A small-sized low phase noise OCXO for lab measurements, microwave IF processing and MXO applications

by SV3ORA

Preface

A reference oscillator is a very important piece of equipment for any RF laboratory. It enables relative measurements and comparisons of the characteristics of other test oscillators, mixers and the rest of the RF building blocks that you may build. In microwave bands, it can be used on MXO applications, IF processing or in combination with low noise mixers, to create high performance microwave signal generation. The importance of having such a piece of equipment at the workbench, led me to the design of a universal miniature reference oscillator.

There are quite a few ways to implement such an oscillator, but most of them, make use of a quartz resonator driven at low current. My experiments are based on discussions I had with other people, amateurs and professionals (especially experts from Wenzel Associates Inc.) THIS TO BE REMOVED, I HAVE GIVEN THANKS AT THE BOTTOM OF THE ARTICLE, and the articles I have read throughout the years. From these articles and discussions, I distinguished some bright ideas, which I am trying to implement in this oscillator.

Key features of the mini reference oscillator:


Project files

Schematic (PDF)
Schematic (ExpressPCB)
Bottom copper layer (PDF)
Bottom copper layer (ExpressPCB)
G101 enclosure (PDF)

G101 enclosure (DWG)
G101 enclosure photo (PDF)

Double clicking on each component on the expressPCB copper layer file, will notify you which value is it. For readability reasons, I did not include the components values on the silkscreen layer.

 

The circuit

The schematic is shown in the next picture. Starting from the left to the right, first there is a simple voltage follower. Despite being simple, this type of follower is of quite low noise and it effectively removes hum from a noisy PSU. I did not use a voltage regulator like a 7811. There are three main reasons for this. Firstly, I wanted the oscillator to operate from different PSU sources and voltages (achieving potentially different output characteristics), as it is indented to be used on the lab bench. Secondly, the use of a typical three-terminal regulator like the 7811, would require at least 14v as an input voltage. Third and most importantly, are noise reasons. Ordinary three-terminal regulators have several hundred nanovolts per root-hertz of white noise and some reference devices exceed one microvolt per root-hertz. The main 12v power source for the circuit can be anything from a 7812 regulator to a lab power supply or ideally a 12v battery. The total current consumption of the oscillator when the crystal oven is on, is about 130mA at 12V going down to about 30mA when the oven is off. It periodicaly goes up and down between these two values as the oven is switched on and off.


Following the voltage follower there is a finesse voltage regulator, which is suitable for cleaning up the supply to the oscillator. A 15 ohm resistor is inserted in series with the regulator's output giving a small voltage drop. The single transistor amplifier has an emitter resistor which combines with the emitter diode's resistance to give a value near 15 ohms. The regulator's noise voltage appears across this resistor so the noise current is shunted to ground through the transistor's collector. The noise reduction can be over 20dB without trimming the resistor values (8.2 ohm emitter resistor) and the intrinsic noise of the 2N4401 is only about 1 nanovolt per root-hertz. Trimming the emitter resistor can achieve noise reduction greater than 40 dB.

Following the PSU, is the oscillator circuit. It is a very low noise crystal oscillator, which drives the crystal at low current and uses it as a filter directly at the output. That's fine as long as the crystal can stand the drive level and the noise will be very low. The circuit uses a FET follower to boost the current so the crystal isn't driven so hard. The crystal current is driving about 500 ohms of reactance instead of a 50 ohm load so the resulting voltage is 10 times higher for a given crystal current. The crystal, filters the noise and harmonics produced by the oscillator, very well. It will contribute some noise close to the carrier but it will act as a wonderful noise filter at the floor. Real measurements at the output of the buffered oscillator, connected through a 60dB attenuator to a spectrum analyzer, show no higher harmonics, even when switching at the logarithmic scale. This could indicate that the filtering of the crystal is very effective.

Following the oscillator, there is a FET follower to boost the current, through a 100 ohm resistor which is used for isolation. The voltage at the source of the FET is enough to bias the single emitter-follower after it, through an isolation resistor. The last stage should drop the signal voltage by a factor of three, but it's capable of driving a low impedance load. Care has been taken so that both amplifier stages do not compromise much further the low phase noise characteristics of the oscillator stage. Transformer coupled output, drives a 50 ohm load and further isolates the oscillator from the load.

Note that there is no series decoupling (resistor or RFC) between the drain of Q6 and the positive supply rail. Everything is done on the lower side of the amplifier transistors (emitter/source), as a try for keeping the amplifier noise down. One could argue that this could lead to a problem because the bypass capacitor C11 is then shared with Q2 and Q3, as well as Q6 - in effect coupling them all together. Actually for Q2 and Q3 this is not a problem, as there are series resistors at the collectors. For Q5 and Q6, this is not a problem as well, as they have different isolated source/emitter circuits. There is not too much difference in taking the signal from the collector/drain or the emitter/source, apart from the phase difference. The amplifier transistors share a common decoupling capacitor but they are too close together on the PCB, so there is no need for an extra decoupling capacitor.
 

The construction of the transformer T1 is simple. Wind 9 turns of 0.1mm-0.2mm enameled copper wire, on a FB-43-101 core. Then wind another 3 turns. Separate the primary and secondary ends as much as possible. As you wind the turns, take care not to remove the enameled paint from the wire by the sharp edges of the ferrite core.

 


The crystal oven

High performance crystal oscillators employ temperature control circuitry to hold the crystal at a precise, constant temperature. A carefully designed OCXO yields temperature stabilities as good as 0.0001 PPM over a temperature range that would cause the crystal to change by 10 PPM. Since size is a constraint in this oscillator, an
LM10 operational amplifier and voltage reference was used as the oven controller. Semiconductor devices are also used for both heating and temperature sensing. This leads to a simple circuit, but with better characteristics than the much simpler thermistor/resistor circuits. The main reason LM395 was used as a heating element, was it's small size compared to other heating elements and the easiness of mounting onto the crystal, resulting in an overall small sized structure. The semiconductor device was though to be more reliable as well. With this oven used, the stability of the oscillator when pre-heated, is better than 1Hz.

The tricky mechanical part of the oven, is the mounting of the crystal to the heater. These pictures show the mounting of the crystal below the LM395, which is placed upside down on the PCB, in order to leave space for the crystal to slip below it.


The whole construction is firmly held on the PCB using a piece of wire, which holds the crystal tightly below the LM395. This wire is used as a clamp, and it is directly soldered on the crystal and the PCB at both ends, to properly ground the crystal body.

The LM35DZ is directly held onto the remaining body of the crystal, using a small amount of thermal-conductive glue.

 
 

Overall construction

The oscillator is composed of two PCBs, connected together in a modular way, so that they can split apart if needed. The bottom PCB is the crystal oven, whereas the top PCB is the oscillator circuit. If one decides that he does not need a OCXO, then the oven PCB can be left out, without affecting the rest of the oscillator circuit. 

The oven PCB has just three connections to the oscillator PCB, one for the positive voltage and two for the ground. The two PCBs are sandwiched together, with these three connections interconnecting the top and the bottom PCBs.

The next pictures show the production of the PCBs in different stages.

The whole circuit is enclosed inside a Gainta G101 die cast box, the smallest of the G-series. It's dimensions and look, are shown below.

Space inside the enclosure is very limited. The bottom PCB pins are just touching the metal box below. To prevent short circuits, a thick insulating membrane was added between the bottom pcb and the enclosure.

When the two PCBs are sandwiched together and just before enclosing them in the box, a small amount of polyurethane foam is inserted between them, such as the crystal and the heating circuits are completely surrounded by the foam, creating a thermal insulation.

I used SMA connectors for both the RF output and the PSU line because of their small size, but any RF connector can be used at these low frequencies. I have chosen to use a coaxial connector for the PSU line, to shield it from external noise. Bare wires used for the PSU line, can pick up external noise and transfer it inside the oscillator circuits. A shielded coaxial cable to connect the OCXO to the PSU is the best.

 

Evaluating the noise performance

For testing the oscillator noise performance, an E5052A signal source analyzer has been used. The results are quite satisfying, although there is a window for further improvements to the circuit. The results of the measurements are shown in the next pictures.

The noise floor is about -170dBc/Hz, which is really interesting. The curve gets "white" somewhere at 10 kHz frequency offset. A cause of the spikes on the plot, could be co-resonances of the crystal or problems related to the bias levels transistors or the output transformer. Adjusting the noise suppression circuit brought no actual improvement. Nevertheless, these spikes were not visible with a different testing equipment and I have been told it's virtually impossible for that circuit to have so much flicker. The instrument showed also that the time jitter decreases noticeably, if one increases the lower integration limit from 10 Hz to 100 Hz, 1 kHz etc.

 

Tuning the crystal oven

Most oven crystals you will encounter are AT-cuts and they have a bowl-shaped temperature curve above room temperature. The idea is to find the bottom of that bowl where the slope is nearly zero so that slight changes in temperature don't change the frequency much.

Remember that AT-cuts dive in frequency when heated, going below the actual curve. Give the crystal time to recover from that thermal shock before determining the frequency. Using Spectrum Lab will give you a plot of the frequency so you can see the exponential recovery from the thermal shock. The opposite is true for suddenly cooling the crystals.

A simple way to see what is the best point for the oven temperature by using a particular crystal, could be to tune an SSB digital receiver to the crystal frequency, then connect its audio output to a spectrum analyzer program like spectrum lab in it's waterfall display. As you tune the oven temperature you will be able to see at the waterfall, the audio tone frequency changing. By trial and error you should be able to find the best point where the curve of the crystal temperature gives the best stability.

Finding the best point for the oven temperature is ideal, but if stability of about 1Hz is enough, you do not really have to get too tired finding this point. In my case I achieved a stability better than 1Hz with no consideration of that point at all and by just adjusting the oven at a temperature of a few degrees greater than the environment temperature. I did not use a crystal designed specifically for an oven, I used an ordinary 10.7MHz crystal, so I thought it would be a good idea not to overheat this crystal too much, for various reasons explained later in this article. 

There are many ways to tune a OCXO but I tuned my OCXO by connecting a 12v power to the appropriate connector and placing the oscillator near a digital shortwave radio using a small SMA rubber antenna.


Switch on the radio and tune it to 10.7MHz with the bfo switched on. Pick up the AF tone signal from the receiver and connect it to a computer running Spectrum Lab or any other waterfall program. You will then notice the AF tone signal in the waterfall. 

Tune the oven temperature in the point where the LED just switches on. Then the oven is set a few degrees above the environment temperature. You may try different temperatures, but remember, the result must be a stable output frequency. Also note, after each setting you have to give the oven much time to stabilize. In my case this time was as long as 30 minutes.

 

Zoom in to that signal and you will notice frequency oscillations like shown above. These frequency oscillations occur because the control loop oscillates, when the temperature sensor is placed far away from the heater. Having the sensor on the opposite side of the crystal for example, can introduce a time delay which makes the feedback loop unstable. In this oscillator, the temperature sensor has been placed very close to the heater, but these oscillations are not avoided and they would occur for some time until the control loop stabilizes.

After about 30 minutes these oscillations almost disappear, as the oven stabilizes.

After about 35 minutes, there are no oscillations. The OCXO is stable to the Hz then.

As you tune the temperature of the oven (and thus the temperature of the crystal), the output frequency of the OCXO changes. When finishing tuning the oven, connect the output of the oscillator to an accurate frequency counter. Then tune the frequency tuning variable capacitor to bring the oscillator to the correct frequency. Each time your OCXO is fully preheated, this frequency must be shown on the frequency counter.


A note of warning about ovenizing ordinary crystals


An oven will not degrade the noise of a crystal designed for an oven. In fact, most crystals will exhibit much better close-in noise when ovenized because tiny temperature changes can push the frequency of a crystal around quite a bit. We are talking about "noise" that is very close to the carrier, maybe below a couple of Hz. Also, an oven allows the crystal to be a very specialized, highly polished type that might very well have bad activity dips below the oven temperature. In other words, crystals designed to operate over a range of temperature have design restrictions that will impact the Q and noise in a negative way compared to a crystal designed to operate at only one temperature.

You can use an "ordinary" crystal in an oven if you don't heat it very hot. The main problem with heating an AT-cut well past its intended temperature range, is that the frequency/temperature slope can get very steep, making the oscillator sensitive to small changes in temperature. But an oven at 40 or 50C should be fine for most room temperature cuts. The symptom of going too hot will be a steep rise in frequency for a small increase in oven temperature.

But is the phase noise compromised when heating on 40-50C? A crystal running off turning point is much more susceptible to thermal variations. So at very low frequency offsets the phase noise can be degraded. If the oscillator wanders around a few millihertz due to thermal noise, that's close-in phase noise. This would only be noticeable at offsets below a Hz or two.

Since there are different crystals with different characteristics, there is not an obvious general rule of thumb to follow. A good starting point could be that, if noise performance is more important than frequency stability and if you find that the crystal you have used, has bad measured characteristics when heated, you may want to replace the crystal with another one or a more suitable type or at the worst case completely switch off the oven, by setting the threshold temperature to the lowest value, so as the oven is never turned on and the crystal operates at room temperature. If room temperature does not change at extreme values (which is often the case) and if a "room temperature" crystal is used, then there may be no need for switching on the oven at all. Even in that later case, the crystal still has to be surrounded by thermal insulation, as a protection against sudden (but low duration) thermal shocks.


Can it be made better?

The oscillator oven is quite good as it is, but it could be made even better. A good starting point would be to heat up all the components of the oscillator and not just the quartz resonator. The thermal insulation should then cover the whole enclosure and not just the oven. Even better, a Dewar Flask could be used instead of foam or armaflex, to give a much greater thermal isolation. The heat-up time might then be much greater, but all components would be eventually temperature stabilized. The oscillator enclosure should then be made much larger and more air-proof, allowing air inside to be gradually heated up and transfer the heating to the various components.

Despite this approach was not followed, mainly for size constraints, the rest of the oscillator frequency critical components used, are of very good quality and low ppm values, so that they are not much affected by temperature variations. All the resistors in the oscillator portion are about +/-30ppm and the capacitors in this portion are NP0 (+/-30ppm). Temperature compensation of the oscillator components (apart from the crystal) was not thought to be necessary, although you may want to try some negative coefficient capacitor combinations and evaluate the results. Also, extreme thermal or radiation bursts were not applied to this oscillator, so if you intend to use it in tough environments, you may wish to do your own experiments, if you have the necessary testing equipment.

Other possible improvements could be made on the power supply circuit or including a discrete voltage regulator if one needs to keep the oscillator voltage stable throughout different input voltages. Accurate low phase noise measurements are quite difficult to perform so I do not know if altering components or their values could improve the phase noise performance even more. Nevertheless, I think the overall performance of the oscillator is very satisfying for such a low cost circuit, possibly far better than many commercial oscillators.

 

Thanksgiving

The design and implementation of this oscillator, took about two years of continuous discussions with engineers from Wenzel Associates, trying to figure out of a low noise approach. Special thanks to Charles Wenzel, who was always willing to help me on this. Also, I have to thank DUBUS magazine and especially DL8HCZ for the communication and DK5LV for the evaluation of the oscillator.

 

References

Time and frequency related files - Wenzel Associates
DJ2LR very low noise xtal osc

 

Back to main site