AbeFM

iTrader: (3)

I'm looking for an excuse to get it out of the car anyway. I think I'll do this.

I also worry I'm (somehow) sinking too much current through the sensors, cooking them.

But I doubt it. There's at least 470 ohms in there, which is, oh, a milliamp. Nothing macroscopic cooks at a milliamp.

AbeFM

iTrader: (3)

For posperity:


My only concern is that the smaller pullup will slow down the rise-rate, making the bigger cap more of a problem than with my current 4.7k/1nf set up

j_man

AbeFM, about the stock '99 sensors - use only the raising (tooth trailing) edges. Those are the correct ones. The falling (tooth leading)ones are incosistent at low rpm. Check the thread at mnet - I just added labels to your scope screenshot explaining which edges are ok and which not.
You don't have to tweak the sensor, move its position, etc. in order to get the falling edges working - just ignore those.

Me and Jason already run our AEM EMS flawless using the raising edges. I suspect the OEM ECU is doing the same.

AbeFM

iTrader: (3)

Very interesting there. Yeah - I had been having issues till I switched polarity. I can't actually independantly control (in software, though I've been asking for it) which edge gets used, so I have to invert one in hardware, and THEN tell the computer to use rising/falling edges.

I'd figured the leading tooth edge would be more accurate. But if I've learned one thing in my life, it's don't argue with what works, figure out why it works. :-) I don't have my laptop but I'll figure something out.

Were you going to post a pic? I'd love to see it. Though the trailing edge is self-explanitory.

Question though:

Here the trailing edge is obviously the bad one, the leading edges seem fine.

Interestingly:

Here the leading edge is clearly wrong, and the trailing is the good one. I saw this in several pictures.

I think the crank sensor was out of adjustment (or on the edge), moving it in, within spec, qualifies as "ok". I think moving the cam sensor is silly and an act of desperation on my part. Leaving the signals all alone, however, and selecting the falling edge should give me the trailing edge on the cam (the good one) and the rising edge on the crank (also the good one).

In practice, I think inverting them both as Arga has pioneered might be the cleanest solution, but I'll try two straight inputs next. The nice part is since the cam pulses fall entirely within the crank pulses, it doesn't matter what you use, as long as you catch the valid pulse.

All this tells me that maybe the OEM ecu is doing the same thing, which might explain why they are flipped from the factory.

Anyway, that's a huge help! It seemed in the process of getting my filtering figured out, I went to a single inverted input, and was watching the wrong edges. Your suggestion is the first really solid thing I've seen in a month on this. Thanks!!

AbeFM

iTrader: (3)

Ok, so, based on this quote:
"I'm 90% sure your Lucent parts are simple comparators. The 177G is the same as a uPC177G which is an early version of the uPC339G which is a very common LM339 Quad comparator."
From someone hacking the Hayabusa ECU, and this schematic:


And a LOT of this:


Then I came up with this:

And I was so proud of myself, I put a fancy black border around it. Isn't that special?

Anyway, all in all, it looks like Arga's inverting inputs are really the way to go. These guys use feedback to keep things stable, a pulse of shorter than X amount won't actually trigger a change. Kinda neat, though it's a lot of componants. I might try it, I dunno. I guess I could build one fast enough.

The interesting thing to me is the high input resistant, and the even higher resistance pull-up.

Lastly, a cap with "102" is a 10nf, right? 0.010 uf?

So, I'm sticking with my theory that the OEM uses the signals opposite to each other, to trigger off the most reliable edge of each sensor.

j_man

Here are the posts with explanations of yours and Jason's scopes and why the raising teeth (the trailing edge for each tooth) are the correct ones:

http://forum.miata.net/vb/showpost.p...3&postcount=56

http://forum.miata.net/vb/showpost.p...2&postcount=63

Arga is also using the raising (trailing) edges with his device.

Here I am also attaching the two pics, just in case if something happens to the mnet links ...

AbeFM

iTrader: (3)

Wow, you're sure putting some time into those! And it's appreciated, too. Here's a question: It's believable when you show it like that, only, there's no reason you couldn't say it was the other parts of the teeth, unless you compared it carefully to a known plot of where the pulses should be.

Either way, it does seem the trailing edge is the way to go. The only differences in the pulse train you showed are 1 pixel, so within the temporal noise.


Anyway, while you've been fighting that (last bit I'll add is that my car ALSO worked a lot better when I switched edges away from the "recommended" ones), I've been seeing other odd behavior.


The bottom two are spark outputs. Holy wrongness Batman

Here, you can see a longer sample


In most of the pics, there's a steady tach signal, and good, solid inputs to the ECU (measured at the ECU pins). But, the ignition is wacky, and the fuel is almost random.

I think, somehow, the firmware on the MS got eaten. don't have any better ideas. There's nothing left to doubt, though, since I measured the inputs and I measured the outputs.

I *guess* the rising edges might be a too flat, but I looked at them before and they are a LOT sharper with the NPN's I'm using than without. Fast rise times.

AbeFM

iTrader: (3)

So, here's the Big News, plus, my own personal Big News.

Big News (general):
The newest JimStim chip ($9 shipped, get a breakout board while you do it for another 20) supports the PROPER 99+ trigger signals. Mine's in the mail already.

Smaller News:
Turns out it IS my firmware. I changed NOTHING, just reflashed the MS-II to b19, and without even opening the box it went from total crap to running perfect.

I NEED this to stop happening. A car is not reliable that might not start three times a week cause the ecu kinda sorta forgot how to run and leaves your coils on so they melt.

Joe Perez

iTrader: (8)

I can't see the MSII spontaneously looking its program just from sitting there. Bit rot, while not complete fiction, is awfully damn rare.

Does this happen (the chip becoming corrupt) during a configuration session while you're flashing it? Is there some external stimulus such as Tesla coil in your garage?

AbeFM

iTrader: (3)

Yeah, but it's off most of the time. Well, I did try a cheapy serial port for a bit, but I don't think that was it. I've been trying to piece it together, and haven't had any luck yet. Sometimes it happens after a lot of burning to the ECU, but sometimes it happens after a lot of NOT burning. I just wish I could get things down to 10 or 15 random, intermittent, psuedo-similarly symptomatic issues so I could get a handle on things.

Anyway, it was kind of a blessing making me bite the bullet and go through my wiring. it didn't need doing, but it didn't not need doing if you catch my drift.


Notice a difference?

I am kinda tempted to load my old MSQ back on and see if the issue is in that, or somewhere else in firmware so that I need to have a full flash to fix it.

AbeFM

iTrader: (3)

Totally failing at the build-journal here.

Latest Updates:
A while ago, I'd breadboarded out a pair of NPN transistor buffered input circuits, using 20k (should be 3.3k) of pull up*, 10k of series resistance, 10nf cap to ground and a NPN which went right to the ECU save for a 4.7k pullup (unsure on that last value)

*I used 20k because I remembered 33k from the OEM input circuit. Turns out I remembered wrong, checking my notes and double checking my photos, it was actually a 3.3k, big difference.

Anyway, the car was running like that for weeks, with it all 'plugged in', I was getting a miss occasionally, but not bad and I thought the circuit was working (the fault of poor logging).

I rebuilt said circuit using the same parts, except I swapped the 20k pullup for the "proper" 33k, this time soldered on the board. Imagine my frustration when my original 5k stumble came back to haunt me. Yes, after months of changes, I'm RIGHT where I started.

Well, I figured if 20k was better, I would make it 20, so I added a 49k in parallel to give me ~19.xk. This was slightly better, but a weird thing was happening: When the car hiccuped now, it goes DEAD RICH (well, 10.2:1) and stays there for a while. It still "runs" but keeps going rich and staying there. I have no idea why, and get no help from the forums.

So I went for a drive, still having that issue, so I figure maybe there's not enough capacitance (though earlier testing showed 10nf was too much and 1nf worked better), and add 10nf more in parallel for a total of 20. This doesn't seem to make much difference, but I notice while I'm doing it one of the solder joints may have been mysteriously open (cold joint). So, I fix that up, add in the extra 10, go for a drive. All seems well for a bit (why it starts off good feeling, I don't know) but then does the same thing: Stumbles so I can't get past ~4,500 or 5,200 rpm, and then goes dead rich for a while.

I go to work to grab the o'scope, while I'm there I just mess around. I find the rise time for the signals is ~3ms, so I lower the pullup resistance further to 5.6k by putting a 8.7k resistor in parallel with the other two. This shortens the latency of the circuit by quite a bit, and has one circuit rising in <500ns, the other in <700. Also the peak voltage is higher on the input side.

Well, I drive it home and of course it didn't really fix anything. I decide to take out the caps, so I pull the caps and swap them out for 2.2nf (my original idea of the "best choice").

So, right now, I have ~5.6k of pullup, 10k of series resistance (1k?), and the same 4.7 pull up, but only 2.2nf of capacitance on the input.

I'm getting pretty annoyed at this whole thing, but I'm just convinced it HAS to work, so I'll figure it out.


Also, I breadboarded up the exact input circuit from the OEM, except I swapped a feedback resistor of 13k for one of 12k since it was all I had, and the 3.3k pullup is approximated to 3.2 by putting a 2.2 and a 1.0 in series. This isn't hooked up (I'm sort of getting the feeling that "by magic" the car works with breadboarded circuits and not soldered ones), but it's in the case so a short time with running some wire and a touch or two with a soldering iron would have me switched over. I can probably find (and hence swap out) the proper 3.3k pull ups, the 13's I would have to make.


In other news: My COP set up fizzled on me a while ago, driving home I thought I was having injector issues, but it turned out that wiggling all my barrel and spade connectors brought the motor back to life, so I went through and soldered every joint. It's fantastic having another thing to doubt.

All in all this isn't SO bad, but it's a little annoying the inconsistancy of it all. Also, the most frustrating part of the build is the lack of support on the MS-IIx forums. Looking at the logs, when the car misbehaves, "bit 8" flashes on and off. Reading the MegaManual, updated a couple days ago, it says what the bits are. But it doens't agree with the MegaLog Viewer. So I'm not sure which is right. I asked, and of course, the guy who wrote the code makes some comment but doesn't answer that question. Pretty annoying that a potentially valuable clue is there but I don't know what it means, even if I do go through the trouble to RTFM.

Well, ok, exciting this isn't but I thought I would make some notes of the process. The car still gets me from a to b, as long as I know how to properly drive it. Boost doesn't seem to bother it at all, though I've only run a few lbs. It's right on the verge of working, and from past experience I know when it works well it works SO well... but now it's juggling bowling ***** in a windy china shop instead of fun.

AbeFM

iTrader: (3)

Huzzah!
I set up my breadboarded OEM circuit today, an exact copy of the OEM inputs. It was worse than the circuit before it - at ~4k the motor would fall apart. I think the sensors just couldn't deal with the extra charge in the caps. With that set up, the JimStim would cut out around 8700 rpm - when the propegation delay matched the pulse width it was canceled.

Well, while it doesn't make a lot of sense, I swapped out the big caps for some smaller ones, using the OEM circuit only with 1nf caps instead of 10nf seems to have it working very well. Time will tell if it's perfect or just very good, but it IS working. I have reved the motor to JUST under 7k, and my boost limit of 175 kpa seemed to be the only thing stopping me. I did hit "knock" at 6900 or so, but this is likely improper adjustment of the knocksense. It pulled 6 degrees in under half a second, and stepped down at 1 degree every few seconds till normal. On the jimstim I couldn't get it to lose sync. I don't have the numbers with me, but at the ecu the rise ("fall") times were 50 ns, the delay was 16us for the cam and 12 for the crank, the voltage 4.4V for the crank, 3.4V for the cam. I might have them backwards. With the bigger caps the delays were somewhat longer, and the voltages lower (that I don't understand).

So, now I just have to 1) Drive it a lot, and 2) commit a copy to solder and see if I'm still jinxed when I do something OFF the breadboard.

Reverant

iTrader: (10)

Fingers crossed man!

Jim

AbeFM

iTrader: (3)

So far, so good. Right now my weird issue is that my "overboost" kicks in at 10-11 psi on the guage, but the MAP reading is 205kpa.

That's just not right!

So, I don't know which one to trust. Guess I need to calibrate them. It's the 2.5bar DIY sensor.

AbeFM

iTrader: (3)

Ok, more notes:

Inputs still working awesome.

My boost guage wasn't agreeing with my MAP, so I decided to hook them together with a T, and T in another boost guage (autometer) and pump on it with a mighty-vac. Push on the back of the lever thing with your thumb to get more pressure. Turns out the boost guage seems to have a constant 0.7 psi low reading, but that they agree pretty well. I've heard rapid pressure changes (like putting pressure on the guage then popping off the hose) can offset it like that. So, I'm attributing my sense of the targets being way wrong to the guage being slow or my watching the road instead of the guages.

I got a new serial port, from Fry's, Parallax 28030 and found it to be MUCH better than the Belkin. It's very small, nicely made, uncased. Takes a mini-usb plug. It has LOTS of options in the drivers and supposedly works with Linux (I didn't try it). In the driver, there's a latency setting. Set to 16ms, it takes 2-3 minutes to write the .s19 file - compared to the 35 minutes it took on the Belkin. Set the latency to 1ms, and it took well under a minute.
Interestingly, the driver reports the baud rate as 9600, and changes don't stick, but it works fine so I suspect the programs working with it set it themselves.

Lastly, I wanted the narrowband signal to be right! I have a Westach guage and it seems way off. So, I couldn't really get the LC-1's output to be where I wanted it, so I set the "warmup" value to different voltages, i.e. 0.45, 0.48, 0.50V, and saw where the guage went, till I got 14.7V and not 14.8 or 14.6.

The process is:
1) Turn on car LC-1
2) Run LC-1 programming program
3) Change value (only one tab at a time)
4) "Program"
5) Close program
6) Power cycle LC-1
7) Check initial value on guage, write it down
8) GOTO 2

After several cycles, you have a table, which turned out to be pretty linear for my guage. Then you line fit it, pick something that goes through 14.7V (in my case at 0.48V), and then pick two points at the end of your guage (in my case 12.0:1 and 17.0:1), enter the corresponding voltages, set the "warmup" voltage to what you want, and hit program. I have a feeling if you have the program running and start the LC-1 it overwrites your values back to default, so don't do it. For a while I was unplugging the serial cable, too, but I found I didn't need to do it.

I wanted to do the same with the MS's WB input, but couldn't, since I didn't want to turn the car on and off a million times - but it seems you can only get a real-time O2 signal when the car in "running"?

WestfieldMX5

FWIW, a cap with 102 = 10 + 2 zero's pF = 1000pF = 1nF
(103 would be 10nF)

AbeFM

iTrader: (3)

Uhg! See, I'd caught this on my own car, but never reined it in on the web. Actually, on my car I ended up running no cap at all, and it works fine.

But I'll go fix my drawing. :-)

huh. I had fixed it and the image stayed while the caption changed. Well, should be fixed now.

AH! It's supposed to autoreplace, but hasn't.

AbeFM

iTrader: (3)

FYI, see it in printed form here:

Marc D

iTrader: (7)

wow abe. looks good. Its virtually PNP from that point on. You ready for some beta testing?

AbeFM

iTrader: (3)

FYI