Updated: Jun 29, 2019
If this blog post was a YouTube video meant to attract attention it would perhaps had a title saying: Silicon Labs lies! Clearly gives false information in their datasheet!
But as this is not an sensationally YouTube video, just a curios and well behaved blog from a guy that likes RF electronics - I will just say that we have worked with assumptions regarding the popular Si5351 that may not be correct. Read on for some interesting new facts that have come to light due to the work of Dr. Simon Schrödle
The Silicon Labs Si5351 clock generator is used in many of my projects and is a very popular PLL for Ham use because of it's low cost and decent performance.
If you search for Si5351 and Ham you will find a myriad of projects that use it to generate clock and RF signals.
It uses a reference input that is usually fed by a crystal or in some cases an external oscillator as in my WSPR transmitters. The stability and frequency accuracy of the output form the Si5351 is directly related to the reference oscillator that is feeding the Si5351 PLL. So in cases were one needs high accuracy one would use a precision oscillator. In my WSPR transmitters I use a temperature controlled oscillator that has a built in circuit to keep the oscillator well behaved as the temperature changes. This gives an accuracy of around 2.5 parts per million. That means the maximum frequency drift would be around 18Hz if the output frequency is 7MHz.
But with an output frequency say on 28 Mhz it would be a bit more, about 75Hertz drift. So that is not to bad but as it gets worse with frequency it means that output on the 2m band can be off by 360Hz. 360 Hz frequency drift or offset is still not a lot for a regular receiver but in the case of WSPR transmission/reception the demands are quite exceptional as the transmission of a single station only occupies a few Hertz of bandwidth.
The entire portion of the WSPR transmission window is only 200Hz wide so a 360Hz drift can potentially land us outside the WSPR window.
The obvious solution to this problem - and the solution for a long time whenever a stable frequency is needed - has always been to use an crystal oscillator in a temperature controlled "oven". This gives accuracy that is in the parts-per-billion instead of part-per-million so our drift would be just a fraction of a Hertz in many cases.
I have a frequency standard product that houses a low cost OCXO that is the perfect candidate for a good stable reference to the Si5351.
I currently use this OCXO for my synthesizers but I am planing to use it as a reference for the Si5351 as well.
A small problem that needs to be solved first is that OCXOs usually operate at 10MHz and the Si5351 needs an oscillator with a frequency between 25MHz and 27MHz. At least that what the data sheets says.
I have built a conversion a circuits that takes the 10MHz and generate 25MHz from it and then feeds the Si5351.
The principle is pretty easy, divide down the 10MHz to 5MHz with a digital divider and filter out the 5th overtone to arrive at 25MHz. The filter can be a conventional LC band-pass or a 25MHz crystal or a combination of the two.
I was planing to use this for a OCXO based Si5351 signal generator and for 2m and 70cm WSPR transmitters.
But now it seems this cumbersome approach is not needed. The si5351 can be directly feed with 10MHz even though the data sheets say input should be between 25 to 27MHz!
Simon Schrödle from Germany has a blog that I follow and in one of his post he writes about test he have carried out on driving the Si5351 with various reference frequencies. Read his blog post here:
By looking at his diagram it seems possible to drive it with anything from 2 to 100MHz! So why is the data sheet stating to use crystals between 25 and 27MHz?
I'm guessing it is because of three factors:
27MHz crystals are inexpensive and easy to source.
They can be made small in size
The stability of a crystal is generally better the higher the frequency it operates at. 27MHz is reaching the top limit of fundamental oscillation of crystals while still retaining a low cost.
So what can we learn from this, not to trust datasheets? :-) Well maybe not but perhaps that reading datasheets is needed and all well and good but sometime you just have to wire something up and try it out!
I haven't tried it out myself but I will soon do so, I encourage you to do the same if you a have an OCXO lying around or if you want to basically insert any frequency in to the Si5351 as a reference, check out the blog post for recommended input power over different frequencies.
//Harry - SM7PNV