# Parallel speakers sound bad, but serials are fine

I have made a simple speaker circuit. It sounds (at least to me) surprisingly good (though an audiophile would probably scream out loud and run away). I have currently run across a problem.

To increase the volume and increase the audio quality (make speakers share the burden) I have tried different configurations for four speakers, paralleling them all, serialing them, and parallel two serialed by the other two. Adding speakers in serial only has good effect to the audio quality, but as soon as I use any one of them in parallel the audio quality gets notably noisier/distorted. I don't really understand why. What is going on? I want to be able to parallel speakers because serialing them is good for audio quality, but it does decrease the volume, and that is a problem, especially if want to use even more speakers.

A "simple" fix would be to increase the power supply so I can drive them with higher voltages, but I am a little scared of that :P and it would also put more pressure on the involved components. Using a rail-to-rail op-amp could perhaps improve the situation a little, but there will still a voltage loss across the push-pull BJTs (I guess), and I don't have any suitable rail-to-rail op-amps available.

It's annoying that it almost works. I don't need the best sound quality or volume, but only something that is somewhat acceptable.

• The speakers are of two different types, but even if I just parallel only two identical speakers do the noise/distortion issue come up (and leave the other two unconnected).
• It is probably not relevant, but my dual supply consists of two 5 V wall warts.
• The op-amp I am using is the quad op-amp LM324AN. It's not rail-to-rail.
• A 100 (to about 220) ohm resistor between the immediate output of the right op-amp and ground decreases the noise in all configurations. I don't know why; I just stumbled across this when I connected a cable incorrectly :p
• And BTW: I don't think the noise/distortion issue comes from paralleling speakers drawing too much current, because the volume has no/small effect on it. It it was a current supply issue then a higher volume would (I think) very much worsen it.
• The purpose of the first op-amp is to center the 0 V to 5 V input around GND i.e. the output from the fist op-amp is -2.5 V to +2.5 V.
• Both op-amps are supplied from the +5 V -5 V rails. That is unfortunately not showing up in the schematic though.

Some more research

Okey. So I did put my oscilloscope to action and probed the voltage before the speakers (after the push-pull).

This is how it looks like with one speaker (and yes there is a lot of >20 kHz noise):

This is how it looks with two identical speakers in parallel under otherwise the same circumstances. The voltage did actually not decrease, but there is this strange thing at the bottom which must be the noise that I hear:

On the following image it is even more noticeable!:

BTW, the second trace is just a marker. It is not showing ground and ground is approximately in the middle of the waveform.

Oh! I am terribly sorry the schematic was wrong on one crucial point. I did have the feedback after the push-pulls! Like this:

================================

Some more research and the solution

================================

First an image of the audio signal on trace1 and GND on trace2 as requested. It does not seem to be very DC biased:

But still it was interesting to take a look at how the rails looked when the audio was noisy. This is picture of the positive rail and GND. It looks noisy, and it gets worse with higher volume:

The negative rail is at the same volume and is notably worse than the positive rail, at least to my eyes (my slow camera blurs!):

The first thing I attempted was adding a 1 µF capacitor between op-amp output and ground, and I was quite surprised to hear that the noise disappeared!!

This is how the negative rail looks like at the same volume, but with the added 1 µF capacitor:

So that simple 1 µF capacitor solved my issue! Thanks to everybody, it would have taken MUCH longer for me to solve this without your help :)

This is how a sinus wave looks after the capacitor was added. Mmm, I can stare at that all day long...:

Update 2016-03-09

I replaced the opamp level shifter with a 1uF AC coupling film capacitor and a 10k ohm resistor to GND after it. I did not really notice any improvements to the sound quality BUT I did notice that the biasing with a capacitor instead of the opamp level shifter is more safe. With the opamp level shifter if it, for whatever reason, would place the signal either too high or too low then the NPN or PNP part of the push-pull output stage would have to carry a heavier load resulting in heating up (not good!). So, I will keep the AC coupling capacitor.

I also added an inductor to filter out the PWM frequencies (~0.3mF), that very much improved the appearance of the sine curve.

Though, even with these two improvements the original issue still arise if I remove the 0.22uF cap between the opamp output pin and GND. Here under are images showing a 440Hz sine-wave with and without that 0.22uF cap.

• Without measurements, who can tell? But at a wild guess, the lower load impedance of parallel speakers is clipping (distorting) due to limited output current from the amp. Increasing the current (not volltage) rating of the wall warts may help, but you may also need to upgrade the output transistors (and their driver stage) to benefit from it. If you can measure the supply volts and it sags when loud and distorted, this is the problem. – Brian Drummond Mar 5 '16 at 13:36
• @BrianDrummond Actually, since OP stated that the distortion when parallelling didn't occur when serialing the speakers, we can be 99.8% sure it's aproblem of the speakers overrunning the amplifier damping factor by presenting a lower load impedance than the amp can effectively drive. - It's a popular problem for 'budding audiophiles' to experience (aka: "been there, bought the T-shirt). ;) – Robherc KV5ROB Mar 5 '16 at 13:57
• Again, GET RID OF THE LEVEL SHIFTING OPAMP. A passive high pass filter will do a better job, doesn't rely on you thinking you know the DC offset and nulling it out, and won't be susceptible to feeback thru the power supply like your circuit is. – Olin Lathrop Mar 5 '16 at 16:22
• No, what I drew makes more sense. I don't get it. I've given you a complete schematic, but you seem to have tried everything except that. First you put C2 between +5V and the opamp output, then between +5V and ground. What is so hard about doing what I suggested? It also seems you still have that stupid level shifter in there, which provides the feedback path thru +5V for the whole amp to oscillate. Why ask for advice if you're not going to at least try it? – Olin Lathrop Mar 6 '16 at 12:16
• @taz0k Looking at your newest 'scope pics, I think you might be seeing some power supply noise on the volume 30 w/o output->gnd cap trace. This is likely from your - rail supply "running low" at higher volume, the imbalance of your supply resevoir caps, or from parasitkcs on your reserve/decoupling caps. It is best to use ceramic, thin-film, or mica caps for all your decoupling, and often suggestable to place one near your (usually electrolytic, or Al-poly) reserve caps. In either case, the lower parasitic resistance/inductance capacitors help filter hf noises more effectively. – Robherc KV5ROB Mar 10 '16 at 2:07

Your amplifier has a fixed output impedance and finite voltage swing. To get the most power out of it, the load impedance has to match the output impedance. Two speakers in parallel have half the impedance of one speaker. This is apparently too low for your amp to drive properly.

Probably your "wallwart" supplies are collapsing under the heavy drain of two speakers in parallel. The lower supply voltage makes the 1.5 V or so deadband in your output a larger fraction of the overall, significantly increasing distortion. You don't say what kind of opamps you are using, other than they are not rail to rail. The supply voltage may be collapsing to the point where there is little active region left between the deadband in the output and the output range of the final opamp.

In addition, parts of your circuit don't make sense and could rather easily be replaced by a better design:

1. You have symmetric ± supplies. That's good. So why why why are you level shifting the input away from ground-centered?

Upon closer inspection, it seems you are level shifting to compensate for the input signal being centered around 2.5 V. This is just plain silly.

You can't hear DC. Even "HiFi" audio only goes down to 20 Hz. The obvious way to deal with input DC offsets is to AC couple the signal. Get rid of all the nonsense to the left of the positive input of the second opamp. Replace it with a 1 µF cap in series followed by a 10 kΩ resistor to ground.

2. You seem to be willing to live with the deadband in the dual emitter follower output stage, but at least include it in the feedback loop so that the opamp can try to compensate for it. All this requires is to connect the 10 kΩ (Argh, no component designators) feedback resistor to the output of the whole amp instead of the output of the opamp.

Here is your basic circuit with the obvious points mentioned above fixed:

Note that this is both simpler and will work better.

There are ways to significantly reduce the final stage deadband. Two diodes is a very common approach.

What I usually do is use a couple more transistors in the output stage to give it a gain of 2. The previous stage then only has to drive to ± half the supply range. That gets around needing a rail to rail opamp, which generally aren't available at the ±12 V or more you want to run them at.

## Added in response to scope traces

You have even more problems than you realize.

Your circuit is oscillating under load, almost certainly by feeding back thru the power supplies. I should have explicitly mentioned this, but that's what C3 and C4 in my circuit are intended to prevent. Try the circuit I posted. It uses mostly the same parts but should perform better.

You can also see evidence of the output stage deadband on the scope trace. Again, including the output stage in the feedback look will help with this, although it won't fix it.

I now see the opamp is a LM324. That's not a good choice for audio. I'd use a TL07x with ±12 V supplies at least. That will probably mean beefier output transistors, possibly with heat sinks.

Once you get this working, I can show you how to get more voltage swing and less deadband from the output stage, but one thing at a time. That would be for a new question anyway.

• All good points, but I think you might have missed the OP's mention that they're using mismatched speakers, in addition to the poor amp design, mismatched parallel loads always == trouble ;) – Robherc KV5ROB Mar 5 '16 at 15:04
• @Rob: Mismatched is definitely bad, but he also said this problem is observed with the same speakers in parallel. In any case, all I said above still applies. I should have mentioned that mismatched speakers are bad, but you've already done that. – Olin Lathrop Mar 5 '16 at 15:10
• Thanks! Oh, that was very clumsy of me. The feedback actually is taken from after the push-pulls (directly before the speaker(s)). I don't understand why my level shifting would be silly. I need to center it around ground somehow I guess? The opamp will add noise and take up space (well now I am using a quad opamp so it won't not really space). I tried to use a AC coupling capacitor but the audio quality decreased (perhaps because of low quality caps?) perhaps because a an RC filter has a (I assume) slower response time... I'll read through your answer again and see what I an/will do :) – fredrik.hjarner Mar 5 '16 at 15:51
• @taz: Passive high pass filtering at the front of a audio amp is effective and how it's normally done. Active level shifting only subtracts off a specific DC level, so isn't as good and is more complicated. C1 in my circuit must be a bipolar cap. If you use a ceramic, make sure it's rated for significantly more than the voltage you will be using. 1 uF 20 V should be fine. That's because some ceramics are quite non-linear at near the full voltage. There are better choices than ceramic, but that will do for now and let you get on to the real problems. – Olin Lathrop Mar 5 '16 at 16:18
• So, the opamp level shifter is now gone. I am now convinced that a AC coupling capacitor is better because of the easy "automatic centering" of the signal. It makes it easier for the output stage transistors that now heat up less because of the better/more reliable centering of the signal. – fredrik.hjarner Mar 9 '16 at 20:06
• And btw. I don't think the noise/distortion issue comes from paralleling speakers drawing too much current because the volume has no/small effect on it. It it was a current supply issue then a higher volume would (I think) very much worsen it.

I like that you applied some good electrical troubleshooting technique/logic here. Unfortunately, however, your logic missed a salient point.

When driving dynamic audio transducers (speakers with magnets & voice coils), your amplifier needs to have a lower source impedance than the load it is driving.

The amount that the amplifier's source impedance reaches below the load impedance of the speakers is called the "damping factor" in the pro audio world in reference to it relating to the amplifier's ability to drive the speaker with accuracy while the extra available current (and directly related ability to fight voltage "sags") fights (dampens) the effects of moving air & external vibrations, etc. trying to distort the motion of the voice coil.

In order to effectively run 4 mismatched speakers from your output without frying any speakers & without causing bad distortion, you'll really need to add 3 more final output staves in parallel to the one you have now. Then, you can connect each speaker to its own output. This will achieve the higer volume/volt capability you wanted to get by paralleling the speakers, while avoiding blowing your amp by drawing too much current, and blowing the speaker with the lowest impedance because most of the power would flow through it if the speakers were all parallelled on one channel.

- - - - - - - - - - Added Schematic - - - - - - - - - -

Here's a schematic of a circuit, using most of your existing components, combining low-pass/high-pass on the audio input to get rid of the DC offset & the PWM ultrasonic noise. Also, it includes parallel speaker drive outputs, along with Olin's decoupling capacitors between power supplies & gnd, as well as decoupling the LM324AN.

Note: It would definitely be better to drive each push/pull unit from a separate opamp output, with its own feedback loop, but that would require 2 more LM324ANs (wire use the feed from 1OUT to run all 4 "final out" sections, with each "final out" wired like OpAmps 3 & 4 to 1 speaker per OpAmp pair)

• Thanks!. Just a comment on your second point. No, the opamps are supplied from +5V and -5V (not showed in schematic though), and the first opamp level shifts the 0V to 5V input to -2.5V to +2.5V, I have confirmed that with my oscilloscope. – fredrik.hjarner Mar 5 '16 at 14:50
• @taz0k ok, then it's just that your schematics are wrong then, because they show everything except for the one NPN "final output" transistor as being grounded between the 2 supplies (might be part of why you found the simulation to be "somewhat inaccurate"). – Robherc KV5ROB Mar 5 '16 at 14:56
• Regarding your first point: I have seen a capacitor added like that on other audio circuits. I also thought that I should add one (so that current would not "theoretically" go to infinity in the speaker, and/or make the speaker "cone" to too much to one or the other direction). But unfortunately adding a capacitor reduces that quality of the sound. I have tested electrolytic ones with values of about 1(or was it 10...) to 220uF. Maybe I should test an even higher value... – fredrik.hjarner Mar 5 '16 at 14:58
• @RobhercKV5ROB Of course you can use a polar capacitor for AC coupling, when you know the DC levels on each side. The capacitor only cares about the voltage it sees across its pins, not the voltage relative to ground or any other point. – pipe Mar 5 '16 at 18:06
• @taz0k Schematic updated & I deleted a couple of comments where I made mis-statements. – Robherc KV5ROB Mar 6 '16 at 21:50