Class A Headphone Amp

[solid]

Hello everyone

I have been wanting to make a fully discreet headphone amplifier for some low impedance headphones, 32ohms, for some time. Since headphone amps are relatively low power, why not make it class A.
This amp is capable of over 70mW into 32 ohms and is heavily based on this amp here.

http://headwize.com/projects/showfile.php?file=balkishan1_prj.htm

comments welcome.

[Wavebourn]

Here is one more example (though, a practical design may be a bit improved against a dirty SIM version - I mean better arrangements of trimmers):

Best sounding headphone amp from Wavebourn to Kubeek :)

http://www.diyaudio.com/forums/attac...amp=1165512602

http://www.diyaudio.com/forums/attac...amp=1165512681

http://www.diyaudio.com/forums/attac...amp=1215726778

THD goes down with lowering of output level, especially higher harmonics disappear, it sounds very natural. Output follower is made of a pair MOSFET+BJT, for better transfer function approximation. It is loaded on a counter-modulated current source that greatly reduces current variations in an output voltage follower. Also, an input stage contains only 2 transistors, very short feedback path. And separate feedback loops in an input and output stages.

[CBS240]

comments welcome.

Hi

A few minor things I see that may not be optimal. The degeneration of the input transistors is a bit large. I think 10 - 20 Ohms should be plenty, assuming you intend to match the input pair. I would suggest a matched array here if you can spare a few extra bucks. Quality is quality. There is missing degeneration in the current mirror. Of course you could use a THAT 340 (or equivalent) complementary matched array, it is a quad and has two matched NPN and two matched PNP. You could use one for both the input stage and mirror, then you would not need degeneration in either.

Q5 and Q7 should have base resistors, ~1K or so, because if the VAS or the output swings to the negative rail, they may saturate and affect the current flow in Q3.

Speaking of Q3 as the tail CCS, 300uA in each leg is a bit too low. At least 1mA is more respectable, I'd go for a couple. Likewise, 600uA VAS bias is too low.

I would bet 100pf is way too large for the miller lag compensation cap. I would look into a two pole comp. scheme, MO.

Missing Thiele network.

I don’t quite understand the use of medium power devices in the output stage. Even if it is heavily bias with 20mA, at 5V, 100mW can easily be handled by a SOT-23. Seems a bit of sacrifice to me.

[GK]

Originally posted by CBS240



Hi

A few minor things I see that may not be optimal. The degeneration of the input transistors is a bit large.

I would bet 100pf is way too large for the miller lag compensation cap. I would look into a two pole comp. scheme, MO.

No way. Problem here is that the closed loop gain is rather low (~4). The LTP emitter degeneration has to be large, other wise a HUGE Cdom will be required for stability.

[solid]

Hi CBS240,

Thanks for taking the time to look at this schematic.

Originally posted by CBS240

Hi

A few minor things I see that may not be optimal. The degeneration of the input transistors is a bit large. I think 10 - 20 Ohms should be plenty, assuming you intend to match the input pair.

My target tail current is 500uA.
The degeneration resistor values chosen are to reduce the open loop gain so as to bring the loop gain at 20Khz down to 30db. Using 20ohm resistors would set the 20Khz loop gain (or NFB) at around 44db... probably too high for stability?

There is missing degeneration in the current mirror.

Yes, the amp may benefit from degeneration resistors in the mirror to balance the currents. Perhaps 100ohms...

Q5 and Q7 should have base resistors, ~1K or so, because if the VAS or the output swings to the negative rail, they may saturate and affect the current flow in Q3.

Perhaps someone else can chime in here. I am not sure on this myself.

Speaking of Q3 as the tail CCS, 300uA in each leg is a bit too low. At least 1mA is more respectable, I'd go for a couple. Likewise, 600uA VAS bias is too low.

I am targeting 500uA for both tail current and VAS current sources.
Increasing the Q3 tail source without any other change will increase open loop gain

I would bet 100pf is way too large for the miller lag compensation cap. I would look into a two pole comp. scheme, MO.

A 500uA input tail current and a 100pF miller cap gives a slew rate of greatar than 4V/us at 20Khz. This seems more than enough for this application which only needs 0.25V/us at 20Khz and 62mW into 32ohms.

Missing Thiele network.

I'm wondering if this headphone amp can do without it??

I don’t quite understand the use of medium power devices in the output stage. Even if it is heavily bias with 20mA, at 5V, 100mW can easily be handled by a SOT-23. Seems a bit of sacrifice to me.

The output bias current is targeted at 80mA; 400mW is a lot for a SOT-23 or TO-92 to handle. Under load, Q10 will need to deliver more than 80mA so dissipation will be higher. I'm actually planning to use a small heatsink for Q10 and Q11.

[GK]

Hmmmm. Why starve the LTP and VAS if the output stage is going to be biased at 80mA?

[Lumba Ogir]

Glen,
maybe to reduce global warming?

[solid]

Here are the node voltages






and device currents






VAS current can be increased to 1mA by changing R4 to 650ohms but this is not absolutely necessary. LTP current may also be increased to 1mA (change R3 to 650ohms) without affecting stability much.

[h_a]

May I ask you why you go for a downscaled standard opamp design?

When I see headphone-threads I somehow always expect to see something nicely simple, as the requirements for headphones are really low. Just a little current and let the preamp do the voltage gain. There were a few interesting designs around this, Grey's source follower and a nice jfet-amp (EUVLs DAO).

Wavebourns design goes then in the more all-in-one solution, nice!

Of course, if it suits you, go for it!

Have fun, Hannes

[Lumba Ogir]

solid,
consider a two-stage design. you don`t need that ugly VAS, all it does is provides a lot of distortion.

Hannes,
right?