[regenfreak]

Originally Posted by G0HZU_JMR

There are commercial phase noise analysers available that use similar techniques to measure phase noise.

If it helps, I had a go at making a low cost, low noise crystal oscillator last year. I used a class A amplifier in the loop and used classic theory to predict the close in phase noise. See the simulated phase noise and the actual phase noise as measured by a phase noise analyser at work. The spikes/spurs on the trace are from nearby equipment in the lab at work. Even with the lab interference, the result looks to be very close to the theoretical result. I tried to design an oscillator that could get close to the limits of the phase noise analyser. This meant driving the crystal quite hard.

The other phase noise plot is of a 102.4MHz VCXO. This VCXO is in a PLL and is locked to a 10MHz OCXO. This is a very low noise VCXO and it was quite expensive. A low-cost 102.4MHz oscillator would probably be 10-15dB noisier.

You can buy a basic 100MHz crystal oscillator from Mouser or Farnell for about £3 although I'd expect the phase noise to be quite a bit higher than the VCXO. It would still be good enough for IMD testing though.

Cheers. I ordered a cheapish 100MHz TCXO from china and it is on its way. Hope I dont need a PLL for that.

The TOI techniques by the two Hams AA7U and N4CY in Siglent application notes (post #307) is the way to go for me. What they did was clever.

Instead of measuring the fundamental and third order distortion products simultaneously, they split the test into two parts. This addresses the tug-of-war between opposing constraints; phase noise due to tone spacing and noise floor. The band-stop filter eliminates the fundamentals before the low-level modulation products going into the spectrum analyzer, making the mixer level of the spectrum analyzer less of an issue. No hooligan can enter the stadium at the gate. We can measure the distortion products with very small RBW and then add the insertion losses of the band stop filter. They used classic bandpass filters for the output ports of a single signal generator and it seems to be effective. Since the Chebyshev has rather narrow bandwidth and steep skirts, the reflected energy of the harmonics back to the signal generator are not escaping. Even you can argue there could be cross-talk leakage two channels of the spectrum analyzer. Their results show it is not the case.

It is easy to draw comparison between hobbyist grade Siglent Spectrum analyzers to HP pro classic workhorse. But they are apple and orange, and their price differences are over an order of magnitude. Secondly, I dont do this for a living in a test lab, no boss tells me off if my measurement is below industrial standard, so we can do whatever we got with limited budget and equipment. Otherwise if it is so much hassle to do a simple TOI measurement, I would rather go to Youtube watching some cat videos...

BTW their resistive-transformer type of power combiner is impressive, close to 78db isolation at 3.5MHz. I have measured my chinese power combiner (attached), it is about 47db port isolation and -3db insertion loss at 10MHz. I am sure it is a hybrid type too. I want to replicate their combiner but can't find any T106-15 toroids for sale in the UK.

I am going to make a 7th order Chebyshev band stop filter with cut-off frequencies at 9MHz and 11 MHz in order to measure the fundamental tones at 10MHz (with a cheapo OCXO) and at 11MHz with a Rigol DG1022Z.

One interesting point they mentioned in the Siglent TOP application notes is that the Siglent spectrum analyzer has a fraction subharmonic spur at 5MHz and they use odd fundamental tones (last two decimal points) in their measurement. I guess the subharmonic spur coming from the frequency multiplier of the spectrum analyzer.