Accurate Internal Oscilator

I've never seen anything that accurate for an internal oscillator.

Have you considered a miniature external crystal ? For example:

We are pondering the same issue for an AVR. Will a watch crystal (32KHz) work for you? We are considering mounting it on top of the AVR with a custom lead frame.

They are very expensive, of the order of $1 each vs. 10 to 20 cents for regular crystal.

Take a close look at the MSP430F20xx series, specifically 2001/2011,

2002/2012, 2003/2013. The internal oscillator is factory calibrated and surprisingly good over temperature and voltage. But the parts are very small.

The next best thing I'd suggest is any of the other MSP430 family, with a 32.768kHz crystal of the required accuracy, and use the internal DCO calibrated against the 32k crystal.

The 50ppm tolerance requires a crystal or ceramic resonator. Among the other suggested solutions, you can use the AC power frequency as a reference. A microcontroller like MSP430 or HC908 can use the PLL to lock on the AC frequency directly. With the other microcontrollers, you may be able to correct either the internal oscillator itself or the critical frequencies derived from the oscillator by using the AC as the ref. clock.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Over what temperature and voltage range ? Does this have any field-signals for Cal-Check, or do you also want this stable over the product life too ?

No one makes that precision, which is one part in 20,000. Trimmed RC osc are in the one part in 100 range, with trims steps to around one part in 250 - so the OP is chasing around 100x the start of the art.

Single package precisions like this, may be possible in years to come, with the Silicon MEMS oscillators. This is likely to be Dual-Die solutions, but with a custom Osc that matches the needs of the MEMS block. See SiTime's web site - their special bridge is 0.8mm x 0.6mm x 0.15mm, so can fit into a molded package.

-jg

This might work for us. We will have three to four die on a hybrid anyway. We will likely buy a life-time (project or my life, whichever come first) supply of AVR169 die to pack with it.

While the AC mains have a very good long time accuracy in most countries, since the generators are driven at a small overspeed at night to catch up any cycles lost during high demand during the day, a mains driven clock can be off by several seconds during the day (frequency error up to 100 ppm or even more). Last time I looked at

the Finnish (and hence also Nordic network) frequency was 49.97 Hz and the time error more than 23 s.

This may be an issue if the time is compared to an accurate time source (NTP, GPS etc.) e.g. one system running of the mains and the other from some NTP source.

However, if it is just needed that two (or more) devices keep the same time, regardless of accuracy, the AC mains is a good clock source, however, there may be a 7-15 ms time offset, if the devices are powered from different phases in a 3-phase distribution system.

Some precaution is also needed to avoid false clock pulses due to all kind of interference riding on the AC mains voltage. For devices intended to work all over the world, some logic is needed to sense if

50 or 60 Hz mains is used.

Paul

The only practical way to get 50ppm absolute frequency tolerance over temperature is with a crystal or an external oscillator that uses a crystal. With the latter you'll get a guarantee, but it tends to use more power. If absolute tolerance isn't important (say 5000ppm) and 15 or 25ppm/K typical tempco is okay then consider a small SMT ceramic resonator.

Best regards, Spehro Pefhany

--
"it's the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

The uC can adjust for this error if necessary, since it needs to sample the frequency anyway. Perhaps you can also adjust it with a light sensor.

However, when space is the constraint, AC is usually not available. This is true in my case and probably true in the OP's case as well.

Not very practical for the mains+RC oscillator case, more realistic in case of mains+crystal oscillator case.

You do not know when the mains is at the nominal frequency and even if you measure the cycles for any 24 hour period, this could still give an error of a few seconds. Averaging such measurements over a period of one week or one month will reduce the error significantly. Crystal oscillators have some initial frequency error and some aging and these can be compensated by the averaged mains reference after a few weeks.

However, the short term (minutes and hours) frequency stability for both the mains and an RC oscillator is so bad, I very much doubt that you could get reliable values even with long averaging, at least you would have to use a temperature sensor to compensate for the RC oscillator temperature sensitivity.

You are going to have to use quite long averaging (years), if you intend to use the light sensor to detect night and day and thus adjust the internal oscillator :-).

If you intend to get the mains frequency from an AC powered lamp, this may work if the sensor is mounted inside a fluorescent lamp using a conventional ballast, however, ambient (solar) light and light from other fluorescent lights driven from an other mains phase will degrade the light variations significantly.

A 15 W/230 V incandescent lamp may also produce significant light variations during each mains cycle, however lamps with higher power or lower operating voltage will have much higher thermal inertia, recusing the light variations significantly during a mains cycle.

Most modern "energy-saver" lamps have a free running switched mode power supply, so you can not determine the mains frequency from the light output.

Paul

come

oscilator is

We were not farmilure with that spacific manufacturer. do you know if the make it also for lower frequencies like 4Mhz?

Thanks,

Roy

No, they don't. The smaller crystals only come in higher frequencies.

If you like the 16 MHz xtal solution, you may be able to switch to a 16 MHz MCU. There are some available in packages down to 3x3 mm, depending on your requirements.

For instance:

Depends on where you get them from.

instance:

I got a couple of 8 MHz samples in 3x2. They are close to $1 for

1000. They are so small that I can't find them anymore, may be in one of my pockets or the washer/dryer.

me

is

Interesting info :):)

Karthik Balaguru