Although DDS sources seem to more-or-less dominate where high resolution / accuracy is needed, there is another route that could be worth investigating:   Fractional-N synthesis with addtional dividers.

For example, the LMX2541  with a  10MHz reference  phase comparator input will generate steps as low as 2.38Hz at its minimum internal VCO frequency of 2.2GHz.  A programmable output divider goes to divide-by 63, allowing steps of 0.038Hz at 35MHz .  Add a single chip binary divider of 256 (for the 137kHz band)  and you're down to 150uHz steps - which are completely deterministic.

Since phase noise and Fract-N sidebands are hardly an issue with this degree of output division, the phase comparator could be operated at a much lower frequency, say 500kHz, allowing steps of  lower than 10uHz.   Which is a significantly higher resolution than the 2.3mHz steps the traditional AD9850 32 bit DDS offers. 

I've used these Fract-N devices extensively at microwaves,  see http://www.g4jnt.com/LMX2541_Synth_Module.pdf ,    but haven't yet turned one into an LF exciter

Andy  G4JNT