These days ECUs handle ignition and they can be programmed any which way a manufacturer sees fit. There's still the issue of lousy gas, of course, and that is generally addressed via our good old friend under the exhaust manifold, the knock sensor. The sensor listens to the engine and his duties are to inform management when things get rowdy: "Gentlemen, excuse me, but we have knock down here. Please retard timing by two degrees at 6,600 rpm at 82 kPA of vacuum. Oh, and it seems like I've noticed that before, so you might want to write it down somewhere and simply take ignition back two degrees next time we hit that spot. Thanks!"
What this means is that knock sensing can work very much like short term and long term fuel trim. When knocking is detected, the ECU reduces ignition advance to handle the knocking at hand (one could call this "short term knock trim"), but it also stores this information and integrates it into some sort of knock table ("long term knock trim") so that next time the engine encounters this condition, ignition will automatically be retarded. So thanks to the lowly knock sensor and some smart ECU programming, we can now run very aggressive ignition timing and then simply retard it either on the fly or based on knocking history.
This whole issue, however, is not without problems. For example, it's pretty noisy down there, so figuring out what is just noise and what is knock can be difficult. Some manufacturers use special pattern recognition routines and filters that reject general noise and only react to the special sound patterns of knocking/detonation.
Then there are the other questions: Should we memorize knocking and include it in tables or should we simply listen to the sensor and always retard on the fly, without any learning? Or should we simply disregard that silly little microphone down there and use programmable ignition maps instead? Each approach is being used, and probably some others as well.
Subaru, for example, uses a method where they have a worst case map and a best case map (i.e. how advanced ignition can ever get), and what's actually used is an ever-changing map somewhere inbetween. What that map looks like is pretty much based on what the knock sensor reports.
My turbo PT Cruiser uses knock information to create short term and long term knock correction tables. Those work the same as our fuel trim correction systems, although I am not sure if the ECU stores an ignition retarding value for each cell or if it invokes entire separate ignition maps.
It is not entirely clear how the stock K-Series engines handle knock sensor input. Based on what I have heard (mostly from Hondata) and seen with my own eyes (in an Acura TSX with the Honda diagnostics computer hooked up), I tend to believe that the stock ECU uses a long-term ignition correction map that's then overlayed over the various cam angle ignition maps. Knock sensor input is probably used to gradually modify those long-term correction maps. It's been said (by Hondata) that the stock knock sensor does not seem to have much short term (real-time) impact on ignition, so for performance purposes, an alternate solution may be better.
Hondata's position is also unclear to me. Doug MacMillan told me they had observed the K-Series engine retarding ignition when there was absolutely no knock, but not reacting much to actual knock signals. Their conclusion was that Honda's knock sensing algorithms were not very good. Doug said it was as if Honda had created a race gas ignition table and then a "negative ignition overlap" that subtracts timing regardless of actual knocking. My own conclusion here would be that Honda uses a long term knock table that is only gradually changed via short term knock sensor input. However, in articles Hondata also stated that the knock sensor seems very sensitive and that they observed power variations between cars that seemed to go away when knock sensor input was turned off. Hondata at some point also said: "The stock ECU does not retard ignition under knock conditions, despite doing so being a smart idea." From that one might conclude that short term knock sensor input, while possibly/probably having an impact on long term knock tables, does not affect current timing as much as it should.
From what I can tell, in the K-Pro Hondata disabled the factory knock algorithms and overwrote the knock tables in favor of boost tables. To my way of thinking this means the ECU does not react to engine knocking at all and everything relies on the actual fixed ignition maps. Why would Hondata have done it that way? Well, perhaps they did not have a way to decode the actual knock sensor algorithm, which means no matter how they reconfigured the ignition maps, the knock sensor would have simply changed the correction tables to pull back ignition to where it feels it should be. Two better ways would be to limit the amount the knock sensor could impact the knock correction tables, or to only allow for short term knock trim.
Yes, this is a big, massive issue. It is one that most car manufacturers grapple with these days, and getting it right, or advancing the state-of-the-art can have a very significant impact on performance, economy, emissions, and longevity. Hondata has answered many of my questions and I related much of my current understanding of things in this brief article. Still, I wish I knew a lot more (for certain and not just as educated guesses) as it is QUITE important to making our tuning decisions. As we go on in this journey of unleashing power via electronics, we'll need to know as much about what the ECU does (both stock and modified) as in the past we knew what cams, turbos and other hardware goodies do. If part of ECU behavior remains in "black box" territory, then that would not be a good thing.
It'd be great to have some authoritative literature out there on this topic, but I haven't been able to find much. -- Conrad H. Blickenstorfer