I am using the AD9517-1BCPZ in an automatic reference switchover mode and also to generate various frequencies. REF1 is a 10MHz signal from a Tektronix function generator (AFG310) but in the actual application it will be a GPS derived signal from a GPS receiver. REF2 is a 10MHz signal from a highly stable VCTCXO (Euroquartz VEM42T33-10). Both signals are LVCMOS 3.3V signals.
The on chip VCO is tuned to run at 2.4576GHz with R divider = 25 and N divider = 6144. The phase detector comparison frequency is 400kHz. The output of the VCO is divided by 4 to generate a 614.4MHz signal that is directly passed to OUT0 (LVPECL). Further two LVCMOS outputs at OUT4 (38.4MHz, divider = 16) and OUT6 (9.6MHz, divider = 64) are generated from the 614.4MHz signal.
The accuracy and the phase noise of the generated signals are good. But the major problem is the false reference switchover that happens even when REF1 (from the signal generator) is accurate and present at all time. The false reference switch over occurs very often but for a brief moment approximately 400ns and it then reverts back to REF1 as it should do in automatic reference switchover mode. According to data sheet Rev-0 it takes four clock cycles of REF2 to determine the presence of REF1 and hence the 400ns time taken to return to REF1 makes sense. But I cannot find the cause of this false reference switchover.
The above mentioned problem happens on our product board and also on the Analog Devices Evaluation Board AD9517-1 Rev-B.
The block diagram of my design, Evaluation Board setup file and the ADIsimCLK Ver 1.3 simulation files are attached. The ADIsimCLK was used to simulate and compute the PLL component values.
I would also want to know the correctness of the statement made on the data sheet Rev-D on page 35. It states that “Automatic revertive switchover relies on the REFMON pin to indicate when REF1 disappears. By programming Register 0x01B = 0xF7 and Register 0x01C = 0x26, the REFMON pin is programmed to be high when REF1 is invalid, which commands the switch to REF2”. In register 0x01C bits [4:3] are reserved and writing 0x26 puts the device into manual switchover mode contradicting the data sheet. On Rev-0 data sheet bit  was used to specify automatic (1) or manual (0) switchover. Has the definition of register 0x01C changed?