Hello all. After "lurking" in the forums for a while my first post here at last 
I have tried to design a headphone amplifier. Mainly to learn how to design with discrete devices. The design now seems to work, but I'm sure there are still many beginners-errors in there. So I was hoping to get some feedback from more experienced people. My current design looks like this:
http://www.omicron.be/projects/amp1/schema.png
The current trough the output stage is set to 100 mA so that the amplifier works in class A for headphone use. The finished prototypes look like this:
http://www.omicron.be/projects/amp1/img_0014.jpg
http://www.omicron.be/projects/amp1/img_0021.jpg
I had some problems with oscillations of the output stage transistors at about 67MHz. However, after a lot of trial and error I was able to cure this by adding C11 and C12. The bandwith of the amplifier is rather large: 10Hz...>100KHz. The output swing is about plus and minus 13.75V (measured when no load was present). My distortion measurement set cannot measure below 0.01% and showed it's minimum reading for all audio frequencies.
To further evaluate the performance of my prototype I used a picotech ADC-216 to obtain a spectrum (test signal 1KHz), which looks like this:
http://www.omicron.be/projects/amp1/spectrum.jpg
The amplifier was loaded with a Senheiser HD475. R27 was adjusted so that the output signal was the same amplitude as the input signal. The red trace is the output spectrum and the blue trace is the spectrum of my test signal. The spectrum suggest a distortion of about 0.0023% (although I don't think this measurement can be relied on very much).
As one can see the distortion is mostly due to the 4th harmonic. Could this point to some mistake in my circuit or layout?
Also there seems to be a lot more noise in the output than in the input signal. I'm not sure how "normal" this is? I did notice that I can reduce this noise a little by adding 2 power supply bypass capacitors (100 uF).
Lastly some circuit related questions:
- The offset correction voltage generator around T14 and T13 is a circuit I "stole" from an old design that appeared in elektor. Does anybody know why one would want to use transistors here and not for example simply 2 diodes?
- In the current source T3 and T3 there is a 1K resistor in the base of T4. I copied that from Randy Slone's book. Does anybody know the exact function of this resistor? Is it there to prevent oscillation maybe?
Any other feedback or suggestions anyone could make are very much welcome.
I have tried to design a headphone amplifier. Mainly to learn how to design with discrete devices. The design now seems to work, but I'm sure there are still many beginners-errors in there. So I was hoping to get some feedback from more experienced people. My current design looks like this:
http://www.omicron.be/projects/amp1/schema.png
The current trough the output stage is set to 100 mA so that the amplifier works in class A for headphone use. The finished prototypes look like this:
http://www.omicron.be/projects/amp1/img_0014.jpg
http://www.omicron.be/projects/amp1/img_0021.jpg
I had some problems with oscillations of the output stage transistors at about 67MHz. However, after a lot of trial and error I was able to cure this by adding C11 and C12. The bandwith of the amplifier is rather large: 10Hz...>100KHz. The output swing is about plus and minus 13.75V (measured when no load was present). My distortion measurement set cannot measure below 0.01% and showed it's minimum reading for all audio frequencies.
To further evaluate the performance of my prototype I used a picotech ADC-216 to obtain a spectrum (test signal 1KHz), which looks like this:
http://www.omicron.be/projects/amp1/spectrum.jpg
The amplifier was loaded with a Senheiser HD475. R27 was adjusted so that the output signal was the same amplitude as the input signal. The red trace is the output spectrum and the blue trace is the spectrum of my test signal. The spectrum suggest a distortion of about 0.0023% (although I don't think this measurement can be relied on very much).
As one can see the distortion is mostly due to the 4th harmonic. Could this point to some mistake in my circuit or layout?
Also there seems to be a lot more noise in the output than in the input signal. I'm not sure how "normal" this is? I did notice that I can reduce this noise a little by adding 2 power supply bypass capacitors (100 uF).
Lastly some circuit related questions:
- The offset correction voltage generator around T14 and T13 is a circuit I "stole" from an old design that appeared in elektor. Does anybody know why one would want to use transistors here and not for example simply 2 diodes?
- In the current source T3 and T3 there is a 1K resistor in the base of T4. I copied that from Randy Slone's book. Does anybody know the exact function of this resistor? Is it there to prevent oscillation maybe?
Any other feedback or suggestions anyone could make are very much welcome.
Looks very cool. I don't know about the base resistor - I haven't seen that before. A current sink isn't going to oscillate regardless of what you do to it. I've seen transistors used as diodes before though - it's usually to obtain better thermal tracking of Vbe, but in this instance I'm not sure that'd really buy you much.
One point though - you mention a 100mA Iq. Isn't that a bit extreme? putting anything like 100mA into a pair of headphones will result in ear-splitting sound levels. Remember that most headphones are better than 100dB/mW sensitivity. You could probably set Iq at more like 20-30mA without ever going into class B.
One point though - you mention a 100mA Iq. Isn't that a bit extreme? putting anything like 100mA into a pair of headphones will result in ear-splitting sound levels. Remember that most headphones are better than 100dB/mW sensitivity. You could probably set Iq at more like 20-30mA without ever going into class B.
I'd agree that R5 doesn't look useful
I don't like coupling the current sinks, at least bypass T4,10 base to neg V with a cap, Self had a slew rate symmetry problem sharing ccs reg/ref
moving T7 collector to gnd will improve VAS linearity - keeps T7 Vcb high and nearly constant reducing Ccb, Early V effects
duplicating T7, R10 buffer in the other diff pair collector to drive the mirror bases improves symmetry
Splitting C7 comp cap with a R to pos V gives a 2-pole compensation which allows for more global loop gain over the audio frequency band, but check stability, and clipping recovery
a little more stability margin may be gained by adding a series RC to gnd at the positive input of the diff pair as discussed recently - as well as a little RF rolloff
C10,11 look a of order of magnitude or more too high to me, alternative CFP compensation can be added as degeneration R in T11,12 emitters
isolating the output from the cable C might help too, 1 uH air coil parallel with a carbon composition resistor (I just use lossy ferrite bead core with >10x sat current rating)
I don't like coupling the current sinks, at least bypass T4,10 base to neg V with a cap, Self had a slew rate symmetry problem sharing ccs reg/ref
moving T7 collector to gnd will improve VAS linearity - keeps T7 Vcb high and nearly constant reducing Ccb, Early V effects
duplicating T7, R10 buffer in the other diff pair collector to drive the mirror bases improves symmetry
Splitting C7 comp cap with a R to pos V gives a 2-pole compensation which allows for more global loop gain over the audio frequency band, but check stability, and clipping recovery
a little more stability margin may be gained by adding a series RC to gnd at the positive input of the diff pair as discussed recently - as well as a little RF rolloff
C10,11 look a of order of magnitude or more too high to me, alternative CFP compensation can be added as degeneration R in T11,12 emitters
isolating the output from the cable C might help too, 1 uH air coil parallel with a carbon composition resistor (I just use lossy ferrite bead core with >10x sat current rating)
Some interesting circuit features here Omicron. Finished boards look great BTW.
The circuit around T3, T4 and T10 looks a little unconventional because of the way T10 is connected. I would expect the base of T3 to move around (ok, by only a small amount) as part of the regulator action of T4. This will modulate the collector current of T10. It migfht be better here to use a fixed base voltage reference for both T3 and T10 to overcome this problem. The reason why a resistor is inserted into the base of T10 with this type of configuration is because if T10 goes into saturation, it will modulate or load the base reference (Doug self talks about this in his book).
JCX, I don't understand this point in your feedback:-
duplicating T7, R10 buffer in the other diff pair collector to drive the mirror bases improves symmetry
If the diff amp was loaded with resistors then this would apply. But in Omicron's design its a current mirror, so the collector of T2 is at 1 Vbe + (Ic T2 x R9) below the supply rail. If you loaded this with a second T7 R10 pair, it would mostly be off since the collector of T1 is at least 2Vbe below rail. Might be if you want to acheive better T2 and T2 balance, you need to insert a 1n4148 between T2 and T6 collectors, and then run this point to a second T7+ R10 pair.
Omicron, what sort of output voltages are you needing to get a reasonable sound level out of your headphones (I assume they are 32 Ohms)?
The circuit around T3, T4 and T10 looks a little unconventional because of the way T10 is connected. I would expect the base of T3 to move around (ok, by only a small amount) as part of the regulator action of T4. This will modulate the collector current of T10. It migfht be better here to use a fixed base voltage reference for both T3 and T10 to overcome this problem. The reason why a resistor is inserted into the base of T10 with this type of configuration is because if T10 goes into saturation, it will modulate or load the base reference (Doug self talks about this in his book).
JCX, I don't understand this point in your feedback:-
duplicating T7, R10 buffer in the other diff pair collector to drive the mirror bases improves symmetry
If the diff amp was loaded with resistors then this would apply. But in Omicron's design its a current mirror, so the collector of T2 is at 1 Vbe + (Ic T2 x R9) below the supply rail. If you loaded this with a second T7 R10 pair, it would mostly be off since the collector of T1 is at least 2Vbe below rail. Might be if you want to acheive better T2 and T2 balance, you need to insert a 1n4148 between T2 and T6 collectors, and then run this point to a second T7+ R10 pair.
Omicron, what sort of output voltages are you needing to get a reasonable sound level out of your headphones (I assume they are 32 Ohms)?
Good schematic, and cool boards
May I suggest to upgrade to BC550C-BC560C, lower noise devices?
Also C4 may need experimentation in bypassing it. Don't use too small value, you could see the parasitic inductance of C4 + pcb tracks going in resonance with the bypass cap, a 10uF value seem right to me instead of 100nF (more commonly used). Same thing with C3.
I admit I would have gone higher with R17-R18, for headphone use. After all, negative feedback will give you the low output impedance needed. How did you choose 1R?
For CFP stability, try using an OLD, SLOW transistor for T15, T16. BD139-BD140 are wonderful devices but a bit fast for a TO126 transistor. Look out for high voltage parts, they are slower (MJE340-350 for example), in CFP there's the rule to keep output devices speed very slow compared to drivers, to get stability. If you feel you have to use caps to improve stability there, I would suggest adding 100pF from collector to base of T12 (or T11), just one of the drivers, not both. Don't ask me why, but it worked to me in the past: I guess it has to do with splitting the poles of the two CFP halves apart, or compensating for slower PNP part...
Let us know
May I suggest to upgrade to BC550C-BC560C, lower noise devices?
Also C4 may need experimentation in bypassing it. Don't use too small value, you could see the parasitic inductance of C4 + pcb tracks going in resonance with the bypass cap, a 10uF value seem right to me instead of 100nF (more commonly used). Same thing with C3.
I admit I would have gone higher with R17-R18, for headphone use. After all, negative feedback will give you the low output impedance needed. How did you choose 1R?
For CFP stability, try using an OLD, SLOW transistor for T15, T16. BD139-BD140 are wonderful devices but a bit fast for a TO126 transistor. Look out for high voltage parts, they are slower (MJE340-350 for example), in CFP there's the rule to keep output devices speed very slow compared to drivers, to get stability. If you feel you have to use caps to improve stability there, I would suggest adding 100pF from collector to base of T12 (or T11), just one of the drivers, not both. Don't ask me why, but it worked to me in the past: I guess it has to do with splitting the poles of the two CFP halves apart, or compensating for slower PNP part...
Let us know
Thanks for all the constructive feedback! I'm really learning a lot from tinkering with this little amp.
With regard to putting a resistor in the base of T10: Doug Self explains in his book that the reason for doing this is to aid in fault-finding (see page 230, I'm still using the 3rd edition). Without this resistor, if the current in one of the current sources drops to 0 then the other one will follow suit as the reference voltage collapses. This makes it unclear wich device has failed. However he also notes that this resistor causes some assymetry in the slew rate.
I'm still at a loss about R5 though. Both Doug and Randy Slone use this resistor in their current sources but I haven't yet found any explanation for it. The reason I was thinking it has to do with stability is that this particular kind of current source has local negative feedback. So there may perhaps be a possiblility of local oscillation?
The inductor in the output is a great idea. I tried this and it definitely helps keep the amplifier stable with capacitive loads. However if I try hard enough I can still make the amplifier oscillate by hanging a large enough capacitor on it's output. I guess that's unavoidable. What would be a reasonable requirement for an amplifier like this with regards to capacitive loading?
The headphone I'm using is a Senheiser HD 600. It's rated at about 300 ohms and 200 mW. So that would amount to about an 11V amplitude. I added a little extra and wound up with a "common" power supply voltage of 15V. The standing current in the output stage I determined based on a theoretical "worst case" headphone of 32 ohms, which would require about 100mA to achieve 200mW.
However in light of suzy's remark about the headphone sensitivity my calculations may be completly irrelevant and 200mW seriously over the top. I must admit that when actually listening the output signal is rarely above a couple hundred millivolts in amplitude. I guess I need to do some reading up on interpreting headphone specs.
About the 1 ohm resistors R16 and R17: I just figured that a voltage drop of about 100mV would be ok for the bias generator to do it's job properly. The drivers T11 and T12 shouldn't get very warm. Since Vbe changes are in the order of millivolts the 100mV drop seemed to fit the 1% rule of thumb. Hence the 1 ohm. A bit of hand waving, I admit. Is there a more rigorous way of determinging the value of these resistors?
I think that in the next few weeks (work permitting) I'll try to design an improved version incorporating some of the excellent suggestions made here. I'm also thinking of making a surface mount version as I'd like to gain some experience in that area as well. An upgrade to BC550/560 is definitly also on the menu
With regard to putting a resistor in the base of T10: Doug Self explains in his book that the reason for doing this is to aid in fault-finding (see page 230, I'm still using the 3rd edition). Without this resistor, if the current in one of the current sources drops to 0 then the other one will follow suit as the reference voltage collapses. This makes it unclear wich device has failed. However he also notes that this resistor causes some assymetry in the slew rate.
I'm still at a loss about R5 though. Both Doug and Randy Slone use this resistor in their current sources but I haven't yet found any explanation for it. The reason I was thinking it has to do with stability is that this particular kind of current source has local negative feedback. So there may perhaps be a possiblility of local oscillation?
The inductor in the output is a great idea. I tried this and it definitely helps keep the amplifier stable with capacitive loads. However if I try hard enough I can still make the amplifier oscillate by hanging a large enough capacitor on it's output. I guess that's unavoidable. What would be a reasonable requirement for an amplifier like this with regards to capacitive loading?
The headphone I'm using is a Senheiser HD 600. It's rated at about 300 ohms and 200 mW. So that would amount to about an 11V amplitude. I added a little extra and wound up with a "common" power supply voltage of 15V. The standing current in the output stage I determined based on a theoretical "worst case" headphone of 32 ohms, which would require about 100mA to achieve 200mW.
However in light of suzy's remark about the headphone sensitivity my calculations may be completly irrelevant and 200mW seriously over the top. I must admit that when actually listening the output signal is rarely above a couple hundred millivolts in amplitude. I guess I need to do some reading up on interpreting headphone specs.
About the 1 ohm resistors R16 and R17: I just figured that a voltage drop of about 100mV would be ok for the bias generator to do it's job properly. The drivers T11 and T12 shouldn't get very warm. Since Vbe changes are in the order of millivolts the 100mV drop seemed to fit the 1% rule of thumb. Hence the 1 ohm. A bit of hand waving, I admit. Is there a more rigorous way of determinging the value of these resistors?
I think that in the next few weeks (work permitting) I'll try to design an improved version incorporating some of the excellent suggestions made here. I'm also thinking of making a surface mount version as I'd like to gain some experience in that area as well. An upgrade to BC550/560 is definitly also on the menu
Omicron said:
Well, folks use to put 2uF at the output. But that's useless in modern output-transformerless solid state amplifiers, because the decreased phase margin caused by the capacitor won't affect Nyquist stability, peak usually happens at mid frequencies (for valve circuits that's a different story, output transformer involved).
For modern design, worst case is actually nearer 100nF than 2uF. You should use the appropriate output inductor value to keep the amp stable with 8r//100nF (ehm... sorry, for headphones I imagine you should use 32r), and in a square wave test with the same load attached, go for little overshoot (eliminating it isn't usually possible without compromising bandwidth with heavy loads - 4r in audio power amplifiers). I guess that your amp, matching with higher - than - usual loads (with respect to audio power amplifiers), won't have too much problems in having wide bandwidth without visible ringing or overshoot on the square wave.
PS a trick learnt the hard way: be sure that your square wave generator hasn't overshoot on its own, and use calibrated 'scope leads
Hi
I would place R5 on the base of T10 instead of T4. T4 doesn't need it, but if T10 were to saturate, it may affect the voltage at the base of T3 and affect the CCS tail current source of the LTP. A small cap across the collector to emitter of T4 will prevent occilation of the current source, T3 & T4.
I would place R5 on the base of T10 instead of T4. T4 doesn't need it, but if T10 were to saturate, it may affect the voltage at the base of T3 and affect the CCS tail current source of the LTP. A small cap across the collector to emitter of T4 will prevent occilation of the current source, T3 & T4.
as a (dynamic) headphone amplifier you only need to drive cable C, likely <1nF will cover most
power amps may expect to see the C of an electrostat panel multiplied by a transformer - even in this case the transformer likely limits high frequency impedance with series R and leakage L so a pure 2 uF load has always been unrealisitc
power amps may expect to see the C of an electrostat panel multiplied by a transformer - even in this case the transformer likely limits high frequency impedance with series R and leakage L so a pure 2 uF load has always been unrealisitc
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.