Hi,
Currently using the LMX2572 synthesizer for clocking some TI ADCs. Is there another IC I could use that would get me to 4 outputs instead of two? Ideally it would keep the same jitter specs as the LMX2572 has (or perhaps be a little better). If not, what TI buffer chip (low jitter) would be a good choice to generate more clock outputs?
thanks
Dennis,
Depends on the frequency you need. It's worth noting that the LMX2572 is a fractional synthesizer with no frequency holes, while LMK04832 is an integer-only PLL with two VCO cores (2440 to 2580 MHz and 2945 to 3255 MHz) and quite a few frequency holes. In many cases LMX2572 wins by default, simply because it can hit the required frequency.
But in cases where LMK04832 could be used, it usually comes out ahead. Comparing an LMX2572 at 3.2GHz to an LMK04832 at 3.2GHz (using the latest PLLatinum Sim, version 1.6.0.0 as of this post):
This comparison favors LMK04832 since the phase detector frequency can be higher and the in-band PLL noise can be reduced, but the floor is inarguably better since LMK04832 can use CML bypass mode direct from the VCO whereas LMX2572 is using the lower-performing divider path. If we pick a different example at 2.5GHz:
Still favorable for LMK04832, because the noise floor is better and the N-divider can be lower.
Now, if I pick both devices to do 491.52MHz, with a 122.88MHz input source:
The divider floor on the LMX2572 just isn't very good relative to the LMK04832, and the higher supported integer mode phase detector clearly improves performance a lot. LMK04832 still performs better in-band due to the lower N-divider.
Finally, picking 156.25MHz output, with 125MHz input source:
Now even with the floor advantage mostly gone, LMX2572 still has significantly worse PLL noise because of the higher N-divider. Maybe over a wider integration bandwidth the floor contribution could add up.
If the LMK04832 can hit the frequency, and has an optimal reference input to get a decent phase detector frequency, it will perform better than LMX2572. Again, I must stress that LMX2572 can hit a bunch of frequencies that LMK04832 simply can't, either because the maximum frequency on LMK04832 is too low, or because the LMX2572 can synthesize arbitrary fractions; there are cases where LMX2572 makes much more sense. Otherwise, LMK04832 tends to be the better option, simply because the VCO frequency and the N-divider are so much lower.
---
Last comparison: the same exercise as above, but with LMX2582 instead:
LMX2582 just barely outperforms LMK04832 at similar operating conditions (LMX2582 scores 86fs jitter), with 4x higher charge pump gain and the doubler disabled to help match 1/f noise and meet phase detector requirements (125MHz PFD). The effect of the higher frequency VCO, and the penalty for the higher N-divider, is still apparent. LMX2582 is the best performing VCO in our portfolio, aside from a few devices with BAW resonators; the difference is almost purely PLL noise.
Even with better performance and lower total solution cost (LMX2582 + LMK00304 < LMK04832), there is still the additive jitter from the LMK00304 floor which would likely make the overall LMX2582 solution jitter higher. It is difficult to beat LMK04832 in a frequency range where it is designed to operate. On the other hand, HMC987/LTC6955 has low enough floor that final performance is probably a wash; but now solution cost is probably higher than just using LMK04832.
Regards,
Derek Payne
