We are having problems with HMC733 in our synthesizers. When the Vtune voltage is at a level corresponding to fundamental freq close to 18 GHz we observe that the VCO generates a particularly strong sub-harmonic signal at ca. 9 GHz This could be even stronger that the fundamental. Needless to say that VCO behaving like this baffles the PLL completely.
A number of HMC733 parts installed in a PLL-based synthesizer develops this strange behaviour either right after installation or after a short while. As a result the synthesizer is unable to generate any reasonable signal close to 18 GHz. Besides this sub-band all seems to work as expected.
To check what is going on we have installed one of the problematic VCOs on a bare PCB with the Vcc, and Vtune connected and the Pout routed with a semi-rigid coaxial line. With this circuit we could observe the behaviour of the VCO only and confirm that a number of chips we were working with exhibited the behaviour like described above. The problem is unlikely to be connected to the mounting/soldering procedure of our chips. Over the last year we built a number of synthesizers using the same soldering techniques and ...all of them are still working correctly. A month ago we built a synthesizer (let's call it SythA) and it also works OK. Using the same PCB layouts, and the same components we built then a copy of it -- SythB. All the problems described above were observed in SythB.
I wonder what operating conditions can cause HMC733 to work in a way described above. We have seen this behaviour in at least 3 chips now. The last one is quite interesting... We had a EVAL board EV-ADF41513SD2Z (with a HMC733 installed). It worked fine so we took its HMC733 out, checked that the harmonics are below -20 dBc and all is within specs. Then we installed it on SynthB, hoping that a verified and good component will bring SythB back to life in all 10-20 GHz band. This helped as the synthesizer was working fine for a week or so. But... after this week we are now seeing again the behaviour described in the beginning of this post quite often. Again, the synthesizer does not work close to 18 GHz, while it works fine outside of this band.
We will appreciate a lot some hints as to what can go this wrong in only one of the sythesizers (SynthB) while the another one (SynthA) working exactly in the same condiftions does not show any problems.
Sorry to hear that you are having problems with some of the HMC733 parts that you have received. The HMC733 parts you have are spec compliant per the datasheet (spur performance isn't guaranteed) however the out of band performance isn't what it should be. It seems that a few parts produce what appears as a spurious signal at half of the RFOUT frequency typically falling between 8 GHz and 9 GHz when operating between 16 GHz and 18 GHz. This spur can appear / disappear over very small tuning range changes, as little as 10mV. Unlike most of the HMC narrow band VCO's which rely on an oscillator topology that results in a consistent 1/2 harmonic, the HMC733 is a fundamental oscillator so this tone should not exist. When this sub-tone appears, instead of a single tone appearing at the VCO / RF input to the PLL, the ADF41513 sees 2 tones and becomes confused and fails to lock. Interestingly, although the HMC733 has been released for nearly a decade, it has really been since customers began pairing the HMC733 and ADF41513 parts together that this problem fully manifested itself. If we review the input sensitivity plot vs temperature for the ADF41513 (Fig 11) we can see why this occurred. Over the out of band frequency range of 8 GHz - 9 GHz the ADF41513 will function with input signals ranging from -37 to -24 dB but across the desired frequency range of 16 GHz to 18 GHz the sensitivity varies from -24 dB to -17.5 or nominally, + 7dB higher. In fact, the 8 GHz - 9 GHz region falls within the best portion of the band for input sensitivity which is not helpful when an undesired spur lands in this region.
Now, if we consider figure 11 in the ADF41513 datasheet you can see that depending on the device, after soldering parts to evaluation boards, the input sensitivity varies ±5.5 dB from part to part over 8 GHz - 9 GHz while at our desired range of 16 GHz to 18 GHz the input sensitivity varies nearly ±10 dB. Some of this behavior is likely tied to the device itself but we also have evidence that indicates that it may be tied to differences in the boards or soldering itself. If we consider these sensitivities along with solder creep this may explain why a part that initially worked failed to work later. If the ambient temperature remains constant, I would expect solder creep to complete within 24 hours if we are dealing with only the evaluation board but it the part were soldered onto a system level board mounted in a chassis and the system were to be run over temperature the solder creep would likely take considerably longer to fully complete. Depending on the spur level and the input sensitivity of the new PLL relative to the spur, I can see how this might happen especially when we consider that the edge of this input signal is used to trigger the phase frequency detector inside the PLL. IF operating near the threshold for triggering, small changes in signal levels can make the difference between a part that works and one that does not.
The root of the issue is tied to process variation and marginal out of band, buffer amp stability in the oscillator. Based on feedback we have received from customers the problem affects just a few devices (as you have witnessed). Customer applications that result in a very good load for the VCO, those not operating at cold temperatures, or those not using the 8 - 9 GHz portion of the band may never have a problem. We are in the process of correcting this issue but until we have a more permanent solution there are a number of ways to address the problem.
1) Install a narrow band filter to attenuate any spurs from 8 - 9 GHz to -40 dB or better, the filter should not significantly attenuate any frequencies in the pass band of 10 GHz - 20 GHz.
2) Provide a very good broadband resistive match at that output of the VCO. Our in-house testing revealed that loads with a return loss in the 28 - 30 dB range were enough to prevent the spur from appearing even at -40°C where gain is high. This was verified using only the HMC733 evaluation board. Unfortunately, the 14 - 15 dB of insertion loss that will come with this level of return loss will result in input levels at 20 GHz that are marginal for reliable operation of the ADF41513 over temperature so I recommend starting with a 50-ohm, 10 dB pi pad attenuator which will provide about 5 dB of margin. It may be possible to increase this to 12 dB, but this would reduce margin to 3 dB. This approach could also be modified and combined with the filtering mentioned above.
3) Another solution is always attempt to lock at frequencies between 16 GHz and 18 GHz by approaching them from the upper end of the band. This prevents the ADF41513 from ever seeing the spur so reliably locks.
4) The input sensitivity of the ADF41513 could be reduced by adding a shunt resistor much in the same way as we do on some of our frequency divider products with differential inputs in order to prevent them from oscillating in the absence of an input signal. Unfortunately deriving this value is bit empirical and very time consuming so currently we have not pursued this option.
UPDATE (1.10.2020): Since posting this we have confirmed that exact frequencies where this can occur will vary anywhere from 15GHz or so to 18GHz depending on the load presented at the output of the VCO. I also found that adding a shunt 120 ohm resistor as close to the RFOUT pin of the HMC733 as possible, improves the stability of the buffer amplifier. I also tried 96 and 150 ohms but 120 ohms was best using the eval board and nearly eliminated the parasitic oscillation at -40°C on a device experiencing significant parasitic oscillations (only occurred a couple of times and over very narrow vtune ranges (<0.1V). The exact value required for your application may vary slightly. This modification only reduces the output power by about 2 dBm and may be a viable solution for some applications. Any further improvement at problematic frequencies should be able to be suppressed by adding a narrow, band-reject filter or a HP filter (ideally downstream from any resistive attenuator that may be in place).
Hopefully this helps,
Many thanks for the detailed answer. Your explanation of why the PLL does not lock in the sub-band 18-19 GHz matches exactly what we ourselves suspected. There are two strong frequencies present in the VCO-generated signal (when tuned to ca. 18 GHz) when this happens the PLL has no means of discerning between them, which drives the loop out of sync.
I do not agree, however, with the explanation suggesting that the reason for this effect to become visible in the 18-19 GHz band is mostly related to the sensitivity curve of the PLL.
The reason for me being rather sceptical is the fact that we were observing the effect using the good, old ADF41020 (not the new ADF41513). When you look at the specs. of ADF41020 and the ref. sensitivity of the chip you see that it is rather flat vs. frequency, and improves... close to 18 GHz (in case of one of the prescalers, the one that we are using). What this means is that the input sensitivity does not matter, apparently! If if were, we would not see the problem with HMC733 as it would be canceled by improved sensitivity of the phase detector in the range close to 18 GHz. We tend to believe that any parasitic signal that is as strong as the observed sub-harmonic at 9 GHz will make it extremely difficult for any PLL to recover (whatever is its sensitivity characteristics).
What this get us to is a conclusion that the major problem is the operation of the VCO, which should not generate the sub-harmonic signal when tuned to 18 GHz. As you suggested, a way out is to eliminate the parasitic signal. We managed to solve the problem using a different approach fromt he ones that you suggested. I guess that it can be added to the list as an alternative. It worked for us, was extremely simple and cheap as it did not require any aditional components added in the signal path.
What we did was to stick a small piece of absorbing material on the top of the HMC733 chip. We used EMERSON & CUMING ECCOSORB BSR-1/SS6M (I guess that anything absorbing would do, a wet paper as well). It is a thin absorber with an adhesive layer on the back, which means that all that is needed to cut out a piece of it (slightly larger that the chip itself) and stick it on the VCO. Believe it or not, but the sheer presence of the absorber on the chip reduced the sub-harmonic signal nearly to 0 (at least by 40 dB), without affecting the fundamental signal at all! This solved the problem in our synthesizer I call "SythB". We haven't seen any problems for over two weeks now. I hope that it will help the others! Enjoy!
Thank you for providing this information as well as a solution that may help others.