Help Needed: Oscillator Survey for High Quality Oscillators

[Gokul43201]

I'm doing a literature survey on high quality oscillators, and since this is peripheral to my area of specialization, I'd appreciate some help.

I'm looking for a low frequency oscillator (~10Hz and up) with a frequency stability of about 30ppm/C, amplitude stability of about 3ppm/C and low harmonic distortion (all higher harmonics at least 90dB below fundamental).

Is anyone aware of anything in the market or in literature that has the above specs? If you know of something that comes close, please tell me where to look it up. And if you are not aware of anything that's been built with these specs, I'd like to hear about that too (and any related wisdom) - negative responses are valuable as well.

 

[berkeman]

To get that low of a frequency and good accuracy, I believe you will need to divide down a higher frequency oscillator. The best way that comes to mind would be to start with a 32kHz watch oscillator, and use an HC4060 or equivalent to divide it down. One of the 4060 series contains an unbuffered inverter to use for the oscillator section -- I'll look it up in the morning when I get to work and post info on it. You should be able to make what you want with just a jellybean 32kHz watch crystal and a 4060-like chip. It will generate discrete frequencies, though. Do you need variable frequencies?

 

[Gokul43201]

I'm looking for an analog circuit where I can change frequencies trivially, by swapping a small number of resistors or capacitors.

I've had to build such an oscillator for low-temperature thermometry, and the project turned out to be non-trivial, in terms of time and effort. I want to know if I've just been reinventing the wheel.

 

[berkeman]

Instead of swapping resistors and capacitors, it would be better to swap divide ratios on a PLL's input and feedback taps. I'll post more tomorrow morning. It would help if you could post more in terms of the specs you want to achieve for output freqency steps, granularity, stability over time and temperature, etc.

 

[NoTime]

If you don't mind spending $$$ then something like this should do it.

http://www.bkprecision.com/www/np_pdf.asp?m=4001

Or if you want to roll your own then a good DAC and Up with a little programming.

 

[berkeman]

Instead of swapping resistors and capacitors, it would be better to swap divide ratios on a PLL's input and feedback taps. I'll post more tomorrow morning. It would help if you could post more in terms of the specs you want to achieve for output freqency steps, granularity, stability over time and temperature, etc.

Here is more info on the frequency synthesizer technique that I was alluding to:

http://en.wikipedia.org/wiki/Frequency_synthesizer

 

[Gokul43201]

If you don't mind spending $$$ then something like this should do it.

http://www.bkprecision.com/www/np_pdf.asp?m=4001

Or if you want to roll your own then a good DAC and Up with a little programming.

TTLs (and other digital creatures) are completely off-limits. The switching noise radiated by a TTL circuit can kill our measurement capability. We measure currents to an accuracy of the order of femtoAmps.

 

[Gokul43201]

Berke, can you point me to a purely analog, low-frequency PLL synth that has specs comparable to those in the OP and a long term amplitude drift that's less that 5ppm? I really don't know anything about these beasts so there's a steepish learning curve I'm climbing.

The wiki article mentions frequencies in the several kHz and up range. Can you reliably dial this down all the way to ~10Hz with losing quality and stability?

 

[berkeman]

Berke, can you point me to a purely analog, low-frequency PLL synth that has specs comparable to those in the OP and a long term amplitude drift that's less that 5ppm? I really don't know anything about these beasts so there's a steepish learning curve I'm climbing.

The wiki article mentions frequencies in the several kHz and up range. Can you reliably dial this down all the way to ~10Hz with losing quality and stability?

How high do you need the frequency to go (sorry if I missed it in your posts)? For low-frequency sine wave oscillators of variable frequency, I agree with NoTime that you should play digital data through a DAC. For 90dB harmonic distortion, you would need about a 15-16 bit DAC, plus some lowpass filtering of the output. I think you can meet your amplitude stability specs, as long as your power supply has that stability.

You could base the design on a microcontroller, or you could just base it on a 16 bit wide PROM and use a CPLD to do the addressing and frequency adjustments. You can still use a frequency synthesizer to generate the sine wave data addressing clock, in order to get your overall frequency adjustments.

 

[NoTime]

TTLs (and other digital creatures) are completely off-limits. The switching noise radiated by a TTL circuit can kill our measurement capability. We measure currents to an accuracy of the order of femtoAmps.

Your specs are a bit of a challenge for any purely analog solution.
Good shielding techniques and some low pass filtering should take care of radiated noise.
Plus the generator does not need to be local.
For example you could use fiber optic cable driving a local instrument op-amp for your setup.

Chances are that your measurement device is at least partialy digital

 

[Gokul43201]

Berke, I'd like to be able to have frequencies from ~10Hz to about 1kHz.

NoTime, our measurement system is entirely analog.

The oscillator I've built cost me about 50 bucks in components (plus about $30 for laying out the board). I'm thinking about writing up a paper for either the Review of Scientific Instruments or for the Journal of Measurement Science and Technology (or a similar journal).

 

[TheAnalogKid83]

what does ppm mean??

 

[NoTime]

what does ppm mean??

Parts per million. It's a measure of stability.

 

[NoTime]

Berke, I'd like to be able to have frequencies from ~10Hz to about 1kHz.

NoTime, our measurement system is entirely analog.

The oscillator I've built cost me about 50 bucks in components (plus about $30 for laying out the board). I'm thinking about writing up a paper for either the Review of Scientific Instruments or for the Journal of Measurement Science and Technology (or a similar journal).

Interesting. It's been a while since I've seen anything with a moving pointer on it.

Your oscillator goes from 10Hz to 1kHz variable?
I might think that you would need a variable inductor to do that.
Plus coils in general might induce unwanted currents elsewhere.

 

[Gokul43201]

Interesting. It's been a while since I've seen anything with a moving pointer on it.

Your oscillator goes from 10Hz to 1kHz variable?

Not continuously. I have to switch between different sets of resistors and capacitors. My board goes inside a box with rotary selector knobs in the front panel.

I might think that you would need a variable inductor to do that.

I doubt you will find a variable inductor (or a variable capacitor) with a 30ppm/C stability in inductance (capacitance).

 

[NoTime]

I doubt you will find a variable inductor (or a variable capacitor) with a 30ppm/C stability in inductance (capacitance).

No argument here.
So what did you use?
Some Wein bridge variant?

 

[TheAnalogKid83]

Parts per million. It's a measure of stability.

I figured that from chemistry but how does parts per million relate to oscillators? I've seen it on crystal specs, but I'm not sure what "parts" per million is representing.

 

[NoTime]

I figured that from chemistry but how does parts per million relate to oscillators? I've seen it on crystal specs, but I'm not sure what "parts" per million is representing.

How much the frequency changes.
While Gokul specified degrees C, in electronics this also applies with time from component aging effects.

 

[Gokul43201]

Some Wein bridge variant?

Yup.

 

[Gokul43201]

I figured that from chemistry but how does parts per million relate to oscillators? I've seen it on crystal specs, but I'm not sure what "parts" per million is representing.

If the oscillator is making a signal with an amplitude of 1V at room temperature and the amplitude increases by 10 microvolts upon heating by 10C, then the mean thermal drift in the amplitude is 1ppm/C.

 

[berkeman]

Your specs are a bit of a challenge for any purely analog solution.
Good shielding techniques and some low pass filtering should take care of radiated noise.
Plus the generator does not need to be local.
For example you could use fiber optic cable driving a local instrument op-amp for your setup.

Chances are that your measurement device is at least partialy digital

Gokul, I know that you've had some bad experiences with digital noise in your instruments in the past, but given your frequency requirements and accuracy requirements, I still think that you could use the DAC based solution that NoTime and I proposed. You will need to do some things carefully in the power supply isolation and shielding and decoupling and star grounding of the setup, but if you want a continuously variable output frequency for the setup, then a digital solution with good filtering and isolation is the way to go. If you only need a few discrete output frequencies, then your passive component substitution route may be the better way to go.

 

[TheAnalogKid83]

If the oscillator is making a signal with an amplitude of 1V at room temperature and the amplitude increases by 10 microvolts upon heating by 10C, then the mean thermal drift in the amplitude is 1ppm/C.

So it has to do with amplitude, or is this just an example because its easier to talk in microvolts than microhertz? In a crystal datasheet, is it referring to the mean thermal drift in frequency?

 

[TheAnalogKid83]

Gokul, I know that you've had some bad experiences with digital noise in your instruments in the past, but given your frequency requirements and accuracy requirements, I still think that you could use the DAC based solution that NoTime and I proposed. You will need to do some things carefully in the power supply isolation and shielding and decoupling and star grounding of the setup, but if you want a continuously variable output frequency for the setup, then a digital solution with good filtering and isolation is the way to go. If you only need a few discrete output frequencies, then your passive component substitution route may be the better way to go.

what's star grounding?

 

[Gokul43201]

So it has to do with amplitude, or is this just an example because its easier to talk in microvolts than microhertz? In a crystal datasheet, is it referring to the mean thermal drift in frequency?

It is not restricted to amplitude. It is just a way of describing relative quantities (in the same way as a percentage).

Berke: I'm still devoting space in the back of my head for the suggestion you and NoTime made. But yes, I do not need continuously variable frequency.

 

[berkeman]

what's star grounding?

Star grounding (and star power distribution) refers to the very important technique of laying out PC boards (and other electronic assemblies) so that different portions of the circuitry do not cause interference with one another. It is required when you have a mixed-signal system, with digital electronics and sensitive analog electronics on the same PCB, for example, and also when you have the potential for digital noise to cause electromagnetic interference (EMI) radiation problems from your product.

For example, consider a wireless sensor device, which contains a microcontroller (uC) and other digital circuitry, and also contains a radio transceiver. If digital switching noise gets into the radio portion of the PCB, it will significantly limit the performance of the radio. So to avoid this, the digital circuitry is placed on one end of the PCB, and the radio is placed on the other end, and the power supply circuitry is placed between them. This forms a star ground and power distribution arrangement, with the power supply at the center of the star. In this arrangement, the power and ground distribution paths for the digital and analog circuitry do not "share any impedance", which helps to prevent conducted crosstalk noise from coupling from the digital circuitry to the analog circuitry.

For another example, consider a wired sensor, which contains a uC and other digital circuitry, and uses a twisted pair transceiver to communicate with a building control network. There are similar reasons to use a star grounding system for this device (as in the radio example, digital noise coupling into the network transceiver circuitry will lower the network performance a bit, but not as much as with a radio), but in addition, it is extremely important to keep the digital noise out of the network transceiver's power supply and grounding, in order to minimize the radiated EMI that is generated by digital noise on the network twisted pair wiring. If the network transceiver circuit and the digital uC circuit share any impedance in their power and grounding, digital RF switching noise can end up causing RF currents to flow out the network wires, and this can cause serious problems with passing FCC radiated emission levels in the 100MHz-500MHz frequency range.

Star grounding is discussed a bit more in this PCB layout advice paper that I wrote several years ago. It is centered on the wired sensor case, with a twisted pair network connection, but also applies to the more general case of mixed-signal PCB designs:

"pc_board.pdf" in "pc_board.zip" at http://www.echelon.com/support/documentation/docs/

 

[Maitch]

I wonder how you got on with your search for a high quality low frequency oscillator.

Looking at the requirements, I expect an analogue circuit (not Wein Bridge, but something not much more complicated), followed by good low-pass filter stages, could be suitable if you don't need to change the frequency much AND if the system is kept in a temperature-controlled box ("oven") to reduce temperature effects. The complicated part is the amplitude stabilisation circuit (that will affect the output voltage stability)... I can imagine a method for this, but it will have very slow settling times.