DIY Headphone Amp - Comments and advice appreciated

Status
Not open for further replies.
T

tobias32

Member
Joined 2015
Hello,

I designed a diy headphone amp based on the LME49600 reference design.

Before building it, I would appreciate the advice and comments of some experienced "veterans" as it is my first project like this. (Not completely new to the audio world, but the first project with high demands to fidelity and quality.)

As indicated my target is to achieve very low THD+N over a wide variety of headphones to attach.

The system is going to run as a multichannel system (16 channels) fed with differential input signals (of about 600 ohms output impedance).

Thanks in advance for every helpful advice and hint (as also quality estimations)!

An externally hosted image should be here but it was not working when we last tested it.



An externally hosted image should be here but it was not working when we last tested it.


(Sry for not having placed the name and value labels)
 
Using LME49720 with its typical 10nA (worst 72nA) input bias current isn't going to get you a particularly low offset voltage in practice when the source impedance seen is 1Mohm. The part looks to have bias current compensation so there's no knowing the polarity of that bias current - it could be opposite at each input terminal. If low offset is the aim, with high impedances you'll need a JFET or CMOS opamp type as your servo.

Incidentally do you care how this sounds or will you be satisfied if the meter reads low enough?
 
abraxalito said:
Using LME49720 with its typical 10nA (worst 72nA) input bias current isn't going to get you a particularly low offset voltage in practice when the source impedance seen is 1Mohm. The part looks to have bias current compensation so there's no knowing the polarity of that bias current - it could be opposite at each input terminal. If low offset is the aim, with high impedances you'll need a JFET or CMOS opamp type as your servo.
Click to expand...

Good point, thanks!
Can you recommend an opamp of this type in same quality of LME47200?

abraxalito said:
Incidentally do you care how this sounds or will you be satisfied if the meter reads low enough?
Click to expand...

That depends on what you mean by "how this sounds". I want to use it as studio monitoring, so high fidelity and linear frequency response are the main design targets.

But I certainly don't want to start a "measurement vs. listening" war 😀
 
I found the OPA2140 is a JFET one with max 10pA bias current and nice audio performance, too, although it has a very poor PSRR and costs about twice as much.
There is also OPA1642 with very similar data (compared to OPA2140) and is only a bit more expensive than the LME49720.
Is there a more favorable one?
 
The quality of the servo amp matters only to the extent of its error contribution to the signal. Normally the servo amp feeds the main amp through an attenuator as it only needs to contribute at worst a few 10s of mV. So its quality is far less of an issue than the main signal opamps. I'd say a TL082 would do a fine job in practice provided its output is attenuated at least 30dB. If you need lower offset then OPA2172 looks to be a good choice at a reasonable price.

I note that your servo amp has only 6dB attenuation (1k + 1k divider) so I'd suggest you raise the value of R16 to 100k or so to render its error contribution to the audio negligible.

When I asked about how it sounds I meant subjectively sounding transparent. Which means reducing all audible error contributions from the amplifier. They don't in general show up on the THD+N reading. If you only care about THD then some of my own practices in design I will stay silent about so as not to waste either of our time 🙂
 
abraxalito said:
When I asked about how it sounds I meant subjectively sounding transparent. Which means reducing all audible error contributions from the amplifier. They don't in general show up on the THD+N reading. If you only care about THD then some of my own practices in design I will stay silent about so as not to waste either of our time 🙂
Click to expand...

No, please don't stay silent 😉 Practical experience is definitly the target of my question.

The reason I am fixated on these numbers is just that I don't have enough experience of myself to intuitively say about the schematics if it sounds completely transparent so I have to use the datasheets to get an impression of the performance.
 
OK then my practical experience suggests that the power supply is the place to focus attention when subjective transparency is desired (as opposed to lowest meter readings). Note that extra attention to power supply won't make the meter readings any worse.

With classAB amplifiers there's noise generated on the supply rails by virtue of the fact that most of the time only one of the two output transistors is actively passing current. In your schematic the primary source of noise is the LME49600 and you want to avoid this noise getting on to your audio signal via less than perfect PSRRs in your various active devices. Ultimately your audio performance will be limited (talking subjectively here) by the PSRR of the LME49600 as that's the source of the noise.

Going then to the LME49600 DS we note that the PSRR plots don't inspire any kind of confidence. Compare and contrast 30029889 and 30029891 (small numbers to the bottom right of the plots). On my copy (Jan 2008) they have the same legend which implies they're of the same measurement. But they're different so perhaps they screwed up the legends?

Assuming though that these are indicative of the PSRR you'll get in practice is an assumption too far. Internally the design's a diamond buffer so its quite possible to run a simulation of this, using LTSpice and discrete transistors. What you'll find is that the PSRR depends on the load impedance connected and also it'll depend on the output voltage (and hence output current). We can safely assume the plots in the DS were made (could even be simmed) with infinite load impedance connected and hence not real-world.
 
The power supply is a conservative LM3x7 regulated (toroidal core) transformer supply.

I must admit, I went without an extra ripple filter as I relied on the PSRR of the opamps. As I thought about possible magnetic injection I favored this approach. But if at least the datasheet of the LM3x7 is correct, it should be attenuated below the noise.

The part after the transformer is shown here:
An externally hosted image should be here but it was not working when we last tested it.
 
With the DSs for the LM317 and LM337 you'd do well to bear in mind the output impedance specification is shown for (I think, if memory serves) 500mA of output current. Most of the time you'll be running much less than this and so the plots are, once again, very optimistic in practice. With say 50mA output current you can safely expect the Zout to rise 20dB.

For your PSU schematic I'd delete the 0R15 resistor and have a modest amount of capacitance fed directly from the bridge. Let's say 2,200uF. Then fit a resistor before the next cap which probably only needs to be 2,200uF. The low frequency ripple rejection of these regs is pretty good when their adj pin is also nicely decoupled so no point in adding too much capacitance at the input. You should pay attention to HF though (where the cap's ESR is dominant over its impedance) because 3 terminal regs suck pretty badly above 100kHz.
 
abraxalito said:
For your PSU schematic I'd delete the 0R15 resistor and have a modest amount of capacitance fed directly from the bridge. Let's say 2,200uF. Then fit a resistor before the next cap which probably only needs to be 2,200uF. The low frequency ripple rejection of these regs is pretty good when their adj pin is also nicely decoupled so no point in adding too much capacitance at the input.
Click to expand...
Alright, sounds good - also in aspects of the board size.

abraxalito said:
You should pay attention to HF though (where the cap's ESR is dominant over its impedance) because 3 terminal regs suck pretty badly above 100kHz.
Click to expand...
What do you mean by HF? Frequencies from about 100kHz to (very approximate) 100 MHz should be handled by the 100n at both sides, shouldn't them?
If frequencies above also should be taken into account, I could add a NP0 cap (sth. from 100pF to 1nF).

abraxalito said:
With the DSs for the LM317 and LM337 you'd do well to bear in mind the output impedance specification is shown for (I think, if memory serves) 500mA of output current. Most of the time you'll be running much less than this and so the plots are, once again, very optimistic in practice. With say 50mA output current you can safely expect the Zout to rise 20dB.
Click to expand...
I could imagine, that's a bit more relaxed because of the 16 channels. The LME49600's alone draw about 220mA idle current (13.8mA each). Then there are at least 32 additional opamps of about 5mA idle current. So there should be a bias current of about 400mA over all.
 
100nF doesn't have a very low impedance at 100kHz. Beyond a few MHz if its a leaded component it'll be mostly an inductor rather than a cap. Some series impedance would be helpful, not just relying on shunting the noise to GND.

Ah I'd overlooked the number of channels, looks like you'll be using the LM3x7 regs to good effect then if they're idling around 400mA. But won't the additional current to power the headphones stand a chance of exceeding their current limit? Depending on the impedances you're driving you should calculate what the peak current draw is likely going to be.
 
abraxalito said:
Ultimately your audio performance will be limited (talking subjectively here) by the PSRR of the LME49600 as that's the source of the noise.
Click to expand...

No. The LME49600 being in the feedback loop of an opamp, any noise making it to its output will be greatly attenuated, provided that the sensing points of the input opamp are not contaminated by the LME49600's return currents. The LME49600 isn't on its own here. It's just a matter of correct pcb layout.
 
abraxalito said:
100nF doesn't have a very low impedance at 100kHz. Beyond a few MHz if its a leaded component it'll be mostly an inductor rather than a cap. Some series impedance would be helpful, not just relying on shunting the noise to GND.
Click to expand...

I'm careful with series impedance after the regulator as it decrases its (effective) regulation performance.
What particularly would you suggest?

abraxalito said:
Ah I'd overlooked the number of channels, looks like you'll be using the LM3x7 regs to good effect then if they're idling around 400mA. But won't the additional current to power the headphones stand a chance of exceeding their current limit? Depending on the impedances you're driving you should calculate what the peak current draw is likely going to be.
Click to expand...

I designed the PS for 1.5 A continuous and about 2 to 2.5 A peak (125mA each channel simultaneously).
I know this cannot fully drive out the LME49600 @ 32 ohm load, but this was the limit before regulation gets dificult and I don't know if someone wants to have more than about 500mW peak in his ears (2 channels).
Above 64 ohms load this is not a problem anymore, as the voltage is limiting.
Additionally if not all 16 channels are handling the same signal, the peaks are distributed and the over all current limit should not be reached.
 
tobias32 said:
I'm careful with series impedance after the regulator as it decrases its (effective) regulation performance.
What particularly would you suggest?
Click to expand...

I'm suggesting series impedance prior to the regulator - perhaps an inductor would suit? Something that's going to give increasing rejection with rising frequency to counteract the inductive nature of caps at higher frequencies.

Moving downstream you'll want to have separate decoupling for the output buffer so its haversine currents (with lots of harmonics extending above the audio band) don't contaminate the opamp supplies in any way. Perhaps even consider separate shunt regs for the opamps?
 
abraxalito said:
The explanation did not pertain to my statement. Perhaps you missed the qualifier - let me point it out for you : 'talking subjectively here'.
Click to expand...

No. You made a technical statement, then followed it with with a meaningless qualifier. I corrected only the technical statement.
 
Status
Not open for further replies.
Back
Top