The goal was simple: design the most simple oscillator for 1 GHz band.
I decided to try the distributed approach, as it seemed fine for the frequency, and I never actually made a distributed oscillator before (compared to many lumped ones, which I designed).
What I mean by saying “distributed approach” is that I wanted that “the engine” of the oscillator be based on distributed elements (a bare microstrip transmission line in particular), and phase difference needed would be obtained by appropriate geometry of transmission line instead of using e.g. LC components.
The other interesting factor is that once you get into microwave things, they scale pretty well, and you could as well take similar approach for 10 or 20 GHz oscillator. When it comes to LC components, this limit is around few hundred MHz to around 6 GHz (depends on geometry and technology of particular component), and I will write a short note about it in future.
There are many ways which one can follow. They typically stick to one of these themes: either provide negative resistance around whatever plays resonator role in your circuit, or provide a positive feedback from the output of the active device to the input. The second theme seemed easier for me.
There’s one universal rule of oscillation, which you’ll find in every course or book about electronic circuits. The total phase shift of signal flow around the active device and the feedback loop is an integer multiple of 360°, and the total amplitude is just 1. The latter, called amplitude condition, will usually hold true due to the nonlinear operation of the active device (though may be true as well due to other reasons).
Once you are decided for positive feedback topology, you could wonder how one couples the active device’s output to its input. There’s whole theory here, but I just decided for a bare transmission line, DC-blocked in the active device’s output.
The “1/3 rule of oscillator design” may be helpful, which you can find in a number of RF design books. The rule will help to determine the right impedances which the active device should see to the input and to the output. That obviously comes in pair with a selection process for the active device. Obviously you’ll need the active device’s input and output impedance at the frequency of interest (1 GHz here), or its S parameters. The S parameters are (should) be published in the device’s datasheet, or can be found as .s2p files in the Net. For transforming S parameters into impedances (Z parameters) you can use one of many available applications (especially Agilent’s AppCAD: http://hp.woodshot.com/) or just do the math yourself. The S parameters depend on transistor’s bias conditions, so you’ll have to decide for some quiescent collector current (Ic) and collector-emitter voltage (Vce), and design right biasing network.
I chose BFR93A NPN transistor, which is around for almost 2 decades, moreover, it is available and cheap. One of the first problems I encountered is that there are many manufacturers of this transistor. Normally, we are so happy with that, as that let’s us have things cheaper. In this case however I noticed, that S parameters vary quite a lot (10..20%) among transistors of this type manufactured by various manufacturers. Moreover, I wasn’t able to identify the manufacturer of the transistors I bought, so I realized that my calculations can’t be really perfect.
To compute the physical length of the feedback line, one needs the electrical length of the feedback line (expressed in degrees or ratio of wavelength), the desired frequency and the signal propagation speed, which can be computed from electrical permittivity of the substrate (around 4.4 for FR-4 PCB). In my case, physical length turned out to be around 35 mm.
Seems there are two last questions to answer: how to arrange the biasing network and how to arrange the output? The biasing network should provide DC power to the output of the transistor (proper DC part of collector current and collector-emitter voltage); the other biasing network should bias the input accordingly (appropriate base-emitter voltage in this case). Obviously, biasing networks should not pass AC power through, and that’s why RF chokes are used. That’s not so easy to find a good choke for 1 GHz band (MiniCircuits offers few types), though you may try to adapt some of the SMD inductances which can have their first resonance frequency as high as around 5-6 GHz. As the goal was “the most simple oscillator”, I simply designed quarter-wave chokes for that. That way both of my chokes for biasing networks are transmission lines of around 36mm in length, capacitively shorted on their ends to the ground plane.
The output of my oscillator is 50 ohm SMA connector. I used quarter-wave transformer to match it to the output of the transistor.
The results obtained when I first powered-up the circuit were satisfactory – output frequency of around 870 MHz and output power level of around 10 mW. I started looking for reasons which made the frequency distant from what I wanted. Obviously, the aforementioned fact of having inaccurate S parameters could have been a reason. I decided to provide a simple fix of “inserting” a bit of inductive-type impedance into the output circuit of the transistor. This is tricky however, compared to an easy bit of inserting capacitance. Obviously the problem is the DC power, which will pass through the inductance, on contrary to capacitance. Therefore, I again used “magical” properties of transmission lines and made an inductive stub by having a small, tunable capacitor at the end of about quarter-wave length transmission line. That results in a tunable inductance for the chosen frequency band. Few turns of the tunable capacitor away there’s been the deeply awaited frequency of around 1030 MHz. Voila.
Here you have PCB layout and photos of the oscillator.