Splitter with a transformer

[R.G.]

Op amps are not that fond of reactive loads and tend to see trouble when there is a capacitive load.  Since the transformer can be modeled as a tuned circuit, there will be a (low) frequency where it sees capacitance.  Perhaps an isolating resistor from the output of the op amp to the transformer will make things better.  It could be anything from 100 to 1000 ohms.

Yep. Opamps and even emitter followers get downright ugly with some capacitive loads. In fact, many opamps today included a maximum capacitance they will drive.

@OP; the geofex versions of these things all use a 100R resistor to isolate funny loads from the opamp and keep it stable.

On other topics mentioned: Active drive for a transformer in applications like this is a way to cure many of the flaws of cheap isolation transformers. It won't so it all, but it helps. The bigger defects in cheap trannies can be sorted into buckets: poor low frequency response, usually about 300Hz; poor tolerance of high signal levels; and peaky, ringing or harsh high end frequency response.

A continuous problem is the low input impedance. Jensen likes their 10K:10K isolator for a passive guitar signal, and maybe it will do OK, as suggested in their app note. But I never wanted to buy a $50 transformer to try it out. Magnetic pickup guitars tend to lose highs badly with impedances under 300K, and really badly into 10K to 100K.

Of course, it's hard figuring out what impedance an isolator transformer presents to a guitar. Transformers don't have impedances (at least not directly), they have RATIOs. The stated impedance and frequency response tell you indirectly about  the internal inductances, and a ltttle about the internal capacitances. A spec on a trannie of 10K:10K only tells you that if you put a 10K resistor on the secondary, the impedance looking into the primary will be mostly a 10K resistance to AC signals >> in the mid band of the frequency response << and that since the impedance ratio is the square of the voltage ratio, the voltage transformation will be 1:1, as the square root of 10K/10K is one.

You can back calculate some things. If the maker tells you the frequency response and the impedance ratios, you can assume that if the trannie is loaded with a resistor on the secondary, that the low frequency rolloff point will be determined by the primary inductance's impedance reducing down to equal the stated impedance. This is a somewhat nuanced thing in terms of transformer practice, so at worst, the transformer's stated low frequency rolloff frequency can be taken as the frequency where the primary inductance impedance equals the stated impedance. So for a 10K:10K with a rolloff of 100Hz, you could infer that the manufacturer is telling you that the primary inductance's impedance is no worse than 10K ohms at 100Hz, and since Zl = 2*pi*F*L, you can calculate L >= 10K/(2*pi*100Hz). L is actually likely to be better than that, or the part would not always meet its specs.

This is the approach I took with the cheapo Xicon trannies. They're all specified at a low frequency rolloff of 300Hz. If you think about it, the primary inductance is eating an equal amount of the incoming AC signal CURRENT as the reflected secondary impedance if they're equal. It is possible to simply give the primary inductance more current to eat and keep the voltage across the reflected secondary impedance the same. That's one thing the amplifier at the input does. It takes the input voltage, then drives the transformer primary (through the 100R to keep it stable) and just shovels all the current out to the primary it can to keep the signal voltage right. As frequency drops, more and more of this current goes into the primary inductance, but as long as the opamp can shovel out enough current, the reflected secondary impedance sees the same signal voltage in spite of the primary inductance current hogging. So the overall actual low frequency rolloff as seen at the secondary is much lower. How much lower depends on where the opamp gives up on shoveling current. My measured result was that I could get those $2-$3 Xicons down to 60Hz for a -3db point.

That brings up signal level tolerance. The little cores can't live with too many ampere-turns shoved into them. The cores will eventually saturate. Not much to be done about this, as the core size is what it is. But one of the artifacts of core magnetization does get better with active primary drive. The magnetic core actually has a non-linear inductance, so that the signal current lost to primary inductance also varies instantaneously depending on how hard it'd driven. Fortunately, driving it with a low freqency source (opamp plus resistor) also partially fixes this, because the opamp will sit there and shovel in current to the mag field's dips and quirks to keep the signal voltlage (more) constant at the reflected secondary impedance. Not a cure, but it helps.

The high end is probably best tamed with that resistor/capacitor load. It is, as speculated, to damp the high end resonance of a cheap transformer and an unknown, unspecified load. The Xicon I messed with was specified +/-3db at 300Hz to 3kHz - that is, voice quality. In fact, it had quite a flat response out to about 14kHz, where response started rising. It ran up to a peak in the mid-20khz region. The cap value was chosen to let the load resistor start sucking energy out of the trannie's secondary above X frequency (can't remember right now) so that it would damp that ring-y peak. The peak wouldn't matter with normal guitar inputs, but somebody is sure to stick their MegaFire Octave Belter and IcePick emulator pedal in front of an isolator and let it rip. The cap/resistor load helps keep the insect-attracting tones down.