LMX2615-SP: Is there an alternative method of changing frequencies that eliminates dead time between frequencies?

Hello Danilo,

I recommend that you point the customer to our application note: Streamline RF Synthesizer VCO Calibration and Optimize PLL Lock Time. This application note discusses the techniques to minimize the time between frequency changes. The fastest "alternative method" is likely manually programming the VCO frequency/amplitude calibration coefficients, i.e. partial assist or full assist calibration.

Unfortunately, LMX2615-SP greatly limits the SPI bus maximum transaction speed, down to 2MHz (vs. 75MHz for LMX2594). Additionally, the LMX2615-SP does not include the ramping feature seen in LMX2594. Consequently, it is very challenging to quickly generate broadband ramps and pulses with the LMX2615-SP. In fact, with the limited SPI bus, full assist calibration could actually take longer than partial assist calibration, due to the register write time adding a dozen microseconds (in the best case) to the dead time.

Regards,

Derek Payne

Hi Danilo,

Even using the "full assist" calibration, we still run into trouble: there's no double-buffering on the frequency registers for LMX2615-SP, so the changes cannot all be loaded simultaneously. Each frequency update could take up to seven register writes:

• Program VCO_DACISET
• Program VCO_CAPCTRL
• Program VCO_SEL
• Program PLL_N[18:16] (not always needed)
• Program PLL_N[15:0] (not always needed)
• Program PLL_NUM[31:16] (not always needed)
• Program PLL_NUM[15:0] (not always needed)

The PLL will most likely lose lock immediately after programming VCO_CAPCTRL, since this will pull the VCO frequency out of range. Until all of VCO_CAPCTRL, VCO_SEL, and PLL_N/PLL_NUM are programmed to agreement, the PLL would be unlocked. And with 2MHz maximum SPI frequency, the unlock time could be anywhere from 25µs to 125µs, minimum, before considering analog lock time.

We are aware of this limitation on LMX2615-SP, and future space-grade devices will be designed to ensure the minimum possible discontinuity between frequency hops - either through double-buffering, or something like LMX2820 instant-cal. For now, this is the best that can be done using LMX2615-SP full assist - hence why I suggested it may be faster to trigger a new calibration with partial assist.

Partial assist calibration uses the state machine clock running at some divide of OSCin frequency to establish the VCO_DACISET/VCO_CAPCTRL/VCO_SEL values, and the starting points of each of those coefficients can be programmed for partial assist calibration without changing the VCO frequency. Now the calibration would only require writing the new N-divider value and writing R0 to trigger calibration with partial assist. For small hops this would be 25µs + calibration time based on how close the starting values are and however much analog lock time is required to make the final ~200mV adjustment.

Regards,

Derek Payne