I have been planning on using the AD9914 to output a 1.5GHz sin wave with 10 MHz of phase modulation.
Of course this is pushing a little too close to the Nyquist frequency to use the AD9914 reconstruction filter, so I planned to take the unfiltered output and use my own external LPF.
Since only 2 samples are needed to reconstruct the sin wave I'm thinking of using a clock of 3GHz, so i really only have to account for harmonics at 4.5GHz and onwards.
Assuming the filter is designed properly will this work as intended?
and as a followup question:
Given the parallel programming rate data rate of 125MHz (1/24 of 3GHz), that gives 12.5 samples for a modulating wave at 10MHz.
Will a reconstruction filter also smooth out the applied phase modulation, as 12.5 samples isn't too good for tonal quality
Generating a 1.5GHz output using a DDS clocked at 3GHz will not work in the way you might think. The output will consist of two amplitude samples that repeat indefinitely. If the initial phase happens to be zero (which corresponds to zero amplitude for a sine function), then the output will be 0V DC. That is, no signal. Similarly, if the initial phase is 90-deg (pi/2), the output will be a square wave.
Note that, operating at the Nyquist frequency (1/2 the sample rate) means the bandwidth of the reconstruction filter would need to be 0Hz!
If a 1.5GHz carrier is a requirement, then you should use a 3.5GHz clock and a well designed bandpass filter with narrow bandwidth (<250MHz). Note such a filter is not easy to come by.
Regarding your "filter" question...
Generally speaking, the bandwidth of the reconstruction filter will be much greater than the 10MHz bandwidth of the modulation and should have minimal effect.