RTC Temperature Calibration -- 20ppm all temperatures feasible?

[chostetler]

Good morning,
I'm developing a device with these requirements:

• STM32F446RET6
• No internet (NTP) or GPS connection
• Real-Time clock, set manually (so, the less often a user has to set it, the better)
• Outdoor -20°C~70°C temperature range (most installations in continental USA)

Ideally, we would like to get our total clock drift under 10 minutes/year (20 ppm) if possible.

I created one prototype using an Abracon 32.768kHz crystal for the LSE, with two C0G 6pF load capacitors. To test the prototype, I routed the LSE to the MCO1 and measured the clock frequency as we swept the temperature from -20°C to 70°C in our temperature chamber. I measured that and compared the ppm deviation to the expected deviation given by abracon (ppm = (T-25°C)^2 * (-0.035ppm)):

Time [wall clock time]	Ambient Temperature [°C]	measured MCO frequency [kHz]	ppm deviation	Predicted deviation (per abracon datasheet equation)
10:10	-20.9	32.76681	-36.3	-73.74
10:13	-20.6	32.76672	-39.1	-72.78
10:24	-11.9	32.76712	-26.9	-47.66
10:34	-3.9	32.76750	-15.3	-29.23
10:45	4.2	32.76785	-4.6	-15.14
10:56	12.6	32.76809	2.7	-5.38
11:06	19.8	32.76818	5.5	-0.95
11:17	28.5	32.76820	6.1	-0.43
11:28	36.2	32.76810	3.1	-4.39
11:38	44.1	32.76785	-4.6	-12.77
11:49	52.2	32.76750	-15.3	-25.89
12:02	62.3	32.76685	-35.1	-48.70
12:14	70.5	32.76620	-54.9	-72.46
12:25	70.1	32.76598	-61.6	-71.19
12:54	70	32.76590	-64.1	-70.88

So, my next steps that I was planning was to:

1. Test various load capacitances to try to bring the room-temperature error as close to 0ppm as possible
2. Use software calibration to modify `CALP` and `CALM` on-the-fly according to the ADC temperature, to counteract (1) the room-temperature ppm offset of the crystal and (2) the offset incurred by high and low temperatures (abracon equation).

However after doing some research, I'm a bit less confident about my approach

• The temperature slope I measured was different from the one given by the Abracon datasheet
• The measurement period for the RTC after calibration (32 seconds) is very wide, and I'm unsure if our equipment is suited for it
• I don't know how I will tune each device if the temperature slope differs for every crystal.

Is it feasible to tune a RTC like this to achieve 20ppm accuracy over a wide range of temperatures using only software? Or am I better off getting something like a temperature-controlled oscillator?

Thanks!

[TDK · Accepted solution]

From the crystal datasheet, the range in accuracy varies by more than 20ppm over the 20-70 temperature range. Note the delta from the blue and yellow lines.

So that's not a viable solution unless you have the tools and time to calibrate each crystal independently.

You're better off with a TCXO.

[Ozone]

In commercial RF transceivers, they used to keep the oscillator section at a constant temperature for this purpose.
And for the ease of control, this was usually a temperature at the higher end of the ambient spectrum.

[chostetler]

Ah, yep. That's a good point, I should have noticed that.

In general, is this a viable strategy, assuming a wider ppm accuracy is acceptable?

Thanks

[TDK]

It's an okay strategy if the temperature is stable. It's not going to give you the accuracy of a TXCO though.

Your code will be more complicated, and you will use more power, it will be less accurate (compared to TXCO), but it will be cheaper in parts cost.

[RobK1]

The TCXO suggested by user AScha.3 might be an option for you:

Solved: Re: Calibrating the STM32's real-time clock (RTC) - STMicroelectronics Community