[time-nuts] questions on uncompensated crystal oscillators

[John Day]

At 02:48 AM 7/5/2006, SAIDJACK at aol.com wrote:
>
>In a message dated 7/4/2006 09:52:33 Pacific Daylight Time,  stans at sycard.com
>writes:
>
>Hello,
>Please excuse the fact that this is not what we all  consider precise
>frequency.  I am selecting crystals to use for a  medium to high volume
>application.
>The basic criterion are:
>- used  with a VLSI inverter based oscillator with
>3.3 V supply, but  somewhat high impedance output.
>- frequency: 27.000 MHz fundamental within  40 (or so) ppm over
>temperature including 7 to 10 years of  aging.
>- HC49S case
>- no production line trimming
>- low cost
>
>After looking at the design issues, I wonder if some of you
>haven't  faced similar designs and have some suggestions
>regarding the following  issues:
>
>
>
>
>Hi Stan,
>
>I have designed a circuit that does exactly this, and published it in EDN.
>See page 92 for the article:
>
>_http://www.edn.com/contents/images/112703di.pdf_
>(http://www.edn.com/contents/images/112703di.pdf)
>
>It's a VCXO 27/32MHz reference for a Video Decoder. You can delete the
>Varicaps if you don't need it to be adjustable.
>
>BTW: The crystals and inverter used in the design are quite good, the
>circuit generates less than 3ps jitter (measured on a 
>Wavecrest  SIA-3000) - jitter
>is really what counts in digital timing for video  etc.
>
>Some caveats to watch out for:
>
>    * for NTSC you can have up to about +-840Hz deviation and  still 
> be within
>limits.
>
>    * Don't overdrive your crystal (see manufacturers spec)  otherwise it can
>age too fast, and get damaged. Typically, use about 100uW or  so. Measure the
>current into the crystal using a fast AC current probe
>
>    * Make sure you have enough gain to guarantee startup 
> (the  circuit should
>work with at least 3x series resistance as will be used  in production). This
>will mostly depend on the crystal ESR (should be as low as  possible), while
>still not overstressing the crystal
>
>    * 10ppm crystals are very(!) expensive, rather use trimming  during
>production testing by removing/adding load cap options. This allows 
>you  to use a
>cheaper crystal (with a higher ppm rating),  and spend more money on lower
>thermal susceptability specs.
>
>Hope this helps,
>bye,
>Said
>

I would agree with you Said, 10ppm is a lot to ask and let's be 
honest, production trimming is not so difficult these days. Without 
going to 10ppm you can get your crystal manufacturer to group your 
crystals by cut angle. This will give you a basic temperature 
coefficient curve - for a given angle the tempco curves are quite 
similar. We used to do this with fundamental mode crystals at between 
12 and 20MHz.

You can then have a set of compensation tables and drive the varicaps 
from a DAC to keep the crystal under control. I have done this using 
small micro's, like the Zilog Z8, or using a bit of really crude 
ADC-Table-DAC circuitry in an FPGA and it works quite well. This was 
done to avoid having to use TCXO's in SCADA/DATA type radios.

For higher performance equipment we used to have the compensation 
table built by the test system during temperature cycling.

John