Class AB differential power amplifier

Dear fellow audio enthusiasts,

A second year electro student here. Since a couple of months I’ve been working on this power amp idea. I somewhat know how to design for good CMRR and PSRR levels but have no idea what to look out for when speaking about THD and whatnot.

Furthermore the JFET input stage voltage gain is counterintuitively smaller than one because it allows for a greater voltage swing at the pnp emitter follower. Any thoughts on this or any other aspect of this design? Any suggestions for nice small signal btj’s or potential power supply options?

Any feedback would be greatly appreciated.
Much thanks and cheers,


Some more info and specs:

  • Vcc is 33 to 35 volts. Ill be using 25VAC, >300VA per channel
  • T1 is tuned so the DC voltage at the output nodes is Vcc/2.
    MultiSim simulation shows:
  • 100W into 4 ohms
  • High output current (>13A peak to peak)
  • voltage gain is about 25dB, rolls off 70udB at 30kHz.
  • input imdedance is in the giga ohm range
  • output impedence is 202mOhms (low enough?)
  • CMRR and PSRR >100dB



If you haven’t already tried, I would try - a lot of designers hang out there.

Nice to see so many bipolars in use rather than all FET (personal opinion only) :slight_smile:
My own thoughts - build it, listen to it, and tune it.

Partly because deep analysis of analogue circuits is beyond me now, and partly because
the lowest THD and whatnot don’t always sound the most musical and enjoyable to audiophile ears.
It would be good to actually listen to multiple iterations and (if you have the kit) aalyse the distortion etc. after you have listened to it.
(and please report back!) :slight_smile:


Will do Joma I posted it there too. I hear you and i’m tempted to order all these parts, though I gotta watch my modest budget, so only if this design is worth it Ill build it. And for sure if any progress I will report back!


1 Like

if it sounds great i might want one :wink:

Hi Rupopulles,

Welcome and thanks for joining our community. I would certainly try to the join diyaudio if you haven’t already.

Here’re a few quick notes after a quick glance at the circuit here:

Add shunt resistors around 100K from your input gates to ground. Without defining the input impedance, the gates will float up and you’ll have problems. 100K is typical for this. No need to use a JFET for Q26. I’d just replace this with a BJT junction based configuration such as Q12/Q10. For these small signal transistors you can use something that is general purpose such as 2n5551/2n5401. You may have temptations to use high beta devices but I’d suggest not using those parts due to the much higher risk of parasitic oscillations and run away.

I’m curious why you’ve decided to float the output up instead of using a bipolar supply and aiming to get around 0VDC on the outputs. My guess is that you wanted to eliminate a supply. You just better hope that both sides stay well behaved in DC and don’t start to slip on you or your speaker will start to possibly see some significant offsets.

The output stage is where I start getting concerned a bit. The configuration is similar to a diamond output where the current running down the driver stage is reflected on the output. You need emitter resistors to define this current since you are essentially using the junction drops to get to 0 voltage on the output. You can start at Q14/Q15 and follow the junctions on each side of the complementary circuit. You’ll see that overall, the ideal DC operating point is 0V from the base of Q14/Q15.

One issue that won’t show up in simulations is thermal runaway on the outputs. I have found these configurations to be challenging to thermally stabilize and one way you can help mitigate this is to also add emitter resistors to Q14/Q15. My guess is that there will be some trial and error here to make sure the output stage is stable.

Finally, you don’t have any negative feedback in the design. Negative feedback is incredibly useful for power amplifiers. Feedback lowers output impedance, dramatically lowers distortion and helps with DC stability. I’m guessing this was a design decision but thought I’d mention that for power amplifiers, feedback is an incredible tool to optimize performance.

My suggestion is to add a negative power rail, add emitter resistors, and look into DC servos using a JFET opamp to servo the outputs in reference to ground. This will help with DC drift on the output especially since you aren’t using NFB.

That was a quick overlook on the circuit and didn’t look into gain/other specifics but I’m impressed coming from someone in their second year of EE. Good work.

If you’d like to talk further, PM me.

Best of luck and happy listening



Best Darren,

Thank you for the warm welcome. I’ll certainly consider your suggestions and will try some more extensive temperature sweeps.

The previous iteration actually did use NFB, though it had allot more top end phase shift and used two diff pairs per channel. And indeed the single supply is chosen so we only need one capacitor bank per channel (and maybe regulation) and save some money. You’ll find a PM from me in the near future and I’ll elaborate more. I’d love to chat further.

Happy listening and cheers


1 Like

I’d be happy to talk more about it! Feel free to contact me. :slight_smile:


Also happy Pi sunday!

1 Like

click on his name above, a mini profile should appear and on there is a message button :slight_smile:

Hi Ruben,

You can also just email me at

Look forward to talking!

The forum software does not allow PMs until you have been around for a bit more.

Dear fellow serious listeners,

First of all thank you for recommending DIYaudio. Very useful conversation kicked off there which led to some small changes.

Now a DC servo is deployed that keeps the VDCout at the center with minimal fuzz and complication.

I followed Darren’s advice and indeed noticed the benefits of using some Re at the output transistors. Even these small amounts linearize the current. Also the input impedance is defined at 200k.

A tiny bit of C is added from base to collector to roll-off openloop gain for the required gain and phase margins. Lastly a modest amount of shunt feedback is added to further extend the possible voltage swing and lower the output impedance a bit while keeping input impedance untouched.

However, now we have about -2.2 degrees phase shift at 20k, just like the design that uses two diffpairs. Additionally I wonder whether I am not missing out on the benefits of other feedback methods.

One recommendation I got is to current source the PNP voltage gain stage and to swap it with the NPN driver. I didn’t get this to work though. How would I define Av? With feedback alone?

Lastly I conceived a plan for the PSU. I accuired a 2k5VA 3x25VAC toroidal bigboy for next to nothing. I have a couple of 8.5 kg heatsinks here. One for each channel? I’m planning to go wayyyyy overkill here. Secondhand high power components seem simply cheaper than new medium power components.

I could also call the 0.5Vcc points ground and connect it to earth and casing. This is not a commercial design by any means, so the amp doesn’t really care.

So far so good. Will keep developing the design for a bit longer before I throw money against less general parts.

All thoughts and opinions are welcome.
Thanks and much cheers,


1 Like