Things to consider: * Current source in place of R6 may help with increased gain and linearity * Darlingtons can be a bit slow to turn off, you can try adding a 480 ohm across BE of Q1 and Q4. Decrease value to taste. * ever heard of a Sziklai pair? Basically it's a darlington but with a base-emitter voltage close to that of a regular transistor. They can be a bit tricky to get stable though. * Given how cheap opamps are, I would have probably cheated and used an opamp amplifier in place of Q2 stage. If it has sufficient slew rate and bandwidth, it will fix any nonlinearty of your output stage. The effects of crossover become limited by the slew rate and large-signal behaviour of the opamp as the crossover will cause a brief period of open-loop while the output stage crosses over. With nearly infinite gain and system gain defined by resistors, output linearity becomes nearly infinite at least in theory. * replace the darlington output stage with N channel FETs, drive them with a bootstrap gate driver IC with a PWM controller, replace C4 with inductor and place capacitor across output load. Congrats you have an irrefutably better class D amp thank me later. :)
@@EEforEveryone I do eventually plan to build such a class B /AB amplifier and maybe experiment with class D's that have taken over the market at this point, and drive some DIY distributed mode speakers (Tech Ingredients channel) to have a true HiFi surround sound system one day😁
More things to consider :P .... (beforehand sorry for the caveman english level ) 1. The input (AC) impedance of this circuit is too low for an audio amplifier wich is usually around 10K to 100K in A.F. band, probably a class A input buffer is needed(or a diff. input stage ideally). 2. Maybe it´s not finished yet but you need a Low Pass input filter for RF (cutoff freq isn´t really critical, the idea is to eliminate the most possible HF content above 20KHz as you can without getting attenuation in the AF band). Also C2 seems not big enough considering the sub 1KOhm input impedance, usually you don´t wan´t a very low cutoff frequency for the HPF to not risk the speakers. 3. The power disipated is too much. Strictly speaking this is not a class AB amp, it´s a: "Soft Non-Switching Class AB amplifier": basically for a good part of the output power it works in class A. The output stage it´s over-biased and you get gm doubling distortion (the output impedance varies between 1 emitter resistor (Re) and 1/2 Re -or less- when the output signal goes from 0V to Vmax). *3.a. The bias voltage should be set to a point where you get a emitter resistance of about one to one half the value of the degeneration resistor conected to the emitter terminal, this is the Oliver criterion in order to avoid static crossover distortion (gm doubling) ... in other words the voltage across each emitter resistor should be around 13mV to 26mV (but this only really makes sense complemented with a good thermal tracking for Vbias). 4. With the bias corrected you can/should put effort into making the collector current of Q2 as stable as you can (maintaining it low to reduce early effect and Cbc impact in the distortion performance), for that a boostrap in this diode-resistor branch or a current source will be needed. The reason to do this is to have a stable the Vbias voltage and avoid modulating the voltage gain. *4.a. Even with a bootstraped current source for D1,3 and R5 a electrolytic capacitor is needed to stabilize Vbias voltage. 5. Diodes, altough better than nothing, provide a really poor thermal track for transistor. Have you checked ThermalTrak BJTs?. When biasing using discrete diodes try to put them as close as you can to the driver or output transistors in the PCB. 6. A resistor across the base of output the transistors is needed, otherwise the load impedance of your common emitter amplifier will vary with frequency in the same way the load does, modulating your voltage gain and that will cause a rise in distortion due to the variying loop gain. A Darlington pair tends to be really slow to turn-on and turn-of due of the increased input capacitance, and this is bad for a class B/AB system, the usual is to employ cascaded emitter followers because the primal idea is not so much the current gain, but to buffer the load and its variations with freq from the VAS.. 7. This NFB scheme reduces the input impedance (i bet you´re getting something around 850 Ohm |Zin| at best ). Appart from that: have you simulated how does the different parasitic BJT capacitances vary with operating conditions and how that affects PM a GM?, maybe some form of compensation will be needed just in case. 8.Zobel newtork? RL filter? these are needed to ensure HF and VHF stability. 9. CE voltage gain can improve using a current source, or bootstrapping the R-D branch, and reducing R9 wich seems too big. Optimal value should be around 10*emitter resistance at Icq and less than 100 Ohm. Because this is a NFB system so you really want A-> infinity to increase the return ratio in order to deal with harmonics, specially the HF ones. 10. (Probably a subtle point) Separate the supply rails of the voltage amplifier stage from the output stage using an RC filter to improve PSRR, low frequencies clipping and THD. 11. Have you considered a CFP output? (Szicklai Pair), because it offers a really big improvement in T.H.D. compared to the darlington used in here (about 4~5 to 1 at best) at the cost of an added base stopper resistor (and some test to ensure you don´t have VHF oscillations going on) Additionally: a. 2Vrms is a really big input you can get that only from a pre-amp, the common maximum output from a professional audio equipment is about +4dBU (~1.22Vrms) and for domestic equipment about -10dBu. b. Saddly, the spice simulation can be way off....just compare your models with datasheets... particularly those power transistors........specially if they came from ONSemi XD. c. You can simulate T.H.D. in LTspice using .four directive and .step over freq (a list of freq = 0, 500, 1K, 5K, 10K and 20K will give a good enough T.H.D. vs freq plot), but dont forget to nullify large time constants and wait for a settling time of at least 10 input cycles before collecting data. There is a great YT chanel called "SW audio" that shows the usual work flow with audio amplifiers in LTspice.
Hi Luciano! First off, awesome thoughts!!! I was so sad when I found this right after finishing our latest video where we responded to the comments. :( This would have been great detail to add, and there's a wealth of great information in your words! I hope that folks find this comment, because it is an absolutely great resource, and points out a lot of flaws in our design. I hope that I find some time to sit down and reply to this comment properly! Thankfully, we didn't miss all of this before we hit fab!
@@EEforEveryone No problem, i didn't know you respond comments in a separate video (which i really appreciate because it shows real compromise to your community) so i really never expected a response XD. P.D.: Now that i´ve re-readed my comment seems that some clarification is needed: when i said that "Darlington pairs are slow" i was refering to frequencies over the AF band. In low power amps they´re usually are not a problem, unless you get some intermodulation/or you´re aiming for a REALLY low THD, but because it´s one of those thing that depends on many factors usually they´re avoided.
Class AB efficiency really depends on how much you overbias or underbias the output stage, AB is basically a class A amplifier that "jumps" into class B when it needs more power. If the bias current and bias voltage is too high it will be more class A than class B.
