After tests the other day looking at the few milliHz error observable using the FDM-DUO locked to an external reference, I decided to make sure it was purely an NCO rounding error and nothing else hapening inside the radio

Assuming a 32 Bit NCO clocked at 122.88MHz,  we need to know the lowest common frequency that is both a multiple of the tuning step and also a binary submultiple of the clock.   this, and all its multiples,  can be generated exactly with no NCO tuning error.   

For the DUO receiver this frequency is 1875Hz, being Fc/65536 and an integral number of Hz.

Therefore, any RF tuning point (LO frequency)  that is a multiple of 1875Hz should exhibit no NCO offset error.

The only frequency in the 137kHz band that meets this criteria is 136.875kHz  so the radio was set to this in USB mode.  (1875Hz * 73)

My DDS low frequency signal generator was set to generate at 137.975kHz so a 1kHz tone resulted.   Using the DUO's internal soundcard, the G3PLX vectorscope software was used to measure the phase of the 1kHz output against a reference NCO of 1kHz derived from samples arriving from the soundcard.  Since the soundcard samples at 48kHz which is an integer multiple of 1kHz there will be no rounding errors.  Radio and sources were locked to the same reference (*) so its accuracy didn't enter into the equation.  

The only potential error in frequency is my DDS source that, due to its internal maths / rounding, was generating this RF 2.17uHz high  (that 's 2.17micro-Hertz)

The vectorscope showed a dot on the screen and a phase readout of this position to a resolution of 0.1 degree.      After 48 minutes of monitoring, the observed phase of the 1kHz tone had shifted from 131.9deg to 130.0deg, ie -1.9 degrees in 2880 seconds.   This equates to a frequency error of 1.9/360/2880 = 1.8uHz

Which equal to the error of the DDS RF source allowing for the 0.1deg phase readout resolution

So, it looks as if the EXACT frequency of the FDM-DUO's local oscillator when set to any particular frequency ought to be calculable:

Work out the nearest integer N value programmed into the NCO for the 1Hz frequency specified, then calculate backwards using that N to get the actual frequency generated.   Multiples of 1875Hz will be exact

Glad that's sorted !

(*) initially, as I was running out of BNC sockets on my Shack master reference splitter (driven from an old GPSDO)  I initially ran the RF source from another GPSDO.  Almost immediately a slight phase wander of the 137kHz signal was seen.   This calculated as being a frequency difference in the order of 10^-9 (1 PPB) between the two references.   Which is exactly the sort of short term error one would expect to see in simple GPSDOs)    A rearrangememt of 10MHz master feeds and all was sorted for the test.

Andy

www.g4jnt.com