LM386 amp circuits

[MrBungle]

Just a tail end comment on this thread; if you're fed up of the horrible THD, hissing and other noise on the LM386, there are some nice solutions here that fit in the same space: http://ludens.cl/Electron/audioamps/AudioAmps.html

I've built the last NE5534 based amplifier with a BC139/140 pair instead of the BC637/640 specified and it's rather good.

[mhennessy]

The LM386 is a bit like the 741 or even the 555. Or the PIC16F84 for those who like micro-controllers...

Been around forever, been eclipsed by much better designs that are often cheaper, and are rarely used professionally. Yet, they hold a certain appeal to hobbyists because there are plenty of circuit ideas and applications for them. Yes, they might be older, inferior and perhaps more expensive. But that doesn't make them obsolete - not by any means.

The 5534 (and the dual 5532) almost falls into the category, were it not for the fact that nearly 40 years on, it's still a near-perfect audio op-amp. Slightly better op-amps exist, that might have marginally lower distortion or noise, but these differences are rarely needed for practical purposes (satisfying though it might be!). Alternatives are often selected for marketing reasons, sadly. A worthy alternative is the OPA134 (or OPA2134 for the dual version), and that's because it has a FET input stage. Not because I think FETs "sound better" - far from it - but because the input bias currents of the bipolar NE5532/4 can cause unwanted DC offsets (perhaps 10s of millivolts). Not always an issue, but when it is, the OPA2134 is a no-hassle cure.

That last circuit in the link will work well, but is by no means optimal. For example, the inverting configuration results in a low input impedance, which might complicated things for the earlier stage. The output swing is non-symmetrical - the previous circuit is better in that regard. For interest, I've attached a circuit that works very well and might be worth experimenting with - naturally, you'd need to adapt it to single-supply operation if required, and scale up the transistors (and reduce the resistors) accordingly.

The "secret" is the 200 ohm resistor which allows the op-amp to drive the load when the transistors are in their dead-spot. If you have a distortion meter, it's fun to play with the value of that to see the effect. Don't go too low as the op-amp has limited drive capability, so it produces more distortion into lower impedance loads. I used an OPA2134 (low DC offset as it's a headphone amp), but the 5534 has slightly better drive capability and DC offsets can be higher when dealing with loudspeakers. Plus the 5534 can run from +/-22V if you're careful, which makes for a pretty loud amplifier (perhaps nearly 20 watts into 8 ohms, but you'd need beefy transistors - probably Darlingtons - along with a regulated PSU set to +/-20V).

Sadly, I ran out of time when developing it (part of a much larger project), but there's lots of mileage to come from it. As it stands, it certainly does an excellent job with headphones, and I'd recommend it to anyone looking the "next step" after the LM386. Don't forget that you can parallel up transistors, and the BC327/337 pairs are ridiculously cheap. Without doing any maths, I wouldn't mind betting that 10 pairs would be good for a couple of watts at least. Watch the temperature rise and bias stability (it's deliberately under-biased because of the 200 ohm resistors).

Hope that's of interest,

Mark

[MrBungle]

You're right about the low input impedance there. I actually stuck an LT1010 in front of it. The main advantage with that amp is it is single supply and therefore easy to power from battery equipment.

Interesting idea with the bc327/337 pairs. Would have to use a current sharing scheme though as they vary wildly from experience. Don't want hot ones as TO92 can't shed much heat.

[mhennessy]

Ah - the LT1010. Quite "esoteric" for this application - an op-amp follower would be cheaper and have much lower distortion (for audio use, the LT1010 is quite bad in that regard). But having been caught out in the past by parts going obsolete, I naturally drift towards the "jellybean" solutions where possible. Less of an issue for home projects, of course.

An emitter resistor per transistor would be adequate - a couple of ohms will do the trick. And if the (short!) collector lead of the transistor goes to a decently thick copper trace, that would probably be adequate for heat sinking - though providing good airflow around the transistors is obviously important... You'd have to monitor it during development before settling on the final design. Luckily, with real music, you get away with a lot because music has a high peak-to-mean ratio (perhaps 20dB, or x100 in power terms). I did a prototype a version that ran from 9V and used 4 pairs driving a 4 ohm loudspeaker, and they stayed pretty cool. Plenty of transistor radio manufacturers did exactly the same with just a single pair of TO92s, after all... I used a chip in the end because of space pressures. The chip measured worse, but sounded good enough.

It's more variations in VBE that cause unequal sharing - not so much HFE spreads - and as a result, these sorts of schemes tend to work quite well in practice because VBE is somewhat less variable than HFE. The voltage developed across the emitter resistor swamps the millivolt variations in VBE.

[MrBungle]

Yes LT1010 wasn't ideal. Input bias current is horribly large and I was somewhat more naive when I did that

Jellybeans here as well usually.

Interesting ideas all in all. I might have a play with that idea. Perhaps a current sharing arrangement would actually give better linearity as well due to the lower variation of Vbe with lower Ic change. I (was) primarily digital and ended up in software (yuck) so I'm a bit rusty on this.

[mhennessy]

Yes, multiple transistors does bring about lower crossover distortion for that reason. And in general, it reduces "beta-droop" induced large-signal distortion, as each transistor sees less current.

The books by Douglas Self are excellent for this sort of thing. I really like his writing style - it's approachable, concise, but with occasional bits of humour. Definitely recommended. The earlier editions can be found on-line, but the later versions are worth buying as they are more complete.

[MrBungle]

Excellent - thanks for the book suggestion.

[Martin Bush]

Just returning to the LM386 and an earlier question I had about combining a stereo input to make a single mono (the answer to which was to use resistors), I wondered if there was a way to do the reverse for running two speakers.

So I simply mean two speakers running the mono signal.

This is simply a learning project for me but I like the idea of ultimately arriving at something that looks like a hand held audio amplifier into which you can plug a stereo source and two speakers.

[paulsherwin]

Just connect the speakers in series or parallel, depending on their impedence and the minimum impedence the amp will drive. Both speakers will obviously be mono as the stereo component has been lost.