FIXING DRIVABILITY ISSUES AND REV HANG:
It drives like a proper manual now.
I did three things, and I kinda did them all at once. So I don't know if it everything is having an effect or just one thing. I should have done more testing in between. I guess I'll update this little section in the future once I learn more.
1) Clean the idle air control solenoid valve - This was all gunked up with PCV blow by goo. Its pretty delicate, and makes rapid changes in position, so it is understandable that being buked up with goo will slow it down. It was operating allot better once clean
2) Cut off the "Air assist" circuit - Something I would ideally not do, but is a large part of the problem I think.
Basically, the "Air assist" circuit bleeds air to some little jets positioned around the fuel injector nozzels. The idea being, a jet of air is fed into the injector spray to help atomise the fuel and aide complete combustion. EJ WRXs have this "air assist" feature aswell.
The problem is, the air assist bleeds enough air for the engine to idle on its own (I tested that) and the computer must act to shut off the circuit. What I think was happening is the ECU waits around half a second to sense whether you have indeed began to coast and then it actuates the idle air control solenoid (which doubles as the air assist control valve) to cut off the air assist suddenly. During this time, there is enough air being bled to hold the revs to around 3000RPM which when shifting delays the upshift and when decelerating the sudden cut creates a sharp drop in engine torque which triggers transmission windup and oscillations (mini "kangaroo hops", or the familiar jerkiness manual subaru drive trains have when you give them a sudden throttle input in a low gear).
With the air assist blocked, (i'm pretty sure) the rev hang is reduced and the sudden uncontrollable drop in torque is pretty well eliminated.
IMAGE: This is how the idle air control solenoid valve controls the air assist bleed. Basically the valve moves "beyond closed" and covers the hole for the air assist witha s econd valve plate. This can only occur when the main valve is fully closed, so the ECU needs to be certain that deceleration is intended and that the manifold vacuum is not to high (read point 3) which is what I think causes the pause and the rev hang
3) Install an orifice plate under the idle air control solenoid valve so it limits how much air it can bleed.
PHOTO: The INLET is on the airbox side of the throttle plate, the outlet is on the engine side of the throttle plate. The "air assist" is not blocked off in this photo. To achieve the block, I just replaced that small hose with rubber plugs.
The orifice plate is a real bodge up with a hand drill. But it'll do. The air doesn't care where the hole is.
The idle air control solenoid valve does two things. 1) It controls the idle speed to maintain consistency whether A/C on or off, whether the engine is hot or cold, whether the air cleaner is clean or blocked, basically any external influence. 2) It acts to prevent a sudden vacuum forming in the manifold when the throttle is snapped closed, which is a condition that produces a temporary lean mixture and a large amount NOx emissions.
Basically in this car, in order to do this the idle air control solenoid valve seems to be in a closed loop control circuit with the Manifold Absolute Pressure sensor and it oscillates between fully open and partially closed while it figures out where it is meant to be to allow the revs to drop in a controlled manner. My theory is it is opening up too much as the throttle is released and the revs are hanging while it oscillates and figures out what its meant to do.
To dampen this effect, I've put an orifice plate so as the throttle is released, the idle air control solenoid valve can't supply enough air to hang the revs. Enough air still goes through to prevent the sudden vacuum forming (although not as well as before) and the idle can still compensate for A/C etc.
This may have an adverse effect on NOx emissions, but really, I need a drivable car first and foremost and I've tested that the revs still drop in a controlled fashion, much better than if the idle air loop is blocked completely.
So ideally air assist is a good thing and so is limiting the vacuum. But they are making the car difficult to drive. When this engine came out as an automatic, almost all of this would have been masked by the transmission and its torque converter, so its possible the systems are not as refined as they could be if this engine was released as a manual.
I am going to continue fiddling with these systems and determine whether air assist has a noticeable effect on fuel economy and whether I have the optimal sized hole in the orifice plate. Or, whether cleaning the valve negates the need for the orifice plate at all.
Andy