In one of the recent posts I mentioned that my next design will be CFA.
Here it comes. First diagram shows the basic concept (well, pretty well-known architecture). Devil in in details.
Some time ago I got an idea of using an OTA (operational transconductance amplifier) topology for an input buffer. OTA is based on current mirrors and acts as a highly linear voltage-to-current convertor. Which is very suitable for CFA. This design is also very suitable for symmetric differential input (no global NFB at this stage makes it very convenient), though it can be used as non-differential as well - just use one of two inputs. Looks pretty elegant to me.
This design shows rather good performance in simulator - minimal compensation, high frequency responce, low phase shift at 20kHz, low THD-frequency dependence.
I started prototyping - already developed PCBs for the prototype - waiting for them to be produced. OTA buffer and the rest of the amp will be built on separate PCBs, allowing to experiment with different designs in the future.
If somebody is interested, I will publish the news on this project here.
Cheers,
Valery
Here it comes. First diagram shows the basic concept (well, pretty well-known architecture). Devil in in details.
Some time ago I got an idea of using an OTA (operational transconductance amplifier) topology for an input buffer. OTA is based on current mirrors and acts as a highly linear voltage-to-current convertor. Which is very suitable for CFA. This design is also very suitable for symmetric differential input (no global NFB at this stage makes it very convenient), though it can be used as non-differential as well - just use one of two inputs. Looks pretty elegant to me.
This design shows rather good performance in simulator - minimal compensation, high frequency responce, low phase shift at 20kHz, low THD-frequency dependence.
I started prototyping - already developed PCBs for the prototype - waiting for them to be produced. OTA buffer and the rest of the amp will be built on separate PCBs, allowing to experiment with different designs in the future.
If somebody is interested, I will publish the news on this project here.
Cheers,
Valery
Attachments
Last edited:
Hi, at the moment I use a sound-card based solution for examining the spectrums and measuring THD. It gives me resolution up to roughly -110db (good pro-grade card e-mu 1616m). I can only measure at 1kHz at the moment because of the software license limitation, though I am going to solve this issue in the nearest future. I used it a lot with different designs, so if I see harmonics somewhere at the noise level (like at the spectrum attached - my previous VFA design) - it will be ok 🙂
I've got a pretty good "hardware" generator and digital oscilloscope with FFT as well, but the picture there is less precise, so I use the card so far...
Kind regards,
Valery
I've got a pretty good "hardware" generator and digital oscilloscope with FFT as well, but the picture there is less precise, so I use the card so far...
Kind regards,
Valery
Attachments
Prototyped. Awesome amp!
Hi All,
Yesterday finished my CFA prototype.
Final schematic is attached. Also see the Bode plot (phase shift @ 20kHz is only 5.3 degrees) and 50kHz square wave form (also pretty small phase shift).
Tested with non-balanced input.
Extremely natural sound, especially audible at fast fronts (drums, bells) and acoustic piano (Steely Dan's "Two against nature" sounds better than with any other amp I listened to in my environment so far).
More than satisfied. Next step - will try it with the input buffer from Damir.
Cheers,
Valery
Hi All,
Yesterday finished my CFA prototype.
Final schematic is attached. Also see the Bode plot (phase shift @ 20kHz is only 5.3 degrees) and 50kHz square wave form (also pretty small phase shift).
Tested with non-balanced input.
Extremely natural sound, especially audible at fast fronts (drums, bells) and acoustic piano (Steely Dan's "Two against nature" sounds better than with any other amp I listened to in my environment so far).
More than satisfied. Next step - will try it with the input buffer from Damir.
Cheers,
Valery
Attachments
Last edited:
Valery,
For me it looks as typical VFA, unusual but with symmetrical LTP. Why do you call it CFA, please explain?
BR Damir
The right side of it is actually a voltage amplifier (VA), driven by current. The input buffer is a v-to-i converter (output is current). The point where the buffer output, VA input and left side of R39 come together is the current summing point (as outlined at the attached diagram).
What I plan to test - is to exchange my input buffer with your gainwire (without your output buffer). In this case your output will also be current. Simulation showed even higher linearity of the whole system.
I thought about it from the very beginning, so I made the buffer as a separate module, so I can change components easily now 😉
Attachments
I haven't build CFA power amp yet, except that line amp, and I suppose that both type VFA or CFA could sound equally good. I very interested to see how you are going to implement the GainWire gain part(it is really current amplifier with v-to-i input convert and current conveyor as the gain stage).
I don't understand use of R39 in your schematic, could you explain?
Thank you in your interest in the GainWire.
best regards
Damir
Last edited:
R39 is shunt-shunt feedback across the output stage half. It reduces both the output and the input impedance, while setting the the transconductance gain of the output stage to R39 (at LF). The total voltage gain is the output stage input impedance (R39=47k), divided by the input stage degeneration (R4=1.8k), since the input stage is a transimpedance stage.
But then by any accepted definition, I don't think this is anywhere close to a CFA. By the same logic used by the author, a classic VFA is also a CFA, since the input stage is transconductance and the "VAS" is in fact a transimpedance stage.
Thanks Waly, but isn't it a kind of positive feedback as it is connected between output and non inverting LTP input?
Last edited:
Hi Waly, great description with regards to R39's role, I would not do it any better, thanks.
With regards to VFA / CFA - it very much depends on how exactly you arrange your global NFB. Your example with classic VFA is correct, but you don't connect your global NFB to the point between the input stage (transconductence) and VAS (transimpedance). You connect it to the input stage's input (funny expression 🙂) and this is what makes the whole system VFA.
In a classic CFA you connect it to the input buffer's output - this is what allows you current sense at that point, so overall gain is controlled by current.
In the topology I used there is no diamond buffer, which you normally see in CFA designs, however it is based on absolutely the same principle.
It is well known that in CFA bandwidth does not depend on the closed loop gain in a wide range of gains. You can see what it looks like in my case on the pictures attached (these ones are simulated, but my practical measurements showed exactly the same). C3 = 1.8 pF, in this case the bandwidth is close to 1 MHz and it's independent from the gain. Testing with the square wave showed a slight overshoot with C3 = 1.8 pF, so in the final design I use 6.8 pF - it reduces the bandwidth to roughly 200 kHz, but the square wave looks beautiful in this case. Phase shift is also very light and slew rate is rather high. Bode plot in the earlier post is a practical measurement at the prototype.
Attachments
Would it not be better to drive the inverting input at the OPS with a fixed and known resistance to GND, then R39 (47 Kohms) and that resistor would set the gain..? with the derived voltage over the resistor added..Frontend runs Non feedback, and the main distortions in the OPS is handled by the inverting shunt FB. Rather clever. and clearly thinking out of the box.
Michael, could you please sketch it briefly for better understanding of the arrangement ) You can use schematic picture from post #5 and just "paint" on it. I will simulate the result.
Thank you,
Valery
Thank you,
Valery
- Status
- Not open for further replies.