In my opinion not.
It completely depends on the cell voltage
And the datasheets says 2% accuracy up tp 3.6V, but the cell voltage can be higher, and when 2 neighboring cells have different voltages then the oscillators will run different
According to Stuart is seems like the communication problems occurs when the module discharges (Balances) the cells, which maybe can be explained with the differences in the voltages on the modules and therefore different clocks
And I agree in what you wrote Smurfix. The required accuracy 3.75% do not depend on the baudrate
There is some articles that can be found, describing calibration of the AtTiny oscillators by use of another microcontroller to get more reliable communication.
A small fw on the āTargetā puts out the clock out on a pin, and the āMasterā measures and calculates the offset, and writes the calibration to the āTargetā
Maybe itās worth considering to see if this makes the communication in the current design more reliable
What also happens in the optocoupler is, that higher frequencies / baudrates makes the opto transistor, in the optocoupler, go more in saturation which effects the Tfall in a bad way, and therefore limits the max speed
I seems like the remedy is to use optocoupler with a opto diode instead of a opto transistor in a setup with serial data, or to just use the transistor as a diode, but it requires access to all 3 pins of the opto transistor
Besides that. The given calculated 3.75% clock accuracy requires a decent signal where itās possible to sample reliable, but the choosen (Relative slow) optocoupler adds to the problem with specially slow falltime due to saturation in the transistor, which can make the sampling of the incoming signal unreliable / difficult on top of the frequency difference of the oscillators due to different Vdd to the MCU
Also Analog Devices recommend 3.75% accuracy and 2% for harsh setups, and normally there is a oscillator driven UART in one end, which is not the case between the cell modules