Junk drawer Headphone Amp - 12v tubes & Germanium

We all have a junk drawer and if you build electronics it could have some interesting leftovers. Mine has some space charge vacuum tubes, germanium transistors and other bits. I'm also in need of a headphone amp to drive 32 ohm cans, so the goal here is to spend close to zero and make something that sounds good.

I'd like to start by acknowledging Pete Millet's 12AE6A space charge vacuum tube headphone amp design as the inspiration for this project. However, I did redesign the cathode follower buffer section so I can use a stable matched pair of AC128 transistors and hopefully they sound good in the circuit. But, because they are thermally unstable, I designed the circuit with an feedback circuit between the Collector and Base which ensures that as the collector voltage drops (or rises), the base voltage is also pulled slightly lower (or higher). This feedback effect helps to counteract changes in the collector current, which stabilizes the transistor's operating point. I also added trimmer potentiometers to tune the Base bias (to 24v rail), Emitter feedback, and the Collector feedback circuits. Hopefully my calculations and design are ok, otherwise I'm sure someone will mention something that was missed.

The regulated 24v dc power supply is currently being designed.
Spice models for the 12A6A and AC128 have to made so the circuit can be tested.

Capture.PNG
 
Last edited:
I had to educate myself because despite years of tube radio restoration I had no clue what a "space charge" vacuum tube is. These are fascinating! For that little tidbit alone I thank you for posting this. Here you go brain, another chew toy!! Here by the way is a link to Pete's design document: http://www.pmillett.com/file_downloads/ax_hybrid.pdf

Um- feeling kind of dumb here- Are the PNP AK128's supposed to be upside down (collector to higher voltage than emitter?) Please clue me in because I have never seen that. Sorry if I don't understand on a short glance what is going on here.

I really like this concept- The higher input impedance and voltage gain and sweetness of tubes, but the lower output impedance of the emitter follower. At lower voltages the germanium transistor performs really well in the follower configuration. I normally think dealing with high filament current and high plate voltage make hybrid amplifiers a real headache, (might as well go all tube) but the space charge tube seems to solve all of this, especially with a regulated voltage on the series filaments. I would think the new problem introduced would be how to drive the low input impedance of the germanium base with a high output impedance tube- I have not done the math, but if it simulates alright I guess it works. I don't bother with spice myself, I either adapt other's designs and "walk it in" through experimentation on the actual hardware. The use of interstage transformers in germanium amplifiers was to solve this problem- providing the impedance matching from predriver to power driver, while simultaneously generating identical out-of-phase signals to be used on the matched totem pole drivers.

I have posted lately on Cheap Germanium transistor amplifiers and FREE service literature for fun and learning pointing to vintage 8-track players as a great source for these parts- it might be fun to tear the transformers out of one of these and see how they do here- The primaries are typically several hundred ohms (impedance) with 4-8 ohm secondaries to drive the bases.

I also notice there is no negative feedback employed in either design- I have not had a chance to read Pete's article, so I don't know what his comments were on this- Was that the intent, or perhaps a further opportunity?

Anyway, excellent job, can't wait to see further developments and updates, and a very interesting build to consider. Thank you again.
 
Last edited:
@wparks Thank you for your reply, I’m glad this post has piqued your interest. It’s funny how you mentioned “brain chew toy”, I have a name in mind for this amp (if it works) that is something along those lines.

No need to feel dumb! I’ll take care of that, it’s my mistake. Thank you for pointing out the upside down PNP transistors, there’s not a logical reason for that. I most recently worked on engineering projects for a company that used power FETs (N-Channel) in most of their designs and I haven’t worked with BJTs in a couple of years. I must have flipped it to what looked “right” to the brain. I’ll go ahead and correct it later today and replace the original image so we don’t have misleading information here (there’s enough of that already).

As for the negative feedback resistor in my design; it should help to stabilize the biasing point of the transistor. If there are any changes to the supply voltage or temperature, the feedback resistor automatically adjusts the base voltage in response to collector voltage changes, which should maintain a correct operating point. RadioFun232 on YouTube mentions this feedback loop in a AC128 preamp design and shows how it produces a “softer sound” (reduced gain), but further research revealed the benefits to stability.
 
Since you are working on the schematic- you are missing the current regulating diode (or plate resistor) in the plate circuit of the 12AE6. (As a side note, I wanted to again thank you for directing my attention to Pete's design paper- I also had no idea that a current regulating diode existed, they are pretty cool, but I suspect unobtainium these days.) Anyway, the plate side of the 1K R1/R8 is at an AC ground thanks to C3/C12 for supply decoupling, so you need an impedance between this AC ground and the tube plate to generate the output signal across that you are driving the output stage with. The plate current of these tubes is so very tiny (under 1mA) that the diode is the best approach- As Pete indicates, you need to present a very high impedance to the plate to generate high enough gain at low distortion, and allow the largest output swing. He found using a large resistor (47K - 100K) did work but gave more distortion and lower performance. Perhaps (if you can't find a diode) you could use a 100K pot here to allow experimenting with this value and listening to the results.

The biggest problem you are going to run into is one I mentioned before- using a high impedance tube output to drive a very lower impedance input of the germanium transistor. This is why Pete used the BUF634- it has an incredibly high input impedance that did not tax the tube, while generating the large current drive needed to drive relatively low headphone impedance. The plate current is a maximum of 1mA on the tube, while looking at the curves for the "medium power" and low beta germanium AC128 you will need several milliamps to drive this base. You could look at doing a Darlington pair out of two 128's, which will multiply the current gain of the two devices, greatly reducing the base current needed, and with the low vbe of 0.2 V of a germanium transistor the addition of the two vbe voltages will still be lower than the 0.6V of silicon transistors.

Pete mentions that the these space charge tubes were typically used in automobile radios- There are schematics and service literature for many auto radios from this time period in the SAM's Auto-Radio series of books posted for free at the World Radio History library as I link to in my "cheap germanium amplifiers" post that I linked above. I plan to do a little digging and find example schematics of how these space charge tubes were used in these larger designs, and maybe learn some more from that. At some point these tube amps needed to drive low impedance speakers, and they most likely did it with transformers. I will mention again that you can dig one out of an 8-track player, and use just one secondary to drive your emitter follower.

Let me know if you try the darlington idea- or other updates. Thanks again!
 
@wparks Wow, great info! Thank you again! I’ll look over the information you linked and the notes from Pete Millet, I still want to stick with Germanium if possible. I do have a pair of AC179 (I think) and some Soviet MP-40 or 41 transistors. I like the idea of using salvaged output transformers also.

I’ll do some digging and brain using, you should see an update soon. Thank you again!
 
@wparks I enjoyed the information you shared and KS-408A amplifier build was very cool! I have a working Zenith transistor radio from 1977 that sounds good, but the 90V capacitive power supply is not too comforting. I plan to rebuild the power amp section along with the power supply (20V for the radio transistors) with something lower voltage and I think the information you provided is an excellent basis for that future build.

I had planned on using a cake tin for my build as well, also with a clever name in mind.
For the headphone amp, I decided to build an AC/127/AC128 push pull buffer stage since I have 2x of each part. I played with the idea of a silicon BJT (BC550) on the input of the push-pull stage, with a bootstrap resistor to increase impedance close to that of the BUF634, but I only calculated 1M or so with that arrangement. I landed on an Op Amp (OPA2134) input section and I believe the input impedance should be close the 5M of the BUF634 in Pete Millet's design.

C13, C14 are set at 10uf which should be more than ok, but a lower capacitance value (~6uf) is fine.
R15, R16, R17, R18 are set at 5R, which is an estimate. These may be changed after testing.

Capture2.PNG
 
Last edited:
Thank you, GypsysFire, for the kind words. I'm glad you enjoyed it. Do you have a model or schematic for the Zenith? Curious why they are using 90V, seems high.

On the headphone amp- The 1N5291 diode is an expensive item. Have you priced them? I don't know that I would want to pay $18 for one diode. There are a range of these however, for different currents- Did you find this datasheet from Digitron? There may be similar flavors that are more affordable- I did not take the time to look each one up. This amp really will not work very well without them, so wondering your thoughts here.

The OPA2134 is being used as a unity gain buffer, so I'm not sure why there is RV4/RV4 and R5/R6? They are not needed. Also, depending upon what the plate voltage is sitting at (I assume far enough below 12V to have an adequate swing) that C5/C11 and R3/R4 are not even needed. In fact, R3/R4 will overtax the tube, as the tube will see a load of 100K parallel the input impedance of the amp, or essentially slightly less than 100K, which is too low. The beauty of the unity gain buffer is that it needs no biasing. The output will follow the input at whatever voltage, so long as it stays between the supply rails of the op-amp, so C13/C14 ARE needed.

As far as using the op-amp to drive the push pull output, this is a very classical design, and a great choice. I would have you check out Rod Elliot's Project #113, as it is the best AB headphone amplifier I have ever built, and he has a really great setup for biasing the transistors to minimize crossover distortion, and I like how he puts 33uF across each bias diode for best fidelity (he says it does make a difference).

The problem I see here, is that you are using germanium diodes, with a vbe of only 0.2V (or less), and biasing them with silicon 1N4148 diodes, with a forward voltage drop of 0.6V or so. Your germanium transistors will be not just a little "on", but pretty much full on, and the quiescent current through the outputs will be really high, the heatsink will get hot, and the outputs will probably go into a thermal run-away as germanium conductance is very sensitive to temperature. When germanium transistors were biased in this way with diodes, germanium diodes were used so they had the same voltage drop. You could potentially use smaller germanium transistors as diodes here. Another option could be to maybe drop to one silicon diode with a capacitor across it and drive the signal to one end, and dial in the emitter resistors (R15-R18) to a value that limits the quiescent current to a reasonable amount. You could also explore a voltage divider ladder as was used in the Kraco Bake King, where the voltage divider positions each base and has a small resistor from base to emitter to control the amount of bias. Notice this scheme also used negative thermal coefficient (NTC) resistors or "thermistors" in parallel these small resistors, so as temperature increased, the resistance of the NTC decreased, dropping the bias voltage and preventing thermal runaway. Either direction you go, more complication is required. Biasing germanium transistors was always a PITA.

I also want to bring up negative feedback, a topic I alluded to earlier. Not just in the form of DC negative feedback in biasing, but amplifier loop feedback. If you study Rod's #113 design, you can see the classical design in the BJT AB output stage is "inside" the feedback loop. The op-amp in this design is just a non-inverting op-amp, but the BJT's are inside it's feedback loop, so the op-amp will literally do whatever it has to, to drive the BJT's so the output voltage is as perfect a scaled version of the input voltage as possible (scaling determined by input and feedback resistances). This is why the distortion of Rod's headphone amp is minuscule. By running your output stage outside of a feedback loop, your waveform will be subject to all of the inherent non-linearities of the transistors, and of the inherent mis-matching between the NPN and PNP transistors, which will have different current gains so the distortion will be higher. (To a disagreeable amount? Maybe, depends upon matching.) In the Kraco Bake King, you will also see this feedback loop, it's the 330 Ohm coming from the post-cap output into the emitter resistor of the transformer driver Q2. The elegance of Rod's design is that the feedback is both AC and DC, holding the output at ground (mid bi-polar rails) as the input of the non-inverting amp is biased to ground, which eliminates the need of an output capacitor. This is not an option unless you can solve the biassing problem, and change the op-amp unity gain follower to be a non-inverting design with the outputs in the loop. Another challenge there is having a high enough input impedance into the non-inverting amplifier.

It seems to me, that the space charge tubes and the germanium transistors are just not a good fit- each has finicky problems on their own, and these two problems are in direct conflict here. I think you are beginning to appreciate just how elegant Pete's design was- matching about the only perfect driver to handle these finicky tubes. Rod's example of that classical design is very elegant too, combining the current drive of the BJT's to the extremely high open-loop gain of the op-amp. Both are excellent examples of pairing devices that have just the right qualities to play so very well with each other that elminates the need for many other bandaid components to make it work. I encourage you to build either Pete's or Rod's design and experiment with them- I have really enjoyed Rod's headphone amp- it is so clean and clear, has enough power to drive small speakers quite well, and has more than enough power for my 32 Ohm Philips SHP9500 headphones, even with 120 Ohm series resistors. I have not built Petes design, but would like to. Please note I'm not saying give up, just keep trying and keep experimenting and learning, and let's see where it goes.
 
Yeah the 1N5291 is terribly expensive, I looked at comparable parts from the datasheet and found the same price point for all parts in the series. I just stuck it there as a placeholder until a substitute can be found.

Thank you for sharing the Rod Elliot schematic, he always has excellent quality designs. It looks like a high quality build just by looking at it. I immediately noticed the feedback loop on his design and the Kraco King and though it was very interesting. Thank you very much for explaining it. I may give up and build something else, but I want to be sure it's a flop (on paper) before pulling the plug. In that case, Ill just build a Germanium headphone amp based on the information you shared.

For the headphone amp that might amount to a mess. The RV4/RV4 and R5/R6 create the offset adjustment network that allows for precise tuning of the OPA2134’s output voltage, ensuring it remains as close to 0V as possible under no-signal conditions. A direct ground connection could cause input offset currents to vary, which could potentially lead to oscillations or DC offsets.

For the resistors R3, R4, from Pin 3 to ground. This sets the input impedance, in this case it would be set to 100k. It also ensures that you don't have a "floating" input that can pick up noise. I believe the BUF634T input impedance is quite a bit higher at about 10M, increasing R3, R4 to 10M would achieve the same input impedance. I started at 100k, but may end up much higher. The 12AE6A has a relatively high output impedance (~30k), so it should have no issue driving a 1M load or more. Driving a high-impedance load ensures that it doesn’t experience significant voltage drop or signal loss. This is the same reason Pete Millet used the BUF634T as a cathode follower buffer.

That's a really good point with the 1N4148s, I have loads of them on hand but based on your advise I will steer toward using Germanium diodes or another solution. Thank you for the suggestions you made regarding the PITA Germanium biasing, I like the resistor ladder you described. I'll look at employing this in the design, as I have the parts needed already 😉 The more I look at Germanium designs the more I start leaning toward building an all SS headphone amp. Tubes do sound nice though.

I'll keep experimenting and post an update either way, very soon.
 
The RV4/RV4 and R5/R6 create the offset adjustment network that allows for precise tuning of the OPA2134’s output voltage, ensuring it remains as close to 0V as possible under no-signal conditions. A direct ground connection could cause input offset currents to vary, which could potentially lead to oscillations or DC offsets.

I'm going to stand by what I said- they are not needed at all. The OPA need not be "biased" to be near ground and there is no direct ground connection. You don't even need the -12V power rail- It's perfectly happy being powered from 24V to ground (it can handle 36V rail to rail), and the output = input will happily swing anywhere the plate tells it to within that range, because with the output tied to the negative input, it's a unity gain follower. Offset currents really don't matter here- The positive input is held by the plate, the negative input is held directly by the output. No more noise will be induced in the op-amp than is already in the plate. Passing the audio through yet another series capacitor, and coupling the inputs to supply through bias resistors will introduce more, not less noise. The plate obviously has some (enough) bounded swing room- it can't swing so low that the tube shuts off, and it can't swing too high or the diode shuts off. The unity gain follow is plenty happy to duplicate this swing at it's output exactly because it will not hit it's supply rails and present minimal noise.

For the resistors R3, R4, from Pin 3 to ground. This sets the input impedance, in this case it would be set to 100k. It also ensures that you don't have a "floating" input that can pick up noise. I believe the BUF634T input impedance is quite a bit higher at about 10M, increasing R3, R4 to 10M would achieve the same input impedance. I started at 100k, but may end up much higher. The 12AE6A has a relatively high output impedance (~30k), so it should have no issue driving a 1M load or more. Driving a high-impedance load ensures that it doesn’t experience significant voltage drop or signal loss. This is the same reason Pete Millet used the BUF634T as a cathode follower buffer.

Again, the input won't pick up noise, any more than the plate already has unless you intend to make this route long like an antenna. I think you answer this yourself- Pete drove this nearly infinite impedance input of the BUF634 directly from the plate with no resistor to ground- Unless the route is long, there will not be a problem- a 30K output impedance should be more than adequate to overpower whatever noise may be coupled into the buffer input. With the unity gain buffer, you really don't even need the output stage for headphones with resistance above 100 Ohms or so but you would have to drive them through an output capacitor. There are higher current op-amps to choose from but in general I don't drive my 32 ohm cans with op-amps. If you wanted to eliminate the output capacitor to drive the headphones directly, then I could see using a symmetrical supply with a coupling capacitor between the plate and a bias to ground resistor.

Anyway, keep going!
 
@wparks I appreciate the suggestions you made, they are actually all correct! By no means do I disagree, I was just pointing out the basis for my design. The way I had it would "work" but it would have higher distortion and it would be finicky. The global feedback suggestion you made (thank you again) does force a design change, but it makes for a more linear amplifier with higher fidelity and reduced distortion, albeit a slight loss of the Germanium "warmth" but I think it will sound better overall.

I have a pair of matched AC187s, I'll have to pick up some AC188s.
Here's a design based on the ESP schematic you shared, with recalculated component values to make use of the Germanium transistors. The Op Amp is now the gain stage, while the Germanium Quasi-Complementary pair are the headphone buffer stage (unity Gain). From my calculations, the output should be about 2.25w into 32-ohm 50mm drivers.
I know what you may be thinking, "R5 shouldn't be there". This is a current limiting resistor, it makes sure the transistor gets enough base current to operate in its linear region without introducing excess distortion or overdriving. I'm starting with 4.7k, but the value may change during testing.

Capture3.PNG
 
@wparks The Zenith Radio is a 8ZT12 chassis from 1971 (date correction), based upon the date stamp on the output transformer.
It uses a single 121-430 for the pre-amp and a 121-436 (NTE124) in the output stage. The chassis lives in a Crosley 52TF shell, so I don't know the exact model of radio.
 
At this point you must be thinking I'm a knit-picky know it all who has nothing better to do than give you a hard time. I'm sorry, I love design, "chew toy" as you know. I'm happy to see you go this way, but as always I'm going to point out issues I see, it's what I do according to my wife so bear with me.

I don't think the 187/188 are up for these voltages. Wasn't familiar with these so I checked the datasheet, and maximum collector emitter voltage is only 15V. (The 139/140 are 80V) If you are banging this rail to rail the emitter of the upper device will reach way down near the negative rail, or 22 volts or more across Vce and vice-versa. These 187/188 devices must have been designed for transistor radios- not much voltage tolerance at all. Now I see the AC127's weren't any better. ouch.

I did not think about the 1N34A (was not sure what you would use) but I think those will work fine, as long as you don't pass too much current through them (50mA max current). I would expect R6/R7 would be increased significantly- only a few milliamps needed to bias the diodes and supply base currents.

You're right. R5 need not be there. It will only hurt performance. The transistors are biased slightly on by the diodes, and they will remain in the active region of operation but their "staying linear" is not what is creating the low distortion. The point of negative feedback is to place within the loop an amplifier that has relatively little delay, and extremely high gain (called open loop gain). The faster the amp, and higher the gain, the more accurate the output will be. Negative feedback provides an "error" signal back to the input of the amplifier that is compared to the input signal, so that the amp can drive the error to zero, hence no distortion. The op-amp is very fast (delay), and has a very fast output slew rate. The op-amp will make small adjustments to it's output voltage in the megahertz range if it needs, to literally do whatever it has to do to make the output transistors to produce an output waveform that is as much as possible a precise match to a scaled version of the input signal. The op-amp can't do anything that will harm the transistors- they take base current from R6/R7, and the op-amp is only driving an AC voltage. The fast, high gain op-amp can overcome a LOT of evil - such as mis-matches in the drivers, and even supply ripple, and still output very little distortion- The noise and mismatch just becomes another source of error to be cancelled out.

Why do you have an output cap (C6)? It's not needed. R1 sets the op-amp bias to ground, so the output will also rest at ground, as long as the feedback loop can see the DC output voltage, which it cannot if it connects after the cap. R10 is also not needed, if C6 is eliminated. The op-amp will rest at ground, no worries.

If you build this, you will want to add the Zobel network (10 Ohm in series with 100nF at the speaker/output terminals) to prevent oscillation if headphone and input wires couple outside of the amplifier. I think you want R8/R9 closer to the specified 10 Ohms unless you have a specific concern. Whether silicon or germanium emitter degeneration resistors less than one ohm are typically only used in higher power amplifiers, and in this little guy only driving headphones 10 Ohms is plenty small. (With them I can still drive bookshelf speakers plenty loud.)

I dropped R4 down to the recommended 3.3K or so to significantly reduce the gain. This amp is really fast, and really prone to oscillation if it is given a bad layout, and such high gain with the 22k is asking for trouble.

If you intend to build one like this, please let me know. I don't fabricate PCB's, preferring to solder everything onto proto-board. I have a pretty compact layout drawing that still minimized coupling that has worked great. I can scan if you are interested. I wanted a compact size, so I found a source for the 10 Ohm resistors in 1W flavor that are as tiny as 1/4W carbon resistors.

I'll check out that Zenith radio- Thanks for the info.

-Warren
 
@wparks No worries at all, I'm the same way in an Engineering review. There's always something that can be improved or even removed. I appreciate your input, especially if it points out potential issues.

You're right, ±12V is at the top end. The transistors can handle it and they would have a much better voltage swing, but thermal stability will be a big consideration. I dropped it down to ±9V, where most designs with these transistors live.

C6 - 1000uf output coupling capacitor:

A global feedback loop can help minimize DC offset at the output and might theoretically allow the AC187/AC188 buffer stage to connect directly to headphones. However, there are some limitations to this method. A global feedback loop that includes the output stage can help fix small DC offsets by sending part of the output back to adjust the input, keeping the DC offset close to zero. This method is most effective if the initial DC offset is low or if the circuit has a biasing network to manage the offset. In favor of simplicity, there's no adjustable biasing network in this push-pull transistor buffer.

Germanium transistors can experience thermal drift, leading to significant changes in their V_BE with temperature. Even with feedback, this drift can cause the DC offset to vary, especially as the amplifier heats up. Feedback can help, but if the offset changes too much, DC current through the headphones could still be a problem. This would not be an issue with Si transistors in the circuit, so direct coupling would be used in that case.
Direct coupling may not present an issue in this circuit and most professional grade headphones can handle a small amount of DC voltage, but I'd prefer to error on the side of caution and keep the cap. Also, I have some Nichicon Muse 1000uf 25V, non-polarized electrolytic capacitors that I really want to use in something.

R5 - 4.7K (~2.2k - 4.7k)

Based on this calculation; with the 4.7k resistor, the current through the transistor base would be approximately 1.84mA. This is a very well balanced current to drive these transistors.
AD_4nXeEIHUnbOPlwtNwt14GqxToJxkPT4A8q0Oc130FzU164mYyBZoh0-jl_cdCi6PXFK5H2TI1X3Anp_vPwEfW1OqRQOI8LTFCH_CueK5kxSDM6va1x5ISTuuDnwYiW6p4bVDUE5eGew

Without the resistor, the OPA2134 could try to deliver a much larger current. The maximum output current of the OPA2134 is around 35mA, in theory it could drive up to that current. The actual current depends on the base-emitter voltage of the transistor (0.2V), but since there's no resistor to limit it, the op-amp would attempt to supply a very high current, well beyond what the transistors can reasonably handle.

R6, R7 - 420R (Corrected from 100R)

Good catch, the original 100R resistors were indeed too low, as they were based on a miscalculation.
These resistors are intended to set a stable current through the bias diodes and establish a proper bias voltage for the transistor bases.
To calculate the resistor values, I started with a reasonable quiescent current of 10mA. In this case, 420R is correct after the adjustment to ±9V rails.
AD_4nXfN2-3Ist_PR0nw1Gt1EiMNAQYxXSIEf8EUhYHb4z-yajczCKWOUVulDjJKhNSwIJl7X4clJNZtfNJKQtm5sC-CZp77My6DUlITfZW3SZNnW4GzP5dFdhAh0COX-4NBX9CeFTY--A


Zobel Network

As for the Zobel Network, that's a can of worms I'm looking foreword to opening. I subscribe to the idea that the capacitor quality in the output and speaker network matters very much. https://www.humblehomemadehifi.com/Cap.html is a good resource for capacitor information in this regard, although it's highly subjective.

Updated schematic:

image_2024-11-13_151642456.png
 
Ok, if you are not getting annoyed, I enjoy the process and am grateful to be able to have these conversations with you. I'm glad you are not offended if I disagree, and vice versa, I like a civil debate, because I learn a lot that way once someone is motivated to explain their rational.

Even with feedback, this drift can cause the DC offset to vary, especially as the amplifier heats up. Feedback can help, but if the offset changes too much, DC current through the headphones could still be a problem.
I don't agree that DC feedback will only correct small offsets. The op-amp is capable of driving the output devices to whatever base voltage necessary to keep the output at ground, unless you are so out of whack that one transistor is saturated and the base voltage is railed at one side or the other. But, as you say, you have some golden capacitors you are jonesing to use- I understand that, but if you do at least take the feedback from before the cap?

Transistor heating is the enemy here, but I think the voltage levels are so low that with any decent heatsink you should be running near room temperature, so long as you can get a reasonable quiescent current. In Rod's design he was not too concerned about crossover distortion because the op-amp is so fast and because the emitter resistors were fairly large. You will have to build this and see what the current is. Why are R8 and R9 still so small? If you are worried about your headphones, and about bias current, I would think the 10 Ohm Rod uses would be just the thing. The one I built has =plenty= of moxie even with these at 10 Ohms, and they cure a lot of evil. I still disagree that the op-amp can do anything to hurt the transistors by overdriving the bases. All it can do is drive one or the other to saturation, in which case the other driver is at it's bias quiescent or more likely cutoff.

Also wondering about the even higher gain? (R4=30K). Shooting for a gain of 31? Again, I suggest the 3.3K or lower- mine is pretty eager to get way loud even at that.

So, I have to ask- what's with the different supplies? You know the op-amp is plenty happy at +/-9V right? Or, since you are OK listening to your music through input and output capacitors why not do a virtual ground and run off of a single ended supply? I'm one of those "capacitors are bad- mmmK" kind of people, and like to eliminate as many from the path as possible. I have gone through the pain of building a really nice +/- 15V @ 1A regulated supply (based upon ESP #5) for running all of my op-amp gear- my ESP #6 phono preamp, my headphone amp (113), and my soon to be system preamp (based upon the Accuphase C-3850 Chinese clone), and it was as involved as any of my other projects in terms of time and pain. Do you have one already built? (I guess what I am asking is, are you such a masochist that you would complicate this effort further by requiring +/-9V TOO? I'd rather add two resistors for a virtual ground)

BTW- I have found it's a real PITA trying to find an elegant way to plug-in-connect +15V, GND and -15V between the power supply and my op-amp accessories. I refuse to buy some funky boutique multi-pin connectors because I prefer panel-mount, and they are big, gaudy and expensive. Can't use a 1/4" or 1/8" stereo phono plug for fear of shorting. I eventually settled on RCA connectors- color coded female panel mount on the devices, with a beefier RCA cable I make myself. Being a consummate dumpster diver, I found that the electrical wires inside of thin vacuum cleaner power cords are great for this- very flexible relatively heavy 16 gauge with thin insulation, in white and black. I braid three conductors and install gold plated RCA connectors with red and black heat-shrink sleeves on them. It looks pretty cool anyway.

I wanted to mention that I have also started thinking of an op-amp germanium hybrid amp. I like what you are doing, but I don't want to have to deal with finding complimentary pairs. I am inspired by the common germanium topology of interstage transformers and totem pole drivers. I'd like to have a similar op-amp front end, driving the primary of the transformer with the op-amp, and use the same type of feedback loop. I could use op-amps to generate opposite phase signals and drive the outputs directly, but I really want to keep the flavor of the transformers in there. Or, maybe explore a single-ended design, either directly, or the interstage transformers could be repurposed as output transformers for a headphone amp. They typically have a several hundred ohm primary and two 3-8 Ohm secondaries, (could be put in series or parallel) and are already designed for the DC bias currents without saturating the core. The transistors out of 8-tracks tend to be along the lines of the 2SB474 or the 2SB481, which are ~35V ~1A devices and I really like the TO-66 package.

Anyway, I've gone on long enough. Keep me posted!
 
Last edited:
@wparks Planet and life lol

You're certainly right about some things. First, R4 must have been a typo, because we're not going for a gain of 30, that value should be 3K and not 30K.

R5: Yes, you're correct! AC187/AC188 have a maximum base-emitter voltage of 10V, so R5 is used to limit the current from the op-amp and protect the transistors from exceeding this voltage. It ensures that the op-amp’s output stays within safe limits for the transistor's base-emitter junction.

The capacitor vs. transformer debate isn't much fun, both have pros and cons. It all depends on the design and desired results. For this one, coupling capacitors make sense because I don't have a matched pair of perfect output transformers in my junk drawer, otherwise I'd use them over transformers. Audio transformers introduce a certain non-linear distortion that gives it a certain sound you can't otherwise get. Anyway, we'll leave the caps but transformers could easily replace them.

For the resistor values chosen for R8, R9 of 0.22R: The circuit's quiescent current is determined by the voltage across the diodes (1N34A) and emitter resistors (R8 and R9). Increasing R8 and R9 from 0.22R to 10R would drastically reduce the quiescent current, likely introducing significant crossover distortion. The value should be correct, but it could be increased to around ~0.47R to slightly increase thermal stability, while minimizing crossover distortion.

As for the power rails, I've worked on worse things. They could run on 12V+ with a virtual ground at mid supply, This could be done. However, the AC187/AC188 output stage relies on symmetrical voltage swing for proper biasing and thermal stability. A single-ended supply shifts the quiescent bias point to 6V instead of 0V, affecting how the transistors operate. If the Germanium pair needs +/- rails, why not do the same for the entire circuit? The OPA2134 could run on 9V+/-, but it may reduce dynamic range a little bit on high energy passages if the amp is driven hard. This might be a good design compromise that will simplify the power supply and reduce size.

I've made some changes to the circuit and I believe it has potential for excellent sound quality. It's on separately decoupled 9V+/- rails to reduce crosstalk and distortion. R4 is increased from 3K3 to 9K due to the voltage change. This is the final revision.

Capture3.PNG


The design could be simplified by removing C6 and R10 if there's no DC at the output during testing. You could run it on single 9V+/- rails with shared decoupling, but individual rails are better.
 
Last edited:
@wparks

Here's the simplified version, If there's little to no DC voltage at the output, we can omit C6 and R10. I think at 9V+/-, the Germaniums should remain stable with adequate heatsinks and thermal drift leading to any DC bias getting through should be no concern.
We can power the gain and buffer stage on shared, decoupled rails. This may increase crosstalk a little bit, but I don't think it would be too noticeable.
Given the simple power supply, this could easily be a portable AC or battery powered headphone amp 🙂
I think I'll end up building this one for that very purpose. Thank you for your help, Warren, you're critique forced it into simplicity and it's very good.

Capture5.PNG


-Zack