I guess it depends on the accuracy you need. If you start with a non-power-of-two frequency,
like 1.000 MHz instead of 2^20 Hz, then you will not generate an exact frequency in Hz, but with
enough bits in the DDS phase, you can come at least as close as the accuracy of your timebase.
For example using 1.000 MHz and 32 bits of phase, your resolution would be about 0.000233 Hz.
For an output frquency of 20 Hz, this represents 11 ppm of error. If you need to enter the frequency
in Hz, then you would need to add a multiplier to generate the correct phase increment. On the
other hand, if for example the frequency is being requested by a microprocessor in your system,
then the frequency could be specified by its phase increment, and the microprocessor could do
the conversion from Hz to phase increment.
If you use the DDS compiler, you can generate a core with a clock enable. You can then run this
core directly from 10 MHz and enable its clock for one cycle out of every 10, effectively running
the DDS at 1 MHz. In that case you don't need a DCM.
Another approach that would give you an exact frequency would be to design your own DDS
that uses a decimal accumulator to directly divide 1 MHz down to 1 Hz. This accumulator would have
6 digits, the phase increment would be in decimal, and the SIN / COS table would have
either 2 or 3 digits (100 ro 1000 elements). While this approach would allow you to have "exact"
frequencies, it would be a lot more complex.
-- Gabor
-- Gabor
