Part Number: LMX2572
We're evaluating synthesizers for a wideband receiver. It's a superheterodyne receiver with two local oscillators, and the first one needs to tune to whatever frequency in a continuous ~3GHz range or in another continuous, non overlapping 2GHz range. (for example, it might have to tune to any frequency from 0.5 to 2.5 GHz and also from 3 to 6 GHz).
We purchased LMX2594 and LMX2572 EVM's as both of these devices looked very good. We've been working with LMX2594 for a while and we've created an optimization routine with several PFD's and we've been able to reduce close-in spurs to acceptable levels (-80dBc typical) in every frequency we try by adjusting MASH order, PFD, and sometimes MASH seed. We also noted that sometimes changing the output power dramatically reduces some spurs.
The issue comes from far-off spurs, as they are quite nasty. We've measured -50dBc VCO-related spurs 1GHz away from the carrier and similar stuff (in LMX2594) and that's not good for our application. Could be solved with an LO filterbank but we first need to know what to expect in order to design such filter.
We recently got the LMX2572 EVM to check if the far-off spur performance is better, as the charge pump current is lower and Kvco is more linear. Here's a measurement taken from an LMX2572EVM set at 4301 MHz from a 125 MHz reference (the one that comes with the EVM) at 800 MHz span with no optimization. First picture is the carrier, second picture 1GHz away from it:
The register settings are these:
My question is wether this behaviour is normal/expected and what should we expect far off the carrier. Are our spur levels normal?
The EVM has been slightly modified to add an on-board ultra low noise, ultra high PSRR LDO. We basically scratch the soldermask on the back and carefully dead-bug a regulator on the backside and a PI input ferrite-capacitor filter, to have a point of load regulator and avoid unwanted extra noise.
I saw there are spurs at multiple of 125MHz, These spurs is probably due to state machine clock, which is also the OSCin frequency according to your configuration.
You can try set CAL_CLK_DIV to 1 to reduce the state machine clock frequency in a half.
Another problem I found in your configuration is the fractional denominator, DEN, it should be set to an irrational number in order to remove fractional spurs. When DEN=1000, I expect there are spurs at 1MHz offset (fraction or multiple of 1MHz). You can make DEN=12345678 to make these fractional spurs go away.
We are glad that we were able to resolve this issue, and will now proceed to close this thread.
If you have further questions related to this thread, you may click "Ask a related question" below. The newly created question will be automatically linked to this question.
In reply to Noel Fung:
These pictures were just a quick setup to illustrate the common behaviour we see with these PLL's, we've tried several strategies with both the LMX2594 and the LMX2572 and we get this kind of "sea of spurs" behaviour.
We were used to other manufacturer's PLLs fractional-N PLLs such as the ADF5355 series of Analog Devices which we used extensively in the past, which gives much more predictable spurs. With them, you get fractional and subfractional spurs and then the only far-off spurs that appear are multiples of the PFD, so you know what you're going to get.They are simple yet very very predictable
The LMX25XX series perform better in terms on phase noise and so far it seems like it gives lower spurs if you tame them properly, but our impression is that the spurs are too "unpredictable". You get the common suspects such as PFD, fractional, subfractional, reference cross talk, etc, but instead of getting lower and lower such that far from the carrier only the ones with stronger generation mechanisms remain above -100 dBm, you get a sea of low-level yet noticeable spurs like in the pictures, and sometimes there are extra far spurs that seem to be VCO-related (they appear at the same level depending on the VCO frequency, impervious to FPD changes and N variations) wich can get as high as -50dB
My question is wether this "sea of spurs" behaviour is to be expected, so that we can better at optimizing them away.
By the way, what do you mean by "irrational number" for the denominator? The denominator is always an integer so it can't be irrational by definition. I thought you meant prime, but 12345678 isn't prime. I've read stuff about numerator/denominator choosing strategies and some people say that making them co-prime helps as it forces the maximum amount of fractional randomization, yet they don't say they WILL reduce the spurs, but that it MIGHT reduce them under some circumstances. Could you please clarify on what would be the best strategy to choose the denominator?
Right now, as we want to move the frequency in 1Hz increments, in our own driver (not in TICS pro) we just set the denominator equal to the fPD in Hz and the numerator as the remainder from fRF/fPD, and then we either expand the fraction, reduce it or add 1 to the denominator to create an unequal large fraction.
EDIT: I've tested and setting CAL_CLK_DIV as you said created even more far-out spurs in between the ones that were already present, which I guess is to be expected. Apart from that, I've tested that a significant portion of the spurs are still present even if you turn off the output with OUT_PD and set the output mux to Hi-Z. The level goes lower but several are still above -80 dBm. Not sure where these could come from really, it must be some sort of crosstalk or coupling but I don't get where it could come from.
The sea of spurs behaviour was observed by us before when we were working with LMX2594 and we did things like putting the chip inside an aluminium box with EMI filters in all data lines and a common mode choke in the DC feed line and we still got similar behaviour. I just want to know if this is normal or expected.
In reply to David Marcos1:
Just a few comments:
1. I saw you mention that the LMX2572 has better far out spurs than the LMX2594. One difference is that the LMX2572 has integrated pull-up, while lmx2594 is not integrated. What this means is that the power supply to this pull-up for the lMX2594 has PSRR of 0 dB. So you might want to see if the power supply has any impact on spurs.
2. If you increase CAL_CLK_DIV, it lowers the state machine clock frequency. This frequency is Fsmclk = Fosc/2^CAL_CLK_DIV. So if you increase this, you get a lower frequency Fsmclk, and this could mean more spurs.
3. If you turn off the output, but still see the spurs, this suggests to me that perhaps it is power supply noise coming through the pull-up resistor as this is always connected to the power.
4. Spurs can come from anywhere, but in our lab, we typically do not see so many spurs that are of this level.
5. For any spur hunt, it is often useful to try to eliminate possible causes. If possible, if you can set the device to integer mode and choose an integer mode channel, then this would eliminate a whole host of potential causes. If the sprus persist, then it points to something else, like power supply. You might also want to compare spurs with minimum charge pump gain to maximum charge pump gain. If they are unchanged, this suggests that they are going around the loop filter. If they get lower at lower charge pump, it suggests that either they are going through the loop filter, or that they are due to noise on the charge pump supply itself.
In reply to Dean Banerjee:
The thing about LMX2572 being better than LMX2594 only happens close to the carrier, where fractional spurs are normally slightly lower. Far off from the carrier we get a very similar behaviour.
Some of the spurs we see are reference related. We inject the reference differentially from an LMX04828 and that creates reference spurs at integer multiples of the reference frequency. Others, we don't know. Most of the stuff seems to relate to known spur generation mechanisms and can be optimized away, as we currently do. But the sea of spur behaviour is where we're not sure about the cause or if it's actually a feature of MASH synthesizers.
See, there are so so many spurs at so so many places that we think it has to be something inherent to the chip and not externally induced.
We're using official EVAL boards which were modified to fit an ultra low noise, ultra high PSRR LDO for point of load regulation. We place the boards in custom-made aluminium boxes with passthrough SMA pigtails, two power banana inputs with an internal common mode filter and feed-thru capacitor and a passthrough signal connector with integrated EMI filters. We supply the thing with a dedicated linear PSU of a reputable brand (TTI). Then we screw the lid and the box is sealed with an EMI gasket.
The pictures I showed were under these conditions. We're now at a point where we've tamed the LMX synthesizers to a point where we've gotten it where we wanted them. We can optimize away most of the highest spurs tweaking the configuration and we've implemented a filterbank in our application to get rid of the far-off spurs below the carrier, which are the worst for us as we're amplifying the LO signal to saturation and injecting it to a mixer at 20dBm in a high-side injection configuration. But we'd still like to know if the behaviour we see could be normal or not under the circumstances described before fully commiting to the final design.
Could you please recreate our scenario (the pictures I posted before with an LMX2572 EVM with stock loop filter, using the settings I posted) and send a capture of the spectrum you get?
I tried your settings on an EVM and drove it with a 100 Mhz reference pro board and I didn't see the spurs you speak of.
Could you try with the reference pro board at 125 MHz like we did? Our settings are meant for 125 MHz instead of 100 MHz and the output frequency should be 4301 MHz instead of 3441 MHz (which is 4301/1.25).
OK, I changed the input frequency to 125 MHz on reference pro and I see 125 MHz phase detector spurs.
I think that it is due to the phase detector because if I keep the output frequency constant, but tinker with the phase detector frequency by making it 125 (x2) / 3 using the OSC_2X and PLL_R divider of 3, then it reduces the spur.
In your condition, looks like the phase detector spurs was about -75 dBc/Hz. For the settings, I note that you have OUTB_MUX=Channel divider. This should be set to OUTB_MUX="VCO". The way with your current settings turns on the channel divider and it draws current and makes a small spur. Changing it to VCO won't fix the phase detector spur, but it's a free 8 mA and reduces the Fvco/2 spur.
Here's my picture
All content and materials on this site are provided "as is". TI and its respective suppliers and providers of content make no representations about the suitability of these materials for any purpose and disclaim all warranties and conditions with regard to these materials, including but not limited to all implied warranties and conditions of merchantability, fitness for a particular purpose, title and non-infringement of any third party intellectual property right. No license, either express or implied, by estoppel or otherwise, is granted by TI. Use of the information on this site may require a license from a third party, or a license from TI.
TI is a global semiconductor design and manufacturing company. Innovate with 100,000+ analog ICs andembedded processors, along with software, tools and the industry’s largest sales/support staff.