Saturday, 8 November 2014

AF Power Amp Experiments

Although VHF focused, I spent some time studying AF power amp design this fall. Even SDRs need audio amplifiers for the ear interface.  If a speaker guy wants some serious wattage blaring in his cottage, then a split DC supply audio power amp with DC -15v / +15v or so garners an easy ~12-16 watts clean average power.

However, back in the land of radio heads, swinging the AC even remotely near rail to rail with a 12 volt single-supply proves arduous. Thus, designing and building AF power amps that cleanly swing as close to the supply rails as possible seems like a good idea.
Before these, my single-supply AF power amplifier experiments usually employed single or Darlington emitter followers arranged as a complimentary symmetry pair. 1 bleak fall day; with coffee and cats, I changed it up and moved away from typical, symmetrical, emitter follower push-pull finals to learn about other topologies.

I tested these amps with an ultra-low distortion, variable amplitude 1 KHz signal generator, an 8 Ω resistive dummy load, a DSO with FFT and 2 DMMs. I'll share 3 progressive amp experiments:

Amplifier 1

First up is quasi-complementary amplifier. Unusual to see in 2014, but great fodder for learning what goes on in a PA.  After making, debugging and analyzing these circuits,  I felt humbled about how little I consider AC signal phase in addition to the easier concepts of amplitude and frequency during my AC signal analysis.

My first quasi-complementary amplifier.

Above in Figure 1 — My first quasi-complementary amplifier.

Q1 = a class A driver with negative feedback that lowers its input impedance. Typically, a small series resistor goes on the input to raise its input impedance, however, I left it off on this experimental amp. 

Q2, an emitter follower [with no phase inversion] forms a standard Darlington pair with Q4 Their asymmetrical compliment are Q3 and Q5: Q3 is a PNP common emitter amp that inverts the signal phase.

When a negative swinging signal enters Q4, it draws current that flows into the speaker because simultaneously, a positive swinging signal enters Q5 and holds it near cut off. When the AC signal flips polarity, Q4 cuts off and Q5 conducts. Thus the quasi-complementary amplifier gives single-ended push-pull output.

Let's discuss applying forward bias on Q4 and Q5 to eliminate crossover distortion.

The simplest method involves placing diode(s) across the Q2 and Q3 bases to forward bias them enough to in turn, forward bias Q4 and Q5 almost to conduction in their quiescent state.

Above — Cross over distortion measured across the 8 Ω load with 2 bias diodes.

Above — Cross over distortion with 3 diodes (quite near to eliminating the cross over distortion).

I could have tried 4 diodes, but just replaced the diodes with an adjustable level shifter or so-called amplified diode to allow precise control of the forward bias on Q4 and Q5. Further, 1 replaced the original DC coupled feedback from the speaker back to the Q1 with AC coupled negative feedback (a 100 pF capacitor):

Figure 1B. My final quasi-complementary amplifier.

Above — Figure 1B. My final quasi-complementary amplifier. 

The 22 µF capacitor in both designs provides essential positive feedback across the 1K resistor. Bootstrapping feedback compensates for the asymmetrical output stage allowing the positive peak signal swing to approach its negative counterpart. The 22 µF capacitor maintains a charge to keep the DC voltage across R1 constant and their time constants must consider the lowest frequency to be amplified.

Figure 1B breadboard;

Above — Figure 1B breadboard; although it still has diodes to set the quiescent current at this point.

Above — FFT  in pink showing strong harmonics. You can see heavy cross-over distortion in the yellow time domain tracing too. I then tweaked the 10K level shifter bias pot to eliminate the crossover distortion:

FFT (pink) with minimal cross-over distortion after adjusting the level shifter pot.

 Above — FFT (pink) with minimal cross-over distortion after adjusting the level shifter pot. 

I'm now routinely setting my level shifter to remove cross over distortion with the aid of FFT plus simultaneous viewing of the signal in time domain ('scope viewing). I usually set the drive so the amp is making about 1/2 its maximal clean signal power when tweaking the bias. Then I increase the drive to "full" clean signal power to confirm that cross over distortion doesn't re-emerge, Usually at this point, harmonic distortion begins to dominate.

While it's possible to set the PA forward bias by just viewing the sine wave and tweaking the 10K pot to find where the cross over distortion just disappears, to me, the FFT takes this to the next level. 

My Figure 1B bench assessment offered a good new-bad news paradox. In the 'scope trace above, the tones are down 63 dBc, so I managed my personal best, single DC supply PA in terms of distortion at 1/2 power. The bad news = it took ~ 200 mA of quiescent current to deliver this performance. 

It seems that the output is stage is starved for gain on the upper half of the AC cycle — which is why it takes so much current and it's peak-peak is limited. Point X, or the collector of Q5 should ideally be close to 1/2 VCC, but in my quest for low distortion and big signal swing, it ended up at 5.68 VDC.

I think back in the day when they used amps like this, builders tolerated a lot more crossover distortion, but I’m surprised that this amp when biased in AB (or maybe deeper) gives such good distortion performance with the mere addition of that 100 pF feedback cap. Further, I liked setting the PA bias with the aid of FFT instead of just looking at the old sine wave.  I'm learning, and for sure, making mistakes.

Above — FFT of Figure 1B's maximal clean signal power = 3.82 Vpp [228 mW]. The worst tone is -53 dBc. Clearly with ~ 200 mA of idle current and only 228 mW clean signal power, this power amplifier is not a keeper, but I enjoyed analyzing + working on it and felt a boost in confidence going forward.

Signal Squaring

A home brew square wave board might help assess your AF power stages in time domain. If your PA can accurately reproduce a square wave, you're on the right track!

A square wave provides a symmetrical waveform that alternates instantaneously between 2 levels and allows you to see rise times, overshoot and other phenomena. I used square wave analysis to help me choose the 100 pF AC feedback capacitor in Figure 1b.

Ensure you watch your DC current so you don't suffer final amplifier thermal runaway during square wave testing as your amp may consume significant power when driven hard. I blew 1 final in my experiments. With PA experiments, heat sinking the finals proves worthwhile. I did this in Amplifiers 2 and 3.

Above — A simple, signal squarer I keep in my lab. With appropriate RF bypass and series coupling caps  it also works great at radio frequency. For example, 0.1 µF capacitors @ HF.

A ragtag squarer breadboard that's seen much use over the past few years.

Above — A ragtag squarer breadboard that's seen much use over the past few years.

Above — Square wave output of Figure 1b

Above — Square wave output of Figure 1b.

Amplifier 2

I converted Amplifier 1B into a full complimentary version:

Above — Figure 2 schematic. A Darlington — complimentary circuit lies on both halves of the push-pull scheme, Function is similar to Figure 1B, but the symmetry adds the advantage of a remarkably low quiescent current for proper AC/DC operation. I added emitter degeneration (series feedback) to the Q1 voltage amplifier. Nearly every resistor was adjusted or swapped to find the sweet spot in order to swing the largest AC voltage 'tween the rails.

Breadboard of Figure 2 connected to an 8 Ω resistive load (2 parallel resistors).

Above — Breadboard of Figure 2 connected to an 8 Ω resistive load (2 parallel resistors).

Bias setup

Above — Bias setup: Even at half-power, the lowest distortion possible gives a 2nd harmonic of only 32 dB down. I tried many schemes to lower this distortion, but failed.

Above —Maximum clean signal power = a disappointment. Even with lower drive, this amp suffered from harmonic distortion. No point in continuing to work on it. Onto Amplifier 3:

Amplifier 3

In review; Figure 2 featured a complementary NPN—PNP driver + PNP—NPN output pair with a level-shifter that sets the bias differential on the drivers which in turn establish the bias for the finals since they're directly coupled. Sadly, Figure 2 suffers from harmonic distortion.

Amplifier 3 fixes these woes:

Figure 3 schematic.

Above — Figure 3 schematic. Optimized for low distortion and quiescent current @ 1W; it's driven with a low noise op-amp and ranks as the best single-supply audio PA I've built to date. 

A combined time and frequency plot at maximum clean power: 1 Watt. The 2nd harmonic is 63 dB down!

Above — A combined time and frequency plot at maximum clean power: 1 Watt. The 2nd harmonic is 63 dB down!  

Following FFT analysis, I performed the most important test of all — listening through a speaker. 

On my bench top sits an old cassette player with line-level output. The audio tape plays the clear — booming — voice of a loud, male Russian professor. In Russian language, only 1 syllable is ever accented and his taped voice peaks rip like thunder — well testing audio amps plus scaring cats. Wow, классный , superb --- for sure, a version is going in my next receiver.

When you connect an op-amp to an AF power stage expect oscillations. Surprisingly, mine were between ~1 and 2.7 MHz. With the output connected directly to the op-amps inverting input you might create a situation where the total phase shift at the feedback loop exceeds 360 degrees plus exhibits a gain > 1  — oscillations — 

Experimenting with all the parts connected to the op-amp's inverting input is the first place to start when debugging higher frequency oscillations. 

Decoupling and bypassing the DC supply with simple RC networks is warranted if motor boating [low frequency oscillations ] arise in any AF circuit.  On my JavaScript RF Tools page,  Section D. Calculate Cut off Frequency for an RC Pi Low-Pass Filter , you may assess combinations of caps and resistors.

Figure 3 breadboard shown in audio test mode

Above — Figure 3 breadboard shown in audio test mode with an RCA female on the output and a "tacked on" 10K volume pot since the voltage gain is pretty high. This Cu board is the same 1 used for Figure 2, so by this time it's looking pretty worn. 

I think some radio enthusiasts feel more impressed by pretty looking circuits rather than well designed and properly tested stages... I thought I got a good exposure though. I have never used Photoshop and prefer to get it done on camera.

Also, we normally don't leave unused op-amp pins floating — I will have stripped parts and trashed this board by the time you read this.

Final Thoughts

I've got some other ideas, designs and advice to assess. Further, some rail-to-rail op-amps will arrive soon. In 2015, I'll see if I can make a linear PA without distortion at signal amplitudes even closer to the rails. Perhaps? Many thanks to my mentors + supporters and

Thanks to you for reading — catch you later!

1 comment:

  1. Cool... likes the 'on air' builds :) I know the geeky pleasure of seeing it finished and working.. in air..! My first electronic circuit was a 'dancing LED' built using no PCB, all the resistors, caps, transistors and LEDs soldered together in a ball shape :)

    Here is a similar attempt by myself: