A minimum 2k bias R to +/- 12 V seems a bit low - that's an extra 576 mW of Pdiss per dual, in addition to the basic ~100 mW for MC33078. At 85 K/W for DIP-8, that would put the die at a nice and toasty ~57 K above ambient. Still tolerable, actually, but definitely a bit warmer than I'd like in the long run.
Ideal class A bias current
* is comfortably higher than expected peak output current (at given maximum level into given load)
* yet does not heat up the IC too much (rule of thumb for total Pdiss: ~500 mW total for a DIP-8, regular dissipation included - note that you can get little clip-on heatsinks which should allow for higher Pdiss).
Note that depending on OP type and supplies, there may actually not be much of any thermal headroom left, such as when operating an OPA2604 at >20 V.
Attractive opamp types tend to be ones with (very) low output stage bias but comparatively generous current capability. I bet the old MC33178/179 would be fun (80 mA source / 100 mA sink, 5 MHz GBW, but idle current = a measly 420 µA per amp). When Jan Meier first championed Class A bias in headphone amps many moons ago, he employed LM617x - very thrifty on current for how fast they are (160 MHz GBW), reasonably powerful. OPA690, though better-biased in stock form, also seems to benefit.
By contrast, there's fairly little to be gained in an old chip of limited output current and existing healthy output stage bias, like the LF356. The '4558 would file under "limited output current" and "lowish bias", so some improvement is likely at moderate bias even if it does remain a fairly slow part.
I'm not sure how something like a TLE2072 would fare, these have a low-crossover distortion output stage but decent current output.
Which rail to choose can be quite a complicated matter - in case of the MC33078, the output can sink more current than it can source, but according to the output saturation vs. load resistance graph, it will only do so somewhat reluctantly, pointing to higher output impedance. Under non-shorted conditions, sourcing (bias to -Ve) actually looks like the better bet in most cases. Things like these may explain ambiguous results.
Going by the results with emitter followers, using a CCS instead of a plain resistor drops distortion by about an order of magnitude. However, complexity wise, the leap to a simple SE buffer stage may not longer be that big at this point... and nobody says you can't combine the two approaches (seen in a headphone amp design around here, Bonsai's I think?). Decisions, decisions...
Ideal class A bias current
* is comfortably higher than expected peak output current (at given maximum level into given load)
* yet does not heat up the IC too much (rule of thumb for total Pdiss: ~500 mW total for a DIP-8, regular dissipation included - note that you can get little clip-on heatsinks which should allow for higher Pdiss).
Note that depending on OP type and supplies, there may actually not be much of any thermal headroom left, such as when operating an OPA2604 at >20 V.
Attractive opamp types tend to be ones with (very) low output stage bias but comparatively generous current capability. I bet the old MC33178/179 would be fun (80 mA source / 100 mA sink, 5 MHz GBW, but idle current = a measly 420 µA per amp). When Jan Meier first championed Class A bias in headphone amps many moons ago, he employed LM617x - very thrifty on current for how fast they are (160 MHz GBW), reasonably powerful. OPA690, though better-biased in stock form, also seems to benefit.
By contrast, there's fairly little to be gained in an old chip of limited output current and existing healthy output stage bias, like the LF356. The '4558 would file under "limited output current" and "lowish bias", so some improvement is likely at moderate bias even if it does remain a fairly slow part.
I'm not sure how something like a TLE2072 would fare, these have a low-crossover distortion output stage but decent current output.
Which rail to choose can be quite a complicated matter - in case of the MC33078, the output can sink more current than it can source, but according to the output saturation vs. load resistance graph, it will only do so somewhat reluctantly, pointing to higher output impedance. Under non-shorted conditions, sourcing (bias to -Ve) actually looks like the better bet in most cases. Things like these may explain ambiguous results.
Going by the results with emitter followers, using a CCS instead of a plain resistor drops distortion by about an order of magnitude. However, complexity wise, the leap to a simple SE buffer stage may not longer be that big at this point... and nobody says you can't combine the two approaches (seen in a headphone amp design around here, Bonsai's I think?). Decisions, decisions...
Good points. I think 4K7 should be OK. Mostly +/- 12 V is as far as most need to go. The CCS diode below should be OK regardless. No more difficult to fit than a resitor. Unlike real diodes they will work the wrong way arround. No idea what you would get then apart from smoke. You can use two if wanting 4 mA I seem to remember.
Current Limiting Diode CRD 2.0mA | Rapid Online
Current Limiting Diode CRD 2.0mA | Rapid Online
One thing you can try is banking up op amps. You will get a reduction in noise that approaches state of the art if using a MC33079. I have three sections as the 75 uS gain stage with up to 120 gain ( Ortofon SPU ). 3 x 2 mA CCS. The final section with gain of 17@ 1kHz is the active 3180/318 uS. My friend Martina Schoener who is a DIY member has a preamp of this type. When handed over she had 6 x 1000 uF and 22 R for the lower feedback arms. She removed these ( 0R ) and lives with a DC offset of 0.15 V that is blocked at the output. The last stage has 22 uF 100 V polyester and 1K8 feedback arm. Personally I find the use of 1000 uF non polar a better sound. She finds more bass her way. The MC33078/79 will work well when other op amps refuse to. Even with a gain of 2000 there were no problems. 75 uS was 3K74 ( MRS 25 series ) and 20 nF, passive. I also have 25 to 100 uS variable sometimes. The 20 nf were 10 nf of +/- tollerence. The thing to rememeber is the voltage swing at this point will seldom exceed 50 mV and often is uV. This is a prime candidate for SE class A. The sound is more open and shows depth. I don't use it on the final op amp as to me it sounds wrong, too soft. I dare say by doing that I stop a build up of identical distortion products? I did try one op amp to plus and one to minus and one standard. That seemed to offer something. I used 3 x 220R summing resitors at the outputs.
As much as a MC pick up maybe down to a handful of electrons at these levels the dividing it three ways works just fine! Unlike adding series stages I can say it sounds more detailed and dare I say like good pentode valves. So it should as noise is lower. Not only is it lower but it has a different sound coulour. Less blue which is logical. It is by statistical noise cancellation so seems a trick. My ears like this trick. It seems so unlikely this should work. It works fine.
I tried a TL074 with all sections together and one set of resistors to all. I seemed to get reasonable results. I didn't listen to it too carefully. It was a headphone amp. It has internal resistor output and JFET inputs. A LM324 went mad/hot when asked to do the same. Sorry to say one section of a NE5532 was better ( ??? ). I wasn't happy to use this idea so may have overlooked a good idea. I was trying a theory and theory seems correct. I am trying a TL074 as a comparator today. I guess the minute leakage is how it works? Never tried it before. One paper a JFET is a blocking capacitor and very high Z. Well it is infront of me so did it just now ( 14 pin DIL holder ). The switching point is slighly different to a LM324. Still works. The input has a 10 K resistor to protect the input as if a LM339. I guess the near infinity of TL074 is the factor ?
As much as a MC pick up maybe down to a handful of electrons at these levels the dividing it three ways works just fine! Unlike adding series stages I can say it sounds more detailed and dare I say like good pentode valves. So it should as noise is lower. Not only is it lower but it has a different sound coulour. Less blue which is logical. It is by statistical noise cancellation so seems a trick. My ears like this trick. It seems so unlikely this should work. It works fine.
I tried a TL074 with all sections together and one set of resistors to all. I seemed to get reasonable results. I didn't listen to it too carefully. It was a headphone amp. It has internal resistor output and JFET inputs. A LM324 went mad/hot when asked to do the same. Sorry to say one section of a NE5532 was better ( ??? ). I wasn't happy to use this idea so may have overlooked a good idea. I was trying a theory and theory seems correct. I am trying a TL074 as a comparator today. I guess the minute leakage is how it works? Never tried it before. One paper a JFET is a blocking capacitor and very high Z. Well it is infront of me so did it just now ( 14 pin DIL holder ). The switching point is slighly different to a LM324. Still works. The input has a 10 K resistor to protect the input as if a LM339. I guess the near infinity of TL074 is the factor ?
As already said 2K4 is a bit low. See how low the 10 kHz distortion now is. The Wien Bridge with NE 5532 and RA53 of 1970's design ( R A Penfold ) is the limiting factor. It needs to be 20 dB better to go further. Hope I got everything drawn right. These are just my notes to me. BC 327/337 will do just about anything. They even have low noise and very high gain.
Very interesting, Nigel. Good to see practice confirm some of what one would expect.
Re: BC337/327 - as handy as they are, I would not want to run them at 800-ish mW of Pdiss for very long. (Typically for high-beta transistors, Early voltage is kinda low, too.) I don't think there's anything comparable in TO-126 or TO-225? Maybe even scaled to twice the size? BD139/140 are decent enough, even the "generic" ones, but no match for beta or fT. And doubled-up transistors != simplicity.
Results for parallel TL074 and NE5532 + SE buffer may have turned out better when running them inverting. Though the TL07x output stage really is that weak and starts running out of breath at ~10 mA rms into 600 ohms. I've seen measurements that suggested first signs of output stage related distortion below 22 kOhms of loading. Somehow the MC33078 manages to be better in practice at low gains with a weaker-biased output stage (5x the GBW may have something to do with that).
NJM4580 still consumes 6 mA (which, granted, is less than a '5532). If half of that is output stage Iq (i.e. 1.5 mA per channel), that's a maximum Class A output current of 3 mA. Or 4.5 mA for the '4556A. That's about how much bias current you'd need, minimum.
Re: BC337/327 - as handy as they are, I would not want to run them at 800-ish mW of Pdiss for very long.
(Typically for high-beta transistors, Early voltage is kinda low, too.) I don't think there's anything comparable in TO-126 or TO-225? Maybe even scaled to twice the size? BD139/140 are decent enough, even the "generic" ones, but no match for beta or fT. And doubled-up transistors != simplicity.Results for parallel TL074 and NE5532 + SE buffer may have turned out better when running them inverting. Though the TL07x output stage really is that weak and starts running out of breath at ~10 mA rms into 600 ohms. I've seen measurements that suggested first signs of output stage related distortion below 22 kOhms of loading. Somehow the MC33078 manages to be better in practice at low gains with a weaker-biased output stage (5x the GBW may have something to do with that).
Not sure. Both of them appear to run a decent amount of quiescent current to begin with, especially the '4556A at Iq = 8-9 mA. You may run into thermal issues before seeing any appreciable benefit.
NJM4580 still consumes 6 mA (which, granted, is less than a '5532). If half of that is output stage Iq (i.e. 1.5 mA per channel), that's a maximum Class A output current of 3 mA. Or 4.5 mA for the '4556A. That's about how much bias current you'd need, minimum.
Yes that's right. I think change to BD135/9 and use 3 K pull down ( green ). These were snap shots for my data. It was at this level that as near as I can tell a zero distortion option up to 20 kHz could be had ( - 80 dB ). As my oscillator won't go there I have to guess from how far it changes. Also a very compact design.
I wish I had kept LM358 ( 324 ) graphs in this same circuit. It was excellent. I posted these examples as it seemed a shame not to. A picture is worth 1000 words. If I find the LM 358 I will post it.
Mostly I suspect what we hear with op amps is crossover distortion. Also JFET inputs do not react to RFI as badly. Somethings people speak about have no reality in descrete circuits. They will insist an op amp sounds better because it has high slew rates, I doubt it is exactly that . Some video op amps are worth trying if so. Fine for MC preamp stage one as +/- 2.5 rails is OK if wanting 50 mV. Some have low distortion which helps and very low noise. They are cheap. High slew rates mean plenty of interstage current perhaps. I recently modified a Quad 33 pre amp to work at higher gain. This to match a Denon DL110. The gain if I remember raised from about 2 mV to about 400 uV, hiss was still excellent. Only two transisors and on paper not able to give correct 75 uS due to lack of loop gain. It sounds great. I would suggest it's lack of complexity can be heard as a positve. It is a very fussy design as it only just works. Gain has to be exact for the job it does ( +/- 3 dB ). I would almost gaurantee most would hate the 33. This is because they will not go to the trouble to get it to work. The clue when right is the 303 power amps clips at volume 8 or 10 when DGG. This is still slightly less than Quad's own radio. If on standard M1 position one need 14 out of 10 as an idea.
I wish I had kept LM358 ( 324 ) graphs in this same circuit. It was excellent. I posted these examples as it seemed a shame not to. A picture is worth 1000 words. If I find the LM 358 I will post it.
Mostly I suspect what we hear with op amps is crossover distortion. Also JFET inputs do not react to RFI as badly. Somethings people speak about have no reality in descrete circuits. They will insist an op amp sounds better because it has high slew rates, I doubt it is exactly that . Some video op amps are worth trying if so. Fine for MC preamp stage one as +/- 2.5 rails is OK if wanting 50 mV. Some have low distortion which helps and very low noise. They are cheap. High slew rates mean plenty of interstage current perhaps. I recently modified a Quad 33 pre amp to work at higher gain. This to match a Denon DL110. The gain if I remember raised from about 2 mV to about 400 uV, hiss was still excellent. Only two transisors and on paper not able to give correct 75 uS due to lack of loop gain. It sounds great. I would suggest it's lack of complexity can be heard as a positve. It is a very fussy design as it only just works. Gain has to be exact for the job it does ( +/- 3 dB ). I would almost gaurantee most would hate the 33. This is because they will not go to the trouble to get it to work. The clue when right is the 303 power amps clips at volume 8 or 10 when DGG. This is still slightly less than Quad's own radio. If on standard M1 position one need 14 out of 10 as an idea.
This was to see where it went. NE5532 BD135/136. I have often played with this circuit hoping one day to find a use for it. Never have as yet. The theory is if the loop is very fast it will close. Not really. At 1 kHz it is doing something. Some place the BD135/6 with the collectors to the load and the power rails as bias via resistors. That I am sure will work if stability can be ensured.
Not sure what the load was . 30 R perhaps?
Not sure what the load was . 30 R perhaps?
4558 sounds bad
1) The 4558 has an un-degenerate bipolar diffpair.
That is, there are no resistors in the emitter to expand
the low-distortion region. If I recall right, +-4milliVolts
will cause 10% distortion in a bipolar diffpair.
2) The 4558 imbalances the HEATING of the two
input transistors. One bipolar ties to a diode-tied load, hence
the collector voltage is a diode-drop above the -15volt rail.
The other bipolar participates in generating a voltage to drive
a Darlington, hence deltaVout is twice the other bipolar, and
the change in heating is twice (altho 2:1 may not matter much).
3) examining various 4558 datasheets, I see differences in the feedback path for the compensation capacitor; different suppliers used different designers; some of the circuits end up with the lower end of the Cap
referenced to negative_RAIL, destroying the negative_PSRR spec.
just some thoughts from a person with some discrete and some silicon
experience
tankcircuitnoise
1) The 4558 has an un-degenerate bipolar diffpair.
That is, there are no resistors in the emitter to expand
the low-distortion region. If I recall right, +-4milliVolts
will cause 10% distortion in a bipolar diffpair.
2) The 4558 imbalances the HEATING of the two
input transistors. One bipolar ties to a diode-tied load, hence
the collector voltage is a diode-drop above the -15volt rail.
The other bipolar participates in generating a voltage to drive
a Darlington, hence deltaVout is twice the other bipolar, and
the change in heating is twice (altho 2:1 may not matter much).
3) examining various 4558 datasheets, I see differences in the feedback path for the compensation capacitor; different suppliers used different designers; some of the circuits end up with the lower end of the Cap
referenced to negative_RAIL, destroying the negative_PSRR spec.
just some thoughts from a person with some discrete and some silicon
experience
tankcircuitnoise
It's just more Linear in General, I'm not trying to Bash BJT they can have their place/uses.
Issues with 4558, compare to UA715 circuit
I saw the UA715, originally from Fairchild, used in high-speed telemetry
systems. So I've attached a link to the Fairchild datasheet.
At very bottom, you'll see two 100Ohm resistors splitting left
and right to feed into emitters of dual-transistor Darlingtons. Those 100Ohm resistors increase the "linear" range of the input stage.
Above the 4 transistors of the Diffpair, there is a 2-transistor cascode
which keeps the voltage/heating of the Diffpair relatively constant which
reduces thermal distortion.
Perhaps more usefully, the cascode reduces Miller Effect to just 2X.
The 4558 does not use Cascode, and the Cmiller of +Vin is much larger
than -Vin. Kinda like examining goat-entrails, these transistor-level
schematics. And the weasel-words "simplified schematic" is always
lurking.
The 4558 distortion is predictable. With Unitygain Bandwidth of 10Mhz
(to make the math easy), the gain at 20,000Hertz is 500. With stage
gain of 10x, that means the surplus gain (1/GH) is 50. Thus to get 1volt
out, we need 1v/50 = 20milliVolt into the 4558 opamp at 20,000Hz.
That is far into the heavy-distortion regions of a bipolar.
The point of interception of 3rd and 1rst is approximately 1/4 volts.
We are operating at 1/10 volt. The distortion, at 20Khz, is only down by
(1/10 / 1/4)^2 == (1/2.5)^2 = 1/6 ==> 16% intermodulation, again
at 20KHz.
HTTP 301 This page has been moved
I saw the UA715, originally from Fairchild, used in high-speed telemetry
systems. So I've attached a link to the Fairchild datasheet.
At very bottom, you'll see two 100Ohm resistors splitting left
and right to feed into emitters of dual-transistor Darlingtons. Those 100Ohm resistors increase the "linear" range of the input stage.
Above the 4 transistors of the Diffpair, there is a 2-transistor cascode
which keeps the voltage/heating of the Diffpair relatively constant which
reduces thermal distortion.
Perhaps more usefully, the cascode reduces Miller Effect to just 2X.
The 4558 does not use Cascode, and the Cmiller of +Vin is much larger
than -Vin. Kinda like examining goat-entrails, these transistor-level
schematics. And the weasel-words "simplified schematic" is always
lurking.
The 4558 distortion is predictable. With Unitygain Bandwidth of 10Mhz
(to make the math easy), the gain at 20,000Hertz is 500. With stage
gain of 10x, that means the surplus gain (1/GH) is 50. Thus to get 1volt
out, we need 1v/50 = 20milliVolt into the 4558 opamp at 20,000Hz.
That is far into the heavy-distortion regions of a bipolar.
The point of interception of 3rd and 1rst is approximately 1/4 volts.
We are operating at 1/10 volt. The distortion, at 20Khz, is only down by
(1/10 / 1/4)^2 == (1/2.5)^2 = 1/6 ==> 16% intermodulation, again
at 20KHz.
HTTP 301 This page has been moved
From a paper I read the use of FET makes it worse or very little difference worth talking about. RFi is a bigger factor. This makes FET even more interesting.
http://www.analog.com/media/en/training-seminars/tutorials/MT-096.pdf
Input Stage Distortion
You all should take a LM324 and put it into class A. Apart from noise most of it's bad sound is then no more. This brings all the fancy ideas into question. Slewing may be just a way of say more ideal stage matching. Some are tempted to think that transconductance is a perfectly workable concept in a typical op amp. I have my doubts.
As far as I can say the critical points of an op amp are. Sophisticated class AB biasing. Critical use of dominant pole capacitor with option to adjust it. I would strongly suggest unity gain stability is a bad mistake. I fully understand why it is liked. I have to say every unity gain stable op amp I know of sound great at unity. A pice of wire with gain. I have great doubts they sound good at a gain of 3. I have no idea why. Gain of 10 seems to take the sound back where it should be yet sweeter ( obvious reasons ). Always consider potting down an op amp if you agree with me. These are small effects I have to say so would forgive anyone who can not agree. One thing I suspect is true is the I to V function of the VAS/TIS is critical.
From building op amp style power amps I find slewing is the brute force way of doing a simple job. The old Quad 303 shows it can be done without any of that and never ever show any musical down side. Some will say it has euphonic traits. It does, but at - 80 dB. A near perfect exponential curve. Another name for euphonic might be linear ? That is reduce the distortion to where only audio-witchcraft has it existing and retain the natural curve of a transistion resitance device. That does rather neatly support FET LTP input stages I guess ? They do look a bit like singletons. As for euphoinic traits I seem to remember I was told I didn't like CD as I had lost contact with real sound. As a part time recording engineer I didn't think that true.
As far as I can say the critical points of an op amp are. Sophisticated class AB biasing. Critical use of dominant pole capacitor with option to adjust it. I would strongly suggest unity gain stability is a bad mistake. I fully understand why it is liked. I have to say every unity gain stable op amp I know of sound great at unity. A pice of wire with gain. I have great doubts they sound good at a gain of 3. I have no idea why. Gain of 10 seems to take the sound back where it should be yet sweeter ( obvious reasons ). Always consider potting down an op amp if you agree with me. These are small effects I have to say so would forgive anyone who can not agree. One thing I suspect is true is the I to V function of the VAS/TIS is critical.
From building op amp style power amps I find slewing is the brute force way of doing a simple job. The old Quad 303 shows it can be done without any of that and never ever show any musical down side. Some will say it has euphonic traits. It does, but at - 80 dB. A near perfect exponential curve. Another name for euphonic might be linear ? That is reduce the distortion to where only audio-witchcraft has it existing and retain the natural curve of a transistion resitance device. That does rather neatly support FET LTP input stages I guess ? They do look a bit like singletons. As for euphoinic traits I seem to remember I was told I didn't like CD as I had lost contact with real sound. As a part time recording engineer I didn't think that true.
Hi Dan. As NE5532 is a 600 ohms device it is tempting to think we can ask more than other op amps. Alas the chip size says no. I think 3 K would be the limit at 12 V . When we say 600R we forget this is at worse a sine wave and usually music ( 1/6 ). The constant pull of resistor is differnent.
NE 5532 has almost no distortion regardless of load and looks almost class A. It can sound very much worse than it measures. This seems to be due to very marginal stability where any other op amps would have no problems. Gains higer than 5 seems to help. Use NE 5532 as part of a power amp to see how easilly upset it is. TL072/OPA2604 is much easier to use. This shows how fussy it is.
One thing people never think of is to pot down an op amp. That is to run it at excessive gain. The advantage of this is to keep the long tail input pair in it's best range by reducing the negative feedback Lets say you need 1 V rms but have 7 Vrms you might be 7 times further from the Nyquist point than you usually would be. I have a hunch this will sound as people want class A to sound ( sweet, open,fast, dynamic, not grey or pinched ). If a power amp is 22K input fed from 2K2 the potting down resistor might be 13 K. This is a nice load and will look as class A in distortion and might even be class A due to bias safety margin ( class AB ) . This could be taken down to 1K3 and 220 R to see if it had more punch. Whilst absolute distortion might be more the low level point is now 16 dB up. Now if you double the power amp gain we are looking > 20 dB rise in low level signal conditions. Not bad I think. These are things which promote stability. One can even high bias a power amp to shift the point where crossover becomes a problem. This will give more absolute distortion ( gm doubling ). It should be where are ears no longer care too much and just like it loud. 2 to 5 watts would be about right.
NE 5532 has almost no distortion regardless of load and looks almost class A. It can sound very much worse than it measures. This seems to be due to very marginal stability where any other op amps would have no problems. Gains higer than 5 seems to help. Use NE 5532 as part of a power amp to see how easilly upset it is. TL072/OPA2604 is much easier to use. This shows how fussy it is.
One thing people never think of is to pot down an op amp. That is to run it at excessive gain. The advantage of this is to keep the long tail input pair in it's best range by reducing the negative feedback Lets say you need 1 V rms but have 7 Vrms you might be 7 times further from the Nyquist point than you usually would be. I have a hunch this will sound as people want class A to sound ( sweet, open,fast, dynamic, not grey or pinched ). If a power amp is 22K input fed from 2K2 the potting down resistor might be 13 K. This is a nice load and will look as class A in distortion and might even be class A due to bias safety margin ( class AB ) . This could be taken down to 1K3 and 220 R to see if it had more punch. Whilst absolute distortion might be more the low level point is now 16 dB up. Now if you double the power amp gain we are looking > 20 dB rise in low level signal conditions. Not bad I think. These are things which promote stability. One can even high bias a power amp to shift the point where crossover becomes a problem. This will give more absolute distortion ( gm doubling ). It should be where are ears no longer care too much and just like it loud. 2 to 5 watts would be about right.
Douglas Self wrote many moons ago already:
Sijosae's headphone driving tests (with no output series R, IIRC) revealed more quirks: NJM2068 was unstable at higher-impedance loads, possibly suggesting capacitive load driving issues at the lowish supplies used, while OPA2604 did not appreciate the low-impedance ones (it's neither a good performer on low supplies nor a good load driver to begin with). OPA2228 was not stable at all, but then again it's a decompensated part and OPA2227 should probably have been used instead.
Incidentally, Mr. Self also implemented this crossover point shifting via Class A bias in a commercial amp (Cambridge 840 something or other, if memory serves?). It's kinda inefficient though and makes me wonder whether one wouldn't be better served just cranking up output stage quiescent current instead. Class A peak output current at a given Iq is twice as high for push-pull vs. SE after all. We're only using SE Class A biasing with opamps because we have no access to the internals of the chip!
Incidentally, I found a schematic for a Beyerdynamic A1 clone - turns out they weren't so clever to combine output stage bias and opamp class A biasing after all. Too bad, it would really have helped the stock MC33078 out. Would be an easy mod though, just leave the current sink transistor (BC550C) out.
It does remain a nested feedback amp; parts with more straightforward compensation may be better behaved inside a feedback loop.
Sijosae's headphone driving tests (with no output series R, IIRC) revealed more quirks: NJM2068 was unstable at higher-impedance loads, possibly suggesting capacitive load driving issues at the lowish supplies used, while OPA2604 did not appreciate the low-impedance ones (it's neither a good performer on low supplies nor a good load driver to begin with). OPA2228 was not stable at all, but then again it's a decompensated part and OPA2227 should probably have been used instead.
Incidentally, Mr. Self also implemented this crossover point shifting via Class A bias in a commercial amp (Cambridge 840 something or other, if memory serves?). It's kinda inefficient though and makes me wonder whether one wouldn't be better served just cranking up output stage quiescent current instead. Class A peak output current at a given Iq is twice as high for push-pull vs. SE after all. We're only using SE Class A biasing with opamps because we have no access to the internals of the chip!
Incidentally, I found a schematic for a Beyerdynamic A1 clone - turns out they weren't so clever to combine output stage bias and opamp class A biasing after all. Too bad, it would really have helped the stock MC33078 out. Would be an easy mod though, just leave the current sink transistor (BC550C) out.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.