I am working with the ADF4351 as a frequency source for a stepped frequency RF sensor. This unit continually sweeps across the frequency band on specific frequencies, starting with the lowest and monotonically increasing until it reaches the highest frequency and starts the process again. I am trying to track down the last noise sources in the system and have recently discovered that when I use a loopback test (just a static attenuator between the Tx and Rx port) that I see varitations from sweep to sweep. I know that this is to be expected and the majority of these variations only measure on the order of 10 to 20 ADC counts in my system which is fine. There are a few select frequencies however that are ranging on the order of 300 to 500 ADC counts. At first I thought that these were unlocked frequencies but when I step through them manually this does not seem to be the case and the system is not reporting a loss of lock condition from the ADF chip (I know that this does not guarantee a lock but wanted to mention it anyway) . When I captured the ADC data and plotted the amplitude over 1300 sweeps it became quickly obvious that there is a bimodal distributuion to the output power. Either of these two power settings varies only 10 to 20 ADC counts, but they are 300 to 500 counts apart from each other. My assumption is that I am close to the edge of two of the overlapping bands for the internal VCO and I am not always locking to the same band on each sweep. Has anybody run across this in the past? Does anybody have any suggestions on what I might be able to do to rectify this situation? At this point i am thinking that I may have to adjust my frequency plan to avoid the problematic steps, but would much rather come up with a better solution as I know that the bands vary from part to part and across temperature. Thanks!