[info] Case Study: I do not like this jitter (solved)

[AndreyB]

I've notices some nasty jitter even on the bench so I've decided to make some tools so that we can keep an eye on all this.

java console has a gauge which displays the signal dwell as measured by out build-in logic analyzer. triggerinfo would tell us that primary logic input is http://i.imgur.com/e34b5vN.png and showconfig shows that one of the injection channels is http://i.imgur.com/ILq2dgd.png

Now let's reset everything to the default state - that's set_engine_type 20, and also enable self_stimulation to turn the whole firmware logic on.

Perfect, now we put a jumper wire between PA8 and PB8 and start the console, set RPM to a higher value, restart the console and look at the numbers.



I do not like all this at all, to be continued.

[AndreyB]

I guess we need some methodology.
Test scenario: rusEfi 4647 compiled in Eclipse 'Debug' configuration, with some funny options applied:

set_engine_type 3
disable engine_control
disable map_avg
disable mil
disable fuel_pump
disable tunerstudio
disable wave_analyzer
disable idle_thread
set_rpm_hard_limit 150000
writeconfig
<microcontroller reboot>
rpm 24000

Export PD1 signal (primary trigger emulation) captured with a logic analyzer into .csv, open Excel, create column C - that would be "ON time in ms"
=IF(B3,C2,(A3-A2)*1000)
and column D - standard deviation, length 46
=STDEV(C4:C50)

That gives us (all times in ms)


These would be our reference values - at least we know that we can produce a clean-enough trigger signal in case we are not busy processing that signal (the signal goes nowhere and some of the utility threads are not running)
To be continued...

[AndreyB]

Now let's make the firmware struggle - same version #4647

set_engine_type 3
rpm 24000
enable self_stimulation

In this scenario we reset settings to default so all the features are active, we generate high rpm signal - the RPM in this test is actually above the RPM hard limit, so no real processing takes place. Yet the numbers definitely look differently:

[AndreyB]

Made some good progress: version #4718

set_engine_type 3
rpm 24000
enable self_stimulation

- a number of improvements to SingleTimerExecutor. note how ON times are not ~2.42ms but back to ~2.5ms where they should be
- fast ADC is now disabled by default if it's not needed for the particular engine configuration. todo: figure out proper fast ADC frequency, maybe re-enable for everybody
- todo? maybe? it's probably worth migrating from getTimeNowUs() to getTimeNowNt(). Instead of using microseconds we would be using CPU native cycles counter thus saving on __aeabi_uldivmod which is a heavy operation

(all times are still in ms)