I'm designing a headphone amp, but I've got some trouble in choosing the output device. I wanted to do some distortion measurements to back up listening tests. But my new employer hasn't got the needed toys. 
So what output device would have the least distortion an largest bandwith in this configuration? I tested the BJT, but the input impedance of the transistor seems a bit small. So I guess I have to choose between mosfet and darlington.
The configuration itself isn't completeley finished. I plan on using a current source instead of R4 to improve linearity. Also I'll check if a resistor/current source from the emittor of Q1 to -VCC improves bandwith/linearity.
So what are your thoughts on this?
So what output device would have the least distortion an largest bandwith in this configuration? I tested the BJT, but the input impedance of the transistor seems a bit small. So I guess I have to choose between mosfet and darlington.
The configuration itself isn't completeley finished. I plan on using a current source instead of R4 to improve linearity. Also I'll check if a resistor/current source from the emittor of Q1 to -VCC improves bandwith/linearity.
So what are your thoughts on this?
An externally hosted image should be here but it was not working when we last tested it.
> I tested the BJT
And what is wrong with it?
> the input impedance of the transistor seems a bit small.
It should be on the order of 1K-2K. If the op-amp is any good, that should not be any problem.
> So I guess I have to choose between mosfet and darlington.
MOSFET input capacitance is VERY difficult to use this way. On simulator I ended up decoupling the MOSFET Gate above the audio band so I could take feedback direct from the opamp. People who built that reported different results, which tells me that it is far less stable in real life than on a simulator (SPICE lives in a fantasy world).
Darlington emitter followers work, but still end up with a double-pole rolloff about the place many chip opamps are gaining excess phase. And IMHO the Darlington is not needed.
> largest bandwith
MOSFET input is a BIG capacitor, enough to embarass the opamp. Or if the opamp can drive the capacitance, it is beefy enough to drive the load, so why have the MOSFET?
Darlington is more subtle. You can do it and it will probably be stable. Input capacitance seems small; problem is that there are two low-passes and above Ft/(√ β) (which may be around 100KHz) the input impedance drops faster than a pure capacitor. You have to ask why you "need" a Darlington.
People who built a similar design with just a TIP41 or similar simple power BJT report that it is very clean and apparently stable.
With a simple fairly-fast BJT, and a half-decent audio opamp, bandwidth will be over 10MHz. If you need more than 10MHz things get a little tougher. But I don't believe we hear even 10MHz all that well. At 20KHz, response should be down something like 0.02dB re: DC or 0.01dB re: 1KHz, pretty flat.
Your diagram is so darn big that I can't see the whole thing on my monitor, but your DC connections don't smell right. You either need a bipolar +/- supply or several DC-block capacitors.
This version will work; I know because others have built it with no help except this picture:
TIP41 is handy in the US; your BC/BD parts will be just fine. Output devices for a small power amp or drivers for a large power amp. 40V, 2A, >20W, Ft>10MHz if possible.
The "0.2A IDC" current source can be replaced with a resistor....
> current source instead of R4 to improve linearity
Linearity is not affected enough to measure. Taking your 47 or 50Ω resistor and assuming 10V rail, emitter current is 10V/50Ω= 200mA. Emitter resistance at idle is about 27Ω/200 or 0.14Ω. This will vary with signal: say 0.2Ω to 0.1Ω. The resulting 0.1Ω variation is very-very-small compared to a 47Ω||32Ω= 19Ω total load. It has 46dB NFB just in the emitter-follower! Plus another 20dB+ NFB inside the opamp loop. 66dB NFB can make gravel smooth.
If we knew the load impedance, and optimized for power efficiency, a CCS would have better efficiency than resistor-coupled. But headphones come in a wide range of impedances, but not such a wide range of efficiencies, which means hi-Z phones need more voltage than low-Z phones, which happens to match the "weakness" of resistor-coupling. Use a 32Ω resistor with +/-12V rails. With 32Ω load, we "only" get about 5V peaks; with 300Ω load we get about 10V peaks. 380mW in 32Ω and 160mW in 300Ω is a fairly good match to typical power ratings.
And what is wrong with it?
> the input impedance of the transistor seems a bit small.
It should be on the order of 1K-2K. If the op-amp is any good, that should not be any problem.
> So I guess I have to choose between mosfet and darlington.
MOSFET input capacitance is VERY difficult to use this way. On simulator I ended up decoupling the MOSFET Gate above the audio band so I could take feedback direct from the opamp. People who built that reported different results, which tells me that it is far less stable in real life than on a simulator (SPICE lives in a fantasy world).
Darlington emitter followers work, but still end up with a double-pole rolloff about the place many chip opamps are gaining excess phase. And IMHO the Darlington is not needed.
> largest bandwith
MOSFET input is a BIG capacitor, enough to embarass the opamp. Or if the opamp can drive the capacitance, it is beefy enough to drive the load, so why have the MOSFET?
Darlington is more subtle. You can do it and it will probably be stable. Input capacitance seems small; problem is that there are two low-passes and above Ft/(√ β) (which may be around 100KHz) the input impedance drops faster than a pure capacitor. You have to ask why you "need" a Darlington.
People who built a similar design with just a TIP41 or similar simple power BJT report that it is very clean and apparently stable.
With a simple fairly-fast BJT, and a half-decent audio opamp, bandwidth will be over 10MHz. If you need more than 10MHz things get a little tougher. But I don't believe we hear even 10MHz all that well. At 20KHz, response should be down something like 0.02dB re: DC or 0.01dB re: 1KHz, pretty flat.
Your diagram is so darn big that I can't see the whole thing on my monitor, but your DC connections don't smell right. You either need a bipolar +/- supply or several DC-block capacitors.
This version will work; I know because others have built it with no help except this picture:
An externally hosted image should be here but it was not working when we last tested it.
TIP41 is handy in the US; your BC/BD parts will be just fine. Output devices for a small power amp or drivers for a large power amp. 40V, 2A, >20W, Ft>10MHz if possible.
The "0.2A IDC" current source can be replaced with a resistor....
> current source instead of R4 to improve linearity
Linearity is not affected enough to measure. Taking your 47 or 50Ω resistor and assuming 10V rail, emitter current is 10V/50Ω= 200mA. Emitter resistance at idle is about 27Ω/200 or 0.14Ω. This will vary with signal: say 0.2Ω to 0.1Ω. The resulting 0.1Ω variation is very-very-small compared to a 47Ω||32Ω= 19Ω total load. It has 46dB NFB just in the emitter-follower! Plus another 20dB+ NFB inside the opamp loop. 66dB NFB can make gravel smooth.
If we knew the load impedance, and optimized for power efficiency, a CCS would have better efficiency than resistor-coupled. But headphones come in a wide range of impedances, but not such a wide range of efficiencies, which means hi-Z phones need more voltage than low-Z phones, which happens to match the "weakness" of resistor-coupling. Use a 32Ω resistor with +/-12V rails. With 32Ω load, we "only" get about 5V peaks; with 300Ω load we get about 10V peaks. 380mW in 32Ω and 160mW in 300Ω is a fairly good match to typical power ratings.
Thanks for looking at my amp. I resized the image and R4 is now connected to -VCC.
I measured the input impedance of the bjt, it was around 800 ohms. On some diagrams on the site of D. Self I saw that the heavier the load, the greater the distortion of the opamp. In order to get a high imput impedance, Hfe of the transistor should be large. A darlington has a big Hfe, so the input impedance large, so low distortion. That`s why I thought that a Darlington could be better.
I`ll sure do some listening tests with some different mosfets. What I don`t get is why the input capacitance of the mosfet is so difficult to drive here.
Or is it just a stability issue?
Some amps are build around a single mosfet, for example the zen amps. In that case the effective input capacitance is the input capacitance is multiplied by the gain of the stage. So if you use such an amp, one can expect that the preamp can`t drive the mosfet well. So to be usable with most preams, you should put a driver stage in front of the zen amp.
I`ll also try the schematic you posted. That is what I had in mind when I began with this amp.
I`ll listen to them all, tweak away and then try to decide.
I measured the input impedance of the bjt, it was around 800 ohms. On some diagrams on the site of D. Self I saw that the heavier the load, the greater the distortion of the opamp. In order to get a high imput impedance, Hfe of the transistor should be large. A darlington has a big Hfe, so the input impedance large, so low distortion. That`s why I thought that a Darlington could be better.
I`ll sure do some listening tests with some different mosfets. What I don`t get is why the input capacitance of the mosfet is so difficult to drive here.
Or is it just a stability issue?Some amps are build around a single mosfet, for example the zen amps. In that case the effective input capacitance is the input capacitance is multiplied by the gain of the stage. So if you use such an amp, one can expect that the preamp can`t drive the mosfet well. So to be usable with most preams, you should put a driver stage in front of the zen amp.
I`ll also try the schematic you posted. That is what I had in mind when I began with this amp.
I`ll listen to them all, tweak away and then try to decide.
One solution for large MOSFET gate capacitance is to isolate it using a small / medium power bipolar. I have found that the largest gate current for IRFP140 class MOSFET is about 4ma during audio frequency (it can be made smaller with higher gate stopper but you run the risk of excessive gain in higher frequencies).
I usually use a bipolar driver stage for my MOSFETs and run the drivers at about 15ma idle. With a minimum beta of 50, that means you are drawing less than 0.5ma from the opamp - something most of them can handle with ease.
With that, you can solve the problem of gate capacitance fairly easily.
I usually use a bipolar driver stage for my MOSFETs and run the drivers at about 15ma idle. With a minimum beta of 50, that means you are drawing less than 0.5ma from the opamp - something most of them can handle with ease.
With that, you can solve the problem of gate capacitance fairly easily.
I made line driver for long cables with NE5532+BC546B+MJE1503x. The bias current is about 150mA, with 100ohms emitter resistor (+/-15V rails). It works fine for long days.
The only difference compared to the first image, that I use resistors between the base and the emitter of the transistors. 4.7kohms for the BC, and 220ohms for the MJE.
sajti
The only difference compared to the first image, that I use resistors between the base and the emitter of the transistors. 4.7kohms for the BC, and 220ohms for the MJE.
sajti
Of course, one REALLY would not want to use an IRFP140 to drive headphones.
Something like an IRF510 will be MORE than adequate, and have FAR lower gate capacitance - 180pF of which you will only be driving a very small part, in parallel with 18pF - so I would say even the venerable NE5532 will not be embarassed given a gate damper resistor (which is prudently included for the sake of the MOSFET anyway). If all else fails, a small cap from OP output to - input will solve this, bandwidth remaining well above audio range.
Something like an IRF510 will be MORE than adequate, and have FAR lower gate capacitance - 180pF of which you will only be driving a very small part, in parallel with 18pF - so I would say even the venerable NE5532 will not be embarassed given a gate damper resistor (which is prudently included for the sake of the MOSFET anyway). If all else fails, a small cap from OP output to - input will solve this, bandwidth remaining well above audio range.
I measured the input resistance by putting a resistor between the base of the transistor and the opamp. Then measuring the voltage across the resistor. Then you have the base current. Then measure the voltage at the base. Base voltage divided by the base current= base input resistance.
Why is this done and how did you calculate the values?
I still have a lot to learn, that's one of my reasons why I like DIY
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.