6-Decade Wien Oscillator Stabilized by Sample and Hold

[ccktek]

This post contains elements of analog circuit design, reverse engineering, boat anchor revival, and probably my eligibility for honorary membership in Test Equipment Anonymous.

The boat anchor is an SIE Model N, serial number X-2 R-C Oscillator covering 3Hz to 3MHz in six one-decade bands, i.e. from 3 octaves below audio to near the 80-meter ham band.  Part of the case and all of the circuitry were missing except for the 4-gang tuning capacitor and range selector switch.  Switched parallel and series R/C combinations of the same values identified the oscillator as Wien.  I’d initially planned to obtain a schematic and restore the original 1950s vacuum tube circuitry but no schematic was available.   As such I designed a Wien oscillator around the remaining frequency-determining parts. The schematics omit relay coil selection circuitry and multiple bypass caps to reduce clutter.  Standard 3-terminal regulators provide +/-12 volts.  The project is strictly analog; some functions may be performed as well or better in software on a microprocessor.   

For the oscillator amplifier I used a cascaded inverting two-amp format rather than the more usual single non-inverting format.  As far as I can determine this design was first suggested by J. L. Linsley Hood in 1981 (Wien-Bridge Oscillator With Low Harmonic Distortion,  Wireless World, May 1981).  Low distortion is facilitated by very low signal levels at the inverting amp inputs. 

Very high Wien bridge resistances in the low bands (50M for 3-30Hz) necessitated FET-input opamps, at least for the left-hand one.  I was unable to find FET-input opamps that would operate satisfactorily at 3MHz; one or two came close but didn’t quite have the ugga-duggas to make it up to 3MHz without high distortion.  I thought about trying a source follower at the inverting input of the left-hand BJT amp but not enough to do it.  Nagging VHF parasitic oscillation and the dreaded squegging in the high frequency BJT opamp circuit discouraged me from adding another stage.  My solution was to switch opamps between FET- and BJT-input types with electromechanical relays controlled by coil voltage selected by a 1P6T wafer added to the range switch.  Seems clumsy but it works.

Photoresistors control the oscillator amplifier gain.  At high frequencies (30KHz to 3MHz) the AGC signal is derived from simple diode rectification and capacitor smoothing.  Ripple on the AGC line introduces distortion, but at these frequencies ripple can be reduced to inconsequential levels by smoothing without inconveniently increasing settling time.  Low frequencies are a quite different story.  At 3Hz, smoothing to reduce ripple adequately would lead to settling time extending to forever and beyond.  I addressed this dilemma with sample and hold of peaks (or slightly beyond them) of full-wave rectified oscillator output.  Full-wave rectification allows sampling of double the rate otherwise possible.  At extremely low frequencies I wanted sampling as frequent as possible and was willing to accept the increase in circuit complexity.  The result is settling at 3Hz within 2.5 seconds or less depending on the type of perturbation; times are proportionately shorter at higher frequencies.  Sample and hold stabilization is not a new idea; for example https://sound-au.com/project174.htm.

The full-wave rectified signal is sent to the inverting input of comparator U2 of the sample and hold circuit.  The non-inverting input of U2 receives a signal modified by U1 and associated components.  The signals at U2 cross shortly beyond the waveform crest and back at the bottom causing U2 to change state at these points.  With changes in frequency and amplitude the sampling point retains a nearly constant relationship to the peak value.  Differentiation of the output of U2 yields brief pulses that when rectified turn on the FET, and the storage capacitor voltage changes to the sampled voltage, up or down, or no change after settling.  At equilibrium after settling, the AGC voltage at the storage capacitor is a flat line free of ripple except for minute blips at the sampling intervals.  A balance pot in the oscillator eliminates any asymmetry in the rectified signal, which would cause AGC ripple.

The very high bridge resistances at low frequency settings lead to high sensitivity to electrostatic noise pickup.  Lots of shielding, including the case itself, keeps mains noise in check; simply removing the top panel ruins low frequency performance.  Abrupt changes in voltages in the comparator and, to a lesser degree the full wave rectifier, produce electrostatic radiation that causes distortion so is shielded from the high impedance portions of the circuit.
 
The bare oscillator operates at 2V p-p output for lowest distortion.  A booster stage raises this to 6V p-p, better but a far cry from the 10V, presumably RMS, indicated on the front panel.  Eking more out of the booster is possible but at high distortion.  Of course 10V RMS is impossible from any opamp running at +/-12V.

THD+N for bands 1-4, 3Hz to 30KHz, measured with an HP 334A ranges from .01% at 350Hz to .035% at 30Hz.  I have no means to measure distortion below 5Hz; at 5Hz it is .02%.  THD+N for 30KHz to 600KHz, bands 5&6, is .06% to .10%.  THD measured with a spectrum analyzer at 3MHz is .35% (.29% before the booster amplifier).  THD (no N) measured with a Keithley 2015 at audio frequencies is lower, for example .005%  at 350Hz.

[moffy]

A very nice 'restoration', though the bench required to fit it would be substantial. We used sample and hold peak detection in the 80s to stabilise a couple of oscillator amplitudes. It's a nice technique that reduces distortion and filtering, cudos.

[ccktek]

moffy, thanks for the compliment and observation about the generator's size.  It is indeed a beast.