LMX2594: IBS of lmx2594

Hi,

For the 30 kHz integer boundary spur, this can be caused as a product between the VCO frequency (Fvco) and phase detector frequency (Fpd).  If this is the dominant cause, then adjusting the phase detector frequency will fix the spur.

But there is also a mixing product of Fosc and Fvco.  I don't have the exact specifics of your design, but PLLatinum Sim suggests that the mixing product between Fosc and Fvco dominates over the mixing product of Fpd and Fvco.  See picture below.

For this spur, the main mitigation is to ensure a high slew rate OSCin signal.  If you change the Fosc signal internal to the chip, you can avoid the spur, but this is extra components.   The programmable input multiplier sort of does this, but does not eliminate crosstalk from the input (Fosc) to the VCO.

Also, for our new LMX2820, we are seeing significant improvements in spurs on the order of 15 dB, although thwe are finializing some of this characterization.

I also encourage you to try our PLLatinum Sim tool   ti.com/tool/PLLATINUMSIM-SW

This models spurs and their contributing factors.  Below has your frequencies entered.

Regards,

Dean

Fvco%Fosc

1.  Fvco%F

Hi,

So lets get some numbers

For the first case:

Fosc = 368.64 MHz

Fpd = 184.32 MHz

Fvco = 10137.63 MHz

Fvco % Fpd = 30 kHz

Fvco % Fosc = 184.35 MHz 

But lets take your second case:

Fosc = 368.64 MHz

Fpd = 221.84 MHz

Fvco = 10137.63 MHz

Fvco % Fpd = 184.35 MHz

Fvco % Fosc = 184.35 MHz

So what we see that for the Fvco % Fpd, we are at 184.35 MHz.  It seems far from 30 kHz.  But not really.  For spurs, the worst case is when you are 30 kHz away from an integer boundary, which we are.  But the second worse case is when you are 30 kHz away from Fpd/2.   So this is 30 kHz away from 184.32 MHz, so this has a 30 kHz spur.  For this second spur, maybe the spur is 6-10 dB lower than it would be from the integer boundary, but still high.  Theoretically, for the MASH engine, it puts out lots of energy at this 30 kHz when close to (but not at) Fpd/2 multiples.

For diagnostic purposes, you might want to try

PLL_R_PRE=9

MULT=6

PLL_R=1

Fpd=245.76

Fvco%Fpd = 61.47 MHz

Regards,
Dean

Hi,

2.  I see what you are saying.  184.35 MHz is not related to 221.84 MHz.   184.35 MHz is suspiciously close to 184.32 MHz. But where could 184.32 MHz come from?   For instance, with 368.64 MHz Fosc, CALCLK_DIV is probably DIV2.  But this gives 184.32 MHz for a mixing product.  You might try to first lock the part with CAL_CLK_DIV=1, but then change it to 0 (DIV 1) to see if it has any impact on the spur.  Also, if there is a factor of 2 in the division between Fosc and input to the programmable multiplier, it could come from there.  I do notice that my PLL_R_PRE=9, which has no factor of 2.

3.  OK, so the phase detector frequency was shifted and the spur was mitigated.  This suggests that this 30 kHz spur is some sort of mixing product involving the phase detector frequency.  You might need more than one phase detector frequency, but I'm not sure.

Regards,

Dean

Z,

If CAL_CLK_DIV is div4, then we have the frequency of Fosc/4 running on the chip.

Fosc/4 = 368.84/4 = 92.21 MHz

Now the div 4 is implemented is with two cascaded div2 stages, so there is also Fosc/2 on the chip as well.

GCD(10137.63,368.84) = 20 kHz

GCD(10137.63,184.42) = 20 kHz

GCD(10137.63,92.21) = 10 kHz

GCD spur can happen when there are two sources.  I'm not sure if the 30 kHz is a multiplie of the 10 kHz case, but it is very hard to debug spurs without trial and error.  So if you set CAL_CLK_DIV=div1, then if this is the issue, it should make the spur at 30 kHz go away.  If not, it is likely something else.

Regards,

Dean

Square wave input is desirable.  The first thing that the LMX2594 does is square up the input, so it is better if it is initially squared up to begin with.

The best results for spurs I have seen are with differential LVDS input.

Regards,

Dean