Poor noise perfromance of VHF VFO

[wixor]

Hello

I'm attempting to home-build a receiver for the weather sats; I've had success with the analog NOAA transmissions, and would like to try the digital METEORs. To make things harder, I'm trying to do zero-if, so I need quadrature signals at ~137MHz, with tuning range of around 1MHz. The purpose of the project is to learn, so ready-made 50 cent single-chip solutions are out.

I've built the oscillator and quadrature generators using LVC U04 unbuffered inverters. It oscillates and tunes, but the noise performance is...



The circuit has two inverters making up a differential oscillator against an LC tank, feeding the phase-shift network, and the remaining four inverters buffering the quadrature signals and driving the mixers. The tank is made of a PCB coil with Q around 80, measured off-circuit by parallel-tuning it to operating frequency, series-matching to around 50ohms, measuring the impedance using NanoVNA and reversing the math. Tuning is done by two 1SV304 varactors, with tuning voltages between 1V and 3V (driven with some filtering from microcontroller's DAC).

The power supply is USB -> ferrite -> PTC -> 10uF (0805, good) -> 3.3V fixed stabilizer -> 10uF (0805, good) -> 2V stabilizer -> 1uF cap (0603 crap) -> ferrite -> 100nF filter cap -> chip.

In the department of things that work, we've got the quadrature network and the frequency meter D-flip-flop.



I've attempted to characterize the inverters a bit. The first problem is that they oscillate with only a resistor between the output and the input, consistent with a stray capacitance of around 0.5pF. I can easily see how that could be distributed throughout the layout, however I do not understand, how does it get the necessary phase shift; a single pole RC would only give it 90deg, so there should be plenty phase margin to avoid suprious modes like that.

This phenomenon does not go away until I reduce the biasing resistor down to 3.3k, which is how the oscillator is built. Thankfully, it also does not occur when driving the gate properly with a VNA. That way, I measured input impedance of around 45ohm + 7pF series, and a signal delay of 1.5ns. The datasheet quotes delay of 2ns typical at 2V, so that's good, except it's also almost 90* phase shift at 137MHz...



I also happily assumed an output impedance of 50ohm, and set up the coupling to the tank such that it's Q is reduced by half, and half the energy is dissipated in the tank, and half in the inverter input. Not sure this is correct though; when I looked up how the ICs are built, it seems they just have the LC tank slapped in the mosftet drains, no impedance matching or anything. They also don't get the helpful ESD protection resistors and diodes on their on-chip transistors...

Any pointers how to improve this are much appreciated. I understand the inverters may not be made for this kind of operation, but I feel they could be doing much better than they are now Alternatively, I could go for transistor-level design, but then I feel I'd need at least a matched pair good for RF, and they kind-of don't exist...

Thanks a lot!

[szoftveres]

I've experimented with PCB VHF inductors before and the Q was nowhere near to what I was hoping for - in my case I even made sure that there was absolutely nothing sitting inside the coil (not even on the other side other PCB). Not sure how you've got Q=80. What is the Q when the tank has the varicaps attached?
My understanding is that on VHF the skin effect dominates the coil resistance, so a decent inductor has to be made out of thick wire, plated with silver.
Might not be the issue in your case but I'd scrutinize the Q again (attached my photos of two versions of the same circuit).

Also, the 7404 has an UB (un-buffered) variant, which -as opposed to the regular 04- is just a pair of transistors and might be more suitable in this oscillator. Some datasheet mention that the regular 04 has a slight hysteresis (despite the 14 being the schmitt-trigger variant); hysteresis will give partial relaxation oscillator characteristics to this circuit (killing the Q).

[KE5FX]

Feed a known-clean signal at a similar frequency into the RealTek dongle and see how the noise pedestal compares.  You may just be measuring the noise of the RTL2832 chip, which (being intended for wideband signals) would not necessarily have great phase noise.

At the end of the day, lacking better measurement resources, the best test of your LO is to see how it works in the intended application.

[LM21]

I once searched "low phasenoise  oscillators" and got several  answers. Like U Rohde and a Colpitts oscillator or Vackar

[wixor]

I've experimented with PCB VHF inductors before and the Q was nowhere near to what I was hoping for

Hey,
I double-checked, and the measurements hold. In fact, I don't recall ever having such precise agreement between sim and reality (tbh, when I simulated the coil in Sonnet Lite, it promised me Q around 120, but I'll take it as it is). To be triple-sure, I also tried adding a 4.7k resistor across the tank, which should halve its Q, and it did exactly that. Also, previous iterations of this circuit were built with Coilcraft 2222SQ coils, which are spec-ed with Q in the hundreds, and they're literally a bit of closely-wound, thick enameled wire. None of that air-gap stuff, etc.



The PCB coil in your photo has soldermask on it; I had my coil tin-plated. I don't trust the soldermask, and the plating makes the trace thicker. Perhaps your trace was not as wide? Or could it be the silkscreen?

The circuit is already built with the U04 variant; my lazy schematic doesn't show that. You're right that the plain 04 is useless here, in particular if you self-bias it with a resistor, then the following dc-coupled stage is not biased in the linear range, and the output hits the ground.

Measuring the Q with the varactors is a bit more soldering, but I can easily check the quality by removing them from the whole circuit. I'm afraid it looks exactly the same, except the frequency is higher.

Feed a known-clean signal at a similar frequency into the RealTek dongle and see how the noise pedestal compares.

This is already done in the measurement I initially posted; the NanoVNA leakage is there specifically for this reason 

We can also do it the way our fathers', by listening to it I am however not capable by identifying the error by the sound it makes

Thanks!

[radiolistener]

RTLSDR clock oscillator noise is not very good. It is definitely better than many cheap Chinese oscillators, but not enough for a high class receiver. So, it looks you're see effect from RTLSDR self oscillator noise.

[rf-fil]

Not sure I understand your first plot - the noise performance. Phase noise - I assume that's what we're talking about here - is measured in dBc/Hz. So you'll need to find a way to express the measurement that way. Btw, measuring phase noise is not easy. As others said, you can't use RTLSDR to measure phase noise, because, to do that, your equipment needs to have much less phase noise than what you're trying to measure.

(Or, subtly, you can use the correlation method to cancel out some of the inherent phase noise of your setup. But that's quite complicated as well.)

Putting on systems engineer's hat - what phase noise are you aiming for? In terms of, let's say dBc/Hz at 10kHz offset.

Back to your oscillator: I would not expect a logic inverter IC to be great. You need a low noise device (at the non-linear VCO operating point), with low 1/f noise, and low parasitics. Also, you need the loaded Q of the oscillator to be as high as possible. And bunch of other things.. Many books have been written about designing low phase noise VCOs :-). One good book that I like is Oscillator Design and Computer Simulation by Randall W Rhea. It's old, but the fundamentals haven't changed.