Hello:
I'm writing a three article series on modern low noise synthesizer design for publication in a major industry journal. I was planning on featuring the LMX2595 as an outstanding example of a low noise synthesizer IC with integrated VCO, along with other options such as synthesizers using the finest external VCO's.
Thus I have been closely examining its noise performance, and have noted what seems to be a consistent datasheet error in the reported normalized noise floor which represents the phase noise limit imposed by the dividers and charge pump. The flat part of the normalized floor above the 1/f corner is given as Pn1hz = -236dB, a very low number. It is the best I am aware of for any part.
Using this normalized floor Pn1hz the flat floor inside the loop bandwidth is supposed to be given by Pnflat = Pn1hz + 20log(Fvco/Fpd) + 10log(Fpd) = Pn1hz +20log(Fout) - 10log(Fpd).
But, when I compare this calculation to the noise shown in the phase noise graphs of Figs 3 to 11, I note the flat noise inside the loop bandwidth from about 20kHz to 200kHz to consistently be about 8 dB too high. It calculates as if the normalized figure of merit should be -228 instead of -236. That would still be an excellent number, just not the best on the market by a wide margin as now indicated by the datasheet.
This could be caused by the crystal reference noise floor, but this possibility is not consistent with the noise behavior shown in the phase noise graphs. It would take a crystal floor of about -151 dBc/Hz from 20kHz to 200kHz in a 100MHz reference to generate this higher in the loop BW floor, which is not typical of a higher grade reference unless its output has been padded down. However, the 1/f noise of the phase noise plots is rising at only 10dB/decade as offset frequency descends below 10kHz down to the graph limit of 100Hz. It is not being noticeably degraded by the crystal reference. That would be indicating a superb 100MHz reference, for example one showing about -135dBc/Hz at 100Hz to degrade phase noise at 100Hz by only 1dB. This is possible, but it is state of the art performance for a 100MHz quartz crystal reference, costing in the range of $400 and up. Such high quality references usually have noise floors in the range of -165 to -185 dBc/Hz over the pertinent frequency range here of 20kHz to 200kHz, not the -150 range that would give rise to the noise floors in the phase noise graphs.
I note that 1/f phase noise calculations based on the 1/f normalized figure of merit of -129 are exactly what the graphs show, so the 1/f model is very accurate.
These questions arise:
1. What crystal reference oscillator was used to take the data shown in the phase noise graphs in the LMX2595 datasheet?
2. Was the output of the crystal reference padded down so that its noise floor was worsened?
3. What normalized noise floor is proven for the part? Is the number -236 actually measured, or is it simulated, calculated, or extrapolated? Can a typical production spread in this number be provided?
4. Is there any plan to bring the outstanding normalized noise floor of the LMX2595 to a part that uses an external VCO? If there was and it is truly around -236, Texas Instruments could have the finest external VCO synthesizer product on the market to supplement what appears to be the best internal VCO product.
Thanks,
Farron Dacus