I was experimenting with an interstage driver using symmetrical CFPs. It was working pretty well so I thought I would see how it performs using the old JLH auto-bias output-stage topology. The simulations look encouraging:
I like the monotonic distribution of harmonics. THD calcs out to around .002%@1KHz, despite the circuit's relatively low OLG. The input stage has a LOT of local NFB, which helps in that regard.
I believe the other helpful factor is that the two CFP blocks use the same NPN' and PNP's so the halves are better matched. At least as far as the simulation is concerned. In a real-life application you'd probably want to match all the NPN's and PNP's: but I don't think the NPN-PNP pairs would need to be matched.
BTW the input capacitor is a relic of some earlier experiments. It really isn't needed in this case.
I tried further increasing the input circuit's transconductance by reducing R11 to 10 ohms but it didn't work out too well. Substantially increasing the OLG would mean that the 50 ohm resistors also would need to be substantially reduced in value, making the idle current less stable w/regard to temperature. And at some point the feedback network's resistance would be lower than the load presented by headphones (or speakers). Given the already-good THD, it just doesn't seem to be a worthwhile tradeoff.
I like the monotonic distribution of harmonics. THD calcs out to around .002%@1KHz, despite the circuit's relatively low OLG. The input stage has a LOT of local NFB, which helps in that regard.
I believe the other helpful factor is that the two CFP blocks use the same NPN' and PNP's so the halves are better matched. At least as far as the simulation is concerned. In a real-life application you'd probably want to match all the NPN's and PNP's: but I don't think the NPN-PNP pairs would need to be matched.
BTW the input capacitor is a relic of some earlier experiments. It really isn't needed in this case.
I tried further increasing the input circuit's transconductance by reducing R11 to 10 ohms but it didn't work out too well. Substantially increasing the OLG would mean that the 50 ohm resistors also would need to be substantially reduced in value, making the idle current less stable w/regard to temperature. And at some point the feedback network's resistance would be lower than the load presented by headphones (or speakers). Given the already-good THD, it just doesn't seem to be a worthwhile tradeoff.
Just a follow-up note regarding the original JLH design. The input transistors' emitter resistors are connected to Vcc/Vee, so there's a pretty hard limit to how high the transconductance can be increased. The HA I'm using right now is a variation that uses +/-4V regulators, so the emitter resistors are quite a bit lower in value, which reduces the THD compared to the original design. The PSRR is quite a bit better, too. The above scheme is even better in this regard.
Some might suggest just juggling the input transistors around to form two differential amplifiers + a cascode. Yes of course that can work. But in my experience amplifiers that employ current feedback are usually easier in terms of achieving stable, non-oscillating operation.
Some might suggest just juggling the input transistors around to form two differential amplifiers + a cascode. Yes of course that can work. But in my experience amplifiers that employ current feedback are usually easier in terms of achieving stable, non-oscillating operation.