My take on a **buildable** Discrete Op-Amp...

[soekris]

I have done some playing around and ended with this pretty nice little schematic, which when using "just" 0402 passives and sot323/523/563 transistors actually should be able to fit a 10x10 mm PCB to match a standard dip8 footprint....

Performance is something like >80 db open loop gain, Gain Bandwidth of 45 Mhz, 2 Khz open loop THD at 7.75V rms into 2K load is 0.2%, 2.5V into 10K is 0.04%. That should result in 20-20Khz, 20 db closed loop gain THD in single digit ppm range....

Should be unity gain stable, although compensation capacitor might need to be adjusted after final layout.

Power is up to +- 26V, with reasonable power draw, can be adjusted down with slight loss of performance.

Input JFET's are sot323, either Toshiba 2SK880 or Panasonic 2SK0662, and the 2SK0662 are actually available from Digikey stock....

Noise with 2SK0662 are 2.7 nV/sqr hz, and offset can easily :-) be matched to 1 mV.

The circuit also have overall great DC, AC and thermal performance, is even short circuit protected....

I might even do the PCB layout and built some when I have some spare time....

Btw, I have actually built, tested an measured it on a breadboard, no simulation here, it's the hardware part of doing hardware that's fun :-)

Comments welcome.

Søren

[soekris]

And the actual schematics....

Søren

[nelsonvandal]

I think the performance will increase quite a lot if some kind of buffering stage atfter the VAS, like a diamond buffer or double EF, are used, or if you use a MOSFET OPS. I think this also make a better sound.

Have you compared the sound of active current sources on top of the LTP to resistors? What's your oppinion? I know from simulation and real life measurements that active current sources vastly reduce THD.

[tiefbassuebertr]

Originally Posted by soekris

And the actual schematics....
Søren

In this case the follow thread could be of interest for you:
AD797 discrete clone - comparison of seven variations
Basicly the same as there one could be perform with the internal schematic of OPA-211 (OPA211) Fig. 44 page 14 about pdf datasheet
http://focus.ti.com/lit/ds/symlink/opa211.pdf

[soekris]

The idea is to have as few parts in the signal chain as possible and to make something that can actually be built, which limit how much buffering you can have after the VAS.... The two output transistors used are high and linear HFE types, so performance is still great into 600 ohm, as long as they stay in class A.

Driving low impedance load will just result in lower open loop gain, basically you load the VAS with Rload*HFE, gain bandwidth and closed loop THD stays the same.

I have experimented with mosfets at the output, like a 2N7002W/BSS84W set, and they're great, especially when driving very low impedance load like 100R, in fact, with more powerful mosfets you can use it as headphone or power amplifier. The problems with mosfets is the biasing, you will need to adjust R4 when building the small modules....

The main reason with the current sources on the LTP is to reduce THD, but also for thermal stability, the 1mV offset should be there over temperature.... If using just resistors then offset depends on Q2A/Q2B, that although being a dual still have different load. You could then add a cascode to Q2A to even the power load on them, but then the parts saved on the current source are lost.... Using a dual part as current source with little power load result in excellent thermal tracking, and not really that many more parts....

I haven't listened to them, but can't see any reason why they should negatively affect the sound.

Søren

[scott wurcer]

Originally Posted by nelsonvandal

I think the performance will increase quite a lot if some kind of buffering stage atfter the VAS, like a diamond buffer or double EF, are used, or if you use a MOSFET OPS. I think this also make a better sound.

Have you compared the sound of active current sources on top of the LTP to resistors? What's your oppinion? I know from simulation and real life measurements that active current sources vastly reduce THD.

Nelson has a valid point, the extra buffering has two purposes. It would be rare for the hfe*Rl to track from n to p side well enough to get "ppm" even harmonics.

In addition the dynamic voltage across the output devices is coupled in through the Cje's. With the MOSFETS mentioned, the Cgd's alone triple the capacitance at the gain node, not to mention almost 90pF of Cgs coupling the Vgs modulation with load current. The n to p side of these capacitances would rarely match and are inherently non-linear anyway.

Buffering reduces these effects especially in a one gain stage amplifier.

I can't comment on most of the components since the part id's are unknown to me. But with a JEDEC theta-j of 300C/W for SC-70 packages and 26V supplies you are running several devices at a significant fraction their limits and certainly a few are 10's of degrees above ambient.

The input FET's would require some selection for Idss > 3mA, and at almost 75mW each quiecient I don't see pre matching to 1mV as feasible since from experience the PC layout will determine the final individual device temp.

[CBS240]

Originally Posted by soekris

I have done some playing around and ended with this pretty nice little schematic, which when using "just" 0402 passives and sot323/523/563 transistors actually should be able to fit a 10x10 mm PCB to match a standard dip8 footprint....

Hi

If you’re really serious about small, then don't forget the SOT963/923 components. Also 0201 size resistors save lots of space, but watch the heat. I believe On-Semi has BC846/856 and 2N3904/3906 in SOT963. BTW, even though all those transistors are monolithic duel packages, it does not mean they are electronically matched.
Here is a nice linear transistor from Toshiba that I like in SOT923. 2SA2154 and 2SC6026


With the super small SOT963/923 parts, you can reduce the current in the VAS saving on the Pd that you need to reduce, and add a diamond buffer so you have the power and impedance where you need it, to drive the output stage. As Scott stated, this greatly reduces the effect of the output transistor's non-linear dynamic properties from affecting the gain stage. Keep in mind that mosfets do require current to drive at higher frequencies. Also the Cdg is not very linear even away from the crossover region. Of course we all know this can be greatly linearized by cascode of the output stage.

[EUVL]

I guess you have seen mine ....

Pass Discrete Opamp in DIP-8 Package

I would second Scott and strongly recommend potting and heatsinking, and using lower rails (I use +/-12V max).


Patrick


BTW I just checked Digikey and 2SK662 is listed but not in stock. It is quite a nice part, but a bit expensive. BF862 is a real bargain, even lower noise, higher gain, lower capacitance, ..... And many have used them in various applications, including Scott Wurcer, myself, and Juma. There are plenty of info at this and the Pass forum.

[soekris]

Size:

The whole point being "buildable", that why I limited it to 0402 and sot323/563, which is what I believe is the smallest I can handsolder.... And there aren't really that many smaller parts well suited for audio use.

VAS loading:

Yes, I'm aware of the issues and a second stage is not a bad idea, but again, physical space.... But I also like to have as few parts in the signal chain as possible. And I have tested with a LOT of parts, ended with Zetex high current, high hfe types as the best performing parts.

MOSFET output:

The Ciss (Cgs) loading is multiplied by forward transconductance, so it's not really that bad, Crss (Cgd) is a direct load and often more critical, but still pretty low, and when using output mosfets you don't need the compensation capacitor :-). It will result in higher distortion at high frequencies, but we're still very low, and I don't believe in targeting just low distortion.... MOSFET's are my preferred output devices, but the main problem is still bias adjustment.

Thermal:

Yes, if both high current and high voltage you would need to pay attention to power loss, potting and heatsinking is then an option.... A practical thing would be to make maybe two version, one for high speed and one for high voltage.

Input fets:

The power dissipation in the input fets is an issue, but my experiments show that the 2SK0662 are very good up to about +-15V, after that you need to limit the common mode range. The best part will be the 2SK880, they spec'ed for 50V. Of course when not targeting 10x10mm then the input fets should be cascoded. But that's a different design....

They don't need selection to idss >3mA, they're just fine being used below their idss as the high Yfs types don't hit the diode conduction point until very current....

Other fets:

There don't exits other very small jfets suitable for audio use than the 2SK0662 and 2SK880. BF861 and BF862 are fine, but sot23. They also have higher low frequency noise. If sot23 then you might as will use the LSK170. And it's the 2SK0662 that digikey stock, not the old p/n 2SK662.

"Pass Discrete opamp in dip-8 package":

One of my inspirations, but my design requirements was different:

High speed: Your 100 Khz looks like my 1 Mhz :-)
Output: I wanted class AB to be able to drive full swing into 600R.


Søren

[scott wurcer]

Originally Posted by soekris

Size:


The Ciss (Cgs) loading is multiplied by forward transconductance, so it's not really that bad, Crss (Cgd) is a direct load and often more critical, but still pretty low, and when using output mosfets you don't need the compensation capacitor :-). It will result in higher distortion at high frequencies, but we're still very low


Not quite sure what you are saying here. The Cgs will appear as 1-(gain of output buffer) so if you have a .9 gain of output buffer 10% of Cgs loads the VAS. The problem is mainly that this is very non-linear.


They don't need selection to idss >3mA, they're just fine being used below their idss as the high Yfs types don't hit the diode conduction point until very current....

Again I'm losing you they are not being used below Idss if Idss is .5mA

BF861 and BF862 are fine, but sot23. They also have higher low frequency noise.

By measurement BF862 is better at 10Hz than 2SK170 or LSK170.

Søren