Low-distortion Audio-range Oscillator

I think one aspect of this thread that is causing friction are the different unstated goals of various participants. Rather than trying to figure out those goals we should describe the target use of the device, REMEMBER ITS JUST A TOOL!. While there is intellectual satisfaction in going further past the edge of the art if you have a wrench adequate to tightening or loosening a nut a better wrench won't do a better job of the task.

Ultra low distortion oscillators are relevant to measuring distortion. Typically these devices are less well suited to general purpose oscillator functions, like finding the resonance of a driver and because of frequency limits, the overall response of a circuit. For me the utility today of one of these is evaluating ADC circuits and to a lesser degree analog amplifiers. I have not encountered (even in ultra high end) amps and preamps that are within 30 dB of these distortion goals. Not even close. So my Boonton with its -100 dB THD+Noise is adequate to those tasks. For measurements of really low distortion chains you don't need to have continuous tuning (I can't think of a mechanism in a linear system that would have the distortion change a lot with small frequency changes). This suggests to me that a few cardinal frequencies with a matched twin t filter would be really useful. Reducing the complexity makes the device much more accessible. Making the source floating would make it useful for balanced or unbalanced applications.

That's my use case for these devices. Who has others? more input would help narrow the actual requirements, which in turn can lead to something that really can be made.
I agree with you recommendation. I have found that using one or two cardinal frequencies with a twin-t make for consistent results that can be compared over time. Lately I've been using 10khz with Dick's twin-t set at the same frequency. If I had to do it over I would have built fixed frequency versions of Dick's twin-t, fewer parts and no touchy tuning when changing frequencies.
Ken
 
It is not correct that one can rely on every circuit having well-behaved frequency dependence of distortion. This applies in particular to filters (e.g. EQs, tone controls, RIAA stuff, crossovers), but also to a lesser extent to plain amplifiers and mixed signal designs (ADCs and DACs). More subtle distortion mechanisms such as power supply related issues and layout effects can take on surprising frequency shapes (e.g. a resonance in the power supply distribution may cause ripple to peak around 5 kHz, with little effect at 1 kHz and 20 kHz). The availability of high resolution distortion measurements with good frequency resolution (say, 6-12 steps per decade) is an incredible help to trace these issues.

Having said that I fully agree that a few discrete frequencies are already very insightful in most cases and surely incredibly more informative than no reasonable distortion measurement at all! And I second that a low distortion oscillator does not replace a function generator with its wide frequency range, essentially instantaneous settling and various output signal shapes.

BTW, I have designed a little PCB for a fixed frequency passive notch filter for very high resolution distortion measurements. The exact notch frequency is tuned to a fixed value during construction by iteratively adding selected resistor values. The easily implemented frequency range is about 10 Hz to 20 kHz (I've gone up to 100 kHz, but the notch depth degraded presumably because of strays) and set by chosing suitable capacitor values. As an additional feature it includes a selectable 10/20 dB input attenuator. This allows measurement verification at lower notch filter operating levels to gain further confidence in the measurement result.

This design is not as easily used as implementations which include notch filter bootstrapping (to avoid significant attenuation of the 2nd and 3rd harmonic; for my design you'll need to correct for frequency response) and center frequency trimming (for my design you'll need to adjust the oscillator frequency if it doesn't match the notch frequency). However, it avoids the subtle and difficult to quantify distortion contributions from the bootstrapping opamps and potentiometers and is thus more suitable when working below -120 dB.

I'm still evaluating this design with respect to its distortion contribution, but it looks like it is below -130 dB. As time permits I'll put together a short documentation and then make the Gerbers available at my website. If someone want's to go ahead without the detailed construction info drop me an e-mail for the Gerbers.

Samuel
 
Last edited:
Bob's oscillator is NOT easy to build unless you know where to get the rotary switches. I do NOT want to buy an old Heath oscillator just to gut it for Bob's oscillator. Bob's circuit is not easy to modify for pc mounted relays.

You are one of the highly skilled designers here that could easily offer a ready to go PCB and BOM that anyone could buy, but you don't. You refuse to take a leadership role here and force people like me to try to learn and understand all of the details of oscillators when I don't want to learn that.

Hi dirkwright,

The rotary switches are indeed a pain. Frequency selection is always a difficult issue, especially for high-performance oscillators. My oscillator is really quite easy to build apart from one's own decision about how to implement frequency selection. I probably would not do it these days with rotary switches - relays are not that bad, especially if one goes for some kind of PIC microcontroller to manage things. A compromise is to use a 2-gang, 11-position rotary switch for the frequency and a small group of relays for range selection.

Cheers,
Bob
 
I wanted to add two more items to the list for the cardinal fixed frequency design:
An attenuator at the output, not stepped, rather smoothly variable.
Frequuency fine tuning ability, so that you can lock it into you desired cardinal frequency to a fairly high accuracy - to match up with your fixed frequency notch filters, and to be centered on the steps of your fft analyzer. ( this probably goes without saying )
 
Member
Joined 2004
Paid Member
I have been researching locking a low distortion oscillator to an external clock. Its not too difficult to do but it can add additional distortion. I have done it successfully with the KH4400. My next goal is to make a divider/dds from a digital audio clock to lock to. It must be very distortion free or it will add distortion compromising the overall measurement.

My other To Do would be to make an IM equivalent. Digital audio in particular and many analog systems have more trouble with two tone HF than THD measurements and the IM measurement process is less sensitive to the waveforms. I got two of Victor's oscillators for this- one at 12 KHz and the other at 11.025 KHz. This is where the locking becomes important since these all drift and enough so to degrade a high res FFT. Now to build a suitable low pass filter without distortion.
 
This design is not as easily used as implementations which include notch filter bootstrapping (to avoid significant attenuation of the 2nd and 3rd harmonic; for my design you'll need to correct for frequency response) and center frequency trimming (for my design you'll need to adjust the oscillator frequency if it doesn't match the notch frequency). However, it avoids the subtle and difficult to quantify distortion contributions from the bootstrapping opamps and potentiometers and is thus more suitable when working below -120 dB.


Samuel

Hi Samuel,

I take it this would be a calculated correction for 2nd and 3rd harmonic. Is this correct?
 
I have been researching locking a low distortion oscillator to an external clock. Its not too difficult to do but it can add additional distortion. I have done it successfully with the KH4400. My next goal is to make a divider/dds from a digital audio clock to lock to. It must be very distortion free or it will add distortion compromising the overall measurement.
I'm wondering if it would be easier and/or better to use the oscillator at just below oscillation as a high-Q filter for a digitally-generated sine wave. Drift in frequency-determining filter components would only change phase, which could be detected and corrected for, similar to (and in addition to) the amplitude determination circuitry. Setting the filter's phase shift to zero would center the passband around the generated sine wave.
 
Member
Joined 2004
Paid Member
That's pretty close in a sense to what is happening. On the KH4400 it locks with about 5-10 mV. I think a filter won't get to as low a distortion as an injection locked oscillator. And you have the digital system coming along for the ride with its junk to deal with. I am transformer isolation on the KH4400 to reduce this issue.

This is only useful if you can access the clock of the digital system to lock to.
 
The B&K 1604 (?) notch filter has a correction that keeps the harmonics within 1 dB of the ideal value and its all passive. There is a 20 dB insertion loss in that mode.

I need to open and reverse engineer it since they did not publish schematics.

Hi Demian,

I can't get this model number to come with anything in google.
There is a B&K 1604A but this is a isolation transformer for power supplies.
 
Disabled Account
Joined 2012
The B&K 1607 is a fine unit and shouldnt be hard to copy/duplicate... uses 1% cap values.... paper.... new PP type will have higher Q and maybe thus greater notch depth. The only issue with old models that dont get regular use is the shafts of main control will be oxidized and hard to turn... needs cleaning and lubrication or the vernier control wont have enough torque to turn the main dial/shaft.
 
That's pretty close in a sense to what is happening. On the KH4400 it locks with about 5-10 mV. I think a filter won't get to as low a distortion as an injection locked oscillator. And you have the digital system coming along for the ride with its junk to deal with. I am transformer isolation on the KH4400 to reduce this issue.

This is only useful if you can access the clock of the digital system to lock to.
Demian,
Does this work by having the ACG skewed so as to just not start from noise/spuriae alone even when the multiplier/FET is full on, then inject a small sinewave at the tuning freq? If so, I might try this experimentally on Victor's Osc that I just ordered. Would be perfect to use with the AP's or a soundcard's D/A sine on a bin center for heavy time domain averaging without needing to trigger on the sent or incoming signal which causes drop of HF from jitter. Is the injection method effectively jitter-free (assumed no RFI-spikes and hum upsetting the injection)?
- Klaus
 
One form of level control I played a bit with is offsetting the frequency, of an injected sine stimulus, from the resonance frequency of an SV filter. With a a high Q filter the gain drops rapidly as the stimulus frequency is changed about Fr. It's a poor approach to ALC because
the shift in stimulus frequency produces strong side bands on either side of the fundamental.
In turn the side bands produce harmonics extending several octaves above the Fr. The intensity of the side bands is proportional to the delta f of the stimulus. FM to AM.

Any form of automatic frequency control will have this effect to some extent.
Jitter and phase noise will have the same effect. All this is in addition to distortion contributions of the stimulus.

I think an automatic frequency control would fight a bit with the ALC.
I think these same effects would be present in auto tuned notch filter as well and
may contribute to a limit of realizable notch depth.

What about arranging the oscillator as a phase locked loop?
The oscillator could be fine tuned by a reference frequency and the reference frequency doesn't have to sinusoidal.
 
Last edited:
Member
Joined 2004
Paid Member
First, injection locking lends itself better to a SVO since you have two low pass filters to exploit to reduce any contributions from the locking signal. There is a lot on serious analysis on injection locking and associated jitter/phase noise in the lit. Usually the close in phase noise is from the external source (XO in my case) and the noise further from the carrier comes from the locked oscillator. So a good low distortion oscillator locked by a good crystal based oscillator should be even better.

Second, even with a PLL you need a frequency control element that is voltage controlled. The Boonton does this with an external CPU controlled PLL and tuning via voltage on an analog multiplier in the cosine section I think. It works quite well actually with the output frequency very precisely locked.

Locking Victor's oscillator can be done with a small signal into the bottom of the AGC (for example). The larger the pull needed the larger a signal is needed with the distortion going up as well. Pulling under 1% should be pretty benign however.
 
"What about arranging the oscillator as a phase locked loop?
The oscillator could be fine tuned by a reference frequency and the reference frequency doesn't have to be sinusoidal. "

This could be done easily by modifying the inverting summing amplifier of an SVO by injecting a small amount of signal into the non inverting input. A dc controlled element can be connected to the inverting node and ground (a second multiplier). The dc off the output of a phase comparator controls the element. The draw back is any ripple on the control signal will produce FM and of course there will be elevated distortion from inserting all this.
 
Member
Joined 2004
Paid Member
B&K 1607 Internal view

Here are some pictures of what is inside the B&K. Tracing this won't be so easy especially with the really high grade construction.

The box is the housing for the tuning pot. The panels are plastic, probably to minimize external capacitance. All of the tuning caps are polystyrene, the input cap is paper (8 uF).

I'll do a little probing to see if I can figure out the harmonic compensation trick.
 

Attachments

  • IMAG0062.jpg
    IMAG0062.jpg
    411 KB · Views: 451
  • IMAG0061.jpg
    IMAG0061.jpg
    933.7 KB · Views: 438
  • IMAG0064.jpg
    IMAG0064.jpg
    504.5 KB · Views: 424