Programmable oscillators?

[jmarkwolf]

Went to purchase some half-can oscillators for the first time in a looooong time. The frequency I need in a thru-hole part seems to be in short supply.

I see that some vendors sell parts pre-programmed to your specified frequency.

Some parts are apparently user-programmable.

Can anyone advise on personal experience with these parts?

[asmi]

I use Time Machine II. Programmer supports wide variety of parts from "regular" low power oscillators all the way to VCXO and SSXO, both single-ended and differential outputs - see the link above to see the full list of supported parts. Some parts allow to "dial-in" drive strength, operating voltage, function of control pin during programming along with frequency. Very versatile kit. It's a bit on a pricey side, but it's one-time purchase and starter kit includes small assortment of blank parts to get you started.
Works great for prototypes as I don't have to keep an inventory of parts for different frequencies, also I can take advantage of volume discounts as I just buy a whole bunch of single part # (blanks) at a time, and then program them into all kinds of generators as I need them. Didn't regret investing into programmer for a second. But for a volume production I'd stick with factory-programmed parts to make things easier. Some distributors (I know of Digikey, maybe others do that too) offer pre-programming service too - in case you need some oddball frequency, and your volume isn't large enough to order at the factory.

[David Hess]

Can anyone advise on personal experience with these parts?

Since the output is synthesized, these parts can have atrocious phase noise compared to the bare oscillator.  For many applications this is irrelevant but they are not suitable for applications which require low phase noise.

[asmi]

Since the output is synthesized, these parts can have atrocious phase noise compared to the bare oscillator.  For many applications this is irrelevant but they are not suitable for applications which require low phase noise.

Some parts have very good phase noise: https://www.sitime.com/products/oscillators/sit8208#magictabs_S81hw_1

[David Hess]

Since the output is synthesized, these parts can have atrocious phase noise compared to the bare oscillator.  For many applications this is irrelevant but they are not suitable for applications which require low phase noise.

Some parts have very good phase noise: https://www.sitime.com/products/oscillators/sit8208#magictabs_S81hw_1

That is very good for a silicon mems oscillator or programmable oscillator but at least 10 times worse than a cheap fixed frequency quartz oscillator.  Its power consumption is high also.  It would be a fun part if you could program the output frequency at will.

I was referring to programmable oscillators in general; some are truly terrible and only suited for driving digital logic.  Some manufacturers have been replacing their fixed frequency crystal oscillators with factory programmed synthesized ones without advertising it leading to very surprised customers.

[asmi]

That is very good for a silicon mems oscillator or programmable oscillator but at least 10 times worse than a cheap fixed frequency quartz oscillator.  Its power consumption is high also.  It would be a fun part if you could program the output frequency at will.

Well there are such parts too. But for SiTime parts it's not really practical as most of their parts are in LGA-like packages making desoldering-re-soldering them royal pain in the butt. So OTP parts are just fine with me.

I was referring to programmable oscillators in general; some are truly terrible and only suited for driving digital logic.  Some manufacturers have been replacing their fixed frequency crystal oscillators with factory programmed synthesized ones without advertising it leading to very surprised customers.

I only use them to drive digital logic (mostly PLLs internal to FPGA/MCU). Never had a problem.
Your statement just made it sound like these things are absolutely garbage and unuseable. They are not. Infact I never designed a project where 3ps of jitter (worst case as per datasheet) would measurably affect anything. But I'm pretty sure such designs do exist. Still - convenience and cost savings from using user OTP programmable parts are enough for me to recommend them as general-purpose generators. Just keep in mind their limitations and use dedicated parts where these limitations make OTP parts unusable.

[BrianHG]

Though it is a crystal PLL synth clock chip, with built in flash so you can program a frequency once and it will power up to that every time in circuit, I use this IC from TI: CDCE925PW, or CDCE913PW
With good PLL tuning variables,  I got quite a bit better than their specified 60ps period jitter as far as my scope was concerned.
With a 27 Mhz crystal, I've generated many clocks with less phase noise that quite a few fundamental 4 pin crystal oscillator and you also get an optional analog PLL crystal tuning pin if you need to lock to an external reference.  Though, you pay for the ultra low jitter, 2 PLLs, and the VCXO feature.  Cheaper versions exist, but I wouldn't use them for broadcast video clock generators.

[mikerj]

Your statement just made it sound like these things are absolutely garbage and unuseable.

What David said is absolutely true; programmable oscillators can have very poor phase noise.  If you need good phase noise performance then these parts would be unusable.  In what way is that saying it's garbage and unusable in a general case?

Just because you don't care about phase noise in your application does not mean phase noise is irrelevant in general.

[jmarkwolf]

Forgive me. Phase noise being "jitter"?

[Leo Bodnar]

Jitter is the time domain and phase noise is frequency domain of the same thing.
There are, of course, millions of subtleties but in essence this is how it rolls.
Leo

[jmarkwolf]

Thanks for the discussion everybody.

My application requires a 1-second time base. With the right frequency oscillator and software I can derive that with my microcontroller, but now
I'm thinking I'll just use the micro's internal oscillator for code execution, and an 8-pin clock/calendar chip and a 32Khz crystal as a time-base for the 1-second interrupt.

It simplifies the code and may even be cheaper than a hard-to-find oscillator frequency for the micro.

[asmi]

Thanks for the discussion everybody.

My application requires a 1-second time base. With the right frequency oscillator and software I can derive that with my microcontroller, but now
I'm thinking I'll just use the micro's internal oscillator for code execution, and an 8-pin clock/calendar chip and a 32Khz crystal as a time-base for the 1-second interrupt.

It simplifies the code and may even be cheaper than a hard-to-find oscillator frequency for the micro.

Yep, in your case RTC IC is the best option. That's what I would've used in these circumstances. If you have RTC peripheral in your MCU, you can leverage it instead of dedicated chip - just get a good quality low-freq. crystal.

[ConKbot]

Your statement just made it sound like these things are absolutely garbage and unuseable.

What David said is absolutely true; programmable oscillators can have very poor phase noise.  If you need good phase noise performance then these parts would be unusable.  In what way is that saying it's garbage and unusable in a general case?

Just because you don't care about phase noise in your application does not mean phase noise is irrelevant in general.

To clarify here, the trash bit is the part you don't see in the datasheet. The close in phase noise (<10khz offset) is quite bad, hence why the phase noise plots if available start at 10-12khz. Stuff that doesn't care about that (clocking a micro or a FPGA, etc) not a problem, and they are handy little buggers.  Trying to use it for an reference in an RF synth, a bit more of a problem.

Used one as a source for a TI pll, and the resulting carrier was comically wide, I almost thought the PLL was unlocked but the frequency center was too stable. The "low jitter" version was improved, but still noticeably worse on the spec an then a plain crystal oscillator.

[David Hess]

With good PLL tuning variables,  I got quite a bit better than their specified 60ps period jitter as far as my scope was concerned.

You have an oscilloscope that can measure 60 picoseconds of jitter?

My best real time DSO only achieves 20 GS/s so would miss it.  My analog sampling oscilloscope can see 60 picoseconds of jitter easily but only under the right conditions.

With a 27 Mhz crystal, I've generated many clocks with less phase noise that quite a few fundamental 4 pin crystal oscillator and you also get an optional analog PLL crystal tuning pin if you need to lock to an external reference.  Though, you pay for the ultra low jitter, 2 PLLs, and the VCXO feature.  Cheaper versions exist, but I wouldn't use them for broadcast video clock generators.

I have read about people getting in trouble when the manufacturer of the crystal based clock oscillator they were using silently changed to a factory programmable one.  The original part did not have a jitter or phase noise specification but they knew from testing how it worked and took for granted that a "crystal oscillator" implied that it was not synthesized.

After reading about various crystal oscillator fiascos, I am inclined to always rely on a discrete transistor crystal oscillator in place of canned oscillators or gate oscillators.  There is just less to go wrong and the performance for a given price is better.

[BrianHG]

You have an oscilloscope that can measure 60 picoseconds of jitter?

My best real time DSO only achieves 20 GS/s so would miss it.  My analog sampling oscilloscope can see 60 picoseconds of jitter easily but only under the right conditions.

Though my scope was 5gs/sec, when creating a video clock, syncing to a hsync and zooming into the clock after 100 clock cycles from the h-sync, the jitter of the different crystal oscillator and different PLL oscillators I was testing was (in a way) amplified 100 fold making it measurable.  Some were pure crap.  And, even the TI chip's performance was hindered strictly following TI's erroneous documentation describing the PLL register settings and formula.  (TI, useful capable cheap parts, but, LOUSY .pdf documentation full of omissions and errors.....)

[Zero999]

Jitter is the time domain and phase noise is frequency domain of the same thing.
There are, of course, millions of subtleties but in essence this is how it rolls.
Leo

Yes, to expand on that. Jitter and phase noise are two ways of quantifying the same phenomenon. Phase noise is the integral of jitter.

[jmarkwolf]

Update from OP:

Excellent discussion everybody.

I've pretty much decided to go with the DS3231 clock/calendar chip as my 1-second time-base. It's got an integral temperature compensated crystal/oscillator feature that does away with the variability of external watch crystals and load capacitors.

Yes, it's overkill, and is a 16-pin pkg, but I want an easily implemented, accurate, 1-second time-base without any pharting around. I think this fills the bill nicely.