Dodgy Power Amp Design

[Raven Luni]

Greetings,

I've been trying to build a small power amp for driving surface transducers. This is loosely based on a couple of designs off the web with some changes of my own. It works (kind of) but behaves very strangely. Sometimes its fine, sometimes theres no sound, and sometimes the sound gradually distorts and gets quieter until it disappears (sounds more like crossover distortion rather than clipping). Sometimes shorting the input for a minute fixes it, sometimes it doesnt. Sometimes it stays stable after that, sometimes it doesnt.

I'm totally clueless when it comes ti amplifiers - dont know why I cant get my head round the stuff. Anyway, what do you make of this:

 

[Laplace]

What if you put a DC bias on the input of IC1b? Many strange things might not happen then.

 

[BobK]

How is out connected? If it is returning to the virtual ground, that is not going to work, it cannot supply enough current and the virtual ground will be varying all over the place.

Bob

 

[Raven Luni]

What if you put a DC bias on the input of IC1b? Many strange things might not happen then.

Worth a try

How is out connected? If it is returning to the virtual ground, that is not going to work, it cannot supply enough current and the virtual ground will be varying all over the place.

Bob

Yes it is. What would be the correct way to do this?

 

[BobK]

Connect though a sufficiently large capacitor to 0V.

Bob

 

[Raven Luni]

Connect though a sufficiently large capacitor to 0V.

Bob

Thanks - will give that a try too

 

[Raven Luni]

Laplace's suggestions works. Bob's doesn't. It could do with a few refinements though. I used a couple of 100Ks to bias the input but it seems to be hurting the frequency response (loss of bass compared to previous results). The output resistor gets burning hot as well. Its a either a 4W or a 7W (cant tell since the batch those came in all look the same)

 

[BobK]

I really can't see how it can work with the virtual ground. All of current in the output circuit has to come either from the opamp, which can supply only about 20mA or from the capacitors, which can supply the current, but the ground voltage would go up and down with it.

Have you tried scoping the virtual ground? I will bet that it swing quite a bit with the input.

What is the impedance of the transducer?

Bob

 

[Raven Luni]

I suspect youre right. The capacitor to 0V definitely didnt work though - tried everything up to 1mF. The only scope I have is ont of those tiny xprotolab things which I doubt would handle power too well. As for the transducer, its 4 ohms rated for 3-6W

I think I'll scrap this one and start again - already blown several components getting it working as well as a couple of the tracks on the board (strip board contains magic smoke)

 

[gorgon]

You should DC bias the IC1b buffer input to the virtual GND. The value of the resistor will load the input and you may need to increase the capacitor.

As far as I remember the normal way to feed a speaker from an asymetrical driver, was to use a big capacitor between the speaker and the output, and terminate the speaker to 0V. The value of the capacitor is depending on the frequency range you want to run through it. The lower the frequency the higher the value.

Since you already waste a lot of power in the 22ohm resistor on the output I'm not sure the capacitor value is very critical, but I would have tried 1000 - 2200 uF as a start.

 

[KrisBlueNZ]

As BobK said, there's a problem with the virtual ground. You can't return the load to it because it's not a strong rail; it's just driven by a little op-amp. Connecting large capacitors to the op-amp's output will only help to a certain extent, and is bad practice anyway (it can cause IC1a to become unstable).

I suggest you get rid of the virtual ground rail. C4, and especially C7, can be returned to the 0V rail. The load must also be returned to the 0V rail, but you'll need to add a DC blocking capacitor in series with it. The value you need depends on the transducer impedance and the lowest frequency that you want to reproduce.

Then you just need to bias the op-amp side of C4 to roughly half of the supply voltage; you can do this by moving R1 and R2 (but not C6) to the op-amp side of C4. The two 220k resistors across the supply rails are equivalent to a 110k resistor connected to a point half-way between the supply rails.

Also, IC1b is not needed. You can couple your input signal directly into the input of IC1c. There is no need to buffer it; IC1c's input impedance is as high as IC1b's input impedance.

With those changes, you only need a single op-amp.

Edit: What's the reason for R8? It just wastes power. I'd remove it, and reduce the gain (reduce R6).

 

[Raven Luni]

These were the main references I was going by:
http://skema-amplifier.blogspot.co.uk/2009/11/18watt-audio-amplifier-tip41tip42.html
http://sound.westhost.com/project76.htm

I didnt bother with the fiddly bias resistors and just went for the 2 diode separation. Is that capacitor across them (C5) even necesary? Adding the buffer seemed to help things initially, but putting a volume pot on the end of it didnt work - any slight change resulted in momentary silence. From what everyones been saying the problems probably started with the virtual ground.

I've also since found this site which explains a few things:
http://lenardaudio.com/education/12_amps.html. So as for that 22 ohm resistor at the output, I'm guessing it was part of one of these zobel thingies without the inductor?

Anyway, going to try it Kris's way, will let you know

 

[KrisBlueNZ]

C5 isn't necessary but it will reduce distortion. I would keep it.

The output series resistor (22 ohm) might help a tiny bit perhaps, if the load is highly capacitive or inductive, but it doesn't need to be so high. A few ohms would be plenty. But with the Zobel network you shouldn't need any output series resistance.

 

[Raven Luni]

Absolutely does not work - only sound is some 50Hz buzz
Here is how it looks now:



I've tried taking the op amp output to both sides of the diode string as wel as the centre. I've also tried increasing the separation to 4 diodes. As for grounding the input, I've tried everything with that as well - divider centre, 0v, capacitor, common etc.

 

[KrisBlueNZ]

R1 and R2 need to connect to the right side of C1, to provide the DC bias at the non-inverting input of the op-amp.

The left end of R6 needs to connect to the inverting input of the op-amp, i.e. the top end of R3, not the bottom end.

C4 would normally be connected in the positive output line, not the ground return. In other words, positive side of C4 to Q2 and Q4 collectors, negative of C4 to the "Out+" point, and "Out-" connected to 0V.

 

[Raven Luni]

Those changes worked, but I found another fatal flaw: I hadnt connected the high side properly to start with to it was only using 1 pair of output transistors. Now that I've fixed that I'm finding that Q2 overheats instantly on power on.

 

[(*steve*)]

That's because the non-inverting input is floating.

I'll try to fix your circuit...

The feedback loop was wrong and the non-inverting input was floating.

The resistor I have drawn in will determine the input impedance of your amplifier to a large extent.

There are reasons for it to be 1k, but as your 0V rail is anything but stiff, it's not likely to yield great performance. Can you spare another op-amp to create a more stable 0V rail?

 

[(*steve*)]

Oh, and you can temporarily wire some 12V globes in series with Q2 and Q4's emitters to limit their current while you set things up.

 

[Raven Luni]

Made those changes, stuck a great big 3.9 ohm resistor on Q2 with 2 heat sinks - it took all the heat but just kept getting hotter and hotter.
Also note all this happened without the op amp connected so it cant be that

 

[KrisBlueNZ]

If you hadn't connected the transistors properly the first time you powered it up, you may have damaged one or several of them.

For each of the four transistors, disconnect the collector from the circuit and measure resistance from collector to emitter in both directions. They should all read open circuit.