If I put my notes here, I might be able to find them again later!
Sapphire 4 headphone amplifier [Rev. 4.1m]
Development thread starts here.
While I consider this my best work to date, the documentation trail has become hopelessly scattered so let me try to regroup in this blog post and make some sort of gateway/portal to the project. [Update: web page write up now finished.]
The followup to the Sapphire 3 started out as a generic musing on stacked diamond buffers and current mirror amplification, coalesced for a bit as the ill-fated Project Unity before becoming part of the Sapphire line initially as a temporary measure as something I could drop in my existing chassis for testing. The first iteration kept the open loop buffer of previous Sapphire iterations, but now I'm running the 4.1m boards with the buffer inside the feedback loop I realize I've ended up at just the classic, dictionary definition current feedback amplifier (CFA). Sigh. So much for originality. It sounds great though, so I'm not as upset about it as I might have been.
*****
The output stage buffer is left outside the feedback loop when R3 is populated in the schematic below, or included within it with R3X populated instead of R3. In both cases feedback is applied to the current input terminal (emitters of Q3,4), so the voltage amplifier is always running closed loop. The difference is in whether the feedback network R3/3X and R2A,B,C is driven by the low impedance buffer output or the high impedance current mirrors Q7,8. This makes a very big difference in how the amplifier performs. To be clear: We put the buffer inside the feedback loop not to improve the performance of the buffer but to improve the performance of the current feedback amplifier.
Performance varies with gain, configuration, and loading. Closed loop: Simulated harmonic distortion is about -100 dB at 0 dB fundamental for 60 Ohm load, 27 dB gain. Measured THD+N unloaded for 25 dB gain is -94 dB A-weighted, with a noise level of -102 dB A-weighted and -95 dB crosstalk (RMAA). I hesitate to give an output power since I haven't measured it, but it's going to be a lot... simulated, into 60 ohms, in excess of 100 mW before distortion breaks -80 dB. Figure minimum 25 mW over 15-600 ohms loads.
Second harmonic is considerably higher open loop, especially at high gains. -82 THD + Noise, dB (A) for 27 dB unloaded, measured (RMAA). This is essentially independent of load as the nonlinearities are generated in the current feedback amplifier not the buffer.
*****
The 4.1m boards feature the option of closed loop / open loop operation, as well as a 3 position gain jumper switch. Further details can be found in the attached BOM.
I run mine closed loop at 25 dB gain with HD-600 headphones.
As noted in the BOM the boards can also be configured as a line stage / preamplifier.
*****
41m3 BOM changes the recommended resistors in the R3X configuration to increase the phase margin a little. R3X 1k was perhaps cutting it a bit fine.
*****
After seeing through quite a few Sapphire 4s built by myself and various others, the basic design and layout appears sound. Everyone has been quite pleased with the results. The only residual issue is the DC offset. DC offset is controlled by the trimmer R6, but the available control range is defined by hfe missmatch of Q1, Q2 and also the missmatch of the Zener references voltage of D1, D2.
Basically the situation is this: if you match the Zeners in advance you are guaranteed not to have problems, if you leave it to chance and draw the absolute worst outliers it's possible that the trimmer is out of range to zero the output. It hasn't happened so far with R6 = 5k but I concede that it's a possible pain point.
Typically you'll find the hfe of Q1 and Q2 are about 60 units apart, but up to about 100 difference in hfe should not present any problem at all and there is no need to match transistors. The Zeners on the other hand sometimes come in at several hundred millivolts different. Ideally you'd prefer to use pairs which are less than 200 mV apart.
Once set the offset remains zeroed and does not drift much with temperature. 50 mV or less seems to be typical as the unit warms up.
While I consider this my best work to date, the documentation trail has become hopelessly scattered so let me try to regroup in this blog post and make some sort of gateway/portal to the project. [Update: web page write up now finished.]
The followup to the Sapphire 3 started out as a generic musing on stacked diamond buffers and current mirror amplification, coalesced for a bit as the ill-fated Project Unity before becoming part of the Sapphire line initially as a temporary measure as something I could drop in my existing chassis for testing. The first iteration kept the open loop buffer of previous Sapphire iterations, but now I'm running the 4.1m boards with the buffer inside the feedback loop I realize I've ended up at just the classic, dictionary definition current feedback amplifier (CFA). Sigh. So much for originality. It sounds great though, so I'm not as upset about it as I might have been.
*****
The output stage buffer is left outside the feedback loop when R3 is populated in the schematic below, or included within it with R3X populated instead of R3. In both cases feedback is applied to the current input terminal (emitters of Q3,4), so the voltage amplifier is always running closed loop. The difference is in whether the feedback network R3/3X and R2A,B,C is driven by the low impedance buffer output or the high impedance current mirrors Q7,8. This makes a very big difference in how the amplifier performs. To be clear: We put the buffer inside the feedback loop not to improve the performance of the buffer but to improve the performance of the current feedback amplifier.
Performance varies with gain, configuration, and loading. Closed loop: Simulated harmonic distortion is about -100 dB at 0 dB fundamental for 60 Ohm load, 27 dB gain. Measured THD+N unloaded for 25 dB gain is -94 dB A-weighted, with a noise level of -102 dB A-weighted and -95 dB crosstalk (RMAA). I hesitate to give an output power since I haven't measured it, but it's going to be a lot... simulated, into 60 ohms, in excess of 100 mW before distortion breaks -80 dB. Figure minimum 25 mW over 15-600 ohms loads.
Second harmonic is considerably higher open loop, especially at high gains. -82 THD + Noise, dB (A) for 27 dB unloaded, measured (RMAA). This is essentially independent of load as the nonlinearities are generated in the current feedback amplifier not the buffer.
*****
The 4.1m boards feature the option of closed loop / open loop operation, as well as a 3 position gain jumper switch. Further details can be found in the attached BOM.
I run mine closed loop at 25 dB gain with HD-600 headphones.
As noted in the BOM the boards can also be configured as a line stage / preamplifier.
*****
41m3 BOM changes the recommended resistors in the R3X configuration to increase the phase margin a little. R3X 1k was perhaps cutting it a bit fine.
*****
After seeing through quite a few Sapphire 4s built by myself and various others, the basic design and layout appears sound. Everyone has been quite pleased with the results. The only residual issue is the DC offset. DC offset is controlled by the trimmer R6, but the available control range is defined by hfe missmatch of Q1, Q2 and also the missmatch of the Zener references voltage of D1, D2.
Basically the situation is this: if you match the Zeners in advance you are guaranteed not to have problems, if you leave it to chance and draw the absolute worst outliers it's possible that the trimmer is out of range to zero the output. It hasn't happened so far with R6 = 5k but I concede that it's a possible pain point.
Typically you'll find the hfe of Q1 and Q2 are about 60 units apart, but up to about 100 difference in hfe should not present any problem at all and there is no need to match transistors. The Zeners on the other hand sometimes come in at several hundred millivolts different. Ideally you'd prefer to use pairs which are less than 200 mV apart.
Once set the offset remains zeroed and does not drift much with temperature. 50 mV or less seems to be typical as the unit warms up.
Total Comments 3
Comments
-
Looks like a nice amp! I am always curious about parts selection, so is there a reason for choosing BC327/337 over BC550/560 or BC547/557? It looks like they may be a little more robust than the 547/557. Are they a type of "double-die" part?
Posted 24th April 2017 at 12:41 PM by needtubes
Updated 24th April 2017 at 01:10 PM by needtubes -
I started by tabling up all possible devices, then shortlisted to those which were commonly available and further limited the list to transistors where the datasheet included audio as an application. High voltage types not needed so those were dropped. I'm pretty sure I ended up with a final selection which included the BC327/337, BC550/560, and BC547/557 and as you know there is not much to choose between them. Someone here at diyaudio recommended the 327/337 to me as being a good sounding transistor ... so I went with that!
Posted 24th April 2017 at 11:39 PM by rjm
Updated 24th April 2017 at 11:42 PM by rjm -
Fair enough! I have found BC550/560 to sound nice, so may add some BC327/337 to my next order to give them a try as well.
Posted 25th April 2017 at 01:09 AM by needtubes