Why is the ADS1115 on I2C so ultra slow ? (solved)

Toit

#Why is the ADS1115 on I2C so ultra slow ? (solved)

Thread channel in help

kaxori 07/03/2023 02:25 PM

Has anyone an idea why?

floitsch 07/03/2023 02:47 PM

Can you define what "slow" means?

floitsch 07/03/2023 02:48 PM

How many measurements do you see?

floitsch 07/03/2023 02:49 PM

I don't have any ads1115 yet, so can't test the code, but the test.toit itself looks good.

kaxori 07/03/2023 03:14 PM

slow means 1 measure per second. In the example the led toggles each step without sleep.

floitsch 07/03/2023 03:30 PM

That's really slow...

floitsch 07/03/2023 03:30 PM

I'm wondering if we are stuck in the sleep loop of the driver (waiting for the value to be ready)(edited)

kaxori 07/03/2023 03:31 PM

i changed the ads1115 and measure now ~140ms !

floitsch 07/03/2023 03:31 PM

Interesting

kaxori 07/03/2023 03:41 PM

the lowest is slowest

floitsch 07/03/2023 03:41 PM

Weird

floitsch 07/03/2023 03:42 PM

Would be interesting to know whether it's faster with other systems or if Toit does something different

kaxori 07/03/2023 03:43 PM

I will check it out

floitsch 07/03/2023 03:43 PM

Thanks

kaxori 07/03/2023 05:58 PM

=> 34ms with Arduino SW

floitsch 07/03/2023 05:59 PM

Even for the 1s one?

kaxori 07/03/2023 06:00 PM

yes measure with the slow ADS1115, the faster one does it in 12 ms

floitsch 07/03/2023 06:00 PM

Strange.

kaxori 07/03/2023 06:01 PM

I guess the used i2c implementation could be improved

floitsch 07/03/2023 06:01 PM

Not that much

floitsch 07/03/2023 06:01 PM

It's mapped to the hardware I2C. Shouldn't be that slow.

floitsch 07/03/2023 06:02 PM

Need to go now, but something that I will need to investigate.
I should get the same sensor soon.

kaxori 07/03/2023 06:54 PM

GPIO Speed test (fast DO toggle):
toit: 5,72 µs, ~ 175kHz
Arduino : 7.6 µs, ~ 132 kHz

floitsch 07/04/2023 08:11 AM

Nice.

floitsch 07/04/2023 08:12 AM

There is one thing that Toit currently doesn't do very well: sleep for very short periods. We are using the Freertos timer underneath and that one only has a granularity of 10ms. So the driver's sleep --ms=1 will only come back after 10ms. But that doesn't explain the delays of ~1s.

kaxori 07/04/2023 08:32 AM

kaxori 07/04/2023 08:33 AM

so for hw close driver hacking small delays are needed

floitsch 07/04/2023 10:00 AM

It would be nice to make sleep work better with less granularity.
It shouldn't be an issue here, though. As I said: I should get an ADS1115 soon, and will investigate then.

Can you confirm that it's the sleep loop that takes so much time?
Would be interesting to count the number of iterations of that loop. Checking how often the device responds with "not ready yet".

kaxori 07/04/2023 06:42 PM

it is not the sleep , i removed it (in a temp copy of the driver).
I also checked the register parametrisation unfortunatly without knowlegde

floitsch 07/04/2023 06:43 PM

strange.
Maybe add some print Time.monotonic_us in strategic places to figure out what is taking so long?(edited)

kaxori 07/04/2023 07:20 PM

ups again: chips needs about 450 cycles (~1ms, without sleep) to get ready

floitsch 07/04/2023 08:43 PM

The sensor should be able to sample at ~1ms. Weird...

floitsch 07/17/2023 10:52 PM

I finally got my Ads1115 sensors.
After a bit of testing I figured out the problem.
The constructor is setting the config register.
It directly ors in the RATES_DR_475SPS_ constant. However, the location of the data rate is bits 5-7. So there is a shift needed:

| RATES_DR_475SPS_ << 5

Alternatively, the RATES_DR_475SPS_ constant should have the bits already shifted. (That would actually be more consistent with the rest of the code).

Once I fixed that, I also changed the sleep --ms=1 to yield.
The driver should probably remember the samples-per-second, and then either just yield (if it's less than 10ms), or sleep otherwise.

With that change the driver is nicely computing new values close to the speed that it should.

floitsch 07/17/2023 10:55 PM

@kaxori

let me know if you want to do the change yourself, or if you prefer a PR.