RomRaider

I had an issue with stumbling when I touched the throttle after idling for a while, so I put all of these tables back to stock, and took another shot at modifying them. My main goal was to minimize the effect of this stuff at high load (fix the rich dip), without screwing up anything else (no more stumbles) so the key changes are in the tables that have a load axis.

I left things stock at low loads and low coolant temperatures where I possible, since I don't drive hard when the engine is cold, and I don't fully understand the intent behind all of this stuff. It seems to me that Subaru's tuners did a good job with these tables at low loads, they just didn't know about or care about what happens during spoolup (even with the stock turbo, intercooler, etc).

So:

The A term has a "rising load activation" table with load and ECT axes, and I zero'd it at 1.2 g/rev when the engine is warm. So at low load, and/or when code, it behaves like stock.

The "falling load activation" for input A only has an ECT axis, so unfortunately it can't be left stock a low loads. So, I zero'd it when the engine is warm. I have a hunch that this aspect of the stock tune (reducing fuel when you lift off the throttle) might contribute to the intermittent FBKC that I sometimes see at the end of a WOT pull, or when I lift to shift, but I haven't yet verified that changing this table makes a measurable difference in that behavior.

The B term has an activation table with a load axis, which is active for both rising and falling load. I just made it taper linearly to 0% at 0.8 g/rev. There's no science behind that change, it just made a nice straight line in the graph. I've included the stock version in the screenshot just for comparison.

I still need to do some more data logging to confirm these, but if the logs confirm what my butt-dyno is telling me, then I'm probably done tinkering with these tables.

And when I define these tables for other ROMs I'll likely just define these three, since they seem to be sufficient to fix the rich dip, without adversely affecting driveability. They're also relatively easy to explain and understand.

NSFW,

I made those same changes on my ROM, I will flash and let you know how it goes.
here's a screenshot for those using RR.

Those changes made my Final Fueling Base stay within 1-2% of the OL Fueling table.
Much smoother power delivery, awesome work!

Cool! I need to go find these tables in some more ROMs.

06 STI is next on the list, and someone with a spec B was also interested but I forgot who, and which year.

If you're feeling generous NSFW the 08's would be great (AZ1G201G)

I do think I've found the routines, by the way - it's just that they look quite fundamentally different from your rom, and it definitely looks like the right tables are in there, these are also slightly different (with different size axes for some - e.g. the JDM 3d table posted earlier in this thread, and different axes in some cases). I just don't trust my disassembly skills enough to call it.

If you've got a chance, here's my IDC file: http://dl.dropbox.com/u/3099895/2%20oct ... MPLETE.idc

The areas I've been exploring start at 0x2B960 - if you do a graph view, it's clearer. One of the branches seems to be the TauB routine (with the sigmoid tables in it) Start: 0x2BBC0
End: 0x2BDCE. The big pain is that there's a ton of indirect addressing (which I've mostly resolved), but also a whole lot of pushing stuff onto the stack and off the stack...

Anybody want to try tuning this for the USDM 06 STI?



Make a backup of your current ROM definition file first. You'll find it at:


Either copy the attached definition file to that location, or insert the XML above right below the "<include>32BITBASE</include>" line in the existing file. The start EcuFlash and open your ROM. You should see three new tables, like the bottom screenshot.

The first screenshot contains my suggested setup for these tables:

* Input A Falling Load Activation - I just subtracted 1.0 from every cell.

* Input A Rising Load Activation - At 1.0 load, with a warm engine, the multiplier is 50% of the stock value. At 1.2 load, with a warm engine, the multiplier is zero.

* Input B Activation - At 0.6 g/rev the multiplier is 0.5, and at 0.8 and higher, the multiplier is zero.

I'm not saying these settings are perfect but I think you'll find that the rich dip is about half what is with the stock settings. However I'd appreciate confirmation of that. I'd really like to see some logs from someone to confirm the results we've had so far with the LGTs.

First, log a couple of with your current tune, starting at around 5000 RPM and just stomping on the throttle until you reach redline. Log both "Primary Open Loop Enrichment" and "Final Fueling Base," and you should see FFB dipping rich while your turbo is spooling. Then make these changes and log a couple more pulls, and you should see FFB tracking your fuel table pretty closely. Hopefully the rich dip will be down to tolerable levels.

I'd really like to see results from both TMIC and FMIC cars, too.

Thanks!

NSFW, just wanted to thank you for your hard work on this issue. Its much appreciated.

You're welcome.

I just stumbled on Megasquirt's page about what they call "x-Tau," which I suspect it what these tables are all about. However I have to admit that I don't understand the way these tables interact anywhere well enough to relate them to the stuff discussed on this page:

http://www.megamanual.com/ms2/xtau.htm

There are many ways to implement wall-wetting corrections.
X-Tau algo with simple RPM/Temp decoupled corrections is not the best but it is much better then nothing.

You should also take into account that those corrections are intake air\intake wall (cylinder head) temperature dependent that is not the same as coolant temp used usually.

Very interesting. I assumed very early in this investigation that the axes on many of those tables were coolant temp because they share the same column values as tables that are definitely coolant temp, but at some point I found code that wrote to the RAM address that I thought was coolant temp, and it was doing more than simply coolant temp. RPM was a factor, if I remember correctly. I didn't figure it out, since it seemed to act a lot like coolant temp I did not spend much time on it... Wall temperature sounds very likely, I had not thought of that, but it makes sense.

Or could it be a fuel temperature compensation?

Also a good theory. I'll look for xrefs between fuel temp and the temperature variable used for this stuff.

Very interesting read this, thanks for the link. I'm very keen to try this out, not 100% confident that the tables I've discovered are these, as the logic is markedly different in the '08 STI's.

Thought I'd put this in here, since I referenced it incidentally in the SD thread:



So I've had a look at the data, I'm pretty confident I've found these for the '08 EDM Sti. I basically made the same changes as you did NSFW, and did some test runs by punching the throttle at RPMs above 4000 or so. What I'm seeing is that open loop fueling target and final fueling base are now basically identical at the point of rapid load change / rapid throttle change. I gather this is the desired effect? I'm running a (whopping large) TMIC, so I never noticed the rich dip until this discussion - but the car certainly feel like it responds to the throttle input much more readily now (which would make sense if the AFRs are not dipping). Hmmm... just noticed that on letting off the throttle, the OL target and final target are also identical now - I gather that's the declining load tables at work.