  The peak-to-peak is much closer to 1%-2% with a much smoother and consistent average.  Also notice that the O2 error voltage peaks are not as great and that there is much more activity, at least twice as much, around the 447 mv crossing point than before.  No more extreme swings into the excessively lean areas.  The motor purrs like a kitten now with no more surging and the exhaust tone is steady.  The loop will still acquire lock just as fast as it did before because the rates outside of the capture range are about the same.