[time-nuts] 32.768KHz from a DDS

WB6BNQ wb6bnq at cox.net
Wed Jul 23 16:34:16 EDT 2008


   If you use a DDS then it pays to use the Analog Devices design tool.
   It is at this URL

    [1]http://designtools.analog.com/dtDDSWeb/dtDDSMain.aspx

   You can open multiple instances in different browser windows.  Also,
   there is an area under the graphs called DISPLAY which allows for some
   selections.  One of them is for a filter.  If you click on the blue
   configure to the right of the filter selection it allows you to set the
   parameters for the filter.

   It is interesting to note just how imprecise these DDS devices really
   are !  For instance, the tendency to use the house standard (i.e., 10
   MHz) as the DDS source clock produces the following values for 32 KHz
   clock.

   1. AD9850 DDS {32 bit tuning/10 bit output} with 10 MHz clock
   produces:  32768.00038293  Hz

   2. AD9956 DDS {48 bit tuning/14 bit output} with 10 MHz clock
   produces:  32767.999999983  Hz

   You would think scaling the clock by [sub] multiples would make no
   difference.  However, this is not true.  You would think using a clock
   that was a binary multiple, just like the 32 KHz number, would make it
   come out precisely.  As well, this was not the case !

   Experiment with the better 48 bit DDS and varying the clock from 10 MHz
   down with the following  results:

     * CLOCK         FREQUENCY
     *
     * 0.01 MHz     32768.0000000000  Hz perfect but no output !  See Mr.
       Nyquest.
     *
     * 0.1 MHz       32768.0000000001  Hz
     *
     * 0.2 MHz       32768.0000000001  Hz
     *
     * 0.3 MHz       32768.0000000004  Hz
     *
     * 0.4 MHz       32768.0000000001  Hz
     *
     * 0.5 MHz       32768.0000000008  Hz
     *
     * 0.6 MHz       32767.9999999994  Hz
     *
     * 0.7 MHz       32767.9999999990  Hz
     *
     * 0.8 MHz       32767.9999999987  Hz
     *
     * 0.9 MHz       32768.0000000004  Hz
     *
     * 1 MHz          32768.0000000008  Hz
     *
     * 2 MHz          32767.9999999972  Hz
     *
     * 3 MHz          32767.9999999972  Hz
     *
     * 4 MHz          32768.0000000044  Hz
     *
     * 5 MHz          32768.0000000008  Hz
     *
     * 6 MHz          32767.9999999972  Hz
     *
     * 7 MHz          32768.0000000115  Hz
     *
     * 8 MHz          32768.0000000044  Hz
     *
     * 9 MHz          32767.9999999972  Hz
     *
     * 10 MHz        32767.9999999830  Hz

   What is interesting is there were few repetitive values.  Also
   surprising is the lower the clock the closer to the correct output.
   BUT BIG SURPRISE !  Dont forget Mr. Nyquest.  To get a reasonable wave
   form you really need to have the clock 3 times the highest output.  So
   that leaves us with 100 KHz as the minimum practical clock frequency.
   The error at 100 KHz is quite small and essentially beyond the range of
   the DDSs 48 bit tuning word.

   Of course one should be mindful of the spurious responses out of the
   DDS.  The 200 KHz selection { an easy sub of 10 MHz} would be better
   based on the Analog Devices program after setting the filter factors.
   I certainly am no expert but the Elliptical filter, which I have some
   small familiarity with, seemed to be the best choice for the frequency
   range.

   Bill....WB6BNQ

References

   1. http://designtools.analog.com/dtDDSWeb/dtDDSMain.aspx



More information about the time-nuts mailing list