Slow time on micro:bit V1 and V2

[kevinjwalters]

I've observed that time is slow in micropython on the micro:bit V1 and V2 using v1.18 from https://python.microbit.org/v/3

This program using sleep runs for longer than 100 seconds.

from microbit import sleep

sleep(15 * 1000)
pause_ms = 100 * 1000

count = 1
while True:
    print("{:06d}".format(count))
    sleep(pause_ms)
    count += 1

It produces this output (borrowed the Arduino IDE as it has timestamping, running on Windows 11)

20:44:16.077 -> 000001
20:45:56.757 -> 000002
20:47:37.501 -> 000003
20:49:18.167 -> 000004
20:50:58.920 -> 000005
20:52:39.635 -> 000006
20:54:20.306 -> 000007

On v1.18

KeyboardInterrupt:
MicroPython v1.18 on 2023-10-30; micro:bit v2.1.2 with nRF52833
Type "help()" for more information.

The 100 second sleep (plus a tiny bit of processing for format/print/increment) is 100.680, 100.744, 100.666, 100.753, 100.715, 100.671. The clock appears to be running about 0.7% slow (at 23C). It's easy to dismiss this as poor time keeping on a microcontroller but the micro:bit V2 has a crystal (listed in BOM) and all is fine in CircuitPython on same board.

I first noticed this when checking the performance of the MCP7940N rtc on a ZIP Halo HD vs MicroPython utime.ticks_us(). Across 24 hours the RTC is running about 18 seconds fast (I think this has a trim feature which I may explore to get it spot on), MicroPython is 552 seconds slow (about 0.64%) and CircuitPython (time.monotonic_ns()) is just 2 seconds slow (extrapolated, I only had time to run the test for ~90 mins).

@JackAtKitronik might be interested in this.

[kevinjwalters]

I wasn't expecting this but just tried on a V1 and same thing.

21:17:05.310 -> 000001
21:18:45.989 -> 000002
21:20:26.738 -> 000003
21:22:07.441 -> 000004
21:23:48.167 -> 000005
21:25:28.859 -> 000006
21:27:09.597 -> 000007

That's 100.679, 100.749, 100.703, 100.726, 100.692, 100.738 seconds.

[kevinjwalters]

For comparison, MakeCode on V2:

13:39:20.760 -> 000001
13:41:00.739 -> 000002
13:42:40.758 -> 000003
13:44:20.778 -> 000004
13:46:00.817 -> 000005
13:47:40.861 -> 000006
13:49:20.870 -> 000007

That's 99.979, 100.019, 100.020, 100.039, 100.044, 100.009 seconds.

[microbit-carlos]

That's really interesting, thanks for the report @kevinjwalters.

I believe the internal RC clock might be used unless BLE or radio is running. Could you turn on the radio and test again to see if that makes a difference?

[kevinjwalters]

My V2 is busy at the moment testing a trimmed MCP7940N. The same V1 is still slow with this code

from microbit import sleep
import radio

radio.config(group=23)
radio.on()

sleep(15 * 1000)
pause_ms = 100 * 1000

count = 1
while True:
    print("{:06d}".format(count))
    sleep(pause_ms)
    count += 1

produces

15:00:50.008 -> 000001
15:02:30.700 -> 000002
15:04:11.390 -> 000003
15:05:52.107 -> 000004
15:07:32.838 -> 000005

on MicroPython v1.9.234-gd64154c73 on 2017-09-01; micro:bit v1.1.1 with nRF51822.

[kevinjwalters]

From my first tests on V2 I have a linear corrected version of graph and that does show MicroPython is all over the place which supports the explanation of a mediocre internal oscillator.

[kevinjwalters]

I think this was discovered back in 2016 too on the V1 in REUK.co.uk: Testing Accuracy of micro:bit Internal Clock, that one was out by (600/598.51 -1) * 1000 * 1000 = +2489.5ppm

[martinwork]

An alternative method to monitor micro:bit's time: millis.zip

Install millis-python.hex in micro:bit V2. Run millis.py on PC. The PC periodically asks micro:bit for it's millis clock and compares it's elapsed time. Disconnect WebUSB or it will corrupt the python serial.

It's clear within a minute that the micro:bit V2 is not quite 1% out.

Edit millis-python.hex to #import radio, and test again. Radio enables the HFCLK in V2, and the micro:bit times seem to match the PC to the millisecond, though I haven't run the test for more than a few minutes,

I think micropython is using a version of CODAL before the HFCLK was always enabled:

https://github.com/lancaster-university/codal-microbit-v2/blob/d159aba47e1797ace83ffbac81009909326f3c4f/model/MicroBit.cpp#L291

lancaster-university/codal-microbit-v2@ed14fa2