Best Transistors at Each Amp Stage

Jump to Latest
R

RussellKinder

Member
Joined 2015
#1
This was a great thread on output BJTs:
https://www.diyaudio.com/forums/sol...istor-families-audio-power-output-stages.html

Has anyone done a similar post on transistors for drivers, VAS, or input pairs?

Back around 1990, I used 2N2920 and 2N3811 matched pairs for inputs, but those are obsolete now. I'm looking at the SSM2212, but I'm wondering about it's performance above 1 mA. The datasheet graphs stop at 1 mA, but the device's ABS MAX for current is 20 mA. I used to run the 2N2920 and 3811 pretty hot. I had little starburst heatsinks on several of them.

I'm looking at BC847/BC857 for a lot of low volt general purpose mirroring and buffering around the amp. Some MMBT5401 and MMBT 5551 for high volt support circuitry. I'm looking at KSA1381 and KSC3503 for VAS, high voltage cascoding, and predrivers. Not sure that's the best choice, but I think it must be close to the best.

I'm looking at MJE15032/33 for drivers. I'm not sure these are the best, but they look good enough. They've been around awhile.

Anyway, I hope this will generate some discussion.
 
#3
RussellKinder said:
Anyway, I hope this will generate some discussion.
Click to expand...

You're the best, very needed topic.

First of, let's talk about useful parameters of transistors for each of stages.

1. IPS.
Input needs as small as possible noise, as low as possible Cob and as stable as possible current. So, low-noise and well-matched transistors and cascoded current source.
Taking as much as possible gain from IPS asks for current mirror loading of the IPS input.
Low drift and low-input related bandwidth limiting asks for smaller dissipation and low parasitic capacitance, so not only current source transistors must be cascoded, but an active input devices also.

2. VAS.
Next gain stages (one, two or more) can be easily differentiated by its output swing. Only last VAS stage exhibits full swing while all other amplification stages works in a small-signal mode.
So, last device are better to have as small Early effect as possible, while most other are better to be low-noise and high-gain. Also, last stage are better to have as small as possible Cob for lowering Miller effect or be wired as common base or be cascoded. In the case of common base such a stage will have low input impedance while picking as much as possible gain are dictates IPS-VAS coupling point to be as high inpedance as possible, so additional coupling stages will be needed, such a point also dictates as small as possible parasitic capacitance, again small Cob.
Cascoded VAS otput can combine high-gain and high-SOA devices but demands for higher supply.

3. OPS.
Common practice there are EF-triple or BJT+MOSFET pair of followers.
In any of them first stage again must have as small as possible Cob because it loads VAS output straightly.
Next follower stages dictates higher current gain and much valueable - as wide as possible safe operating area.
Also there are huge aspects about EF3 stability, but this is a theme for dedicated topic.

Choosing of proper devices becomes hard each year. IPS can be easily built from widely available BC550/560, 850/860, 2n3504/3506. Inner VAS stages can be build on them while in case of high supply rails those devices can be cascoded by MPSA or MJL15 transistors. Last VAS stage and OPS first follower stage are better to be built with again widely available KSC3503/KSA1381 transistors. Second OPS EF3 stage can be built with widely available KSC2690A/KSA1220A, most of home users will not be limited by its voltage rating.

All other musings are better to talk with known schematic because there are very much optimisation points, not only just transistors models itself.
 
Last edited:
  • Like
Reactions: altec9440, inputdejv and savu
#5
A few more thoughts on input pairs.

I generally find that the input devices that have optimum low noise and high gain have Vce breakdown that is too low to use in that location without cascoding. I generally put a cascode pair above the input pair, and make them float 2 or 3 volts above the input common-mode voltage. The characteristics of the cascode transistors are far less critical.

Of course, I get other benefits of the cascoding, too, such as moving the input pole to a higher frequency, linearization because the input pairs now have no Vce swing, and boosted output impedance (and thus gain) of the input stage.

Matching of the input load current mirror is equally important to balancing the input stage, and a high beta there helps with current matching. I don't like to use Wilson or other fancy mirrors there, because I want that mirror to be as fast as possible. Mirror poles should be way out there in frequency, and I want no high frequency weirdness, either.

I also agree the tail current should come from a supply-invariant current source. Tail current that has any dependency on the rail voltage causes supply ripple to mix (multiply) with the input voltage, which gives you modulation distortion. Signal and AC mains components are present on the supply rails, so high PSRR is important.
 
Last edited:
#7
BesPav said:
1. IPS.
Input needs as small as possible noise, as low as possible Cob and as stable as possible current. So, low-noise and well-matched transistors and cascoded current source.
Taking as much as possible gain from IPS asks for current mirror loading of the IPS input.
Low drift and low-input related bandwidth limiting asks for smaller dissipation and low parasitic capacitance, so not only current source transistors must be cascoded, but an active input devices also.
Click to expand...

Thinking about this a little further, I want to add that low noise and low Cob are competing specs. A low noise transistor has a relatively large die, and some diffusion dimensions are tiny (base thickness) to get high gain, resulting in larger capacitances.

Cob of the SSM2212 is 23 pF, typical.
Cob of the KSC3503 is 2.6 pF, typical. An order of magnitude lower, and the input only sees 1.8 pF of that (Cre = 1.8 pF).

With cascoding, there is no miller multiplication of this Cre to the input. This pole is also easily canceled with an feedback capacitor. In short, I don't really care much about the Cob of the input pair. I care more about beta and Rbb.

Meanwhile, that very low Cob and Cre are what make the KSC3503 a good choice for VAS.
 
#12
stocktrader200 said:
Ksa992 and ksc1845 form the front end of my amp
Ksc3503 ksa1381 are vas and drivers for irfp240 & 9240
Outputs
Click to expand...

IRFP240 and 9240 are 3 decades old MOS devices. The industry has been working hard, meanwhile, to reduce the Rds(on) * Qgate product for buck converters and class-D motor control.

I quit using MOS long ago, but it strikes me that the gm / Qgate ratio must also have gone up massively in the meantime. I don't want to derail my own thread, but I think there would be better devices than the 240/9240 now. OTOH, the 400W @ 4ohms amp I made in 1990 using those devices in the output stage (the TO-3 versions, actually) sounded great!
 
#14
A lot of what we are seeing is a move to SMD packages, not necessarily the abandonment of good BJTs. For example, MPSA92 are obsolete (by ON Semi, at least), but the MMBTA92 (I think that's the part number) aren't.

For new diy builds, it's not really a bad thing. They aren't that hard to solder and I find that the small size and being close to the board means that thermally coupling input pairs isn't the issue it once was.

It sucks for servicing, and that's where I'm hoping that smaller manufacturers will keep producing limited quantities of the common ones (MPSA42 / 92, 5401 / 5551, etc).

As a note, the BC850 and BC860 are pretty marginal for higher-power amplifiers because of their voltage rating. I would be inclined to think that the FJV1845 / FJV992 would be a good option for those applications.

Cascoding the input stage works fine, it just adds complexity so I like to avoid it if I can.

As for MOSFETs, I'd be concerned about low Rds(on) devices when it comes to current sharing and SOA in the linear region. There's an IR application note somewhere on this. Current sharing with MOSFETs always seems to be a nightmare compared to BJTs, unless they're matched (but I hate matching transistors).
 
#15
I moved to SMD for small signal back in 1998 for my production stuff (before I moved into IC design, professionally). I appreciated how close together I could place all the devices, and the PCB seems to draw heat from them better than the through-hole. I used a lot of MMBTxxxx devices. Now I'm happy to see a lot of DMMT duals available for cheap. (I've been away from discrete audio design for a couple of decades. I used to do it as my job, and now I'm coming back to it as a hobby.)

I can see using d-pak instead of TO-126 for VAS and pre-drivers.
 
#16
One big advantage of SMD is that they are much less prone to mechanical failure. I'm sure we've all seen TO-220 and TO-126 packages where the leads sheared off from vibration. Obviously that's avoidable with good engineering, but the SMD parts make that a fair bit easier.

Of course, one does need to be careful about board flexing.
 
#18
H713 said:
One big advantage of SMD is that they are much less prone to mechanical failure. I'm sure we've all seen TO-220 and TO-126 packages where the leads sheared off from vibration. Obviously that's avoidable with good engineering, but the SMD parts make that a fair bit easier.
Of course, one does need to be careful about board flexing.
Click to expand...

Agreed, but good lord, do your designs end up in Baja 500 vehicles or SpaceX launches? I designed for car audio, and I never saw anything like that.
 
You must log in or register to reply here.
Top Bottom