The Objective2 (O2) Headphone Amp DIY Project

[MrSlim]

I have just opened a new "official" O2 Front Panel Group Buy thread here: O2 Front Panel Group Buy The original one was for interest and design discussion. This one is for orders..

[RocketScientist]

Originally Posted by abraxalito

where the O2 has just two paralleled 4556 stages in its drive stage yours will have 3. That'll allow you to raise the supply voltage to 15V without compromising the thermals too much.

Nice idea, but the devil is in the details. The O2 output stage works well because the the two paralleled op amps are on the same silicon die and inherently well matched. But once you try to parallel another package, all bets are off.

The offset voltages of the 4556 (or any similar op amp) become an issue when paralleling them with just 1 ohm in series with each output. The worst case spec for offset is +/- 6 mV. So you could have one op amp at +6 mV and another at -6 mV. That yields 12 mV with 2 ohms of total resistance resulting in 6 mA of DC current between the op amps. That exceeds the bias current of each output stage and the distortion rises dramatically (around 100X!).

Even if the offsets differ by only 2 mV you still get 1 mA of extra DC current between the paralleled op amps. That might not seem like much but it's still about a 20% change in the output stage quiescent current and enough to throw off the bias current and increase distortion.

You could increase the value of the isolation resistors to at least 4.7 ohms and probably be OK with most samples of the 4556. But you still would need to check to make sure the offsets were reasonably close. Otherwise 10 ohms might be necessary. Using even four op amps in parallel with 10 ohm isolation resistors would give a 2.5 ohm output impedance compared to the 0.5 ohm output of the O2--5 times worse. So it comes down to how high of an output impedance you're willing to live with in exchange for more voltage swing.

Again, I think the better solution is to use the BUF634, LME49600, or similar, and not try to parallel more than two 4556 op amp sections in the same package. But if you're OK with a higher output impedance or perhaps hand matching IC's for similar offsets it could work. But it's not really a one-size-fits-all production ready solution.

[abraxalito]

Originally Posted by RocketScientist

Nice idea, but the devil is in the details. The O2 output stage works well because the the two paralleled op amps are on the same silicon die and inherently well matched. But once you try to parallel another package, all bets are off.

So you found that because the parts are on the same die, their offset voltages are extremely well matched? And track well with temperature? Doug Self did a design of a speaker amp using literally dozens of NE5532s and although I read his article a while back, I don't remember him reporting this effect as a problem. Did he overlook something vital? Perhaps he got away with it because the NE5532's typical offset spec is lower? Or by averaging over a greater number of parts the distortion generated in the high offset parts got attenuated by the majority of parts that had more typical offsets?

The offset voltages of the 4556 (or any similar op amp) become an issue when paralleling them with just 1 ohm in series with each output. The worst case spec for offset is +/- 6 mV. So you could have one op amp at +6 mV and another at -6 mV. That yields 12 mV with 2 ohms of total resistance resulting in 6 mA of DC current between the op amps. That exceeds the bias current of each output stage and the distortion rises dramatically (around 100X!).

100X distortion only at low level or even at levels where the output current considerably exceeds the inter-device current?

You could increase the value of the isolation resistors to at least 4.7 ohms and probably be OK with most samples of the 4556. But you still would need to check to make sure the offsets were reasonably close. Otherwise 10 ohms might be necessary. Using even four op amps in parallel with 10 ohm isolation resistors would give a 2.5 ohm output impedance compared to the 0.5 ohm output of the O2--5 times worse. So it comes down to how high of an output impedance you're willing to live with in exchange for more voltage swing.

This is all interesting stuff, thanks. So I might just seek out a lower offset opamp to pursue the idea with. Or just go with LT1010 and have done with it

Again, I think the better solution is to use the BUF634, LME49600, or similar, and not try to parallel more than two 4556 op amp sections in the same package. But if you're OK with a higher output impedance or perhaps hand matching IC's for similar offsets it could work. But it's not really a one-size-fits-all production ready solution.

Yeah I was considering that since only 1% or less of guys would want/need this, its just a roll-it-yourself idea, not for full production. Going to BUF634 or similar is rather a quantum leap in expense compared to your (very elegant) solution.

[RocketScientist]

@abraxalito, I'd have to go look again at the Doug Self amp, but I thought it had isolation resistors of at least 4.7 ohms? Paralleling op amps is a very real problem and has been documented by TI, National, and others. As Doug Self has well documented, the performance of Class AB output stages is very dependent on the DC bias. And any DC current in the output throws off that bias.

The increase in distortion is mainly crossover distortion so it's worse at low output levels. The 4556 has almost zero visible crossover distortion on its own, but if you push its bias point much higher or lower it becomes a significant problem. The paralleling creates an asymmetric DC current in the output stage which only has a few mA of bias current to begin with.

There's also an issue with the gain stage. I optimized the design close to the NJM2068's limits for output current at low distortion. If you add more op amps to the output, it increases the load on the gain stage. So I would suggest giving up 3 dB of noise and using the NE5532 instead in the gain stage if you want to add more output op amps.

As I suggested earlier, those needing more than 7 Vrms might want to wait for the desktop version of the O2 or to use a design that can manage more output dissipation without any worry about bias, increased distortion, gain stage loading, etc. The LT1010 is my least favorite output buffer. It has lower current capability (150 mA) and performs significantly worse than either the BUF634 or LME49600. Still, it might be better than trying to push the 4556 further than makes sense.

[abraxalito]

I've just downloaded the JRC4556 datasheet and found nothing within it to suggest that devices in the same package will have better matched offsets than between packages. So I'm curious what are you basing your claim of :

The O2 output stage works well because the the two paralleled op amps are on the same silicon die and inherently well matched.

on, if not the datasheet? Do you have a more comprehensive edition than that available on the web?

[RocketScientist]

Originally Posted by abraxalito

I've just downloaded the JRC4556 datasheet and found nothing within it to suggest that devices in the same package will have better matched offsets than between packages. So I'm curious what are you basing your claim of :

The O2 output stage works well because the the two paralleled op amps are on the same silicon die and inherently well matched.

on, if not the datasheet? Do you have a more comprehensive edition than that available on the web?

It's common knowledge that semiconductors made on the same die tend to be much better matched than those made on different dies. I've tested a few dozen different 4556 parts with half a dozen different date codes, and every single one has been well matched between the two halves of the same die.

It's no different than the dual FETs often used in input stages in discrete designs, They're inherently well matched as they're made on the same die. It's just the way semiconductor manufacturing works. It's not something you find in a datasheet.

[abraxalito]

Originally Posted by RocketScientist

It's not something you find in a datasheet.

Ah so not the datasheet then, thanks.

<edit> I just realised there's no real reason for three opamps because the limits are stated as being thermal. Just two opamps in separate packages will do the trick. Empirically chosen for low enough offset

[Pars]

Originally Posted by RocketScientist

It's common knowledge that semiconductors made on the same die tend to be much better matched than those made on different dies. I've tested a few dozen different 4556 parts with half a dozen different date codes, and every single one has been well matched between the two halves of the same die.

It's no different than the dual FETs often used in input stages in discrete designs, They're inherently well matched as they're made on the same die. It's just the way semiconductor manufacturing works. It's not something you find in a datasheet.

The dual FETs I've used (and tested, at least for Idss) aren't matched all that well oftentimes. I can get much better matches using singles. Thermally coupling them can be an issue (hence the heatsinks sold on a GB here awhile ago). FETs include 2SJ109, 2SK389 (and their singles, 2SJ74 and 2SK170).

[qusp]

agreed, i will use well matched thermally bonded j74 over j109 every time for this reason. j109 are only specced to match 10% from memory, its possible to get far better than that selecting pairs of singles and if you use a common heatsink such as EUVL's or just silver epoxy you can get them to track well thermally too.

[musical]

The 9V NiMH battery is usually either 8.4V or 9.6V. Which is more suitable for O2? Also Thomas Distributing has a Lithium Polymer 8.4V battery iPower 9V 500mAh Lithium Polymer Rechargeable Battery Can that be used?