The speed of an AC synchronous motor is by definition set by the frequency of its AC supply, so a variable speed drive is a variable frequency source stepped up to the required power level. This will also work for any single phase induction motor. The standard AC single phase synchronous motor actually requires two phases in a sine / cosine relationship (called quadrature because the two phases are separated by one quarter of a cycle) for it to operate reliably in one direction. In many applications the cosine drive is faked with a capacitor so all that is required is a single phase oscillator / drive of the correct frequency range.
Single phase drives can be either digital or analogue. I am going to concentrate on analogue as it is easier for the hobbyist to implement than digital of equivalent quality. (This situation is reversed if you are considering larger scale production - the cost is in the programming). The single phase oscillator I present here is based on the one I built for my Garrard 301 Drive and although it works quite well it does not achieve the performance levels of the later and more complex drives.
The circuit consists of a basic Wien bridge oscillator with a very simple PTC gain control which creates the variable frequency sinwave, running into an LM1875T power amplifier and then into a step -up transformer. It could easily be improved by inserting one or more Sallen and Key low pass filters between the oscillator output and the power amp IC, also by running the power amp IC in inverting mode. I have not made either of these modifications so there may be pitfalls I haven't foreseen.
Click pic for larger version
The values shown will give an output frequency variable either side of 50Hz. For 60Hz use 120nF capacitors. Also note that this requires a +/- 18V supply to achieve 15 watts at 110 V and that means using regulators for the oscillator ICs supply rails, +/- 12V is fine. For lower powers you may run the whole thing from these voltage supplies but the transformer step up ratio must be changed. +/- 12V and a 6 volt transformer will achieve approximately 5 watts.
A more sophisticated approach is to generate both phases electronically rather than “faking” one. There are many avenues to doing this, using both analog and digital techniques. Possibly the easiest is to simply run a digital crystal clock output through a divider to generate a square wave of the correct frequency, run that through a PLL to generate a locked signal in quadrature then clean both of them up to give the sine / cosine pair. This works, but not necessarily that well. The PLL will generate the two phases in exact quadrature only when the VCO signal is exactly in the centre of its range, making the LP filter design very tricky. Even if we manage this to keep the two signal in quadrature way we have to ensure that they see equal phase shift in the subsequent circuits – any relative phase shift will knock them out of quadrature. Given that we have to institute a huge amount of low pass filtering to get a passable sine wave out of a square wave and that all low pass filters give phase shift in the pass band, this means taking a lot of pain in the filter circuits.
An alternative is to use one of the new direct digital synthesis chips from Analog Devices but these require microprocessor control and the high precision versions are definitely not cheap. Since I haven't put in the time to learn to program PIC or Atmel AVR chips I can't go that way but in any case there's something NQR about using a digital solution when a neat analogue one is available. The neat analogue solution is simply to build an ultra low distortion oscillator which produces the required waveforms and then just amplify them as needed with no filter components.
Fig 1 Fraser Oscillator
This circuit is based on the old Fraser oscillator, which uses a pair of all-pass phase shift networks to achieve a pair of 90 degree delays at the specified frequency then an inverting amplifier to bring this to 360, ensuring oscillation (see figure 1). Because the all-pass networks pass all frequencies without attenuation, it can suffer from parasitic oscillation at a frequency where the sum of the parasitic phase delays in the three amplifiers also sums to 180. With my prototype this occurred at about 600kHz, not difficult to remove but still inconvenient. I decided to rebuild the circuit with passive two pole low pass filters as the 90 degree phase delays, making a hybrid between the Fraser design and the Dippy phase shift oscillator (see figure 2). Using low pass filters as phase shift elements also reduces higher order distortion as the distortion products are above the cutoff frequency of the LP filter.
Fig 2 Modified Fraser / Dippy Oscillator
Using pairs of two pole filters would normally require four adjustable elements to vary the frequency. I didn't want to go that way, partly because the two poles are implemented at different resistance levels spaced a decade apart to improve accuracy. Instead I decided to make one pole in each pair have 45 degrees phase shift very close to the desired oscillation frequency and the other so that the point of 45 degrees phase shift can be altered by about 20%. This means that as long as the components are reasonably well matched we can vary the frequency by about 10% using only two variable resistors rather than four. A second set of trimmers in the circuit also allows the phase delay of the two filters to be altered, giving us the ability to alter the phase difference between the outputs either side of strict quadrature. To make the matching of components easier the capacitors were chosen to be parallel pairs of common values, it is much easier to sort within the tolerance spread of the caps to find pairs which sum to close tolerance. Similarly the resistors are tested and closely matched although this is icing on the cake as resistors on the same bandolier are generally close anyway. Since a pair of passive 45 degree phase shifts will give a total of 6dB attenuation, the gain section of each filter is set to 6dB.
Like all oscillator circuits this one requires careful balancing to achieve stable output voltage. To this end the inverting amplifier uses a Vactrol optocoupler, in which a light dependent resistor is fed by an LED. The LDR is placed as the shunt element so that controlling the current to the LED controls the gain of the circuit. To feed the DC equivalent of the AC circuit output to the LED driver I used the multiphase rectification / differential amplifier scheme I had developed for my three phase oscillator project. Since the Fraser oscillator gives three phases at 0, 90 and 180 I used a fourth amp as an inverter on the 90 degree output so I had four equally spaced phases to reduce the ripple level. These four phases were each rectified and fed to a summing amplifier, the output of which is fed to a differential amplifier. The diff amp has a series capacitor in one input leg so the difference amplified is the DC level shift between the summed rectified waveform and its level shifted equivalent, thereby removing most of the ripple by the differential action.
This is then fed to a Sallen and Key low pass filter using a 2 nd order Chebyshev characteristic, chosen for its low level of phase shift below the cutoff frequency. The output of this filter drives the Vactrol through a variable resistor for gain control. There are a couple of other resistors which serve to stabilise the circuit by reducing the loop gain of the feedback.
Oscillator Schematic
(click for full size version)
As expected this circuit gives very good distortion results if audio grade op-amps like the LM833 are used for the primary amplifiers and the output is kept to a sensible fraction of the supply voltage. The output voltage has to be high enough that the half wave rectifiers pass at least half of the positive half sinewave (eg ¼ of the full wave) to reduce the ripple, I chose to set the output voltage at 2.5 volts. The amps in the feedback circuit are three sections of a TL074 because they give me better stability, note the presence of loop stabilisation capacitors on the shunt elements in the inverting amps.
To preserve these good results the output amplifiers are LM1875s connected as inverting amplifiers, thereby reducing the common mode voltage at the input. This can make trimming the gain more difficult as the low impedance of the inverting configuration will interact with the gain setting potentiometer but these are basically set and forget devices so this doesn't overly concern me. There is a timed relay connected between the oscillator and the output amps which allows for two output voltage levels. The LM1875s can give up to 20 watts steady state per phase with the right power supply (2 x 24 volts at 2 Amps). The timed relay can be used to give full output voltage for the start up then drop to a lower running voltage to reduce noise. In fact this feature can be used to run greater than rated output during the start phase so a higher power motor can be safely run as long as the steady state power is less than 20 watts per phase.
The outputs of the LM1875s are connected to two toroidal output transformers chosen for their very low magnetisation current. This is an important point – I tried this with a pair of IE core low voltage lighting transformers and they overloaded the LM1875 outputs which overheated very rapidly, one even let the smoke out. The transformers are run “backwards” since they were designed for use as step downs not step ups. Don't forget that the voltage outputs of transformers generally have a percentage of regulation built in, so the actual turns ratio is slightly lower than the advertised voltage ratio. Accordingly, I use 10 volts RMS into a “9 volt” transformer.
Output section
(click for larger pic)
The independent voltage controls on the outputs allow the voltage to the two phases of the motor to be altered with respect to one another and the phase control allows small variations either side of strict quadrature. On the motor I tried for my prototype these controls allowed the drive to be tuned to give a distinct sweet spot at a voltage ratio of 10:9 and another sweet spot on the phase control which appeared to be at around 89 degrees phase difference. These two spots appear to be independent of one another. With the values given the frequency is variable from about 57.5 Hz to 63 Hz, about +/- 5%.
Measured output distortion into a resistive load equivalent to a 5watt motor was -80dB (0.01%) which I believe is better than the commercial equivalents. To match this spec the equivalent load on the LM1875 must be kept above 8 ohms. This sets the output power limit, with 9 volt output transformers and +/- 18 V supplies at 2 Amps up to 30 watts is achievable, 6 volt transformers and +/- 12 Volt supplies at 1 amp will run a 10 watt motor.
Parts cost for my prototype was around $100 but that was without a power supply or an enclosure.