LMX2595EVM: LMX2595 frequency drift

Callum Deakin said:

I'm having some unexpected issues with frequency drift when operating the LMX2595 EVM. I am using a high quality Wenzel oscillator at 100 MHz as my reference input (single ended) and attempting to get a 12.5 GHz output directly from the VCO.

What is the output frequency you are getting? The phase noise plot looks like the frequency is locked? This would mean that the frequency drift is related to the oscillator drift. Can you double check to make sure whether the device indicates that it is locked? Either through muxout pin or register readback?

Are you talking about drift,  as in the output frequency is not centered at 12.5 GHz? Or are you saying that the phase noise at a certain offset is higher than expected?

Callum Deakin said:

1) There are large fluctuations at <100Hz due to the previously mentioned frequency drifts. The phase noise analyser also warns me of the frequency drifts.

Can you send me a screen capture of your PNA?

Callum Deakin said:

2) The phase noise power is around 10 dB higher in the reference section (<1kHz) than the expected (i.e. reference oscillator multiplied by 125^2).

The phase noise of the LMX2595 in the < 1 kHz section should closely match the LMX2595. Are you taking this measurement on the EVM? Have you disconnected the onboard oscillator?

Thanks,

Vibhu

Thanks for your response Vibhu.

The frequency drift seems to have solved itself over the past couple of days. It is curious because I did not change anything in the setup (that I am aware of). But yes when I said frequency drift I meant that the phase noise was unexpectedly high and varying quite dramatically between measurements in the <100Hz region, sorry for the imprecise language. I also think that the higher phase noise in the reference section is perhaps due to the noise floor of the spectrum analyzer OCXO.

On another note, do you know what could be the cause of the excessive number of spurs in the 1kHz - 30kHz region? All measurements were taken using the EVM, which I have configured for singled ended reference input exactly as in the diagram below from the manual

Also here is the screen shot of the phase noise measurement

Hello Callum,

These spurs are related to the SPI clock. Can you try unplugging your USB connector and plugging it back in?

At least when it is unplugged you should see the spurs disappear.

Thanks,

Vibhu

Thanks Vibhu.

Indeed unplugging the USB does remove most of these spurs, see below. Can you elaborate a bit more why the SPI clock causes these spurs, especially considering the SPI clock rate is 10s of MHz yet the spurs appear in the kHz region? Is this just a design fault/limitation of the EVM? Or is it something inherent to the chip itself?

Also, what is the cause of the remaining spurs in the plot in the kHz region below?

Hi Callum,

Vibhu is OoO for a while, so I can answer your questions moving forward.

First, the USB2ANY is more than just a SPI bus - consider that the spurs are still present even when the SPI clock is not actively running. Inside the USB2ANY is a microcontroller, and a connection to a PC with a (most likely switching) power supply generating 5V. So there are quite a few signals which could be coupling through the USB2ANY pins.

The spurs created by digital controllers or switching power supply feedthrough can show up in many strange and unexpected places within the loop bandwidth if they become mixed into the charge pump output, the VCO tuning pin, or the VCO bypassing pins. The EVM design couples many of the digital signals alongside the power routing for the device, and the return current path for the low-frequency SPI signals travels underneath the VCO bypass capacitors and tuning pin; unfortunately this couples digital noise in while the USB2ANY is connected. Fortunately, there is a straightforward solution to minimize this contribution: unplug the USB2ANY, as you've seen.

Second, the remaining spurs. I see what I suspect are power line harmonics in the 100Hz to 1kHz range, so perhaps the power supply for the LMX2615EVM or for the reference requires some additional filtering? We usually put a substantial L-C choke filter on the power supply outputs when taking precise measurements.

As an additional thought, I see that you have 30dB/decade rolloff up to 100Hz, which surpasses the expected 1/f noise for the LMX2615 at 13GHz. Can you confirm that the reference oscillator does not have spurs in similar locations?

Similarly, the measured noise is about 10dB higher in-band than the predicted PLL noise. Would you happen to have the fractional divider enabled (i.e. mash order higher than "integer")? Fractional spurs can be introduced into the output, even when the fractional numerator is zero.

Regards,