opamp IS, CFP VAS, lateral FET OS

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Several moons ago I designed a PCB for one of my experimental amp designs, and at some point braved ordering parts and sending the pcb design off - today the PCB arrived in the post and I've spent a fair few hours putting it together and doing the very first, low power, tests.

The principle is to take advantage of an opamp as both input stage and to close the feedback loop.

The limited voltage output of standard opamps means using a voltage amplification stage to magnify the opamp output somehow, and I've selected a class B CFP stage with gain of about 5, using 4 small-signal transistors.

The output stage is a pair of Exicon lateral MOSFETs, avoiding the issue of thermal compensation for the bias - less complexity with a new design seems a good idea.

R6 and R14 form the feedback network for the top CFP pair, setting the gain, and the bias voltage is amplified along with the signal by the CFP sections. The bias arrangement is a bit rough and ready I think, but emulated OK.

The implementation is nearly all surface mount, on a board 3" by 1.5".

So far I have got the amp basically working, with the minimum of bias needed to prevent instability, about 15mA are taken overall I think. The bias pot I chose was 11-turns which turned out to be a wise choice, but I had to hunt around for a small enough screwdriver to adjust it!

Unloaded it seems to work fine so far (using siggen + 'scope), I need to arrange some form of heatsink before going any further. Its always a relief when a new board basically works without modification.

This design predates my opamp-tripler circuit (see this posting: https://www.diyaudio.com/forums/sol...stacking-voltage-operation-4.html#post5954865)
I will be building that on a PCB that's also just arrived at some point (parts need ordering). It will be interesting to compare the two techniques for leveraging an opamp as integral part of a power amp.

In retrospect I should have use DIP8 for the opamp in this design to allow experimentation with different opamps.

I think the choice of the discrete transistors is fairly non-critical in this arrangement which is a bonus as good audio transistors are getting harder to source. Great opamps at bargain prices are too good a resource to ignore!

[ oh yes, and I know the input network is DC coupled, I think I forgot to fix that ]
Last edited:
Looks nice.
Found your jewelers screw driver kit for that pot
A simple bare Cu wire formed around the SOIC8 leads should make it easy enough to remove. that and some solder wick to clean it up. maybe not as easy as a dip socket.
left it kinda tight to get an iron in there = hot air machine :)
some aluminum for a HS and you are off to the races
good luck with the design
Last edited:
thanks for sharing.
looks like fun :)
just curious: if you investigated at all the the Acoustat/Hafler (Transnova) approach with the floating supplies and let the MOSFETs themselves do the heavy lifting?


Its interesting, but for multi-channel amps the need for multiple supplies seems a show-stopper, being way more expensive than a few small signal transistors needed for a VAS.
Nice design. Can't understand why output stages with gain are demonised; in practise they work pretty well, especially with mosfet outputs where the greater than unity output stage gain gives a degree of hf stabilisation. Designed and built something similar to this a while back - sold a couple of hundred of them and they were pretty well received.
Nice design. Can't understand why output stages with gain are demonised
Because they tend to distort. The hope here is the opamp has enough gain to counter this. In a good 3-stage amp the IS and VAS are very linear to start with and the feedback is mainly working at reducing the cross-over residual from the OS.

A unity gain CFP seems to be about -80dB distortion without effort for instance, but adding the gain resistors eats into the local feedback that gives that performance, and they also dissipate significant power (in a CFP OS this might be 10W or more, here being only a VAS/driver stage its manageble)
So I've mounted the output devices on a copper strip I had, not a great heatsink but enough thermal mass for testing at higher powers. I'm limited to 0.5A from the +/-45V supply of mine, so with 8 ohm load I'm limited to about 7Vrms (6W) output.

I've measurements from the feedback divider output and from the output terminal. Due to less that ideal layout there's distortion from the feedback network take-off point not being separately routed from the output terminal - I'm improving the PCB layout for the next version.

Anyway here's the 1kHz plots (absolute dB values aren't calibrated):


The green trace is from the output terminal via a breadboarded 11:1 divider, the red trace from the on-board feedback divider output, blue is loopback for the Focusrite Solo.

The technique of probing distortion from the divider output node is useful as there is no gain or attenuation, and its easy to share the ground point with the injected signal. However it does assume the feedback components are linear.

Anyway I've written a bit of Python to analyze distortion and plotted:

The residual is in ppm of the fundamental plotted over a single mains cycle period - considerable hum and noise are present, I do have a fair bit of breadboard spaghetti and nearby mains cables...

The red lines are the individual harmonics directly correlated from the samples, the green spectrum is a Welch-style spectral plot with a higher noise floor. The horizontal red line is the THD, horizontal blue is the noise and distortion.

The calculated THD of 0.00173% is by summing the harmonics, the SINAD is computed from the rms of the entire residual.

So overall I'm reasonably happy - I should have addressed the feedback take-off point in the layout. 10kHz data will be next.
For 10k testing I choose 10k/11k intermodulation rather than chase harmonics beyond the audio range.


This time green is the loopback, red from the on-board feedback divider output, blue is via the output terminal and external divider (lower noise floor as lower divide ratio).

Similar results in that the take-off point issue pushes up the distortion for the output terminal, basic distortion level is comparable. For the on-board feedback node the detailed peak levels are:

Again this is at a few watts output into 8 ohms.

I've yet to write some Python code for analyzing two-tone distortion data automatically, its a slightly harder problem than for single-tone.
I'm limited to 0.5A from the +/-45V supply of mine,
I can understand using this supply for initial bring it up, but why not use or get something that would be closer to what the final design would use = a std unregulated supply with big ecaps? that way you could listen to it too, as far as evaluation goes.

Interested to know what you are using for test equipment/software?
Because at this stage I'd like for it not to burn up - with a new design its best to proceed cautiously and be on the look out for problems before they become expensive ones. Modest current limits really save you time and money, I'd rather be building the second design to compare against this before taking things up to high powers.

Currently the circuit has no current limiting on the VAS transistors which is something I would fix with a re-spun pcb before using in earnest. I'd like to figure out all the issues and fix them all in one iteration.


'Scope, Focusrite Scarlett Solo, Audacity and R.E.W. to drive the thing, but most of the analysis I do offline with Python, easy to code up whatever I want using scipy.signal library, very powerful.
Okay makes perfect sense to proceed with caution, those fets ain't cheap, I know I burnt a few Hitachi's of late, now I have to replace them. Having a PS with current limiting is a good idea.
Even when I am testing out something that has been wrung out, still nice to power up with a DIM bulb tester, (except when I burnt the Hitachi's = zealous idiot) it has a switch to bypass the lamp, once the lamp dims & I know there is not a gross failure. Even saves me from having to change a fuse.

I'd like to learn about your testing environment, sounds interesting, for another time, I got enough on my plate right now.

I am still working with an old Amber 3501 audio test set, quick n easy for regular THD+N and some simple IMD measurements but it is nice be able to look at the spectrum for sure. Maybe one day I'll have that capability. In the mean time I'll look up the items that you mentioned, out of curiosity.
Bob Cordell had mentioned to me he has a QuantAsylum QA401. I am going to build up a few of Victor's oscillators too (Home | ULD audio)

As a suggestion, make use of the 100x100mm area for the ultra cheap pcbs that they are offering, so that you are not so tight on space, esp for a prototype, can always scale it down after all the kinks are worked out.

Cheers Rick
The TO264 Exicon FETs are not actually expensive at a few pounds each - remember each lateral replaces one BJT output transistor, one BJT driver transistor one power resistor, half a bias compensation transistor, and requires less pcb area and less mounting hardware/heatsinks.

Overall cost is probably about the same.
Disabled Account
Joined 2002
Hi, a dual opamp would make choice in opamps wider and the second opamp could be used for DC servo. An 5 mm pitch film input cap plus bleed resistor would be nice, often the power amp is the place where DC blocking is wished for in the chain (to avoid defective woofers).

Instead of using Zener diodes voltage regulators would be nice. Or a combo of Zeners followed by regulators. When using for instance OPA454 the supply voltage of the opamps can be quite high...
Last edited:
The TO264 Exicon FETs are not actually expensive at a few pounds each
I wish they were only a few pounds :), TO-264 package is for the dual die, £10.70
£5.17 for the single die in the TO-247 is not too bad, but one would need four of them for a higher power app.
In the Cordell DH-220C design I am working on with Bob, he uses a driver between the VAS and the fets, but the KSC3503/KSA1381 are pretty cheap and are used for the VAS as well.
The output devices cost a lot less than the power supply though, keep things in proportion!

Yes, sorry, I got TO247 and TO264 mised up there.

Zeners seem fine for generating +/-18V rails in an amp, simple, reliable, cheap, and the series resistors protect against over-current if the opamp goes west. Shunt regulation has its place and this is a good example.
Disabled Account
Joined 2002
Yes Zeners seem fine but fail more often than one likes and most of them are noisy. When they fail in this design (either as open circuit or a short circuit) results are fatal for woofers as no DC protection has been implemented. The chances that an opamp fails are very small.

I once worked for a manufacturer that stopped using Zeners (after a few recalls) and went for extra secondary windings with a slightly higher voltage for the VAS. This voltage was simply rectified and filtered with RC filters. Later they used cap multipliers.
Last edited:
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.