How to design an audio power amplifier circuit?

Question

Design an audio amplifier that takes input from a small microphone and can drive a small speaker. A condenser microphone can be used. The audio amplifier must be based on discrete devices and audio transformers may be used if needed. However, op-amp or IC amplifiers CANNOT be used. Moreover, base your design on locally available MOSFETs and BJTs. Justify the selection of transistors for each specific stage. Design Specifications:

1. AC output power = 2 Watt (8 Ω Speaker) (Ground Referenced preferably)
2. The amplifier's frequency response should be 100 Hz to 20kHz or above at 3 dB level.
3. The audio amplifier should qualify for the given experimental testing standard. DC Power Supply: ±12 V (dual power supplies only allowed in the last stage of the pre-amplifier stage) Overall Amplifier Gain: 1000 V/V (60 dB) Pre-Amplifier Specifications: • Zi > 100kΩ if BJTs and 1MΩ if MOSFETs, Zo ≥ 5kΩ Power Stage:(Only Class AB Stage) • RL = 8Ω • Vo = 5 Vrms

Image 1 Description: This is the Oscilloscope View of the output signals. There are 3 BJT Pre-Amp Stage and Class AB with Driver Stage Power Stage. The red signal is output after 1st transistor, Blue after 2nd, Green after 3rd and Yellow at the Output.Do note that this is the oscilloscope view for a 1k sine wave input with 5mV amplitude or 4mV RMS.

Image 2 Description: This is the Frequency Response of the Circuit. 20dB Gain after the first transistor, 40dB after 2nd and 60dB after 3rd. The Pink line is the gain at the output.

Image 3 Description: This is the pre-amp stage. The first two BJTs are biased based on the 1/3 Vcc rule. These are connected with only a +12V power supply. Meanwhile, the last transistor is biased so that it allows a +6V and -6V signal swing. It has +12V power supplies attached to all stages and a -12V power supply only attached to the final stage of the pre-amplifier stage. The capacitors have been set so that the Fl is set at 100Hz. I have tried to achieve impedance-matching of the Rout of one stage with the Rin of the next. The coupling capacitors have resistors in series so that the biasing remains intact but the gain can easily be modified.

Image 4 Description: This is the Class AB with input going in after passing through 20uF capacitors. There are 15k resistors for the biasing of the BJTs. There are two TIP22(NPN) at the top connected in Darlington form and two TIP127(PNP) connected one after the other in Darlington form. The biasing has been done by 6 diodes. A 5 ohm feedback resistor is attached between the power stage input and output.

Question: I want to know why when I send a signal of 1kHz or less, the sound is clear but as soon as the frequency goes higher the sound becomes like pulses or it starts sort of motorboating. This is apparent when I just play sine waves on a speaker. And when I play a song, some low-frequency sections play okay but others are motorboating/distorted. I have tried decoupling capacitors etc but that just makes it worse. I would need a complete explanation and calculations of any changes involved or for any new values of resistors and capacitors.

Answer 1

your questions

Why is the signal having an offset and improper signal swing after the preamp stage

I think you mean the green curve and that this trace was taken at a point just after the series output capacitor and before your power amplifier stage. If so, then the reason it is biased with a DC offset is because of \$R_{16}\$, \$R_{17}\$, and \$R_{18}\$ in your power amplifier stage image. That resistor divider does work out pretty close to your DC bias in the green trace.

and why is the signal so distorted at the output?

There's an RC time constant operating here on one side of the cycle (no active device doing anything) and an active device working the other side. It's hard to know anything more, because we don't know if the output is diminished (attenuated) so the exact RC timing is obscured from us.

My guess would be your \$R_{26}\$, \$R_{27}\$, and \$C_{11}\$ are the cause. This could cause one half of your 2-quadrant drive to go inactive and that would explain the RC timing.

It would be easy to test. Just change \$C_{11}\$ by a factor of 10, either way.

What seems to be the issue if I biased the class AB at 20mA and have resistances of 565 ohms and 0.7V biasing of TIP31 and TIP32?

I'm not sure what this means. I almost think I do, then I find I know I don't. So I'd need more explanation from you.

a thought about the pre-amplifier stage

For very high input impedance with BJTs (I prefer JFETs here... but just saying), I really do like what Elliot presents in Figure 4 here. It's a beautiful design. I like it still more when I see it each time, again.

Let's set it up to meet the indicated high impedance:

^{simulate this circuit – Schematic created using CircuitLab}

The above schematic does meet the input impedance requirement that you were given. \$I_{_{\text{C}_\text{Q}}}\approx 200\:\mu\text{A}\$ and the output impedance of this circuit is \$\approx 1.1\:\text{k}\Omega\$. (If you really need \$5\:\text{k}\Omega\$ then just add a \$3.9\:\text{k}\Omega\$ series resistor to the output.)

This gets halfway there: \$30\:\text{dB}\$. The rest can be done in a simple class-A or class-AB amplifier stage. This breaks up the circuit into two explainable parts: a pre-amplifier with gain (as above) and a power amplifier with gain (unshown.) Half of the total gain performed with one, half with the other.

Here's the output using LTspice:

I'm using a \$6\:\text{mV}\$ peak input signal as this is about what your input should be given the \$60\:\text{dB}\$ gain specification and the output power into \$8\:\Omega\$.

The output above includes overlapped outputs over the temperature range from \$-20^\circ\:\text{C}\$ to \$+45^\circ\:\text{C}\$ and using BJTs that vary in \$\beta\$ by a factor of \$2.5\times\$. It just works. No cutoff occurs.

I did add a \$10\:\text{M}\Omega\$ resistor load at the output so that the capacitor there wasn't floating. But that's all. So it's the circuit as presented above.

Here's a simulation that uses a very cheap electret microphone:

That is using a typical and cheap low sensitivity electret together with an ambient SPL of about 60 (a normal conversational level of audio.)

By the way, I said I really like Elliot's design.

complete design

You've said that you don't want the speaker to be bootstrapped. So I'll toss out that idea. (The benefit is that I can use one resistor instead of using a current source structure. But that's your call here.)

You've added that you must use a class-AB. So I'll go with that, as well.

I've already provided what I consider to be a very good voltage amplifier topology, above. It will work using varying supply voltages (constrained by their choice, but without failing to continue operating reasonably well, regardless.) It performs the same over widely varying temperatures of operation. It performs the same over widely varying transistor parameters. It's just... good.

So why not re-use it as far as possible?

Because there is so much voltage overhead to work with, I'm going to go with Sziklai pairs for the class-AB power quadrants and I'll use the usual \$V_{_\text{BE}}\$-multiplier to maintain their voltage separation.

But my goal here is to illustrate re-use of good design sections.

Let's look at a Elliot's design but this time as a block diagram:

^{simulate this circuit}

All I did was to hook up the NFB input of the left section with the NFB output of the right gain-setting section and then tie ISRC, FB, and OUT together. And that was it.

But these are still useful individual blocks and they can be used with the class-AB stage without change!!! (Except, possibly, adjustments for the reference voltage and/or current source magnitude and associated small details. The block concepts remain.) These simple building block concepts are just as useful there as they were in the pre-amplifier. So this is exactly what I'm going to do.

Here's what the resulting block diagram will look like:

^{simulate this circuit}

Note the re-use of good tools. We only need to add the creation of a \$V_{_\text{BE}}\$-mulitplier and two quadrants of class-AB drive for the speaker. The rest was just replication of what has already worked so well.

Let me fold things together a bit:

^{simulate this circuit}

It's still the same block diagram. And the voltage rails can be changed to, say, \$\pm 8\:\text{V}\$ and the output will be just as before. Same gain, nearly the same THD, etc. It handles power supply ripple with ease. And you can change out various BJTs with varying key parameters on \$\beta\$ and \$I_{_\text{SAT}}\$ and it still works just as before. It is elegant.

Here's the final schematic:

Here's the bode plot:

(Note: I reduced the pre-amplifier input capacitor by a factor of 1000, to about \$10\:\text{nF}\$, to dampen the gain at very low frequencies and produce the above bode plot.)

At \$1\:\text{kHz}\$ the gain is:

So for the \$-3.0103\:\text{dB}\$ points we are looking for \$57.33\:\text{dB}\$.

I've circled these approximate \$-3.0103\:\text{dB}\$ points. So you can see the bandwidth. This betters your specification on the high frequency end. You didn't provide one for the low frequency side. But you can see what it is here. Roughly speaking, from \$50\:\text{Hz}\$ to \$22\:\text{kHz}\$.

And here's the input and output signals for a \$1\:\text{kHz}\$ input signal:

A nice gain of about 1000.

Beauty and functionality!

One remaining detail. Apparently there's still that \$5\:\text{k}\Omega\$ pre-amplifier stage output impedance specification. To solve this (if you need to) then stick a \$3.9\:\text{k}\Omega\$ resistor in series with \$C_4\$ in the full schematic I provided above and call the first stage the pre-amplifier and the 2nd stage and class-AB 2-quadrant drive as the power output stage. That breaks it up into two parts and satisfies all requirements. Leave a jumper there so you can just replace the jumper with the \$3.9\:\text{k}\Omega\$ resistor when and if the teacher complains. That should shut them up, then. And it won't change the output even the tiniest bit.

Unless an error is found, this is my final contribution to your question. I'm done.

(The above doesn't depend in it, but for a good process to follow in designing thermally stable current sources using the specific structure used above, see here).

Answer 2

I'm wondering what level a course this is, and what the lecturer seeks to find out about the apprentices. If there's one particular desired solution in the curriculum, or rather, if the lecturer would like to learn what the folks are up to, how advanced topologies they would venture, how far they'd fetch their designs, how many of the classic hurdles they'd be able to overcome / see in advance (based on study of the topic) etc.

Do I understand correctly, that you've actually tried building some of the circuits? Or are you still working with a simulation only?

In my practical experience: "motorboating" may be a sign of a feedback loop coupled via the power supply. Do you have a potentiometer for volume control, likely someplace between the preamp and the power amp? The "motorboating" starts when you turn the potentiometer some way up, thus increasing the overall gain of the amp cascade, thus possibly getting into unstable territory (phase and gain vs. frequency, see the Nyquist's stability criterion. Explanation available.)

Another thought: if you say that you get distortion above 1 kHz, in general this could be a sign of "crossover distortion". This does get worse at higher frequencies. And, it may be a sign that your Darlington power totem is not properly biased = that its quiescent current is zero. But: it can only "almost vanish" at low frequencies IF: if you have a strong negative feedback, from PA output to its inverting input. I cannot see that in your schematic.

Having your prototype as a physical rats-nest makes undesired coupling stronger / more difficult to avoid. Keep wires short where they're necessary, and prefer to work with a breadboard (a dedicated PCB design would be impractical at this stage, I guess). Put PSU blocking capacitors on the breadboard or close to it.

Historically I'd get into a sort of feedback loopey trouble (in the bass / subbass range), if my input amp stage had some degree of coupling of the power rail voltage back into input. Such as, using a DC-biasing resistive divider between 0 and +Vcc in a single-supply topology, serving to keep the DC potential of my (AC) signal input halfway between the power rail and GND. That is why, I strongly prefer dual=symmetrical power supply from the very first stage, which allows me to have an inherent solid signal reference ground "somewhere in the center" between the power supply rails. The ref.gnd potential is inherently within the operational range of all the silicon in the circuit, and I can reference the input signal source, and any feedback networks, and the output, to that very center ground. If I wanted to be 100% sure that I don't get a feedback through the PSU, I'd use a pair of linear stabilizers for the preamp (consider 7809/7909 for your assignment) if you need to use the same mains PSU for both the preamp and the power output.

Hmm. This symmetrical (split) supply really only makes sense for circuits using the op-amp concepts, i.e. differential input etc. The very basic feed-forward "common emitter and friends" tend to rely on single supply, where GND serves as the supply (-) and a signal reference ground at the same time. Which in turn gives you a "not soo good" PSRR.

@periblepsis has suggested a handful of neat ideas, including a nice topology for a simple feed-forward preamp. I can see it containing some constant current sources... looks like a really nice circuit, albeit marginally incomprehensible to me in all the details (I'm too young, and a hobbyist).

As mentioned, I'm past the stage that "trivial is good". I'd suggest using a handful of discrete BJT's to implement a minimalistic op-amp topology. Gives you quite a bit of self-adjustment of the various operating points, a negative feedback to help you set the closed-loop gain and frequency response, and quite a bit of "power supply rejection" (immunity against feedback and EMI ingress via the PSU).

I'd use one op-amp-like circuit for the preamp, and another one for the power amp - to stick to your assignment, to follow the usual "macroscopic modular arrangement of gain blocks", and to have some room to tilt the gain this or that way.

There's a free online PDF book that I'd like to recommend: Designing Analog Chips by Hans Camenzind. Mentioned to me before, by others here at SE. You can probably skip all the silicon doping and etching for the moment. I am referring specifically to chapters explaining these topological building blocks:

1. the long tailed pair
2. the constant current source
3. the current mirror

This will allow you to study the most basic op-amp internal topologies and come up with discrete-based circuits along those lines of your own, as a starting point. See for instance this simplified topology or this practical example, allegedly a TL082 When trying this with discretes, you may need to throw in a resistor here and there - although surprisingly the self-biasing circuits often need very little in the way of extra resistors. To prevent premature release of smoke, exercise in your simulators.

In the preamp block, the output stage should work just fine in class A = open emitter into a pull-up resistor, or into a constant current source, or some such. The required output impedance and amplitude mean that neither heat nor current gain are too much of a concern.

As for the power amp: The complementary class AB output stage that you have suggested - looks nice in principle. Next, you need a voltage gain stage to drive it. The simplest would possibly be a single N-BJT connected with common emitter, working into a P-BJT constant current source. Or, rather than a stand-alone CE N-BJT, I'd suggest to use the open collector output of a differential input stage = combine a long-tailed pair, with a current mirror to couple its output to the bases of your Darlington power stage (you can flip N- and P-BJT polarities to correspond to your previous stages).

Also, the cascade of diodes between the darlington bases (to bias the quiescent current of the power stage) seems like a bit of an overkill... Are they schottky? Principally, each B-E junction (two of them in series in each Darlington) should correspond to a somewhat open Si diode, so like 4 Si diodes (1N4148 ?) should have a forward voltage drop roughly on par with your power transies.

I've seen a design where this cascade of diodes was replaced by a single N-BJT connected in "common emitter" basic topology, with a trimpot between C/E, the adjustable pin connected to the Base - allowing you to adjust the quiescent current of the power Dalington totem. (Use a multi-turn trimpot, and start with the Base pulled all the way to the Collector, for minimal voltage between the Dalington bases.) If you thermocouple this "bias setting helper BJT" to the power darlingtons, you get automatic quiescent current adjustment to prevent thermal runaway.

Further sources of inspiration: 1 2 3 - ordered from the simplest to the most comprehensive. This is not to humiliate you, or to suggest that this is the way to go - I mean to suggest that you try to find repeating concepts, circuit stages and tricks in each of those amps. All of them are your basic class AB with a diff input, just with a different "depth of attention to detail".

Note that the Darlingtons have some forward voltage drop from B to E, especially when you drive them to seriously open. And you need some more "voltage headroom" for the stage that's driving the Darlington bases. You should strive to squeeze the most out of that little "headroom near the rails" - by using a current mirror against an active constant current source, to drive the Darlington bases. Even that way, the inevitable voltage drops down the cascade of the B-E junctions in the very output, principally limit the efficiency of this style of a power amp. If you have +/-12V to play with, and you need 5V RMS = approx. 7V peak on the output, I believe that this is achieavable without additional radical tuning. (There are designs with CE-connected power output, able to drive "rail to rail", that are however more sensitive to setting the quiescent current.)

Does it feel way too complicated?

Based on my historical early experiments with "simple" circuits, just individual transitors (up to 3 per amplifier block), all I can say is that reducing the transistor count ends up bringing more headache than design clarity. Those "simple" circuits easily end up requiring more tweaking (based on experiment and measurement), and get you stuck in a swamp of compromise and inefficiency, compared to a circuit making use of the "three friends of op-amp magic" listed above.

EDIT:

with respect to the note about using a transformer in the PA, if the lecturer really wants you to dwell on the very basic principles, how audio used to be built like a century ago, maybe try a query into google images along these lines. Note that these old topologies, with a minimum of valves and lots of transformers in the signal path, initially were adapted for / adopted in transistorized designs. In those days, discrete transistors cost an arm and a leg. What was that - sixties? Maybe still into early seventies? Guess what the ratio is today, comparing the cost of a transistor to the cost of a custom transformer. Especially when prototyping.

Next came transformer-less audio PA designs, initially with a single-transistor CE N-BJT feed-forward input and a quasi-complementary output totem. Back in the day, the quasi-complementary output totem was popular for cost reasons, and remains in use in on-chip integrated designs where the process favours N-polarity transistors (P-polarity are difficult or impossible to achieve, or their parameters lack luster).

Overall, class AB with a differential input stage (excess open-loop gain and a negative feedback), is like eighties to nineties. About a decade ago, if not two decades, there's been a massive exodus towards class D in audio PA. Even consumer radio of any kind (long wave to WiFi to 4G/5G) hardly has any near-discrete analog front end anymore. Pretty much everything is done by single-chip solutions using broadband sampling (rather than superhet mixing). You do meet some LNA's and mixing in TV antenna technology (satellite and terrestrial), but that's a swan song.