What's in the box?

[wrenchone]

This is a circuit I generated to try and sort out the harmonic content issues posed by the early version of the first circuit. It uses a jfet front end, violating the "all bipolar" criterion of the Schitt circuit. Distortion is higher than in the first example, but still not shabby, and harmonic content looks ok.

I used an input capacitor, but it is not necessary for proper circuit operation. However, I would normally insist on input cap, as I don't trust outside-world sources.

[wrenchone]

This next circuit is more in line with the design criteria stated by Schitt for their Magni 3 headphone amp, being all bipolar, and potentially with an all-DC path from input to output.

I'm pretty sure that Schitt just uses resistors instead of current sources to bias their input stage in place of current sources I1 and I2, as I see only 8 active devices/side in the picture of their PCB. I use the current sources to increase power supply immunity, and would implement them in practice using a couple of current mirrors and a master current source.

I chose to use all Darlingtons from stem to stern, to reduce the number of line items on the BOM. I also have reels of both the SOT-23 and SOT-223 versions of MPSA14 and MPSA64, another motivation.

I used a couple of tricks that Schitt probably doesn't use in their circuit. The VAS stage transistors use strings of GaP green LEDs for emitter bias (voltage drop without a lot of degeneration resistance). These are represented by V7/R10 and V8/R8. They allow me to increase the value of load resistor on the previous stage for more open loop gain.

I also use a GaP green LED to bias the output stage (V6).

Circuit and simulation results are shown for guidance only. THD is not shabby, and harmonic content looks ok.

[wrenchone]

The last circuit illustrates my love affair with current mirrors. In a real implementation of this circuit, I would use another pair of current mirrors to bias the input Darlingttons.

THD is a little higher than in the previous example, perhaps due to lower open loop gain. Also the high order harmonic content is a little higher than I would like. This might be fixable with a little more effort. Still, this menagerie is not a bad haul for three days of off-and-on diddling with PSpice.

[Mark Tillotson]

I'm confused, you have distortions of 0.001% or thereabouts, why does the harmonic profile/structure matter, they are all extremely low, its basically linear for all possible purposes?

[wrenchone]

None of these circuits have a whole lot of open loop gain, so distortion will rise with input excitation. The specified 1V RMS into 32 ohms is only a little more than 30 mW output.
I want to see a distortion spectrum dominated by 2nd harmonic, with higher odd-order harmonics diminishing with order. I get antsy when I see something like a 7th harmonic nearly as high as the 2nd, even if the THD is low. Higher excitation levels will bring up everything, and may result in listening fatigue. I ran into a situation like that this weekend with a simulation, (2nd and 7th almost equal in amplitude) though I suspect part of the issue is using a version of PSpice at home a lot older than the version I have at work.

[wrenchone]

Here is a stripped down version of one of the circuits resented previously, with only 8 transistors, just like the Magni. I'm using LED bias tricks described previously to bias the output and VAS stages. THD and harmonic balance look respectable. I may end up building this one once I figure out the optimum placement for the servo.

I suspect the topology is very similar to the Magni, except for the LED tricks.

[wrenchone]

I've had an entertaining time in PSpice stabilizing this design, as current feedback amps are different when it comes to compensation. I'll re-post the design when I'm finished. It will be interesting to see if the compensation on the real-life amp turns out the same as the simulation. It's different a lot of the time.

[wrenchone]

Here's the circuit of post 16 tweaked for frequency response. The simulation shows a fast, optimally damped response. I suspect things will look different when I actually build this one up, as the compensation in real life rarely matches that of the simulation, due to strays in real construction, as well as model vs. reality...

[wrenchone]

Don't consider this the final take. I'll be doing some massaging of the design to optimize both THD and bandwidth/stability. The design currently crosses over at ~40 MHz according to PSpice, with no funny business on the way down.

Increasing the feedback network resistance to optimize stability caused an increase in THD. I managed to wrestle THD back down to the ~0.001% region, but it's not as nice as what I started with. Also, there is offset at the input that will require use of an input coupling cap. As always, food for thought and grounds for further research.

[Mark Tillotson]

Originally Posted by wrenchone

None of these circuits have a whole lot of open loop gain, so distortion will rise with input excitation. The specified 1V RMS into 32 ohms is only a little more than 30 mW output.

Why not simulate at a higher level then?

I want to see a distortion spectrum dominated by 2nd harmonic, with higher odd-order harmonics diminishing with order. I get antsy when I see something like a 7th harmonic nearly as high as the 2nd, even if the THD is low.

I prefer to see all the harmonics in the noise floor, then it doesn't matter, and that's not impossible to achieve.


You can use a weighted sum of harmonics to get a final figure that better represents the thing you want to minimize.


Looking at the distortion residual is a very powerful technique for telling you what's happening too.