Input/Output Caps on simple Headphone Amp.

[Chaos_Klaus]

Hey EEVblog forum,

I am mocking around with simple headphone amps. Schematics are attached. The Supply is bypassed with 100nF and 220µF caps in both cases which is not shown.

There is one that I built around the LM386 and I am not really impressed by both the sound colouration/clarity and the noise level. The LM386 is from TI. I used metal film resistors. The zobel-network should prevent oscillations. The hiss-reduction circuit works, but colours the signal and still leaves enough noise.

Now I just threw together an amp on my bread board using two NE5532, like I saw it in Elliot Sound Products - Project 109. I actually just did it for fun to see it fail. To my supprise this sounds waaaaay better than the LM386, while using less passive components.

The NE5532 obviously can't source a lot of current, but my headphones are quite efficient and range from 32Ohm up to 250Ohm. I don't plan on using any lower Impedance speakers, so I guess I can live with that.

Now for the questions: Are there downsides to using regular opamps like the NE5532 to drive headphones directly? I can't find a definitive number on the current source capability of the NE5532. The short circuit current is given 38mA typical. But I don't know if that tells me anything.

If I use a dual supply and my signals are referenced to the supply ground, do I really need input and output caps? (Ultimately I want this amp at the end of the signal chain in a more complex audio device.) Input caps, I could propably use to make the amp more versatile when used with different sources. But if I kill any DC at the input, why would I need an output cap?

If I needed caps: How would I figure out polarity with an electrolytic cap when the input signal ranges from negative to positive voltages?

A headphone presents at least some inductive load. With the NE5532 amp, do I need a Zobel-network to prevent oscillation aswell?

Thank you!

[krivx]

I think you need to make some measurements.

If you need to know how much current you can get out of the NE5532 then measure it. Apply a test tone, turn up the volume, connect your headphones with an ammeter in between.
If you want to know if you need caps then measure the DC offset for each headphone driver.
To find out if a Zobel network is necessary look at the output with and without.

[Zero999]

The LM386 is normally fine for this. Make sure you're using the LM386N-4, as it's the only varient specified to work at 18V. The other versions are only specified to 12V, 15V absolute maximum.

The datasheet for the NE5532 says it can typically drive 26V pk-pk into a 600Ohm load with a +/-15V supply. Using Ohm's law that's a peak current of just under 22mA. It later on says the short circuit current could be as low as 10mA but it could be as high as 60mA.

As this is just a one off, assume it can supply about 20mA when Vout is 2V below the supply, which would mean with the current shared between the op-amps the it would be 40mA, which is 1.28V peak into a 32Ohm load. The current sharing resistors in your schematic (R4 & R3) are far too high. Try dropping them to 22R or even 10R.

[tooki]

This topic actually comes at an opportune moment, as I've been working on something similar.

Now for the questions: Are there downsides to using regular opamps like the NE5532 to drive headphones directly? I can't find a definitive number on the current source capability of the NE5532. The short circuit current is given 38mA typical. But I don't know if that tells me anything.

If I use a dual supply and my signals are referenced to the supply ground, do I really need input and output caps? (Ultimately I want this amp at the end of the signal chain in a more complex audio device.) Input caps, I could propably use to make the amp more versatile when used with different sources. But if I kill any DC at the input, why would I need an output cap?

I've assembled this circuit: http://www.paia.com/proddetail.asp?prod=9206KP (direct PDF to schematics). It is a minimalist design that does not use any caps outside of the power supply, and as far as I can tell, it works fine, at least with the low-impedance headphones I've tested it with (Sennheiser HD555, Audio-Technica ATH-M50x, Beats Studio [don't hate, I got them cheap for airline travel ], and various ordinary iPhone earbuds) and also as an output to go to amplified desktop computer speakers. Mind you, I've not done any kind of scientific analysis, just "sanity testing" to see if I can hear any obvious problems. As far as my ears can tell after many hours of listening, it's totally transparent.

It definitely cannot drive anywhere near as much current as the Tascam MH-40 Mk II studio headphone amp I've got, but the Tascam inserts quite a bit of noise, whereas I cannot detect the NE5532 adding any at all. (And the Tascam uses lots of power and thus dissipates lots of heat, which I'd like to eliminate.) I'd say that as a sort of rule of thumb, the NE5532 can drive stuff similar to what an iPhone can drive.

A headphone presents at least some inductive load. With the NE5532 amp, do I need a Zobel-network to prevent oscillation aswell?

None of the several NE5532-based headphone amp circuits I looked at appear to use one, so I'm gonna say it's not necessary. The main differences between circuits are:

- one or two NE5532's per channel
- caps or not
- power supply configuration



Anyway, the pix show a single amp block (and the power supply on a separate board), and then a 5 amp unit (with power supply on same board). These work great on 12V DC that's then split up to ±6V with a virtual ground. To my surprise it also runs perfectly on a 9V battery. And yes, that's an LED crimped to a female dupont plug for testing. :p

The NE5532 tech specs recommend using 100nF bypass caps on the power rails, but since I didn't have any with 0.1" lead spacing on hand, I haven't installed any yet and haven't noticed any problems. I'll probably add them for good measure once I have small ones to do it with.

Anyhow, if anyone has any comments or suggestions (or questions if you want me to test something), let me know!

[Chaos_Klaus]

minimalist design that does not use any caps outside of the power supply, and as far as I can tell, it works fine

Well. The problem with DC is that it essentially takes away headroom. So at least at the input, there should be caps I guess. Output caps ... well, I don't know. DC though the headphones could heat them up and might damage them. Don't really know if that could be a long term problem.

None of the several NE5532-based headphone amp circuits I looked at appear to use one, so I'm gonna say it's not necessary.

Well, given the nature of the internet ... I wouldn't make that assumption.

How do I find out if the opamps are oscillating? I have an old 20MHz analog scope and didn't really see any difference with or without a zobel network/boucherot cell.

Make sure you're using the LM386N-4, as it's the only varient specified to work at 18V

Did that already.

I attached a new schematic for the NE5532-Amp.

So far it performed well with all my headphones.

The current sharing resistors in your schematic (R4 & R3) are far too high. Try dropping them to 22R or even 10R.

In the new schematic, I used 10Ohm resistors but then a 100Ohm Resistor in series with the output to get the output impedance right and make sure that different headphones won't give dramatically different levels.

I don't really see why it is bad to have this resistance inside the feedback loop.

[Zero999]

There's no need for a 100R Ohm resistor. The output impedance should be as low as possible to give a good damping factor.

[Zero999]

Why not just add a couple of transistors to boost the output drive?

EDIT:
The first circuit won't work so I've removed the picture from the post but it's still in the attachments. See post linked below for an explanation of why it doesn't work.
https://www.eevblog.com/forum/beginners/inputoutput-caps-on-simple-headphone-amp/msg785332/#msg785332

[Chaos_Klaus]

Hey Hero999,

thanks for providing these schematics!

I thought about transistors already as most designs use them for the output stage. However, there is so many types that I don't really know which ones to choose.

With the two NE5532 I was able to get almost 3V peak across my 32Ohm headphone with a 1kHz sine wave. That would mean about 90mA peak current already. I actually had to take the headphones off way before I hit 3V peak. Actually, with normal mixed/mastered (=compressed) music, I was getting about 2V peak when I was listening really quite loud.

I totally get the second of your schematics, but the first one puzzles me. The outputs of the opamps don't seem to do anything.

I noticed you use a resistor (R7) on the input. Why? Aren't the inputs of an (ideal) opamp said to be of "infinately high" Impedance anyway?

Cheers!

Edit: After reading this article on clipping asymmetric waveforms, I am a little concerned about my Headphones being subjected to DC voltages. In the designs in the article, caps are used at the input (obviously) and with the feedback circuit to get a DC gain of 1, while retaining more gain for the actual audio signal. Where would I put this kind of cap in my design? In series with the 1k resistor? Wouldn't it have to be kinda large? Or is the power so low that I don't need to bother about a few hundred mV of DC?

[Zero999]

I totally get the second of your schematics, but the first one puzzles me. The outputs of the opamps don't seem to do anything.

To understand the second schematic you need to be aware of what's going on inside the op-amp, especially the output stage. This was discussed in another thread fairly recently and I posted a schematic showing a simple op-amp, made from discrete components, connected to output transistors in this configuration.

You'll need to be logged in to see the schematic below:

https://www.eevblog.com/forum/beginners/non-standard-op-amp-configuration/

Q1 and Q2 are the transistors inside the op-amp's output stage. Normally, when the output is taken from the usual output pin, they behave as emitter followers but when the output is connected to 0V and the output is taken from the power supply pins, they act as common emitter stages: one for the positive and the other for the negative part of the signal. Now the signal will be inverted and superimposed on the power supply pins so is not useful for driving earphones or a speaker so the external transistors, Q11 and Q12 are again connected as common emitter amplifiers which invert the signal and bring it back to the middle of the power supply.

In the first circuit I posted in this thread, I added R9 and R10. Firstly to reduce the gain of the output stage to avoid oscillation and secondly because the NE5532 contains two op-amps which need their output stages to be connected in parallel. The NE5532 was originally designed with its output to be configured as an emitter follower which has a gain of slightly less than 1, not a common emitter amplifier with a large gain driving yet another common emitter amplifier with a huge gain. The gain of a common emitter amplifier is equal to Rc/Re. Re = R9 & R10 connected in parallel which is 50k and Rc is R1 or R2 so the amplifier will have a gain of 330/50000 = 0.066 which compensates for the gain of Q1 and Q2 which would be 150 or so as they have no collector resistor.

EDIT:
I've realised this won't work because R9 and R10 limit the drive current too much.

I noticed you use a resistor (R7) on the input. Why? Aren't the inputs of an (ideal) opamp said to be of "infinately high" Impedance anyway?

In an ideal op-amp yes but in a real op-amp there will be bias currents due to the fact the transistors need to be biased into conduction. In order to minimise the effect of bias currents on the output offset voltage, the DC resistances connected to both of the op-amps should be as close as possible. On the - input R6 goes to 0V in parallel with R5 which is connected to the op-amp's output which sits at 0V so the resistance seen between the - input and 0V is equal to R5 and R6 in parallel which is just under 910 Ohms so it makes sense to use 1k for R7 because it's a round value.

Edit: After reading this article on clipping asymmetric waveforms, I am a little concerned about my Headphones being subjected to DC voltages. In the designs in the article, caps are used at the input (obviously) and with the feedback circuit to get a DC gain of 1, while retaining more gain for the actual audio signal. Where would I put this kind of cap in my design? In series with the 1k resistor? Wouldn't it have to be kinda large? Or is the power so low that I don't need to bother about a few hundred mV of DC?

Yes, you could AC couple R6 so the gain is unity at DC. Notice how R7 now becomes 10k because the capacitor blocks DC so the - input now has a DC resistance of 10k to 0V. You could also connect R7 to 0V and connect the signal to the + input via a 1µF film capacitor. The disadvantage is it reduces the input impedance to 10k.

[dom0]

The LM386 is a very old and rather poor amplifier that is by no means "high fidelity".

A 5532 driving a - properly - biased complementary buffer (a small class AB output stage) will already be a very good amplifier for low output power. Technically you can get better performance with more recent op amps, but that anyone can actually hear the difference is doubtful.

The circuits mentioned above with a "current dumping"-like scheme were discussed lately and there's basically no reason at all to use them in this application. (The main advantage of this topology that it allows almost arbitrary output voltages, up to several hundred volts.)

[Zero999]

I've just realised some of the circuits I posted won't work. The idea of adding R9 and R10 to reduce the gain was flawed. The resistors stopped the oscillation but they also preventing the op-amp from providing enough current to drive the output transistors. It would work when there's no load, but as soon as the 32R load is connected the output would clip at a low voltage.

If R9 and R10 are reduced, it will oscillate when the output stage isn't loaded. One way to get round this is to add a zobel network so the output always sees a load at high frequencies. The whole circuit could easily be reconfigured to work from a single supply voltage. All the nodes which need connected to 0V at DC are biased at half the supply voltage by R7 & R8 and anything which needs to go to 0V at AC is coupled via a suitable capacitor.

[dom0]

Sorry, but that's bodge. Connecting R11 to the output instead of ground will linearize the output stage much better than that. Quiescent output stage current in the attached sim is 200 µA or so.

I still don't get why you still dwell on with this circuit, it's just not well suited to this task.

[Chaos_Klaus]

Ok. I did some reading on transistor output stages. So if I added one, would it work like in the attached schematic? I don't really understand how to properly bias the transistors so they work in AB mode.

How do I find the values for R1, R2, R3 and R4? R1 and R2 should somehow make sure that the voltage across the "diodes" Q1 and Q2 is 0.6V (so that the transistor stays turned "on") How do I do that exactly?

I read that on can use use Q1 and Q2 in a diode configuration. This way, with Q1 and Q3 (and Q2 and Q4 respectivly) thermally coupled, I would get better temperature stablity, right? I could mount them on the same heat sink, or with the low output power I could propably just glue them together?

R3 and R4 are somehow there to provide further temperature stabilization. Are the values of R3 and R4 somehow related to R1 and R2? Wouldn't they change the base emitter current of the transistors?

How do I find out which transistors to use? The designs I saw used BD139/BD140 which seem kinda large and overspecced for me. Others used 2N3904/2N3906.

[SeanB]

Q1,2 could be the same type of transistor, or even use 2 low power diodes ( 1N4148, 1N914) in series, preferably coupled thermally to the heatsink, though just placing them on there will do at this low power. C3 is not needed, as you are AC coupled in the input, and the feedback loop will handle the small DC imbalance from real components fine with only a small DC offset on the output.

Distortion will be measurable, as the output stage is definitely class B, you would do better using the Q1, Q2 transistor to make a Vbe multiplier, which will give a good voltage offset and also allow you to set the output stage current so the output stage runs in class AB, probably with around 50mA of current flowing, and this will reduce crossover distortion by a large amount.

[Zero999]

Sorry, but that's bodge. Connecting R11 to the output instead of ground will linearize the output stage much better than that.

Quiescent output stage current in the attached sim is 200 µA or so.

Sorry, that's not good either. You've removed the emitter resistors from the output stage so it will no longer be thermally stable: there will be a risk of thermal runaway.

Ok. I did some reading on transistor output stages. So if I added one, would it work like in the attached schematic? I don't really understand how to properly bias the transistors so they work in AB mode.

How do I find the values for R1, R2, R3 and R4? R1 and R2 should somehow make sure that the voltage across the "diodes" Q1 and Q2 is 0.6V (so that the transistor stays turned "on") How do I do that exactly?

I read that on can use use Q1 and Q2 in a diode configuration. This way, with Q1 and Q3 (and Q2 and Q4 respectivly) thermally coupled, I would get better temperature stablity, right? I could mount them on the same heat sink, or with the low output power I could propably just glue them together?

R3 and R4 are somehow there to provide further temperature stabilization. Are the values of R3 and R4 somehow related to R1 and R2? Wouldn't they change the base emitter current of the transistors?

How do I find out which transistors to use? The designs I saw used BD139/BD140 which seem kinda large and overspecced for me. Others used 2N3904/2N3906.

That circuit is the same as the other circuit I posted.


As well has having some crossover distortion, it also doesn't have much voltage swing.

[Chaos_Klaus]

Ok, So I just did some more reading and found out that a silicon diode will be at 0.6V with 1.5mA flowing through it and at 0.7V with 10mA flowing through.

Without R3 and R4 present that would mean:

(15V-0.7V)/(2*0.01A) < R1,2 < (15V-0.6V)/(2*0.0015A) 

715Ohm < R1,2 < 4800Ohm

R1 and R2 could be anything between 715Ohm and 4.8kOhm? Is that true?

Distortion will be measurable, as the output stage is definitely class B

But wouldn't biasing the transistors make the amp work in class AB mode?

C3 is not needed, as you are AC coupled in the input, and the feedback loop will handle the small DC imbalance from real components

That's only true when the signal is not clipping. This article shows how C3 can help prevent DC in the output when an asymetric waveform is clipping. The headphones I use were way more expensive than this little amplifier, so I think I will include C3 for safety ... unless someone tells me that it is ridiculous to do that with this low power design.


EDIT1:

That circuit is the same as the other circuit I posted.

Ok, that is true. Sorry.

EDIT2:

Corrected mistake in calculation above ... current though R1,2 has to flow though both the diode and the transistor.

[dom0]

Sorry, but that's bodge. Connecting R11 to the output instead of ground will linearize the output stage much better than that.

Quiescent output stage current in the attached sim is 200 µA or so.

Sorry, that's not good either. You've removed the emitter resistors from the output stage so it will no longer be thermally stable: there will be a risk of thermal runaway.

Yeah, I forgot those in the sim. Although at 32 ? one can probably get away with zero Iq in the output stage while using a standard op amp that might limit at 30 mA or so.


But again, I'm repeating myself here, but this circuit topology is not really well  suited for this task. Output swing is lower with the follower, true, but hardly relevant here ; OP wants to use dual supplies, maybe +-10 or +-15 V as usual anyway.