DIY Headphone Amp - Comments and advice appreciated

[johnc124]

If I can rely on the work of the DC servo, I can remove the output cap and maybe have enough space for the input buffer opamp. But at this point I tried to avoid this by using a high impendance input (100k) and a FET input opamp (to avoid excessive noise).

But I'm currently neither satisfied with the choice of input opamp (OP1652 seems most promising, but has a significant current noise, ISL28210 looks very good, but is much expensive and a bit rare.), nor with the choice of DC servo opamp (TL072 is cheap and fits with medium audio performance, but has a high offset voltage (some mV), other FET-opamps are more expensive, bipolar opamps have too high offset currents for the 1Meg resistor).

Do you have some good proposals for these two opamps?

Was lurking on your thread and thought I'd chime-in when I saw one of my products mentioned. The current noise of the OPA1652 is NOT 0.5pA/rtHz, that would be pretty ridiculous for an op amp with MOSFET input devices. Yes, I said MOSFET, it's not a JFET. The marketing people at TI got scared that audio folks wouldn't give a CMOS op amp a chance, and so they just called it a "FET-input". To be honest, I'm not sure where the 0.5pA/rtHz number came from, the real number should be pretty close to the value given by the shot noise calculation of the input bias current. For 10pA of input bias current, that would give a current noise of sqrt(2*1.6e-19*10pA) = 1.8 fA/rtHz.

Datasheet issues aside, *face palm*, I'm a big fan of the OPA1652. I think it performs above its price, especially in terms of noise. Last week, I measured one to have a broadband noise spectral density of 3.8nV/rtHz, that's pretty good in the realm of CMOS and JFET op amps and even lower noise than an OPA827!

Other good options are the OPA172 family, or the new OPA1688. Since the OPA1688 was just released it might be a few weeks before distributors are carrying it.

 

[00940]

A few additional points:

- with an opa1652 and the resistors value as is, you have an offset of about 1mV with the "typical values" from the datasheet (of about 3mV with worst case values). You really don't need a DC servo at this point, especially since your input is AC-coupled.

- if you're concerned with lack of pcb space: I'll repeat myself, use a quad opamp. It makes a lot of things easier. And it's cheaper too.

- I didn't notice earlier: it's good practice to have a high value resistor from the + input of your amp to gnd, in case the pot goes bad.

 

[DPH]

I would also try the OPA1652. It sounds really good(for a JFET op-amp) and is less than $2 at Mouser.
In my opinion, it's the best sounding of the OPA16x2 series including the bipolar input OPA1662.

OPA164X = jfet and lowest noise. John Caldwell says they're in the process of revising its datasheet to reflect some amount of process changes that affect its performance (in a good way). Lowest bias current of the lot and, as result probably going to be the lowest noise of the 3. Highest DC gain, but lowest GBW (which is probably the most material @ 20kHz) of 10 MHz vs the ~20 MHz of the other 2. Could change with the update. Most expensive of the 3 parts.

OPA165X = CMOS part. Higher knee in the noise curve than the OPA164X series, even if a slightly lower HF noise level (according to the datasheets we have in front of us today).

OPA166X = BJT input part. 600 nA input bias current vs the <=10 pA of the other 2 = highest noise part under most circumstances, but the lowest noise frequency knee. I don't know if you can get a feedback network low-enough impedance to nullify the slight advantage it has in e_n, see figure 3 on the 1662 datasheet.

They're all pretty similar and all darn good parts. Hard to see where I'd pick the 166X part over the 165X part, especially given the 1652 is a little bit cheaper than the 1662, but someone can correct me if I've missed a notable advantage. The LM4562 also makes it hard to recommend the 166X unless you need the low idle current. 164X is the low-noise champ here, at least at higher source impedances, but also the most expensive of the 3.

Edit to add--looks like John himself beat me to the punch. 🙂

 

[tobias32]

Was lurking on your thread and thought I'd chime-in when I saw one of my products mentioned. The current noise of the OPA1652 is NOT 0.5pA/rtHz, that would be pretty ridiculous for an op amp with MOSFET input devices.

Yes it is. But I wonder why you don't just revise the datasheet. I'm a bit worried, as the "error" is present in every aspect of noise current data, including the graph estimating the resulting noise voltage density for varying input resistors.

Other good options are the OPA172 family.

Most of the data reeds very good. But as a rail-to-rail opamp i'm just a bit worried about the very low PSRR.

 

[johnc124]

Yes it is. But I wonder why you don't just revise the datasheet. I'm a bit worried, as the "error" is present in every aspect of noise current data, including the graph estimating the resulting noise voltage density for varying input resistors.

Most of the data reeds very good. But as a rail-to-rail opamp i'm just a bit worried about the very low PSRR.

"Just revise the datasheet" is a bit easier said than done. The OPA1652 datasheet is on my list to revise, but I'm currently working on the OPA1642 as mentioned above. Basically, any time you "just revise a datasheet" you have to convert the whole document to the latest standards mandated for datasheets. Which triggers quite a bit more work for me than just correcting the noise data. Trust me, errors like this in the datasheets for my products drive me INSANE, and I am working to fix them.

I have a feeling the reason the error propagated through the to the total noise curves is that the curves were most likely made by an apps engineer who was just given the noise specs and nothing else.

As for the comments that were made above about OPA1642 and OPA1652. In a lot of circuits the OPA1652 is a lower noise option. 1/f noise usually contributes little to the total RMS noise in audio bandwidths. The big caveat to this is in phono preamp circuits where the gain is also much higher at low frequency. Here the JFET input of the OPA1642 really shines, it makes a FANTASTIC pre-amp for MM phono cartridges.

Also, the three-stage topology of the OPA1642 preserves high-frequency linearity better into low impedance than the two-stage folded cascode topology of the OPA1652. But if the load impedance is above about 2kOhms, this really isn't a factor.

For bipolars, OPA1662 is really for low power applications or low supply voltage applications. A higher performance option would be the OPA1602, and as mentioned before the LM4562 is really a great part as well.

 

[DPH]

^ Thanks, John, for setting things straight. 😀

 

[tobias32]

- with an opa1652 and the resistors value as is, you have an offset of about 1mV with the "typical values" from the datasheet (of about 3mV with worst case values). You really don't need a DC servo at this point, especially since your input is AC-coupled.

Yes, this would be a great pro for the FET + high input impedance solution.

- if you're concerned with lack of pcb space: I'll repeat myself, use a quad opamp. It makes a lot of things easier. And it's cheaper too.

I'm a bit cautious atm, since I'm not shure about the opamp. As not all opamps discussed are available as quad I could shoot myself in the foot if I redesign the layout for a quad opamp and lateron decide for an opamp only available as dual. That's the reason I'm pushing the opamp question, as it is vital for progress of the project.

I'm even considering the THAT1200 + buffer solution, as this results in really great CMRR behaviour. But I think I would need the DC servo again (with up to 10mV offset and no AC-coupling.)

- I didn't notice earlier: it's good practice to have a high value resistor from the + input of your amp to gnd, in case the pot goes bad.

Ah ok, that's a great idea. Thanks.

 

[tobias32]

@johnc124:

Alright, so I can assume the OPA1652 to have a noise resistance >>100k (what indeed makes it very interesting again).

Btw: Thanks for clarification.

 

[abraxalito]

I'm even considering the THAT1200 + buffer solution, as this results in really great CMRR behaviour.

Is great CMRR something you really need when you have a balanced source? Since most people get by fine with unbalanced (which has zero dB CMRR) I'm wondering how good CMRR is good enough on line level signals?

 

[00940]

I'm even considering the THAT1200 + buffer solution, as this results in really great CMRR behaviour. But I think I would need the DC servo again (with up to 10mV offset and no AC-coupling.)

The question is: do you really need to go to such length ?

You're at line level and at the end of the chain. It's not quite clear what your source is but if you're in pro application, you probably can have a strong signal. The signal to noise ratio will be easier to manage by this fact alone. Plus, you don't need to worry about noise being amplified later on.

 

[tobias32]

Alright, I think I got it 😀 No dedicated line receiver 😛 So I'll stick with 54 dB CMRR (0.1% resistor matching) which will tolerate line inbalances up to about 100 ohms (with 100k resistors).

Yes, the regular input will be pro audio equipment with balanced 75 ohm ouputs, connected by about 3m of shielded patch cable.

So if I can drop the DC servo, too, I should even have enough space for a quad opamp (for each channel).
Which leads me to 2 final topologies:

1) based on dual OP1652 or ISL28210 (sry, I couldn't decide yet 😛)
AC coupling -> CMRR stage (4x 100k 0.1%) -> pot -> driver stage for LME49600 -> LME49600

2) based on quad LME49740
AC coupling -> 2x buffer (Gain +1) -> CMRR stage (4x 2k 0.1%) -> pot -> driver stage for LME49600 -> LME49600

Which one would you prefer?

 

[johnc124]

There is also a quad of the OPA1652, the OPA1654, which would be cheaper per op amp than the OPA1652.

 

[00940]

If you want to use a quad per channel, your second option can offer very high input impedance, which is nice if you want to // all your 16 amps on one source.

I'd use the opa1654 anyway.

 

[tobias32]

If you want to use a quad per channel, your second option can offer very high input impedance, which is nice if you want to // all your 16 amps on one source.

I'd use the opa1654 anyway.

I don't think I will ever do this 😀 This would be much easier with 1 input stage.

No, it's only sound quality which matters for me in this decision. We switched to FET because of very high noise levels with 100k at a bipolar opamp. Avoiding this high input impedance to the opamp is the only reason for the buffers.

So I thought, LME49740 has a slight better sounding than OPA1564. Basically it comes down to this for this decision.

 

[00940]

Out of curiosity: in which application do you need 16 high quality headphones amps ?

 

[tobias32]

Oh, haven't I mentioned it, yet?

The Amp is determined for pro audio IEM. Because I can't afford a personal monitoring station for every band member (I could, but only if massivly cutting down on the quality of the other components), I'm setting up my own solution.
Current digital mixing consoles include the capability of personal monitoring mixing, I just need an amp afterwards. So here we are 😉

I know I'm going a bit too far with my quality demands. Surely that's a bit ambition, too, but basically the idea behind the high quality demands is, that vocalists or players of instruments without discrete tones (classical instruments) need to strike exactly the right note. And a crystal clear sound makes it much easier for the musician to distinguish their own sound from the mix and focus on singing/playing instead of listening.
But again, I admit that there's also some ambition in this project 🙂

 

[abraxalito]

Ah so you mean the sources for these amps are going to be the DACs included in a commercial digital mixer for individual personal mixes? In which case the bottleneck is almost certainly going to be those DACs and their following analog signal processing stages, not your amps. Discussions over SQ of opamps are really rather moot in this context - almost for sure there's going to be a loaded down CMOS opamp in the chain upstream.

<edit> If I've understood your needs correctly, what you really would like to have (funds permitting) is one of these setups ? - http://www.soundonsound.com/sos/sep11/articles/hear-technologies-hear-back.htm

 

[tobias32]

Ah so you mean the sources for these amps are going to be the DACs included in a commercial digital mixer for individual personal mixes? In which case the bottleneck is almost certainly going to be those DACs and their following analog signal processing stages, not your amps. Discussions over SQ of opamps are really rather moot in this context - almost for sure there's going to be a loaded down CMOS opamp in the chain upstream.

Yeah, I'm aware of that. But I would like to put it into perspective a bit:

As I know, the D/A part is the by far most important and often weakest part of a digital mixing setup with regard to sound quality I'm not going to use "DACs included in a commercial digital mixer" but a special D/A box for all audio interfacing with the console. I tried to select the box with the most promising sound quality as claimed by professional reviews (based on listening and measuring). I ended up with a Midas box, who have one of the best reputations among pro audio business. (Just to clarify, I mean Midas, not the Behringer "designed by Midas".)

Certainly you're still right and the DACs will be the weaker part of the system, but maybe not that far behind this amp and I definitely don't want to contribute noticeable distortions, so it should stay better than the DACs.

But the most important point is that I already stated towards 00940:
Even if initiated by a real world need, this is a project out of curiosity and ambition!
I want to gain experience, how to build a very high quality system (with limited invest), have the fun building it and be proud about the result.

I hope it was understandable 😛 Sorry for the long explanation 😉

If I've understood your needs correctly, what you really would like to have (funds permitting) is one of these setups ? - Hear Technologies Hear Back

Yes, something like this, even if this particular setup seems rather basic for me. You can achieve a real high convenient setup if you integrate it with the mixing console (f.e. direct access to each input, mixing channels, ...).

So I had a look on Allen & Heath ME1 or Aviom A360.

But I'm very contented with my solution, too, as I nearly get the same convenience (same functionality, only controlled by smart phone instead hardware), but with a self built amp and a certainly better sounding DAC than it is integrated in these boxes.

 

[sgrossklass]

I wouldn't get hung up on DAC performance too much. If all that you're after is audible transparency in D/A, a lowly Realtek onboard sound chip will do fine, or an internal soundcard around the 30€ mark, or an external one starting at around 50€. Of course none of these have to output 0dBFS at around the +20 dBu mark (balanced), and dizzying distortion and SNR measurements will cost extra, but still, you get the point. The diminishing returns curve in the D/A world is far steeper than in analog.

Which leads me to 2 final topologies:

1) based on dual OP1652 or ISL28210 (sry, I couldn't decide yet 😛)
AC coupling -> CMRR stage (4x 100k 0.1%) -> pot -> driver stage for LME49600 -> LME49600

2) based on quad LME49740
AC coupling -> 2x buffer (Gain +1) -> CMRR stage (4x 2k 0.1%) -> pot -> driver stage for LME49600 -> LME49600

Which one would you prefer?

The latter would have a lower-noise input, but is still lacking a gain control. You will probably be running pro levels into the amp (i.e. +4 dBu nominal), which are even higher than consumer CD player level. Even for the latter, an amplifier gain of unity has been found to be ample for IEMs - heck, it'll generally do fine for 250 ohm DT880s! You should do the math with levels, gains and sensitivities of all involved, upon which you'll probably discover that the setup will go to hearing damage territory easily.

It is very likely that you will want to implement a few different input stage gain settings, usually with an inverting amplifier stage after the line receiver and a set of DIP switches in some user-accessible spot (note implications on board layout). Steps of around 8 dB tend to be most common. Since you already have 6 dB of post-gain, I'd shoot for maybe +2 / -6 / -14 / -22 dB or something like that. You can also go bigger, like +4 / -6 / -16 / -26 dB. It also tends to be useful to include a bit of attenuation in the line input stage already, for somewhat more sensible levels in following stages, typically -3 to -6 dB.

Sometimes it really isn't a bad idea to take a look at some commercially available designs (and their manuals), like, say, a Lake People amp. They're not necessarily perfect, but should give you an idea what sort of features may prove useful in practice.

 

[tobias32]

Sometimes it really isn't a bad idea to take a look at some commercially available designs (and their manuals), like, say, a Lake People amp. They're not necessarily perfect, but should give you an idea what sort of features may prove useful in practice.

Oh, that's a good idea. I read through some of their manuals and they really have a sense for modularity.
I don't think I need the possibility to switch that much options during use (so by jumpers, etc.) - beside Pre-Gain, which is a very good idea.

Where I definitely want to be modular is during assembly. I nearly finished a layout, which is - depending on assembly - able to implement all discussed schematics/topologies.
Thus I can simply order the parts for different topologies and have a side by side test of both configurations.

The latter would have a lower-noise input, but is still lacking a gain control. You will probably be running pro levels into the amp (i.e. +4 dBu nominal), which are even higher than consumer CD player level. Even for the latter, an amplifier gain of unity has been found to be ample for IEMs - heck, it'll generally do fine for 250 ohm DT880s! You should do the math with levels, gains and sensitivities of all involved, upon which you'll probably discover that the setup will go to hearing damage territory easily.

It is very likely that you will want to implement a few different input stage gain settings, usually with an inverting amplifier stage after the line receiver and a set of DIP switches in some user-accessible spot (note implications on board layout). Steps of around 8 dB tend to be most common. Since you already have 6 dB of post-gain, I'd shoot for maybe +2 / -6 / -14 / -22 dB or something like that. You can also go bigger, like +4 / -6 / -16 / -26 dB. It also tends to be useful to include a bit of attenuation in the line input stage already, for somewhat more sensible levels in following stages, typically -3 to -6 dB.

As mentioned above, this is a very good idea!

Just as a side comment: Whereas it is called "pre-gain" I think it would be the best to implement the damping directly before the post-gain (last opamp, excluding the output buffer).
[I once learned to always keep the levels as high as possible (if not driving an amp into compression) for as long as possible. Thus all opamps prior to the last will operate in the best possible THD+N conditions and all distortions or induced signals will be damped as well. It would be completely the other way round if the "pre-gain" would be positive.]

One comment only @sgrossklass:
Did you read my reply to your last post? http://www.diyaudio.com/forums/head...omments-advice-appreciated-6.html#post4472944
I would be curious about your answer to my questions about your comments - especially points 3 to 5.