Charge pump mismatch in Fractional-N frequency synthesizer

[hmsheng]

charge pump mismatch

What's the effect of charge pump mismatch in a sigma-delta fractional-N frequency synthesizer?

Can you help me get this paper?

An analytical phase noise model of charge pump mismatch in sigma-delta frequency synthesizer
Xiaojian Mao, Huazhong Yang and Hui Wang

 

[ieropsaltic]

fractional n frequency synthesis

Mismatch at charge pump causes the up and down currents to be unequal. This causes the slope of the I (or Q) Vs the phase error to be unequal for positive and negative phase errors. This causes non-linearity at the origin point. This linearity can cause shaped noise of the delta-sigma modulator to get mixed and folds back to low frequency i.e high frequency shaped noise falls to low frequency causing significant in-band phase noise. Note that delta-sigma noise doesn't get filtered before the loop filter and that the charge pump is the first block seen by this noise inside the loop so its linearity is very important. One way to avoid this non-liniearity is to intentionly apply a constant current at the node of charge pump output. This is equivalent to biasing the I (or Q) Vs phase noise curve away from the origin.

 

[hmsheng]

charge pump nonlinearity sigma delta noise

Im very interested with what you said. That's exactly what I want to know. But it's hard to understand the non-linearity of charge pump will fold the out-band quantization noise to in-band. Could you please give a more detailed explanition or give some reference? Thanks a lot!

By the way, as you said a constant bias current at charge pump output can reduce the non-linearity of charge pump mismatch. But it's known that in the integer-N situation this constant bias current will cause large reference spur! Will this also cause reference spur for fractional-N frequency synthesizer?

 

[rfsystem]

fractional n spur

Yes! Offset biasing create refernce spurs. It is the nature of the chargepunp to minimize the simultanious on time of positive and negative current. that is counteracted by using offset current. I habe seen it in some papers that a switchable current source ist used to etstimate or measure the effect; reference vs. inband noise. It is not for production!

The high frequency to low frequency folding ist practical simulated with a AHL model instead of a circuit model.

You can imagine that the chargepump nonlinearity creates difference tones out of two high frequency tones. In this case second and third order nonlinearity is not of much difference.

The modelling curve is not a typical two gain linear segment curve with a little sharp edge at zero. Because of the minimum ontime of both current pulse there is more smooth transition curve. That should be considered in modelling otherwise the foldback effect is to big.