DC to AC converter. - can this be done?

The 45Hz makes it challenging, too, if you're trying to put it through a transformer that doesn't break your back.

One way to approach this would be to use transformers that pass well beyond the desired frequencies, then condition the frequency response of the circuit with the feedback components. This would help with the instability, but it would make transformer selection difficult.

Ultimately what you want is a circuit which has a high common-mode impedance and a low differential mode impedance. Your transformer idea is just one way to make this happen. Another way would be to build a pair of current drivers, with feed back coming from a differential amplifier. This would ease your stability difficulties (and shrink your circuit), yet it would still have enough pitfalls to make it really fun.

You could mix & match, also, by using a diff-amp feedback but a transformer in the forward path. This would ease the need for such an absurd transformer that I suggested in the all-transformer solution.

No matter what you do you'll have to deal with stability issues. This all sounds very doable, but not trivial.

---snip---

Thanks. Looks like we were thinking along the same lines. Transformer has a shield for common mode HF noise grounding. I don't thing 120Hz will jump from coil to coil unless the coils are wound very badly.

the Diff and for feedback looks promising.

Any good reference books for stabilizing beasties like these? There is one example in AoE but it is grounded input and ouput which makes it somewhat easier.

Beyond a 2nd-year networks book and a 3rd-year op-amp book -- not that I know of. I always approach these problems with a healthy dose of first principals leavened if necessary by the judicious use of SPICE.

Whenever the transformer starts coming near saturation, this method will drive it to the rail, where it hangs until the next zero-xing. Not a good idea. Works only for very small levels. It also requires your amp to be compensated much more, because of the additional phase shift at roll-off of the transformer. But you could take the feedback from the primary and implement a negative output impedance which equals the combined ohmic losses of primary and secondary windings. This will be stable unless shorted. To make the negative impedance, you can put a small resistance in series with the primary and add an appropriate level to the voltage feedback signal.

There are more possible cures, avoiding the transformer at all and also the floating supplies...

Just bought running shoes from Alpine Design and this model was called "Analog". Seriously. So I guess I qualify ;-)

If the size of the transformer is a concern one could take a different tack: Build a switch mode power supply, basically taken from a TI/Unitrode app note or similar. Make sure its regulation bandwidth is well in excess of 200Hz, usually not a big deal.

Now transfer the power at over 100kHz, rectify at the output, easy on the output cap value, load that pretty good so the cap will bleed down fast enough for a 200Hz upper limit and provide the usual TL431+optocoupler scheme for the feedback.

You would modulate in via the FB input of you regulator. It's not really trivial to build a switcher that goes from truly zero to full scale but it's been done before. SEPIC is what I usually do but that's not isolated. Flyback or push-pull may be your ticket because IIUC the output ground side needs to be floating to inject that other 2Vpp signal.

There is a series of really tiny lab bench supplies from Lascar Electronics. Don't know if they can be hacked to modulate the output but AFAIK they are fully isolated and quite cheap:

It sounds like the OP needs to source AC to his load; this rules out a 'regular' switcher. One could consider a switcher with synchronous rectification, although things would get complex that way.

I wonder if one could use a class D amplifier chip intended for subwoofers? You'd have to apply external feedback, but that should be straight forward. You may even have to power it from an isolated supply, but you'd still have something that was built of individually straightforward blocks.

Does the value of the output load move quickly or slowly? If slowly perhaps your colleague could keep the overall setup above but rectify/smooth both input and output, compare DC's, and use the average DC error to adjust AC error. Time constants will be in the order of seconds.

output can move fairly rapidly. Not a step changes but it will not be smooth lazily varying load.

The switching power supply and the ClassD are intereseting but switching noise will be an issue as we are making high resolution measurements in the vacinity.

I appreciate the input. Really gives me some ideas for tackling this problems on similar projects.