I appreciate those notes; the data sheet for the DDS part also explains it. It has a 12-bit LUT driving an internal 10-bit DAC (the 5.3 MHz sine is still a bit coarse, with < 4 samples per cycle, but in the 5680 it is cleaned up by an external filter).
"Because phase information maps directly into amplitude, a ROM LUT converts the phase information into amplitude. To do this, the digital phase information is used to address a sine ROM LUT. Although the NCO contains a 32-bit phase accumulator, the output of the NCO is truncated to 12 bits. Using the full resolution of the phase accumulator is impractical and unnecessary because this would require a look-up table of 2^32 entries." ...from http://www.analog.com/static/imported-files/data_sheets/AD9832.pdf However, my question was actually about how the remainder of the circuitry in the FE-5680A combines the 5.3 MHz from the DDS (at 4 mHz tuning step size), and the 60 MHz VCXO, to reference against the 6.835 GHz Rb frequency and ultimately achieve 0.18 uHz (micro-Hz) tuning step size at the final 10 MHz output. I don't think a simple multiplier-mixer-divider chain (for example) could give you such a small tuning step size at the output, the frequency ratios don't work out. I've heard of fractional-N PLL synthesizers but I'm not sure if that's the principle here. > -------Original Message------- > From: Graham / KE9H <[email protected]> > > The AD9832 is an Analog Devices DDS which has a 32 bit tuning word. > The way a DDS generates the output, is that it (effectively) has a cosine > wave look-up table, with 2^32 entries that comprise a single cosine wave > cycle. > > The tuning word tells it how many entries the DDS should advance every > reference input clock cycle, then it pushes that amplitude value in the > look-up table > to the output D->A converter. > > So, if the input reference is 20 MHz, then the DDS can generate frequencies > with a resolution step of > > Vref/2^32 = 20,000,000 / 4,294,967,296 = 0.0046566 Hz. > > The DDS output frequency is (tuning word /2^32) times Vref. > > In the actual implementation, rather than a 4 billion entry look-up table, > I am sure they have some algorithm that calculates the amplitude of > a cosine wave, or a much smaller table with a sophisticated interpolation > routine. > > --- Graham / KE9H _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
