I'm assuming this has been covered before in the past, but I could not find it with searches.
I've been reading on and off for a number of years and have followed a number of threads with great interest, there is some amazing work shown here. But one thing that I haven't seen (and probably just missed), is how to pick a good transistor for audio?
There are a number of parameters regarding transistors, and some of them I assume don't apply when you think about 'sounding good'. For example, Vce I would think would only be important to make sure the transistor can work in a given design.
I would assume just using the transistors that Nelson Pass uses would good choices (I'm a big fan of his since the Threshold days). Or copying one of the many designs found here would a good starting point.
But what if you wanted to pick some transistors? Where do you start? I would assume something claiming to be low noise would be a good start.
So, where does one start? What makes a given transistor a good selection for audio use?
Thanks in advance for any advice you can offer.
I've been reading on and off for a number of years and have followed a number of threads with great interest, there is some amazing work shown here. But one thing that I haven't seen (and probably just missed), is how to pick a good transistor for audio?
There are a number of parameters regarding transistors, and some of them I assume don't apply when you think about 'sounding good'. For example, Vce I would think would only be important to make sure the transistor can work in a given design.
I would assume just using the transistors that Nelson Pass uses would good choices (I'm a big fan of his since the Threshold days). Or copying one of the many designs found here would a good starting point.
But what if you wanted to pick some transistors? Where do you start? I would assume something claiming to be low noise would be a good start.
So, where does one start? What makes a given transistor a good selection for audio use?
Thanks in advance for any advice you can offer.
Hi,
It is very simple, just use J-Fets and Lateral Fets. They are "good for audio".
The rest is selecting enough power handling, voltage handling, current handling and transconductance for the Job and if you design looped feedback circuits you also need get enough Ft.
Bipolars have so many issues, you really need to characterise the part yourself (as most datasheets no longer show H parameter variability with voltage/current) to be even able to make good guesses, their parasitic capacitances are all over the place with signal, their thermal behaviour throws further spanners into the works, so why deal with all this pain?
For Bipolars look at Hfe, Hie, Hre and Hoe and more critically how they change with the Voltage across the device and the collector current. You also need to look at how Ft changes with current changes and parasitic capacitance changes with voltage change. And then you may weep and go and use something else for audio, like tubes, that is actually suited for the job.
For J&L Fets you only really need to look at Gm vs. current and voltage and possibly the parasitic capacitances. VFets are just plain bad news in terms of parasitic capacitances and their modulation with signal. After looking at Fets you may also weep and go and use something else more suited for audio use, like tubes...
For Bipolars it is transistors that show minimal variation of H parameters with Signal (current/voltage) and have low parasitic capacitances and high and stable Ft with signal changes.
For J/L Fet's is a GM that is constant with signal changes and low capacitances.
Ciao T
It is very simple, just use J-Fets and Lateral Fets. They are "good for audio".
The rest is selecting enough power handling, voltage handling, current handling and transconductance for the Job and if you design looped feedback circuits you also need get enough Ft.
Bipolars have so many issues, you really need to characterise the part yourself (as most datasheets no longer show H parameter variability with voltage/current) to be even able to make good guesses, their parasitic capacitances are all over the place with signal, their thermal behaviour throws further spanners into the works, so why deal with all this pain?
For Bipolars look at Hfe, Hie, Hre and Hoe and more critically how they change with the Voltage across the device and the collector current. You also need to look at how Ft changes with current changes and parasitic capacitance changes with voltage change. And then you may weep and go and use something else for audio, like tubes, that is actually suited for the job.
For J&L Fets you only really need to look at Gm vs. current and voltage and possibly the parasitic capacitances. VFets are just plain bad news in terms of parasitic capacitances and their modulation with signal. After looking at Fets you may also weep and go and use something else more suited for audio use, like tubes...
For Bipolars it is transistors that show minimal variation of H parameters with Signal (current/voltage) and have low parasitic capacitances and high and stable Ft with signal changes.
For J/L Fet's is a GM that is constant with signal changes and low capacitances.
Ciao T
Glowing bulb fantasy ...? ....🙂
Show me an published IEEE or AES paper showing the superiority of tubes over solid state devices for audio.
-Peter Aczel
Show me an published IEEE or AES paper showing the superiority of tubes over solid state devices for audio.
-Peter Aczel
Hi,
Why should I?
A single look at the various datasheet suffices to illustrate the points. Please just peruse them, they are all over the internet.
Ciao T
Why should I?
A single look at the various datasheet suffices to illustrate the points. Please just peruse them, they are all over the internet.
Ciao T
I think a.wayne's post was probably at least somewhat tongue in cheek, noting who he discloses as the author of that remark. 😀
Hello Mr. Wayne Sir,,I always enjoy your posts..But sometimes you may be taking what other people are telling you a little to seriouly. Please enjoy the sprit of things someone is telling you.😛 This person did not tell you anything bad In fact. 😉So enjoy Regards Evette. Forgive my bad spelling Youi know I was brought up BAD.
Hi
I’m no pro but here is some info.
Well it honestly depends on the intended application and what stage/type of amplifier or equipment you're designing. I don't know of any Holy Grail transistors, they're always trade off's with spec’s price availability high tide maybe low tide.
If I’m designing a standard 3 stage negative feedback amp I use low noise small signal pnp's for the input comparator that have vce's above at least a single rail. I'm sure hfe, capacitance, and bandwidth are important to somebody but I've never noticed much difference between your standard low frequency small signals’s at this stage. I will however make sure that the transistors intended working current versus gain bandwidth (Ft) is somewhere on the data sheet curve.
I'll throw a couple of low voltage high gain small signal npn's at the bottom to make a current mirror which will help in overall gain/lineararity in the feedback loop. Voltage and current seen by these is minimal so again check your datasheet curves. A couple 2n3904's or the like are usually ok.
Next an upper PnP CCS is a good practice, again a vce above the upper rail is required and current is minimal 3~8ma so some more small signals will work fine.
VAS/VBE multiplier stage: you're going to need some transistors you can physically mount to something and in the case of VBE multiplier transistor it will need to be mounted to the main power transistor heat sink. So first off its case type, I usually go with medium power to-126/sot-32/82 types. More then likely you'll also use these as power transistor drivers, but not always, so I buy complementary PnP/NPN pairs here to save cost later. Your upper PnP completing the upper ccs will see both rail voltages across it so its vce needs to be above both rail voltages added together and maybe throw in a 20% error of caution. The same goes for the lower class A NPN driver. Currents here will be between 10~20ma for this hypothetical amp stage so check your datasheet curves. The vbe multiplier: just use whatever you got in the through hole NPN variety even though I have seen to-92's used here mounted to the main heat sink so?? The voltage here is low and current 10~20ma. The ac characteristic's are not important due to this being bypassed with a cap and is only used to set the forward vbe of the output stage and thermal tracking.
I like multi element Darlington’s for outputs because of how simple they are and they include a reverse bias diode. But many do not because of the lack of control and tailor ability you gain with individual drivers and power trannies. These are load/cost decided for me. Data sheets are quite important here. You need to project worst case operating scenarios for you output stage including: vce above rails with at least 30-50% for emf talk back, pp current for intended load must under the worst heat condition never be aloud to touch the secondary breakdown curve on the data sheet. Complementary Devices are best but that NFB can really fix a lot. Then make sure your drivers can handle the same vce as your outputs. PP current divided by the HFE of your output transistors must not exceed the secondary breakdown curve of the drivers and don’t forget to de-rate both curves for heat.
In a negative feedback loop amplifier you can really get away with a lot of short cuts in the ac characteristics of transistors but heat volts and amps will burn you every time. This is but one of thousands or tens of thousands of audio circuits out there with many having more specific needs.
I’m no pro but here is some info.
Well it honestly depends on the intended application and what stage/type of amplifier or equipment you're designing. I don't know of any Holy Grail transistors, they're always trade off's with spec’s price availability high tide maybe low tide.
If I’m designing a standard 3 stage negative feedback amp I use low noise small signal pnp's for the input comparator that have vce's above at least a single rail. I'm sure hfe, capacitance, and bandwidth are important to somebody but I've never noticed much difference between your standard low frequency small signals’s at this stage. I will however make sure that the transistors intended working current versus gain bandwidth (Ft) is somewhere on the data sheet curve.
I'll throw a couple of low voltage high gain small signal npn's at the bottom to make a current mirror which will help in overall gain/lineararity in the feedback loop. Voltage and current seen by these is minimal so again check your datasheet curves. A couple 2n3904's or the like are usually ok.
Next an upper PnP CCS is a good practice, again a vce above the upper rail is required and current is minimal 3~8ma so some more small signals will work fine.
VAS/VBE multiplier stage: you're going to need some transistors you can physically mount to something and in the case of VBE multiplier transistor it will need to be mounted to the main power transistor heat sink. So first off its case type, I usually go with medium power to-126/sot-32/82 types. More then likely you'll also use these as power transistor drivers, but not always, so I buy complementary PnP/NPN pairs here to save cost later. Your upper PnP completing the upper ccs will see both rail voltages across it so its vce needs to be above both rail voltages added together and maybe throw in a 20% error of caution. The same goes for the lower class A NPN driver. Currents here will be between 10~20ma for this hypothetical amp stage so check your datasheet curves. The vbe multiplier: just use whatever you got in the through hole NPN variety even though I have seen to-92's used here mounted to the main heat sink so?? The voltage here is low and current 10~20ma. The ac characteristic's are not important due to this being bypassed with a cap and is only used to set the forward vbe of the output stage and thermal tracking.
I like multi element Darlington’s for outputs because of how simple they are and they include a reverse bias diode. But many do not because of the lack of control and tailor ability you gain with individual drivers and power trannies. These are load/cost decided for me. Data sheets are quite important here. You need to project worst case operating scenarios for you output stage including: vce above rails with at least 30-50% for emf talk back, pp current for intended load must under the worst heat condition never be aloud to touch the secondary breakdown curve on the data sheet. Complementary Devices are best but that NFB can really fix a lot. Then make sure your drivers can handle the same vce as your outputs. PP current divided by the HFE of your output transistors must not exceed the secondary breakdown curve of the drivers and don’t forget to de-rate both curves for heat.
In a negative feedback loop amplifier you can really get away with a lot of short cuts in the ac characteristics of transistors but heat volts and amps will burn you every time. This is but one of thousands or tens of thousands of audio circuits out there with many having more specific needs.
Hi,
I must admit my comments of "just go and use a tube" are a bit tongue in cheek as well, even though as in a joke a jester may make, it also contains a great truth.
Transistors, Fets and tubes are what we have to amplify a signal. No device is "perfect". They all suffer from problems and hence there is no "universally good" solution.
We want the the signal to be music. And we want to listen to it. So if we had an absolute ideal amplifier and speaker we would all agree that that is is well suited to the Job.
With real devices and real amplifiers and speakers we have problems and all is a compromise and not everyone agrees what the best compromise is.
And so we get people championing each and ever technology.
If we bend our will to dealing with each devices problems and are willing to compound circuits out of many different devices to maximise the gain from their strength and to minimise their drawbacks, we may be able to get closer to at least some version of the ideal that is close.
Ciao T
I must admit my comments of "just go and use a tube" are a bit tongue in cheek as well, even though as in a joke a jester may make, it also contains a great truth.
Transistors, Fets and tubes are what we have to amplify a signal. No device is "perfect". They all suffer from problems and hence there is no "universally good" solution.
We want the the signal to be music. And we want to listen to it. So if we had an absolute ideal amplifier and speaker we would all agree that that is is well suited to the Job.
With real devices and real amplifiers and speakers we have problems and all is a compromise and not everyone agrees what the best compromise is.
And so we get people championing each and ever technology.
If we bend our will to dealing with each devices problems and are willing to compound circuits out of many different devices to maximise the gain from their strength and to minimise their drawbacks, we may be able to get closer to at least some version of the ideal that is close.
Ciao T
I design with semi-conductors during the work day and just as studiously avoid them at night when I am designing my own toys. Hard to imagine a much more linear device than a D3A without feedback - so you know what camp I fall into. 😀
Hi,
If "get away" means "obtain a certain set of objective specifications" this may be true (though I'd probably take exception here too), in terms of subjective sonic qualities you will find each shortcut costs.
IF I was making such an amplifier as you describe (which I probably would not) and was limited to using Bipolars (which I would instantly give up as a bad job anyway) I would uses:
1) a NPN Buffer before the PNP Input Stage, using really high Beta transistors (>= 300) and select the PNP's carefully for low early effect, good Linearity and all and preferably use a single die dual transistor PNP for the job
2) Heavily degenerate the input stage (several 100 ohm) and cascode the CCS for the tail
3) Cascode the input stage (Hawksford style).
4) Use a heavily degenerated improved Wilson Mirror (4 transistor) as input load
5) Use a PNP Buffer with high Beta and CCS Load (to the positive rail or ground) in front of the VAS Transistor
6) Select the VAS transistor for low distortion (see Samuel Groners comments on D. Selfs book for quite detailed measurements), generally means low early effect
7) significantly degenerate the VAS Transistor (probably at least 100 Ohm) and run a lot of VAS current
8) Cascode the VAS Transistor with a Hawksford Cascode
9) Load the VAS with a Bootstrapped resistor over a CCS or use a cascoded CCS possibly with being connected to act as PP VAS at high frequencies.
10) Use an output stage that is at least a tripple or possibly even a quad to ensure the VAS is not exposed to a non-linear load
11) Use some form of "thermal track" output transistors though even those have too much of a thermal timeconstant.
12) Measure the thing, listen briefly, be disappointed and plug my Tube Amp back into the system...
Ciao T
If "get away" means "obtain a certain set of objective specifications" this may be true (though I'd probably take exception here too), in terms of subjective sonic qualities you will find each shortcut costs.
IF I was making such an amplifier as you describe (which I probably would not) and was limited to using Bipolars (which I would instantly give up as a bad job anyway) I would uses:
1) a NPN Buffer before the PNP Input Stage, using really high Beta transistors (>= 300) and select the PNP's carefully for low early effect, good Linearity and all and preferably use a single die dual transistor PNP for the job
2) Heavily degenerate the input stage (several 100 ohm) and cascode the CCS for the tail
3) Cascode the input stage (Hawksford style).
4) Use a heavily degenerated improved Wilson Mirror (4 transistor) as input load
5) Use a PNP Buffer with high Beta and CCS Load (to the positive rail or ground) in front of the VAS Transistor
6) Select the VAS transistor for low distortion (see Samuel Groners comments on D. Selfs book for quite detailed measurements), generally means low early effect
7) significantly degenerate the VAS Transistor (probably at least 100 Ohm) and run a lot of VAS current
8) Cascode the VAS Transistor with a Hawksford Cascode
9) Load the VAS with a Bootstrapped resistor over a CCS or use a cascoded CCS possibly with being connected to act as PP VAS at high frequencies.
10) Use an output stage that is at least a tripple or possibly even a quad to ensure the VAS is not exposed to a non-linear load
11) Use some form of "thermal track" output transistors though even those have too much of a thermal timeconstant.
12) Measure the thing, listen briefly, be disappointed and plug my Tube Amp back into the system...
Ciao T
Beauty is simple
The beauty of this amp is its simplicity. I've built geezz I don't know, maybe 100 or so of this style amp with cheap minimal parts with great success. Granted I don't own any fancy distortion analyzers but I've got my ears and my friend’s ears. I have built a few amps from slone and others that imho were over cascoded mirrored and degenerated to death. They must be better or why would anyone go through the trouble, but I couldn't really hear any difference. I currently have a couple el84 12ax7 and some edcor's I'm fiddling with, and being my first tube attempt wish me luck.
The beauty of this amp is its simplicity. I've built geezz I don't know, maybe 100 or so of this style amp with cheap minimal parts with great success. Granted I don't own any fancy distortion analyzers but I've got my ears and my friend’s ears. I have built a few amps from slone and others that imho were over cascoded mirrored and degenerated to death. They must be better or why would anyone go through the trouble, but I couldn't really hear any difference. I currently have a couple el84 12ax7 and some edcor's I'm fiddling with, and being my first tube attempt wish me luck.
Even some of the top designers use vertical MOSFETS. They are very much maligned. I have heard all the arguments before but the proof is in the listening and the amps I have designed with vertical mosefts sound great.
Ah look Ken't ain't she a beaut.
An externally hosted image should be here but it was not working when we last tested it.
Oh yes, NP is very good in choosing a transistor. I also think that tube amp designers should be good in designing transistor amps, because they are used to put the amplifying devices (tubes) to operate at their best operating condition. This is what should be done with transistors also.
It means that how the amp is designed is related to what transistor to be used.
Low noise transistor is always good, as long as they are usable (see the max. operating condition, Vce) and have also good feature for the job (hfe, cob, fT). With low Vce, low noise transistors can be used in the input stage of low supply rail circuit, or in a cascoded input stage. But noise in the input stage may come also from active elements (ccs) and power supply, so target overall noise can be achieved by perfecting the supply and ccs instead of using low noise transistors. The strength of a bjt is imo the transconductance. It is this feature that will lead to great sonics, hence enjoyable sound reproduction. If this feature can be achieved from other stages of design, then it is okay not to prioritize the transconductance at the input stage (but I found that in most situation high hfe will tend to bring better "result" than low hfe low noise transistor).
The lowest noise can be achieved by JFETs. Unfortunately they have a low transconductance. I always suspect any JFET input amps. They have a clear audible positive performance, but I think they have something negative that is not so audible, which is the flat, uninteresting sound (especially when driving low transconductance output). I suspect this is because of the low transconductance. Better imo is a JFET that has MOSFET characteristics. They are rare, but I believe they exist. It is like parallelling JFETs where input capacitance is lowly maintained but transconductance is improved.
Low Cob transistor (bjt) is very critical in the VAS (No, I'm not a theorist but a tweaker. I listened to transistors, ranked them, and see what parameters that they have to sound as they do). If a circuit has a big capacitance (compensation), I will reject it. I will not build an amp showing a miller cap greater than 100pF. And an internal capacitance of a transistor does have the same importance.
Related to capacitance, high fT output transistor is important. Of course the circuit must be designed to take advantage of this fT, but from experience it seems that most circuit will benefit from high fT output transistor.
I'm not a fan of bipolar amps, so I don't have much experience tweaking them, but from my experience building bipolar amps so many years ago, the one with Sanken 2SC2922/A1216 was the one who was close to my expectation (in the low frequency region, as a result of linearity I guess). Lately when I examined its parameter (datasheet) I found that it has an fT of 50MHz! (even bigger than the newer Sanken). It brought me to believe that fT is the most important parameter for bipolar output stage (and the amp must be designed for wide bandwitdh I believe). Unfortunately though, I couldn't find an "original" (or up to spec) Sanken in my country nowadays.
For mosfets, especially the smaller ones, especially in class-B, I found that Cis is the weakest link. No matter how nice the other parameters, if the Cis is bigger then the overal performance is lower.
NP and others have tried to increase current to tame this capacitance, but soundwise I cannot see it work.
Many also have thought that MOSFET circuit can be made shorter than bipolar circuit (look at the Citation mod by NP). Look at the so many designs including the one suggested by Hitachi. I am against this. Mosfet (laterals) have far lower transconductance than bipolars. Shortening the circuit is a mistake imo.
Look at the famous crescendo for example. It measures very well, but how much enjoyment will it bring to the music listener? Driving a latfet with such cascoded VAS is the only suspect I have for its "inferior" sound quality.
For class-A design, the higher capacitance of a VFET (compared to LFET) is no longer a weakest link so that its higher transconductance become a big benefit. I couldn't find for example, a class-A circuit where LFET perform better than VFET.
To summarize, I think we need a high transconductance, linear amplification.
Which top designers? Top designers are usually not good listeners. NP is a good one (he will be a legend like JLH), but where has he used VFET for class-B? Citation, yes, but it's his biggest "failure" after PLH 😀
Hi,
I build Amp's like that in the 80's. They did not sound very good then and I doubt they have improved. Even in the 80's I went quickly to less usual topologies (Cascoded VAS and Circlotron Outputs).
Make sure your Tube Amp has enough power and the correct damping for your speakers, or you will find the results disappointing. If you get it right the results are great.
I grew up with old Tube Radio's in Germany (there speaker and amplifier are designed together and the results can be exceptional) and ever since have tried a lot of stuff with Solid State and Hybrid Amplifiers but I always come back to real tubes...
My current Amp is very simple, and Tube.
It combines a EL34 Push-Pull output Stage operated as what Menno Van Der Veeen Super-Triode (and which I call "ARC-Connection" after the first documented use) run at 60mA/380V per tube.
Gives around 32 Watt before clipping and around halve that in Class A1. Damping Factor is around 2, so the speakers need to be designed to run with low damping factor amplifiers...
The output stage is driven from a low impedance (4.7K in cathode and anode) split load phase splitter with two sections of what you might call a russian 5687 connected in parallel, running at around 190V/20mA.
Sole Voltage gain stage is a ECC83 as SRPP.
No looped feedback.
No electrolytic Capacitors in the signal circuit anywhere.
Silver Foil coupling capacitors. Good quality RCA's and Binding Posts.
Ciao T
I build Amp's like that in the 80's. They did not sound very good then and I doubt they have improved. Even in the 80's I went quickly to less usual topologies (Cascoded VAS and Circlotron Outputs).
Make sure your Tube Amp has enough power and the correct damping for your speakers, or you will find the results disappointing. If you get it right the results are great.
I grew up with old Tube Radio's in Germany (there speaker and amplifier are designed together and the results can be exceptional) and ever since have tried a lot of stuff with Solid State and Hybrid Amplifiers but I always come back to real tubes...
My current Amp is very simple, and Tube.
It combines a EL34 Push-Pull output Stage operated as what Menno Van Der Veeen Super-Triode (and which I call "ARC-Connection" after the first documented use) run at 60mA/380V per tube.
Gives around 32 Watt before clipping and around halve that in Class A1. Damping Factor is around 2, so the speakers need to be designed to run with low damping factor amplifiers...
The output stage is driven from a low impedance (4.7K in cathode and anode) split load phase splitter with two sections of what you might call a russian 5687 connected in parallel, running at around 190V/20mA.
Sole Voltage gain stage is a ECC83 as SRPP.
No looped feedback.
No electrolytic Capacitors in the signal circuit anywhere.
Silver Foil coupling capacitors. Good quality RCA's and Binding Posts.
Ciao T
Hi,
There are reasons for that other than sound quality. You can design around device limitations if you must. The question in DIY is however, should we?
In a commercial Amp I can save a lot of money in production by using IR Vertical Fet's over the second source copies of Hitachi Laterals (as Hitatchi long stopped production). I can use fewer output devices and these fewer devices cost a tenth or less. If one has to accept compromises in performance and spend more time on design, so be it, with this kind of return on investment.
But if I am going to build an Amplifier for myself, my personal satisfaction, with no particular (sensible) limit on parts costs (once you count the cost of your time it becomes obvious why making a cheap amp from cheap parts does not give value for money), why should I hamstring myself by unnecessary compromises?
Ciao T
There are reasons for that other than sound quality. You can design around device limitations if you must. The question in DIY is however, should we?
In a commercial Amp I can save a lot of money in production by using IR Vertical Fet's over the second source copies of Hitachi Laterals (as Hitatchi long stopped production). I can use fewer output devices and these fewer devices cost a tenth or less. If one has to accept compromises in performance and spend more time on design, so be it, with this kind of return on investment.
But if I am going to build an Amplifier for myself, my personal satisfaction, with no particular (sensible) limit on parts costs (once you count the cost of your time it becomes obvious why making a cheap amp from cheap parts does not give value for money), why should I hamstring myself by unnecessary compromises?
Ciao T
- Status
- Not open for further replies.