I prefer a diamond buffer (DB) on the output of the opamp. You then also have the option of taking the feedback from the opamp output and excluding the DB if that's your persuasion (not mine I might add). This set up is fast, easy to comp (output pole is much higher in frequency than with common emmitter op stage) and thermally stable with the right choice of component values. And mid fi it ain't. if you have power to spare, there's always the opportunity to set the output stage up in class a as well with the DB
The other option is to use a current source loaded EF. I have done this and it's surprisingly good. 3V rms distortion at low single digit ppm into 32 ohms. Full class A. Opamp was half an LM4652.
However, Ed's representation is compact and cost effective. But, it's compromised in my opinion and as you point out, HP amps for hi fi generally have to be extraordinarily good.
The other option is to use a current source loaded EF. I have done this and it's surprisingly good. 3V rms distortion at low single digit ppm into 32 ohms. Full class A. Opamp was half an LM4652.
However, Ed's representation is compact and cost effective. But, it's compromised in my opinion and as you point out, HP amps for hi fi generally have to be extraordinarily good.
An absentee vote here (post 62 if link doesn't quite work):
http://www.diyaudio.com/forums/solid-state/145190-return-my-differential-vas-2.html#post1848220
120 ohms is often quoted. These measurements are relevant:
http://www.diyaudio.com/forums/headphone-systems/109161-headphone-distortion.html
My reading of the ant-audio paper is that larger signal at the headphone produces larger distortion. I really don't see their point.
Headphones vary so greatly as loads that either a build-out or a brute force approach needs to be chosen. Is there any reason (except cost, etc.) to choose the build-out?
Thanks,
Chris
Headphones vary so greatly as loads that either a build-out or a brute force approach needs to be chosen. Is there any reason (except cost, etc.) to choose the build-out?
Thanks,
Chris
I suspect it would be best without the optional resistors. Then Vbe nonlinearity of the output devices becomes irrelevant. The transistors will have to be chosen carefully though, to give reasonable idling current and linear current gain.
Hi,
I used this style with the extra transistors (TIP35/36 or such) on a TDA2040 as driver. TIP Bias current set via a "current mirror with gain". In my case the extra outputs are class AB, however, perhaps uniquely, this arrangement is non-switching.
Thank you for reminding of it, I nearly forgot.
Maybe I'll do one using a LM3875 plus external error correction op-amp and a bunch of 2SC5200/2SA1943 before these parts too are nixed...
Should be a fun Amp as it will do EVERYTHING wrong.
Ciao T
I used this style with the extra transistors (TIP35/36 or such) on a TDA2040 as driver. TIP Bias current set via a "current mirror with gain". In my case the extra outputs are class AB, however, perhaps uniquely, this arrangement is non-switching.
Thank you for reminding of it, I nearly forgot.
Maybe I'll do one using a LM3875 plus external error correction op-amp and a bunch of 2SC5200/2SA1943 before these parts too are nixed...
Should be a fun Amp as it will do EVERYTHING wrong.
Ciao T
The way I read x-pro's description is that the signal level at the amplifier's feedback point (junction of C4, R8 & R12) is the same in all cases. The voltage across the headphones would therefore depend on whether they were plugged into the 22-ohm or 47-ohm output; lower when plugged into 47-ohm yet this gives higher levels of harmonics.
Looking at it that way, it appears that harmonics appear across the headphones that are not present at the feedback point, so presumably are not present in the input signal either.
Given x-pros findings, I tend to think a low output impedance is important. However, I also suspect a lot of phones frequency responses are tailored with a 120 ohm source impedance. I have no proof, but I suspect this may be the case.
This presumes that the opamp (as yet undisclosed) is itself with high impedance between the rail inputs, probably a quite-reasonable assumption compared to a big-guy base-emitter impedance. In other designs along these lines, one sometimes sees common-base stages between the V+/- and the power devices, particularly when higher-than-opamp-max rails are needed.
But with horizontal-ruler-flat beta of ideal output devices and biased as richly as your chip temperature will permit, there will still be gobs of thermal distortion open-loop, due to the tempco of beta. With sufficiently large geometry devices these signal-induced changes may only affect things at very low frequencies, and the overall loop gain may render them a relatively small effect. Note that the thermal effects are not made less important by running at rich bias, unless somehow the temperature coefficients of the betas themselves change significantly with temperature.
Recent suggestions are legit, but they throw off the point of the design, and usually make things sound worse. That would include the 'optional' resistors or current sources.
This is a design that is simple, but not 'idiot proof'. Complementary matching will seriously help performance, even if only to reduce output offset. This design really likes to work in Class A or at least, rich A-B, like 200ma or more. Try to run it near Class B, and it will be seriously worse than just using a complementary follower. Can everyone see WHY this is so?
This is a design that is simple, but not 'idiot proof'. Complementary matching will seriously help performance, even if only to reduce output offset. This design really likes to work in Class A or at least, rich A-B, like 200ma or more. Try to run it near Class B, and it will be seriously worse than just using a complementary follower. Can everyone see WHY this is so?
why not wrap a OPA827 feedback loop around each channel of a TPA6120 and be done? - see what mere decade old and less parts can do
you can see more high frequency specs for the TPA6120 in the THS6012 datasheet since TI just rebadged it for "audio" - but did a uncommonly thorough job of audio distoriton measurements vs level, load, frequency, IMD
my guess is that some of the sub -90 dB audio frequency distortion plots are showing thermal effects
with 15 mA Iq likely 5-10 mA is output Q bias - after all the part manages -90 dB distortion at 100 kHz into 25 Ohms - seems like an indication that crossover distortion at audio isn't an issue
use some local gain in the TPA6120 and the input op amps can be ran from sub-regualted (shunt) supplies
you can see more high frequency specs for the TPA6120 in the THS6012 datasheet since TI just rebadged it for "audio" - but did a uncommonly thorough job of audio distoriton measurements vs level, load, frequency, IMD
my guess is that some of the sub -90 dB audio frequency distortion plots are showing thermal effects
with 15 mA Iq likely 5-10 mA is output Q bias - after all the part manages -90 dB distortion at 100 kHz into 25 Ohms - seems like an indication that crossover distortion at audio isn't an issue
use some local gain in the TPA6120 and the input op amps can be ran from sub-regualted (shunt) supplies
Last edited:
- Status
- Not open for further replies.