Why is the offset of my PLL so large?

[mssong]

Why is the offset of my PLL so large?

The waveform of the TCXO and the divided waveform
UP and DN signals
UP current and DN current

I have attached the waveforms.

I have also attached the charge pump, vco type is LC. varacator cap is 200fF to 800fF.

 

[crutschow]

Why is the offset of my PLL so large?

Offset of what?

 

[dick_freebird]

Looks like phase detector says you are offset, and nothing is being
done about that.

Could be tuning range, could be that you flipped the logic and are
locking up, not locking-to. Could be that your loop filter is missing
and you just don't get proper tuning voltage for the VCO. And
generally you would have a divider stage between HF output and
the feedback clock, to get a frequency and phase match against
a low frequency high quality reference (these are not available
in GHz range, and granularity wants a low clock freq as the min
frequency increment I think?).

Maybe open the loop and see what VCO and @phaseDet freqs
you get from VTUNE pinned to either rail, and whether the UP/DN
result would then be trying to correct or worsen, as a start at debug.

 

[mssong]

Offset of what?

Sorry, Phase Offset

 

[mssong]

Looks like phase detector says you are offset, and nothing is being
done about that.

Could be tuning range, could be that you flipped the logic and are
locking up, not locking-to. Could be that your loop filter is missing
and you just don't get proper tuning voltage for the VCO. And
generally you would have a divider stage between HF output and
the feedback clock, to get a frequency and phase match against
a low frequency high quality reference (these are not available
in GHz range, and granularity wants a low clock freq as the min
frequency increment I think?).

Maybe open the loop and see what VCO and @phaseDet freqs
you get from VTUNE pinned to either rail, and whether the UP/DN
result would then be trying to correct or worsen, as a start at debug.

I added a loop filter and locked the Vctrl to the desired frequency, but the waveform shows a large phase difference between the divided waveform and the TCXO's waveform (about 30ns).

The target frequency is 2.4ghz and we divided it by 240 to get 10MHz, so the period is 100ns.

I thought that the phase error of 30ns from 100ns is big, so I posted a question.

What could be the problem?

When I disconnected the up and down nodes from the charge pump and re-checked the current, I found that the mismatch was very small (about 0.05uA out of 50uA).

I also put an Inverter Chain on the reset node of the PFD to give it a delay (about 3ns) to avoid the dead zone issue.

 

[FvM]

A charge pump should have integral behaviour, strongly asymmetrical up/down pulse widths means integrator in saturation.

 

[dick_freebird]

If the DC analysis gives you a railed-loop initial condition
and the VCO has large regions where more or less VTUNE
makes no difference, you might end up stuck or needing
way more tran time than the loop time constant.

Might check for equal-delay-to-phase-det through the
REFCLK and FBCLK paths. Might have to compensate
the divider-chain delay. Although 30ns seems a long
timescale for anything GHZ-y, maybe the counter is a
dumb ripple with enough stages that it adds up.