We are having trouble with some individual frequencies where the SNR is very poor (maybe 12 out of 100 for 5MHz channels). A subset of those we tracked to poor phase noise on the transmit side (using external equipment). We assume similar problems exist for the rcv and hope this accounts for all 'bad' frequencies.
In some cases we could get good SNR by using the bleed circuit (Regs 0x23c/0x27c) to move the VCO operating point away from signal ground. We monitor this by means of the thresholds in regs 0x24b/0x28b and the comparator output in regs 0x247/0x287.
There are 2 problems with this:
1) the settings are not consistent: xmt & rcv within the same unit have different vtune for the same frequency and bleed and there are differences across units as well. So the 'correct' bleed value needs to be found on the fly for each individual synthesizer.
This brings us to the second problem:
2) the steps of the comparator are very coarse, specially around signal ground (500 to 800mV), so most of the time we really can't see the effect of the bleed setting.
We did play a little also with the VCO cal offset (regs 0x238/0x278), but we weren't able to find a reliable solution.
Other than those tests, we are using the ADI table for FDD, 80MHz reference without modification. Most of the problem frequencies are between 5250 and 5450 MHz.
We need a document where the Synthesizer theory of operation is detailed so that we are able to control the cal process, or an algorithm to modify the parameters so that good performance is guaranteed for all frequencies.