discrete op amp

[Bonsai]

Originally Posted by dirkwright

Well, not to throw stones, but the Pass B1 buffer isn't exactly complex and elegant.

I suppose I can delete the C4 capacitor. Offhand I'd say it shunts HF to -32V rail, but what do I know. It's 5pF from what I can gather by looking at the board.

I think this is a mistake. Conside the rails as short to ground at AC (because the power supply impedance is usually considered low to ground, and you nornmally have decoupling caps).

I could understand this if the cap was connected between the output and inverting input (although there would probably be some stability issues . . . )

[dirkwright]

Originally Posted by Bonsai

I think this is a mistake. Conside the rails as short to ground at AC (because the power supply impedance is usually considered low to ground, and you nornmally have decoupling caps).

I could understand this if the cap was connected between the output and inverting input (although there would probably be some stability issues . . . )

OK, I can leave it in, no big deal.

I designed a push pull output stage for it.

[dirkwright]

I built this in Tina, a SPICE program, using different transistors because they didn't have models for the original ones. I added a simple push pull output stage. Distortion is 0.0016% @ 350mV and 1kHz input. Gain is about 5. I think it's pretty good considering I'm a newbie at this.

[dirkwright]

I worked on this some more and came up with this. I had neglected to place a load on the previous design, and the input voltage generator was in the wrong place, as well as other things not optimum. I decided to use a diamond buffer output stage also. With 200mV in, 10kHz and 10kohm load, distortion is 0.0025%. At 500mV it's about the same. At lower frequencies, the distortion is about half (0.00107%). Of course, this is an idealized, simulated circuit, but it's still fun to fool around with it.

I've twiddled with just about every resistor and capacitor in this circuit, as well as adding and removing components. I figured out that the class A voltage gain stage is extremely sensitive to load and the constant current arrangement is extremely sensitive to the voltage divider bias network. I had to add the 28kohm resistor between this stage and the diamond buffer to reduce the load on the voltage gain stage. I guess the input impedance for the diamond buffer isn't that high. Anyway, I'm open to comments and suggestions on this circuit.

Even with 2volts input, distortion is still only 0.004%. dang.

[dirkwright]

It looks like the 2SD667/2SB647 are similar to the BC337/BC327 that I simulated with. I don't have SPICE models for the former pair. So, hopefully if I build this I can use the transistors I have on hand instead of buying new ones.

[dirkwright]

Originally Posted by Bonsai

I think this is a mistake. Conside the rails as short to ground at AC (because the power supply impedance is usually considered low to ground, and you nornmally have decoupling caps).

I could understand this if the cap was connected between the output and inverting input (although there would probably be some stability issues . . . )

It turned out that this little capacitor reduces distortion, so it's in. I've played with the value and I think it's dependent on the kind of transistor used for Q2. With the BC 560 I got 15pF as ideal, lowest distortion at 10kHz.

[Bonsai]

The distotion levels you are getting are high given the signal levels and the type of circuit. Any good the little cap is doing is masked by other serious problems.

The good news, is the cures are easy to implement.

Your LTP tail current is set to 300uA and the LTP load resistor is 22k. This is far too high and means that the LTP is grossly unbalanced, leading to distortion (mainly 3rd). Remove R6 (470 Ohm resistor) - you dont need it in a small signal amplifíer like this. To balance the LTP collector currents, R8 will be in the range of 3800 - 4200 Ohms. To do this, measure the LTP collector currents, and adjust R8 until they are within 0.5% of each other - the closer the balance, the better.

The base resitors in your diamond buffer are on the high side. In a diamond buffer, you can often get away without these, but if needed (base stoppers to prevent HF parasitic oscillation) then 10 ohms is normally enough.

Similarly, the driver standoff resistors (R4 and R13) are also too high. In most designs you can do away with these, but if needed to set up the correct output stage current, they will tend to be 10's of ohms and not 270.

The output stage emmitter degen resistors are shown at 47 Ohms - this is very high, and necessitated by the high value of R4 and R13. Typically these would be 3.3 to about 5.6 Ohms. To optimze the diamond buffer output stage, set R1 and R2 to 3.3 Ohms, remove the base stoppers, remove the stand off resistors. Run a sim and check the output stage current. If you want class A operation for typical 600 Ohm loads, I'd set the output stage current at 30-50mA (note: much higher than an IC design - but thats one advantage of going discrete!). If its much higher, increase R1 and R2 value - if much lower use the standoff resistors in the driver emitters - you may need 10-30 Ohms at most. Another option here is to tweak the driver current sources.

Your diamond buffer driver current sources are shown at 2mA - this is ok. In a well optimized design using this topology, you can replace the current sources with resistors (10k on 15V supplies works well) and still get outstanding distortion performance.

With these adjustments, you should be able to get distortion down to 30ppm at 20KHz at 10V into a 1k load, and about a quarter of this at 2-3V into 1K.

How to get distortion to below 1ppm (2-3 Volts into 600 Ohm load):-

1. Use a current mirror load in the LTP
2. Add a beta enhancer transistor between the LTP current mirror and the VAS amplifer

One final point:- between the diamond buffer output and your load (after the feedback take off pount), you need to insert a 22 to 50 Ohm resistor to isolate the amplifer fro capacitive loads. without it, you will likely run into stability problems.

[Bonsai]

Note, if you are using small signal devices for the output stage (TO-92) you will have to keep the output stage class A current to 15-20mA

[godfrey]

Originally Posted by dirkwright

I figured out that I can probably replace R7 with a transistor, connecting the base to junction of the emitter of Q3 and R10.

Yes, that would make a much better current source. Any small PNP transistor will be fine. There's an easier way to improve that current source, though: just replace R7 with a constant voltage reference like a zener or LED. That will be easier to fit on the existing PCB.

An even bigger improvement is to replace R4 with some sort of constant current source. I see you did that in your later Tina circuit - good!.

Originally Posted by dirkwright

I figured out that the voltage divider R7 (910) and R8 (62k) draw 1mA, so I can replace R7 with a 1mA constant current diode to improve PSRR.

No, that's the opposite of what you want to do - it will make PSRR terrible. To get the effect you want, you need to leave R7 as is and replace R8 with a constant current diode. Oops - got that one wrong in the Tina circuit.

Originally Posted by dirkwright

I built this in Tina...

See comments above. Also; C4 in the original circuit (C3 in the Tina circuit) looks to me like a really bad idea. I can't see that it can be doing anything useful, but it will make stability and high frequency PSRR worse.

Good luck!

[dirkwright]

Wow, thanks everyone for the feedback! I've tried some of the ideas already but none have reduced distortion yet. I tried reducing the LTP load resistor from 22k to 10k but it made the circuit impossible to balance. I'd have to make changes somewhere else to make that work. Removing the 470 ohm resistor in the voltage gain stage made the distortion much worse also. Thanks for the great ideas!