New wideband controller ALM compared to Innovate LM-2

[dr.mike]

Gotta call BS on this one. Anyone who knows how the LC-1 works internally, knows exactly how those traces were made. The ALM sensor was in a bung near the engine. And, the LC-1 sensor was at the end of the exhaust system, probably stuck in the tailpipe. Thus delaying its readings by the length of the exhaust track.

You see.... the LC-1 takes a series of INDEPENDENT measurements. Usually at about 200Hz. That's a new completely independent measurement every 5ms or less. And there is no bleed-over from one sample to the next. It is, therefore ,IMPOSSIBLE to get traces like that unless the LC-1 is placed much further from the engine than the ALM and NB sensors.

Unless you are saying that the LC-1 is storing about 10-15 samples in a delay buffer and spitting them out FIFO style ( which is doesn't )... this just physically can't ( doesn't) happen. It LOOKS like an attempt to mislead prospective customers. If I bought an ALM based on these claims and found that the traces were obtained this way, I would claim fraud.

You can see the slope of the LC-1 signal is much steeper (faster) than the ALM. And, more closely matches the NB signal. It is just delayed several ms because of the sensor placement, further down the exhaust stream. Reverse the positions of the LC-1 and ALM sensors and you will get a very different result.

The variability in the delay that happens "sometimes" is based on the exhaust gas velocity at different engine speeds. No mystery here.

As for the LM-2, I don't remember if it has a full-speed signal output.

All wide-bands that use the Bosch CJ125 ( or CJ110,CJ120, or CJ135 ) have about the same SLOW performance. Which is quoted by Bosch ( t63 number ). If I remember right, the t63 is about 50ms. that is... it get 63% of the way to the right value within 50ms. That's the slow slope you see on the ALM trace. While the LC-1 gets 100% of the way to the correct value in 5ms or less.

To make my point. I have PERSONALLY used an LC-1 to tune individual cylinder carburettors on 12-cylinder ( 12 barrels of Weber ) engines. I can resolve individual cylinder AFR values at up to about 1500-2000rpm, using a scope. Try THAT with any CJ1xx based unit

BTW.... A LOT of times, that "noise" on an LC-1 at low rpm is variations in AFR between individual cylinders. Not something you want to "filter out".

[jharvey]

If there is a pipe length delay issue, why do the traces start at the same point, but reach a result at another point?

From the link posted with that scope trace, I found this PDF which seems to show the bungs are all close together, in a fairly straight piece of exhaust tube.

ecotrons.com/files/FM_WB_Shootout.pdf

I would agree the above posts are missing some massive amounts of data. If you have equipment and such, perhaps you could fill in those gaps by posting a youtube, pictures, or similar that will show a the details of the tests, not just the testers opinion.

[jharvey]

I see the datasheet claims a slew rate of 1 V/μs. I haven't found your mentioned t63 spec yet. What datasheet (or what ever) is that from?

[dr.mike]

I can't quote the document directly, due to NDA. But, if you look at the traces in the picture, you will see that the scope is set to 20ms/div ( minor div ). The ALM output takes about 3 divisions ( 60ms) to climb to about 63% of the value that it eventually settles on. This is not a coincidence

To get to 90% of the final value would take 2.3 times as long, or, about 115ms. Or, just less than 6 divisions. A quick peek at the traces confirms this as well. No magic here

The 1v/us is the slew rate of the amplifier. Not the whole system.

Also look at the "noise" of each trace. The ALM trace seems to have a lot of high frequency noise. Not too bad. While the LC-1 output seems to be made up of a series of steps that are about ~3ms long and mostly flat over those periods. Those are individual AFR measurements. The LC-1 does a measurement, sets the DAC value, then repeats at a rate between 100Hz and 300Hz or so.

Each measurement is self-contained at 100% the target value. i.e. the previous sample has zero effect on the current sample. So, the response time of the LC-1 is t100<10ms. In practice it is closer to t100<5ms.

It is therefor impossible for any device to be 50ms faster than a device that has a response time of only 10ms or less. To do so, that device would need to know the result at least 40ms before it actually happened. Neat trick, if you can pull it off. Then go collect your Nobel prize in physics. Or, at least, call the Amazing Randi

[dr.mike]

Also... if you look at that wide-band shoot out article ( ALM was not there ). You will see that all of the units that use the CJ125 ( or its kin ) scored 300ms or slower in the latency tests. And... only 1 scored an accuracy of 0.10 AFR point or better.

[jharvey]

I guess a first question would be, what's a reasonable/desireable response time?

I see these as the good features of a O2 sensor.
-- Having one good sample once per rev, at a full RPM of say 10,000, would be good, as you could potentially detect lean fire in one cycle, which could prevent lean on the next cycle.
-- If you could get 12 or more good samples per cycle, you might be able to detect which cyl was lean in a V12.
-- Having long term knowledge of lean or rich is also good, as you can tune out fuel rail drops based on fuel flow, and other longer term delivery problems.

So I would wager a guess that a 6mS response time would allow you to get good data at full RPM, and would allow for some very detailed data at lower RPM's. If we toss the concept of per ignition detection, and look for just the longer term effects of rich/lean, even at 50mS it would allow for a fairly fast reaction to rich/lean. However a faster response time would allow that detection with less lean fires, and could decrease engine stress. At full RPM a 50mS response could allow lean operation for nearly 10 cycles before it's detected. Of course lean fires are bad.

Does that sound reasonable? I see the CJ125 as fairly good, but perhaps not quite the best. At 10kRPM, it could lag by up to 10 cycles, where a faster design could lag by perhaps ? cycles.

A slower response time shouldn't really be an issue, unless you are running your tank low and get vapor bubble or have some short term fuel problem. I think it would work just fine as long as your system is functioning properly. What benefits would one get with with a faster response time, when the engine is running after it has been tuned?

I don't follow why the bump at 50mS in the yellow trace. I understand the "noisy" claim is due to a fast sample rate, and that could easily be smoothed out, if so chosen. I think that trace is trying to claim the signal has wondered, and after 50mS it removes the wonder. I also don't know what the actual gas presence was, so I don't know if either trace is actually correct. Certainly lots of missing data.

[dr.mike]

Remember that that 50ms only gets you 63% of the way to the correct value, from where you started. To get 90% of the way takes ~120ms and 95% ~150ms etc.

The trick is not so much to read individual cylinder data, although that can be extremely useful. It is that you get data that actually correlates to the rest of your data during acceleration/deceleration. for example, if the RPM and AFR are constantly changing during acceleration, then the AFR data will never match up if it is delayed by 300+ ms. And it's not that you will get the right data 300ms later. You will get the WRONG data 300ms later.

If you rev from 1000rpm to 6000rpm in 1 second thats 5000rpm/second. If your AFR data is delayed by 300ms, then UNDER IDEAL CONDITIONS your AFR reading would be off by 1500rpm. But that is only if the AFR was constant over the whole period. If the actual AFR changes during the transition period, the AFR reading is NEVER right during transition. Not until it has been steady for more than 300ms.

For this reason , almost all wide-bands are really only useful for measuring steady-state AFR. i.e. constant speed / constant load.

This is why speed is more important than absolute accuracy for a wide-band. In fact, on an inertial dyno, where the speed is never constant, they are just about useless.

As to single cylinder measurements, you need at least 2 measurement per cylinder cycle and preferably 3 to know where to draw the line separating one cylinder from another. So, if we have a V12 with 2 banks of 6 cylinders feeding dual exhausts... ( exactly the same as an even-fire 6-cylinder engine w/single exhaust )

at 1200 RPM, the engine turns 20 times per second. That's 10 complete cycles of a 4-cycle engine per second. So.

Total period = 100ms

We divide that by the number of cylinders ( in an even-fire engine ), in this case, 6. we get

16ms per cylinder.

If we want 3 good measurements per cylinder, we need about 5ms per reading.

Even the CJ125's 50ms t63 time would be smeared across more than 3 cylinders of data. And the actual response of about 300ms would be smeared across 3 whole engine rotations. AT IDLE.

An LC-1 running at 3ms/measurement (common) gives 3 good samples in 9ms. Multiplied by 6 cylinders, thats 54ms per complete cycle. Or 27ms per engine rotation ( 2 per cycle ). 1/27ms = 37Hz. 37Hz x 60 = 2222rpm.

So, above about 2200RPM, you no longer get 3 clean samples per cylinder firing. Nyquist says you are good out to 2 samples per cylinder firing or, about 3300rpm. But, the separation algorithm gets complicated.

And this matches up exactly with experience.

[Fred]

Thanks for your input Dr Mike and welcome to the forum! :-)

[jharvey]

Oh also I forgot, it's 4 cycle, not 2 cycle typically, so that would be 5 cycles of lean with a 50mS delay. Not 10 cycles.

[Fred]

Mike, I changed the thread title:

Was: "New wideband controller ALM beats Innovate LM-2"
Now: "New wideband controller ALM compared to Innovate LM-2"

A little more objective. The rest of the thread speaks for itself.

Fred.