Limiting Power of TPA6120A2 for Headphones

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

Bo Deadly

Well-known member
Joined
Dec 22, 2015
Messages
3,266
Location
New Jersey, USA
I want to try TPA6120A2. Here's my schematic (click to enlarge):

GWxlFF8.png


Some parts are optional. The mono input and volume control is just for making a quick and dirty bench amp. The output inductors are just to see how heavy of a load I can drive without it becoming unstable. I will refine this when I figure out how I'm going to use this exactly.

The problem is this chip is a little too powerful for headphones. It's nice to have that power but it scares me a little to think I am attaching speakers to my ears with an amp that can put out more than 1 watt per channel. Eventually something is going to break and I don't want to loose an ear in the process.

So the question is, is there a simple way to make the output power adjustable or just limit it to a certain level like say 250mW?

I'm using +-15V in my designs but I could regulate down to +-5V. That would significantly limit the damage it could do but I think it could still do damage nevertheless.

What if I put diodes in the feedback path of the op amp (meaning parallel to the 1n cap)? Like back to back zeners or a bidirectional TVS diode like SMBJ5.0CA?

Or would that ruin the wonderful specs of this chip?
 
What headphones will you be using and what's their impedance and sensitivity? The max output voltage (converted to RMS) and the impedance determine max power into the headphones you're using, and the sensitivity will tell you how loud they will go with that much power.

I'd consider a resistor divider on the output of the amp, with the bottom resistor (or at least their parallel equivalent) scaled to be no more than 10 percent of the headphone impedance. This gives good damping, which some headphones (thinking of my Sony MDR-7506) really need.

I looked up that model, impedance is 63 ohms (10Vrms into 63 ohms is 10*10/63, about 1.5 watts), sensitivity is 106 dB/W/m. Presuming I understand that, that's 1 watt at three feet (1 meter).  I bet that goes up 10 or 20dB, maybe more at 1 inch. I'd definitely want it to clip at lower power into the headphones.
 
> Presuming I understand that, that's 1 watt at three feet (1 meter).

No. Put the phones on a dummy head with SPL meters in the head's ear-holes. That's what it means.

(Also 106dB SPL at 1m from one Watt almost violates conservation of energy. 1 omnidirectional Acoustic Watt is 108dB SPL at 1m, a small source can't be much directional, so 106dB means over 50% efficiency, which does not happen in small wide-range speech/music transducers.)

AND --- that spec is *wrong* (but I see the same typo you see). The real spec is milliWatt !
Sony sales-sheet, page 4.
 
7V RMS behind 27 Ohms will put nearly all headphones VERY loud. I did once find this useful while doing live recordings in the same room as loud groups.

Arguably 2.2V behind 27 Ohms should be enough for anybody else.

So 6.3V peak-peak, plus 2V each side for chip drop, is 10V total or +/-5V supplies.

Use +/-5V supplies, remove L37, make R262 27 Ohms. See how you like that.

BUT-- if you clip it, it may be worse on your ears/drivers than an unclipped signal at the same average level.

Real ear safety is same as lifting safety-- learn what is safe and what causes damage.
 
I would look @ adding input limiter/compressor and keep the rails voltages higher. This will keep the amplifier working in its best LINEAR MODE. By adjusting the input limiter the output level would be clamped or compressed to meet your needs.
Duke :)
 
Scenarios that I'm thinking could be potentially descrutive to ears or phones would be power supply failure like one rail goes down or a bad solder joint causes the chip to flutter between open and closed loop gain. A solution that requires power to function or that is limiting at an input, might not work with these scenarios.

Passive limiting like a diode network directly on the output would work. Unfortunately something like D. Self Small Signal Audio Design p. 511 or similar (but directly on the output) would have a threshold based on voltage which would not work equally well with headphones of different impedances.

However, I know the value of the buildout resistor (lets call it RO) so the voltage drop across that is directly proportional to the current. Perhaps that could be used to throttle the supply rails in some way?

For example, the supply might go through mosfets but if the current through RO is high, the mosfet turns off. The following partial schematic is supposed to be a vague framework but of course the important bits are left out as they are unknown.

mGbz6Iz.png


More specifically, if the current through RO is positive the voltage across it would be reflected on the Vbe of the NPN and in turn somehow pull CPLIM low. Visa versa for the PNP with negative current.

If this is possible it could allow for maximum performance with +-15V supply (and I'm not wasting power dropping 10V across regulators to get +-5V) but takes away the source of power that would do damage if current coming out of the chip is too high.

If anyone has any ideas as to how to complete this circuit, I would love to try it in LTSpice.
 
Adding complication rarely improves reliability.

Within limits, a dumb series resistor WILL throttle power.

I am not aware of any headphone worth protecting that will be killed by 7V behind 27 Ohms.

If you believe your low-Z phones need heavy damping, put the 27 Ohms inside the NFB loop.
 
Two questions here. How much do you need? And how much is safe?

I've written thousands of words on the topic. Much was lost in a forum collapse.

Here's one image which plots what many headphones (from Rane's 1983 paper) need to hit 120dB SPL (very loud!), in terms of Voltage, Current, and Power.
 

Attachments

  • I-V-Z-P.gif
    I-V-Z-P.gif
    17.4 KB · Views: 28
You can see that your +/-15V supply (near 10Vrms output) can easily put 3 Watts in 30 Ohm phones.

> proportional to the current

You can see that many 32r phones need ~~100mA to get loud, but 100mA in 300 Ohms is 3 Watts.

So if you limit both voltage and current, there is one magic value (V/I) for maximum power power, and significantly less power for any other load. And the envelope is "square". In a many-impedance application, this seems to ensure sub-optimal results in many cases.

If you understand the Power Match math, the envelope is curved. There is a broad maximum around the magic load. Here is about 7Vrms through 27 Ohms:
 

Attachments

  • I-V-Z-P-powermatch.gif
    I-V-Z-P-powermatch.gif
    13.3 KB · Views: 14
PRR said:
You can see that your +/-15V supply (near 10Vrms output) can easily put 3 Watts in 30 Ohm phones.

> proportional to the current

You can see that many 32r phones need ~~100mA to get loud, but 100mA in 300 Ohms is 3 Watts.
Careful with your quoting here. I said "the voltage drop across [the 39R buildout resistor] is directly proportional to the current". So that resistor could be used as a current sense that would indicate power regardless of the load impedance.

But your Power Match math graph is interesting and I will be using a simple buildout resistor as you previously suggested so your comments are reassuring. I am about to make another batch of boards so I'll just try it and see how loud it is.

The output swing of the TPA6120A2 with +-15V supply is limited to a little more than +-12V. The board has footprints for series resistors on the rails. I will see how large I can make them without getting appreciable distortion / instability. That could shave a few volts of the max swing as well.

Incidentally my primary headphones are Sony MDR-7506 just like benb. Although they so old that the rubber has stopped flaking off.
 
squarewave said:
Careful with your quoting here. I said "the voltage drop across [the 39R buildout resistor] is directly proportional to the current". So that resistor could be used as a current sense that would indicate power regardless of the load impedance.
Power is current multiplied by voltage; you need to monitor both and use a multiplier if you want to control power.
 
abbey road d enfer said:
Power is current multiplied by voltage; you need to monitor both and use a multiplier if you want to control power.
True. Don't know what I was thinking with this one. More voltage more power. More current more power. Same difference.
 
Last millenium I designed headphones boxes that were used in studios for musician's foldback. The "challenge" was to allow using on the same system headphones with varying impedances. Many headphones were 8 ohms at the time, but also 200 ohms. So I used the a very simple formula that allowed about equal power in both cases, i.e. inserting resistors which value is the geometric means between these possible values. Geometric means of 8 and 200 is 40, so I put 47R resistors in the boxes, that were driven by a small power amp (Crown D75 was quite popular in this role).
Delivering about 20-25 Vrms, the resulting power in the headphones was about 1.5W!
One could daisy-chain 10+ boxes without risk of overloading the amp.
The question has been raised that putting a resistor in series would change the damping and thus the frequency (and transient) response: that is true, but at the time most headphones were designed to be driven with a non-zero impedance, because most of the h/p outputs on HiFi amps were obtained by derivind the output signal via resistors.
Today, the situation would be different, because there are no more 8 ohms  headphones, so the range would be 32 ohms to 200 ohms.
What's more, modern h/p's are designed to be driven by almost zero-ohm source. I recently helped Audio Technica solving a problem; they had to adjust the sensitivity of one of their products to be in accordance with the new regulation that limits SPL at 100dB(A). The proper response would have been to modify the impedance of the drivers to decrease sensitivity, however, they did a quick and dirty fix by inserting resistors in series. They had a lot of complaints from the testers that the sound was not right. Indeed, the frequency response had changed a lot, particularly at LF.
I recommended them to make the attenuator a proper L-pad, which would provide the drivers with a more adequate source impedance. To my knowledge, that is the solution that has been approved and put into production.
So, if I had to do these headphone boxes in the new millenium, I'm not sure I would find a satisfactory solution...  :(
 
Mm, this actually seems to work:

6yWwjBn.png


R4, R7 monitor current and R3, R8 monitor voltage. The ratio is almost exactly 1:1 as the load changes.

This shows that as the load increases, the output is clipped at a lower voltage:

G4PJg1V.png


Of course cutting of the supply like this would probably make something as fickle as TPA6120A2 go nuts.
 
This has been well explored (beat to death) but I have one more anecdote that more resembles limiting power.

Back last century I designed a headphone amp at my old job, to melt the wax from musician's ears using differential drive for more signal swing... but today I'm not talking about that aspect.

Power limiting... My headphone amp was strong enough that it sounded good driving loudspeakers, but there was a problem with that, it would open the thermal fuse inside the 1A wall wart. Since the customer is always right even when they connect loudspeakers to a headphone amp, I figured out a way to still kick while protecting the wall wart.

My solution was to add a resistor in series with the transformer winding. At low current draw like from higher impedance loads the rail voltages would charge to full voltage, but when driving a lower impedance speaker load the resistor would limit the reservoir recharging current such that the rail voltage would sag, indirectly limiting the output voltage and current but only when heavily loaded. This increases the typical voltage sag caused by transformer winding resistance making it actually useful for something.

JR

PS: Abbey's theme of higher output voltage swing with series resistance is widely used in studios over here too for generic headphone distribution where you need to accommodate multiple different headphone loads. I designed one 4x headphone splitter box at that same old job, actually I inherited a design already in process using 22 ohm resistors that I adjusted up to a couple hundred ohms. 22 ohms IMO is too low Z to be effective at any of the sundry tasks the headphone build out resistance is expected to accomplish. 
 
> Careful with your quoting here.

I assumed you would understand the reference.

I am sorry if you didn't.

> this actually seems to work:

Complicated and fickle. The opamp is surely class B. If you set a 30mA current limiter in series with each rail, you would get the same plot. You will need rail bypass caps AT the opamp pins. This does allow short bursts of over-current. However headphone melt-down is a factor of time.

Side-note: your selected 31mA current is just about what common opamps self-limit at. A TL072 will meet this spec and far cheaper than your super-amp plus current control.

But do we want Power dropping with impedance? NO! Many low-Z phones are hungry. And to a first approximation we want "same" maximum power in "all" impedances, since efficiency/sensitivity has nothing to do with impedance. (Take the copper out of a 32r driver. Hammer it to half diameter. Re-wind. You have a 128r driver. Conductor length is 4X, but resistance is 4X, so the Power Efficiency is un-changed.) As a secondary correction to "same power in all impedances" we note from accumulated specs that the higher-Z drivers "tend" to have higher power sensitivity, for historical reasons. These were often "bridged" across existing transmission lines and wanted good acoustic level with little line loading. The <100r phones have generally aimed at dedicated phone jacks and amps, so could be less thrifty, which opens cost-savings.
________________________________

> my primary headphones are Sony MDR-7506

If that were your only phones, you should ask if you need a fancy amp at all. The 7506 will touch 120dB SPL (LOUD!) with 1.2V 20mA RMS. TPA6120A2 can deliver 7X the voltage, 25X the current. That's _50_ times the power you could reasonably want in a 7506! If you want a 100W lamp on your bench, do you get a 5,000 Watt lamp with a dimmer? Further, a hypothetical 20 Ohm phone (not so uncommon)

Also it is easily seen that ~~1V audio signals don't need a slew rate over 1V/uS. '6120 slews 1,000X faster. Since you don't directly pay for this, it isn't a "cost". But it is not a bonus you will use. That spec derives from the 10MHz DSL chip that the '6120 is re-specced from.

Yes, you may have other phones and they may suck more.
_______________________________

Voltage limiting gives more power in low impedance.

10 Volts-
10r == 10 Watts
31r === 3 Watts
100r = 1 Watt
310r = 0.3 Watts
1K === 0.1 Watts

Current limiting gives more power in high impedance.

0.1 Amps-
10r == 0.1 Watts
31r === 0.3 Watts
100r = 1 Watts
310r = 3 Watts
1K === 10 Watts

Using both gives the lower of the two limits, and a "best load" where power is high, but 1/10th this power for impedances 10:1 either way. As phones do run from 4r to 2K, we have too much for some and maybe too little for others.

10 Volts or 0.1 Amps--
10r == 0.1 Watts
31r === 0.3 Watts
100r = 1 Watt
310r = 0.3 Watts
1K === 0.1 Watts

At the "best load" of 10V/0.1A= 100r we can get 1 Watt which I assure you is too much for Koss Pro4. If we cut-down to say 3V 0.030A, an adequate 0.1W in 100r but a low 0.010W limit in 10 Ohms. Any strict V/I limit gives very large variation over the typical range of phone inpedances.

One alternative is a Multiplier to compute and integrate Power from V and I. This is historically absurdly complicated.

Is there a simpler scheme? Recall your Power Ttransfer Theorem. Resistive source and resistive load. Max power when matched, OK. But note how slowly the load power changes as the load resistance is varied. 5X load impedance, power is not down to half. (Would be 1/5th for voltage limiting.) This is even more clear if load R is plotted log-scale (because headphones cover several decades of resistance). Yes, efficiency sucks, but at part-Watt levels this may not be fatal.
 
Here's a comparison of V/I limiting and series resistor. I used V+R values I think are a suitable first-crack for a LOUD phone amp, and picked V/I limits to give the same power in extreme impedances. You see that V+R limiting gives a fairly flat 250mW-450mW power over a w-i-d-e range of impedances (6 to 130), while V/I limiting can deliver over 1,000mW in 17 to 45 Ohms (a very popular range), without doing better for the extreme cases like 8 and 300/600 Ohms.

load ------- 7V 27r ---- 6.7V 0.24A
8r ------- 321mW ------ 460mW
16r ------ 425mW ------ 920mW (2.1X!
27r ------ 454mW ---- 1,560mW (3.4X!!)
32r ------ 450mW ---- 1,400mW (3.1X!
64r ------ 378mW ------ 712mW (1.8X)
128r ----- 261mW ------ 360mW
250r ----- 160mW ------ 184mW
500r ------ 88mW ------- 92mW
1K -------- 46mW ------- 46mW
2K -------- 24mW ------- 23mW
 

Attachments

  • Headphone-power-limiting.gif
    Headphone-power-limiting.gif
    14.7 KB · Views: 14
Historical reference: iPod and viral Cmoy. The Cmoy was clearly better in the higher Z phones, which for historical reasons were often the "better" phones when Cmoy was invented. Cmoy with the usual opamps was worse on 32, so hi-current Cmoy derivatives appeared. Both have amused many listeners. Interesting that 25mW-50mW at 1V-2V is widely acceptable.
 

Attachments

  • Cmoy-iPod-power.gif
    Cmoy-iPod-power.gif
    10.3 KB · Views: 12
PRR said:
> this actually seems to work:

Complicated and fickle. The opamp is surely class B. If you set a 30mA current limiter in series with each rail, you would get the same plot.
The low current of the simulation is because the the threshold of the circuit is low. Adding diodes somewhere and playing with the resistor values might raise it. But it's hard to test with the UniversalOpAmp2 model because it is also significantly current limited with a heavy load. Actually the 25 ohm trace might be wrong for this reason.

At any rate, the point of the sim was to show current and voltage could be used together to throttle power directly on the output which I believe it is. But I agree it is a little complicated and I will not be pursuing it further.

Regarding your last post, it is interesting that an iPod is such low power. It also makes it clearer than ever that TPA6120A2 is way over powered for headphones. Then again a powerful op amp like NJM4556 falls short (and it doesn't have the thermal pad on the bottom). So I'll just use a biggish buildout resistor and see how it goes ...
 

Latest posts

Back
Top