Discrete Opamp Open Design

Disabled Account
Joined 2012
and at 8mA bias and 600 Ohm load still only had .01% distortion.

The Sziklai pair has feedback and in my experience is much more prone to oscillate, but the added gain typically reduces the distortion by a larger factor.

I love that bootstrapped diamond output stage of your's. It's an instant classic.

I tried the Sziklai pair applied a little differently.... Most of the time the first transistor has at least a 100 ohm or greater resistor in its collector for local gain. This can lead to unstable operation. Instead, I used the pair just as compliments in an attempt to lower Output Stage distortion. Low gain so it is very stable (from just a 15 Ohm sampling R in the collector). Each Sziklai is made with matched compliments and have same emitter resistors and same currents thru each. Then run two pair in push-pull to any level needed (class A or AB). It just makes a super linear emitter-follower. Seems to work very well. Thx-RNMarsh
 
Last edited:
Look at signal-to-noise at a given frequency to determine how sensitive it will be to impedances, and then make the determination about how the filter components will load the amp output. Then fold in the loading thereafter.

Every situation is different, and I know of no rule-of-thumb. Lower Z is always lower noise, but depending on where the circuit is in the system you may be well-dominated by noise in the input signal already, so there may be no reason for heroic measures in the filter.

To be clear, SNR versus frequency for the opamp? Like this? One is at gain of 10 and other is configured as a buffer.

I'm assuming a pre->buffer->opamp->opamp->buffer->amp config.
 

Attachments

  • Screen Shot 2012-12-12 at 1.00.03 PM.png
    Screen Shot 2012-12-12 at 1.00.03 PM.png
    154.1 KB · Views: 541
  • Screen Shot 2012-12-12 at 1.02.32 PM.png
    Screen Shot 2012-12-12 at 1.02.32 PM.png
    146.6 KB · Views: 532
Last edited:
diyAudio Member RIP
Joined 2005
To be clear, SNR versus frequency for the opamp? Like this? One is at gain of 10 and other is configured as a buffer.

I'm assuming a pre->buffer->opamp->opamp->buffer->amp config.

So you are cascading two Sallen-Key filter sections with buffer ahead and after? Are you defining a large passband in this fashion, thus a low-frequency highpass followed by a high-frequency lowpass?

Circuits using opamps, and in particular filters, have "noise gain", and Sallen-Key topologies are no exception. Actually some S-K ones are quite low noise gain compared to other topologies. In any event, lower impedances (smaller resistances, bigger capacitances) to realize a given transfer function will always be lower noise (lower thermal noise due to the resistors), but will require more current to drive, both from a given driving source and from the opamp itself.

The noise gain, meaning the frequency-dependent multiplier of the opamp voltage noise (the opamp input current noise effects are treated separately), is dependent on, but not identical to, the realized overall filter transfer function. Higher Q will generally entail higher noise gain. For most audio work with lowpass and highpass filters, the Q will be modest (little or no peaking in the frequency domain transitioning into the stopband).

Off topic (you may stop reading now):

Boy it's nice to be done with jury duty. Six days, and at the end hopelessly deadlocked. Wouldn't you know two of the three for acquittal had engineering backgrounds ;)
 
Disabled Account
Joined 2004
My experience as well with Sziklai. And the poorer ability to turn off an output device translates into a dynamic asymmetry that is somewhat bothersome at high frequencies. Bryston has used a triple compound that they have managed to make work, but it looks very temperamental in simulation.

I used it in one of my xDSL drivers, it worked too well and the lowest of the 4 bias settings met all the specs, 400mA peak at 250uA bias and still maintaining -65dbc at 2MHz. Pain to stabilize though. It was too bad the lifetime was short because it was true differential and everyone decided they wanted plain dual op-amps (multi-source) even though the application suffers from common-mode oscillation/peaking issues.
 
So you are cascading two Sallen-Key filter sections with buffer ahead and after? Are you defining a large passband in this fashion, thus a low-frequency highpass followed by a high-frequency lowpass?

that, or two LPF or two HPF, etc.

Thanks for the explanation. It would still be nice to have a rule of thumb. It sounds like the answer is to construct a model for the entire filter and evaluate it en masse. Perhaps, stepping through the model over multiple xover frequencies. Drop in another opamp model and repeat, rinse, etc.
 
Last edited:
Disabled Account
Joined 2004
Nice ref.

It does contain the most common mis-conception about CFA's. I rewrote Black's basic equation for VFA's and CFA's to create equivalent terms and it is clear that the CURRENT into the inverting input is the feedback signal.

Or take the simplest case, a follower, in a VFA and instantaneous step appears at the input and the voltage across the inputs determines the current delivered to the gain node, in a CFA the instantaneous current is determined by the output voltage across the feedback resistor or the current into the inverting input.
 
Last edited:
diyAudio Member RIP
Joined 2005
It does contain the most common mis-conception about CFA's. I rewrote Black's basic equation for VFA's and CFA's to create equivalent terms and it is clear that the CURRENT into the inverting input is the feedback signal.

Or take the simplest case, a follower, in a VFA and instantaneous step appears at the input and the voltage across the inputs determines the current delivered to the gain node, in a CFA the instantaneous current is determined by the output voltage across the feedback resistor or the current into the inverting input.
Good point.
 
Disabled Account
Joined 2008
It does contain the most common mis-conception about CFA's. I rewrote Black's basic equation for VFA's and CFA's to create equivalent terms and it is clear that the CURRENT into the inverting input is the feedback signal.

Or take the simplest case, a follower, in a VFA and instantaneous step appears at the input and the voltage across the inputs determines the current delivered to the gain node, in a CFA the instantaneous current is determined by the output voltage across the feedback resistor or the current into the inverting input.

Absolutely correct.

BTW: I have designed two new discrete op amps (or building blocks), one CFA and one VFA. They both have (to quote Brad from the BH thread) "precisely equivalent positive and negative slewing/settling behavior"

I have also designed two new preamps using these building blocks, the simulated results are very promising. Have to build them to confirm the simulated results.
One has balanced differential input, balanced "singel ended drive differential" output. The other has balanced differential input, balanced differential output. Both can of course also be used as se input, se output, se input, diff output and diff input, se output.

Attached some simulated results:
Preamp 1: (10V pp sinus out into 600ohm (less than 0.1ppm THD))
Preamp 1: 20k Square se out
Preamp 2: 20k square diff out

Both are using BF862 with BF545C as cascode


Cheers
Stein
 

Attachments

  • Preamp 1 10Vpp into 600ohm.pdf
    39.1 KB · Views: 179
  • Preamp 1 20k square se out.pdf
    29 KB · Views: 97
  • Preamp 2 20k square diff out.pdf
    31.6 KB · Views: 79
Last edited:
diyAudio Member RIP
Joined 2005
On this general subject of current versus voltage as variables, I was reminded of some help I gave to a visiting Bell Labs physicist at UCLA whose specialization was plasma physics. It was a very strategic bit of benevolence on my part although I didn't know it at the time.

He was lamenting, as I loaned him my copy of the RCA photomultiplier tube handbook that was missing from the engineering library, that he just couldn't get his equipment up and running for long, if at all. So I paid a visit to his lab and looked over some of the setup. It was difficult to suppress a few chuckles, as there were so many things effectively booby-trapped. For openers, the high voltage power supply was a huge and quite lethal thing capable of iirc a few 100mA up to several kilovolts :eek: Yet it was to be used to drive dynode voltage divider strings for the PM tubes, which drew at most a mA or so.

Then, the tube base he had turned out to have a 50 ohm termination resistor buried inside, anticipating pulse-counting photometry, but he was doing low-frequency "d.c." photometry. He initially didn't believe me when I said I measured 50 ohms. We got rid of that, or found a tube base that didn't have one.

But the funniest part: all of the connectors in the system were BNC. High voltage, opamp supply voltages, signal inputs, signal outputs :eek:. So of course the grad students were frying everything on a daily basis.

Anyway, I said Let me make something up for you that's a bit more bulletproof. I wired up an opamp (I think it was an ADI module) and used a feedback R and a little feedback C, tailored to the cable capacitance. I found an incompatible connector for the HV, and reserved normal BNCs for input and output. And I put a little safety resistor in series with the input, along with some clamp diodes to common ---I think it was 1k. If someone still managed to get HV to it, it would fuse before the opamp blew.

The problem was, that the physicist could not wrap his head around the appearance of the schematic, which, with the input protection resistor, looked like a voltage-input inverting amplifier with tons of gain (I think the feedback R may have been 1Mohm). I explained repeatedly that a PM tube had about the highest intrinsic output impedance of anything in routine use, and in that sense looked very close to an ideal current source. But even so, if he wanted to persist in viewing the amp as having a voltage gain of -1000, that still "worked": the, say, -10uA of phototube current would develop about -10mV across the 1k, and the gain of -1000 would then cause the output to be about +10V (given adequate opamp open-loop gain). It wasn't the most felicitous way to view things, but that explanation, and the fact that everything worked and had good signal-to-noise ratio, seemed to satisfy him.
 
... I could volunteer to look at higher output bias version. As you can see in my prototype, I'm using DIP14/16 heatsinks to bolster the heat dissipation of a SOT223 device. Does anyone have a recommended complementary SOT223 device? I have tried using BCP53-16 and BCP56-16 from On ...

Marc, I hope you haven't frozen your design. As I posted earlier, the NXP BCX53/56 are available in 0.5 W SOT-89 packages.

I added a few circuit elaborations to the basic folded Kaneda, and the performance improvement is dramatic.

1. Added a Class-AB push-pull buffer to reduce loading on the totem-pole VAS.
2. Added a Hawksford Cascode to the PNP driving the upper NPN of the totem pole - this is the game changer for the VAS.
3. Cascoded the LTP JFETs - contributes a few dB in THD20 reduction.

It's now at 13 actives, but all the additional devices are worth it. Compensation tweaks, clipping/overload, stability, etc. haven't been investigated.

Here's the basic circuit, which I'll call the Folded Kaneda-Hawksford.
 

Attachments

  • kaneda_hawksford_5jf.asc
    7.4 KB · Views: 56
  • kaneda_hawksford_5jf.jpg
    kaneda_hawksford_5jf.jpg
    89.4 KB · Views: 465
Last edited:
... Can you post your transistor models too.

The JFET model is the same as the 2sk117/184/209/2145 posted earlier - you can use the 2sk170 model in a pinch.

Here are the BJT models, downloaded from Infineon and Rohm respectively.

Edit: The output stage models are not critical - I cross-checked with NXP BCX56 and BCX51 models: Icq is within 15 per cent, and THD20 is within a few dB of the Infineon models attached.
 

Attachments

  • kaneda_bjt.txt
    3.2 KB · Views: 71
Last edited:
diyAudio Member RIP
Joined 2005
But even so, if he wanted to persist in viewing the amp as having a voltage gain of -1000, that still "worked": the, say, -10uA of phototube current would develop about -10mV across the 1k, and the gain of -1000 would then cause the output to be about +10V (given adequate opamp open-loop gain). It wasn't the most felicitous way to view things, but that explanation, and the fact that everything worked and had good signal-to-noise ratio, seemed to satisfy him.

The need for loop gain in this particular case had more to do with a fairly low open-loop input impedance for the opamp, not because of the apparent closed loop gain. Scott W., do you recall a module from ancient times that was a sort-of-general-purpose "budget" part? I think the number was the AD118, but of course that number has probably been recycled a few times since --- I couldn't find anything on the web. Maybe it was advertised in Analog Dialogue.