2

In a class AB follower push pull amplifier output stage* operated open loop and driven from a low impedance source, these are, from top to bottom, the emitter currents of both transistors, their individual and total base currents, and the output voltage and current of the amplifier.

enter image description here

The last graph is not to scale.

These transistors have more than 10V Vce, they are not saturated.

When one transistor turns off, it behaves a bit like diode recovery: it keeps conducting a bit too long, then abruptly turns off. Its base current behaves in a similar way.

Two questions:

  1. I would like to minimize this, so what is the relevant datasheet parameter to look for? There is "Storage time" but I believe it is about coming out of saturation, which is not the case here.

  2. Harder physics question: what is the relationship between dIe/dt and the peak amplitude or total energy of the turn-off spike?

(*) Schematic of the class AB push pull amplifier output stage:

enter image description here

Transistors are 2SC5200 and 2SA1943. Current is about 2.4A peak, and frequency is 48kHz. For reference, this is the same setup, with a pair of IRFP240/9240 MOSFETs instead.

enter image description here

bobflux
  • 70,433
  • 3
  • 83
  • 203
  • 1
    I would imagine that, just like in FREDs, a bit of platinum doping would improve turn-off speed. I wonder if anyone's ever made devices like that? – Hearth Jul 24 '21 at 21:59
  • That depends, do you have access to a silicon foundry? – Hearth Jul 24 '21 at 22:01
  • 2
    Heck, where's your schematic of what you are showing graphs for. What biasing arrangement did you use? How much bias? What supply rails? What input signal levels blah blah. Schematic would provide answers and stop me asking dumb questions. – Andy aka Jul 24 '21 at 22:01
  • @Andyaka if you say you need schematics for a push pull emitter follower, I will add one, but really? – bobflux Jul 24 '21 at 22:02
  • 1
    @bobflux What do you mean "really" as far as the schematic question? There are many topologies for a class AB amplifier. You could use a VBE multiplier or diodes for biasing, you could use a Darlington output stage or a single BJT, you might have any number of things that could affect the output waveforms. So I would say yes, really. – John D Jul 24 '21 at 22:12
  • @bobflux How much is Re? How is the base biased? How much is the DC bias current through Re? The sweet spot of how the output behaves depends heavily on biasing. – Justme Jul 24 '21 at 22:13
  • 1
    If you want to shorten the turn-off time, you can do that with a reversed base current. Drive each power push-pull stage transistor with its own base driver push-pull stage instead of bias resistors. – Janka Jul 24 '21 at 22:23
  • @Justme Idle current in this class AB output stage is 50mA and base resistor is 20R. I did not put it in the question because bias current determines at what output current the transistor that should turn off will turn off, but it does not affect the behavior of the transistor when it turns off, which is what the question is about. – bobflux Jul 24 '21 at 22:24
  • 1
    @Janka Yes... but... if I wanted fast switching I would use MOSFETs. It's a curiosity question, I don't know what is happening inside a bipolar transistor when it turns off and I want to know. – bobflux Jul 24 '21 at 22:31
  • 1
    @bobflux Is the load on the output really a constant-current sink? How much load there is? – Justme Jul 24 '21 at 22:39
  • @Justme I fixed the image, sorry about that – bobflux Jul 24 '21 at 22:43
  • So, what does the simulation model for the transistor contains? Which transistor is it, and does it model any parasitic capacitances and/or inductances? – Justme Jul 24 '21 at 23:07
  • @bobflux The little bit of ringing that you've helped emphasize in the notched output -- doesn't it suggest some inductance that is ringing with some capacitance, to you? If this were the BJT's emitter-base junction capacitance (order of 10-12 pF or so) then I'd expect to go look for something on the order of 8-10 mH, just looking by eye. Since your emitters are facing the load, consistent with that idea, that makes me wonder if you really have a resistor there or an inductor load. Could you correct me on that point? – jonk Jul 25 '21 at 02:50
  • @Justme it isn't a simulation, the traces come from a scope. – bobflux Jul 25 '21 at 08:13
  • @jonk I don't think it is ringing, the spikes are well damped, the network analyzer doesn't show anything nasty, and spice simulation with just a resistive load shows the same result as this little experiment. If I use MOSFETs instead of bipolars (I updated the graph) it is much smoother. Still I will check what the network analyzer says about the phase of the load impedance, thanks for pointing that out. – bobflux Jul 25 '21 at 08:33
  • I believe it is indeed the same principle by which a pn diode reverse-recovers. the BE junction is such a diode. – tobalt Jul 25 '21 at 09:42

3 Answers3

3

This is typical BJT behaviour, and there are a couple of things going on.

Firstly, the turn off time thing, yea, it is carrier decay time, cure is to take Vbe slightly negative as the device turns off to help sweep the charge out, commonly this is done by placing a resistor between the two device bases to ensure some standing current in the driver transistors and to allow the thing to pull down. A cap between the base connections (outside of the base stoppers) is also sometimes a thing for much the same reasoning.

Secondly, you appear to be slightly under biased, as you change the bias setting, you will go from a dip in gm at crossover, to a situation where gm actually rises at crossover, the sweet spot is generally when you have a voltage developed across the emitter resistors equal to Vt (~26mV) but this can cause annoyingly large amounts of heat if low value emitter resistors are in use, so many amps are run rather under biased.

I would note that there is some EXCEPTIONAL measurement work on output stage topologies here https://www.diyaudio.com/forums/solid-state/374367-power-amp-output-stage-measurements-shootout.html well worth the time, that whole thread is gold.

Dan Mills
  • 17,266
  • 1
  • 20
  • 38
  • Hey thanks, that's my thread :D I will post those measurement there shortly. And... there is already a 100nF cap. Yeah I know it's typical BJT behavior, but I'd like to know what datasheet parameters to check to find a BJT that does it less than the big chunky TO-247 ones. Probably in the "storage time" zone on the datasheet, but they don't reveal these secret specs for transistors marketed as "audio" so... bummer! lol Probably if they put these parameters in the datasheet there would be less buyers cause it doesn't look very good... – bobflux Jul 25 '21 at 11:05
  • A good rule of thumb is that components are marked "Audio" because they don't measure well enough for anything that actually matters! Being selective about what goes on the datasheet is an annoying habit some of the vendors have, 'switching' mosfets with nothing about SOA or how Vth changes with die temperature for a recent example on my bench. – Dan Mills Jul 25 '21 at 11:18
  • yeah, you're right... btw I measured the Vgs tempco of MOSFETs, it depends on current (gets lower in absolute value as current increases) and IRFP240-9240 datasheet gives completely wrong numbers... – bobflux Jul 25 '21 at 11:40
1

First:

Most relevant characteristic is bandwidth gain product, or alternatively any of the time properties like turn off or turn on time. Faster - better.

https://en.wikipedia.org/wiki/Gain%E2%80%93bandwidth_product

Second:

If you mean 'how can I simulate it' then capacitances and industances nearby the base, emitter and collector, allow to do somewhat accurate simulation of the real behaivor.

If you mean 'what actually happens on physical level that causes these effects?' then it is likely related to population of charges in the junction. Amount of charges located in the semiconductor, that has some inertia if you try to change it, as it takes time for them to enter or leave that area.

If you want to know what physically to change to make it work faster - usually it means making the semiconductor smaller. Smaller transistors are usually faster. Big transistors could be made of many smaller ones to get power and speed.

Surprised Seagull
  • 406
  • 2
  • 6
  • Yeah I think you're right. MOSFETs don't do this at all, so by "differential diagnosis" it must be something BJT specific like charge storage... – bobflux Jul 25 '21 at 08:38
1

GBW is certainly important in high current unity gain CE’s but it is the linearity during crossover that is highlighted in thus question with some storage latency effects.

The both examples shows a significantly greater base current than expected during crossover. I see that as a result of Hfe changes from Imin to I max for base current while Ic does not follow the same curve when conducting and Cbe capacitance has greater latency effects as Rbe rises rapidly during crossover. The minimum current gain at Ic max Hfe =2.3A/30mA= 77 in 1st and 2.4A/0.6mA =400 in the 2nd perhaps from a Darlington arrangement. I see the Class code sorting of hFE with not only higher hFE more dynamic range of flat hFE as the better choice.

These dynamic variations of Zout/Zin are BJT specific in this circuit. The 1st has far greater spike harmonics while the 2nd has more 3rd harmonic content with triangular input current and dominant 3rd voltage harmonic on the output. Unlike Class A which is only 2nd harmonic, this 3rd harmonic I believe will be Iq dependent and Hfe vs Ic flatness dependent. While capacitance will increase with power ratings, hFE effects on Ic during crossover going from Amps to milliamps.

The more linear the power transistor (flat hFE) , the better. BTW power FET’s have an advantage here as a voltage follower.

Tony Stewart EE75
  • 1
  • 3
  • 54
  • 182
  • I agree about MOSFETs. I discovered the emitter resistors had about 100nH of inductance. Replacing them with much lower inductance SMDs resulted in cleaner traces, but the base current spikes typical of BJTs are still there of course. hFe changes during crossover as it switches from top transistor hFe to bottom transistor hFe are a source of distortion indeed. But the base current spikes strongly depends on frequency so it's not just hFe (your hFe calculation looks weird btw). – bobflux Jul 25 '21 at 20:03
  • And yes good catch the PNP in this output stage has quite a bit of hFe fall off as current increases, while the NPN does not. – bobflux Jul 25 '21 at 20:04
  • Weird but hFE is >> 1k max from your 2nd plots – Tony Stewart EE75 Jul 25 '21 at 20:57
  • Oh sure, the second plot is from a pair of MOSFETs so it has a lot more current gain! – bobflux Jul 25 '21 at 22:43
  • It looks Ok for open loop. What did you expect for THD? – Tony Stewart EE75 Jul 25 '21 at 22:50
  • I'm going for lots of zeros. Right now I'm testing various devices and output stage topologies, going for lowest high harmonics. I think the MOSFETs will win due to smoother gm in the crossover region. – bobflux Jul 25 '21 at 23:06