LMX2594: Spur difference between LMX2592 and LMX2594

Hi Mario,

My coworker will get back to you tomorrow.

Regards,
Hao

Hi Mario,

Do you have a specific test condition to compare the spurs performance between these devices? As you know, the VCO frequencies of these devices are different, their spurs performance will not be identical. For example, if the output frequency is 5GHz, you could have it with the direct VCO output from LMX2592, but you have to make the VCO to 10GHz in LMX2594 and have it divided down to 5GHz. 

Hello Noel,

As I said, there are many frequencies and not just some specific ones. In other words, sweeping it in a range of few GHz, a lot of much worse than LMX2592 spurs will pop up. You say "...will not be identical" but note that I am not talking about 1 or 2 dB  but more than 10 dB. And of course I do know that the frequency of the two VCOs are not the same (otherwise we wouldn't have changed), but the only thing which counts is the final output cleanness (whether divided or not). The comparison has been made between 5 to 7 GHz which was taken directly from LMX2592 while the LMX2594, whose VCO runs 10 to 14 GHz, is divided by 2 (and physics suggests if ever, thanks to division, purity should improve rather than worsening).

Regards,

Mario

Hello Mario,

You are correct if I pick a device say the LMX2594 and compare the spurs at a particular VCO frequency with no divider vs the same VCO frequency with a divider the spurs should decrease with increasing divider values. However, I cannot say the same applies to a comparison between two different devices.

It's also important to note that PLLATINUMSIM-SW has a default loop filter that populates for each device. For the LMX2594 the default BW starts at 300-400 kHz while the LMX2592 starts at 20-30 kHz BW. The BW will change with the frequency of the VCO/output frequency.

I checked simulations for both these devices with a 6000.1 MHz output frequency (to create a lot of spurs). I also changed the BW of the LMX2594 to match closer the the default of the LMX2592 at this frequency (~21 kHz). The simulations show that the spurs from LMX2594 are actually -20 to -30 dBc lower than the LMX2592 when using the same BW.

Thanks,

Vibhu

Hell Vibhu,

good to know. As for  the first two paragraphs you wrote my comment is just... "of course". The loop has been changed accordingly (if you read the begin of this thread you can read "the loop has been changed as well as all the other parameters in order to fit the differences").

As your result are completely different from ours and as you say that you have PLLATINUMSIM-SW simulations, could you please send me the two files (projects .sim) so that I can see all the parameters you have used and replicate them? You know, you simply said that you set 6000.1 MHz and it is impossible  to understand how you made all the rest.

Thank you,

Mario

Hello Mario,

Please see attached. If there is a specific frequency or a specific spur you are concerned about please let me know. My comments above were more of an observation based on one frequency.

Thanks,

Vibhu

https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/48/5226.2592_2D00_6000.1output.sim

https://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/48/7673.2594_2D00_6000.1output.sim

Hello Vibhu,

I used your files but first I would like to be sure I am seeing exactly what you did as I am a bit perplex (if this is what you simulated). Therefore, before considering anything can you please confirm my below simulations are as per yours?

Regards,

Mario

Hi Vibhu,

I do not know whether the two images have been uploaded... If not let me know to do that.

Mario

Hello Mario,

Yes, these are the same simulations I sent over e2e. Again, before you read too much into it, this is just an example to create spurs. The idea was to show the difference between the two devices. If you select a specific frequency and requirements you can optimize phase noise on the tool.

Thanks,

Vibhu

Hello Vibhu,

Then I doubt the simulations you sent are meaningful for the following reasons.

1. The comparison you chose refer to a spur performance which is not usable at all: can you fancy having a synthesizer which has -17dBc or, as you said “a better between 20 ro 30 dB“-43 dBc? Perhaps for a toy, definitely not a professional clock. Minimum acceptable should be -80 dBc or better. In other words, you made an improvement but already much much worse than what can be obtained.
2. The Fpd used in the two simulations are different (100 MHz vs 200 MHz) as well as Fden is different: both of them do make different in spur. Moreover, because of the division by 2 the resolution is twice as much which, once more, cannot be comparable.
3. The loop filters do not have the same phase noise (and has an influence on spurs as well).

I will give you an example of what I meant here below using the LMX2592: can you make the same performance by LMX2594 (note that we are talking of making that frequency not in integral mode (Fnum <> 0) but fractional in order to have a frequency step (resolution) no more than 1 MHz)?

Regards,

Mario

Hello Vibhu,

Then I doubt the simulations you sent are meaningful for the following reasons.

1. The comparison you chose refer to a spur performance which is not usable at all: can you fancy having a synthesizer which has -17dBc or, as you said “a better between 20 ro 30 dB“-43 dBc? Perhaps for a toy, definitely not a professional clock. Minimum acceptable should be -85 dBc or better. In other words, you made an improvement but already much much worse than what can be obtained.
2. The Fpd used in the two simulations are different (100 MHz vs 200 MHz) as well as Fden is different: both of them do make different in spur. Moreover, because of the division by 2 the resolution is twice as much which, once more, cannot be comparable.
3. The loop filters do not have the same phase noise (and has an influence on spurs as well). Although it is not as important as for the previous points, in order to make a comparison, the loop filters should be tailored to have similar phase noise.

I will give you an example of what I meant here below using the LMX2592: can you make at least the same performance by LMX2594(note that we are talking of making that frequency not in integral mode (Fnum <> 0) but fractional in order to have a frequency step (resolution) no more than 1 MHz)?

Regards,

Mario

Hello Vibhu,

the thing is that we should talk about reasonable figures and definitely a -51 dBc spur is not acceptable for a synthesizer as at least -85 dBc should be the target (of course not in integral mode). What I saw is that if I set both devices to have similar phase noise and same frequency resolution, on 2594 spur are higher (talking about spurs < 85 dBc).

Anyway, if you look at the spur @ fpd (Phase detector), you will see that on 2594, in spite of the divider, that spur is worse than the one in the 2592.

Last, as I saw that there was a new release of PLLatinum Sim I downloaded it and observed some difference in result from the previous one and that worries me a bit. Moreover you can see from the attached picture that there must be a mistake as 78.125 multiplied by 79 does NOT make 6093.75 as shown.

I attach also a sim file where the 2592 can move the frequency with a resolution of about 1 MHz having acceptable spurs for different settings (also attached the picts in case simulations will get different results): maybe you can show the 2594 can perform as well or, as you said, even better.

Thank you,

Mario

Sorry Vibhu I do not why but but the *.sim file is not accepted as an error is reported...

Mario

Hello Mario,

I wanted to reproduce the multiplication issue, where you were seeing the Fpd * N - divider != Fvco but couldn't do it. The way I enter frequencies is by typing in what I want my Fosc to be followed by what I want my Fout to be and the multiplications comes out correct. Can you reproduce the same thing multiple times. As you can see here my N-divider is 78.

I agree with your comments above, so what are the parameters that you are willing to change, the LMX2594 VCO frequency will be higher and we will use a divider to get down to the same frequency. Are you looking for a comparison (Fnum != 0 fractional mode of course), with LBW and Fpd the same, or am I free to use anything in my disposal?

Thanks,

Vibhu

Hello Vibhu,

thank you for your reply. As there are several points, let us start.

1. The calculation of N_Divider is correct if you enter the frequency directly into the Fout Box, as you say indeed. However, if you want to reproduce the anomaly, on the same design try to write 80 in Fnum, then write 0 again and you will see that N_Divider is, wrongly, 79 instead of 78 as it was, correctly, at the beginning. This is an anomaly which is clearly visible. However, I said I was concerned because if a mistake like this happens elsewhere, where there is not any evidence of it, such as spurs for instance, there would not way to know it and thus mislead.
2. Your question about the freedom of changing parameters can be simply answered : every parameters can be changed to allow reaching the best result. Of course hardware cannot be changed therefore for LBW  there is very little margin. In other words any of registers can be changed for each new frequency if necessary.
3. Yes, I am familiar with MASH seed and I tried many combinations to see its effect. By the way, as the PLLatinum SW calculates somehow the effects of this, is there the possibility to know how this figures is used? You know search for the best figure, for each frequency, would be an unmanageable task.
4. The simulation you posted, the one you said is comparable to the one i posted using 2592, shows worse performances. If you  compare the figures you will see that they are at least 10 dB higher on 2594. And here we are back to the original object of this thread: it looks that LMX2594 has worse, something between 10 to 15 dB, spur performance than LMX2592 and your simulation seems to confirm this.
5. Last point: is there any possibility to have the formulae by which PLLatinum calculates spurs? If I had it I could write a SW which, recursively, could optimize all parameters which influence the result. Unfortunately, PLLatinum does not offer such an utility (moreover, the button "Find Optimal Seeds" looks as is not working correctly as manually better seeds can be found).

Regards,

Mario