Question about output stage used in open loop

[Banfi T.]

Hi there,

I have a problem with designing a power amplifier which does not use multistage feedback. While I was able to desing and build voltage amplifiers without multistage feedback I just couldn't answer the question of the last stage.

First, I tried to learn why the last stage is nested in the feedback loop most of the times and I came to the following conclusions:
1. Without the feedback the non linear natur of the (power) BJTs would lead to a distorted signal.
2. The output impedance of the amplifier (or should I say "resistance"?) is decreased by using negative feedback. This is - of course - essential to reach good damping.

According to my knowledge the first issue can be handled.

The 2. point is where my knowledge seems to give up. I read many times that connecting several power BJTs in parallel should solve this problem, but I do not see how this practise is capable to deliver the goods.

I understood that the output resistance of the BJT should be calculated by divining the thermal voltage (generally 26mV is used by such calculations) with the bias current. (Of course I am talking here about an emitter follower configuration.) If the latter is about 200mA then the result is 130 mOhms. If we connect an other emitter follower with 200mA bias current then that will have 130 mOhms output resistance in itself too and becasue of Ohm's law the two together would have half of the output resistance, which is 65 mOhms. As I see it is because of the higher overall bias voltage and if we apply 400mA to a single emitter follower we will have the same result.

What it isn't clear for me:
- Can the equation I used above applied to calculate both AC output impedance and DC resistance?
- Which of the two parameters (AC and DC output impedance/resistance) is really important?
- How much bias current shall be used to drive real life speakers? (In other words: how much current is enough?) The queastion is not about the power handling capacity now. How much bias current is needed to achive low enough output resistance.
- I read in some articles that by applying parallel output transistors or class A (wich means using higher bias current as I see) would lead to decreased distorsion. I can not see how this is possible. (I am talking here about simple emitter followers, no CFP.)
- Is there a recommendation regarding how much the damping factor should be?

Thanks for any help.

bt

 

[JonMarsh]

If there's some way you can get your hands on a copy of Doublas Self's book on audio power amplifier design, you can get a wealth of information on conventional biopolar solid state design. He also does some analysis of output stage configurations. The Emitter follower triple T circuit can be quite linear open loop, especially when using the extended beta range transistors like those available from Toshiba and On Semiconductor. Paralleling transistor increases the SOA and ouptut current capability, and lowers the net output impedance.

Other approaches like the CFP output stage measure well under some conditions, but may not be as robust with reactive loads, and usually have issues about the turn-off speed for the power devices- due to low turn-off drive current available. For this reason driving them at high frequencies (20 kHz and above) at high power levels can be risky...

I did a high power CFP stage a few years back with Magnetek MSOFETs that worked quite well, but the driver was a "full size" MOSFET, and the turn-off resistance for the driven banks of MOSFETs was quite low, so it was more the exception that goes against the rule due to design innovation, rather than the typical bipolar implementation.

And if you want to do an amplifier without feedback loops, technically speaking, using a CFP pair is cheating, as there is a two stage loop contained therein.

The emitter follower triple isn't glamorous, but it gets the job done... with good design it's possible to get distortion numbers in the output stage under 0.005% open loop... isn't that low enough?

~Jon

 

[fab]

... with good design it's possible to get distortion numbers in the output stage under 0.005% open loop... isn't that low enough?

~Jon

Low enough yes, but how do you achieve it without feeback?

 

[Banfi T.]

JonMarsh,

I am aware of the drawbacks of the CFP and I don't plan to use that configuration

I haven't read the book from Self, maybe it's time to get one. Although I have to mention that most books do not speak about open loop solid state designs, but maybe this one is an exception...

 

[Jocko Homo]

You got down to 0.005%!

Only got mine down to 0.01%. But then I ran out of room on the heat sink.

Yeah......if I had room for a few more.

I used to build amps with CFP outputs. The open-loop output versions sounded better, and did not have problems oscillating in the 2-5 MHz range.

Jocko

 

[jan.didden]

[snip]The emitter follower triple isn't glamorous, but it gets the job done... with good design it's possible to get distortion numbers in the output stage under 0.005% open loop... isn't that low enough?

~Jon

Hello Jon,


0.005% open loop? Wow, that's impressive. Care to elaborate on this?

Jan Didden

 

[peranders]

Jan, a diamond buffer (open-loop) can also produce low distortion. When I measured my buffer with my soundcard it was less than 0.008% (the soundcard itself)

 

[jan.didden]

Yeah, sure, but this is a power output stage, I guess with load! That's a different kettle of fish.

Jan Didden

 

[traderbam]

Good questions.
You need an output Z of 0.2ohms for good performance.
Assuming the voltage driver stage has very low Z, the output Z will be determined by the transconductance of the output device(s). Forget the maths. Look at the curves in the datasheet for the power transistors. From these you can draw the output Z vs Ic and see how much Ic is needed to achieve 0.2ohms or less and you can then see if there is an optimum Ic for minimizing Gm vs Ic.

You won't be able to estimate the imaginary part of the Z without great difficulty. So don't try. Just use low capacitance devices. Don't think of dc and ac separately - both are important.

I agree with you questions about paralleling devices. It doesn't really help in any way except to allow devices to be operated on a more linear part of the Gm curve or to distribute power. I do not recommend paralleling devices unless you have compelling reasons to do so.

 

[ashok]

The caps.

Peranders,

The caps are wearing mini skirts !
Did they come that way or did you cut away the rest of the plastic sleeve? If so ,why ?

Ashok.

 

[JonMarsh]

I agree with you questions about paralleling devices. It doesn't really help in any way except to allow devices to be operated on a more linear part of the Gm curve or to distribute power. I do not recommend paralleling devices unless you have compelling reasons to do so.

Well, let's just consider what Doug Self found- paralleling sets of extended beta devices brought the open loop distortion down as well as the open loop impedance.

Now, in the case of an amplifier dedicated to only moderate or high impedance loads, and with low output power (low rail voltages) and relatively high bias current, there would be no advantage to paralleling output devices, and then I would agree with you.

By your own calculations, achieving a low open loop output impedance will require a combination of operating bias and paralled devices, so your last comment seems a little at odds with the rest. I'd say lowering output distortion and output impedance, as well as increasing SOA are fairly "compelling reasons", no?

With regards to achiving low open loop distortion (the range I'm quoting being the output stage ONLY, connected directly to my HP8903- this is nothing special, read Self's book) a few factors seem to be necessary. Besides the use of a properly biased EF triple, the right output transistors are mandatory(the extended beta Toshiba or On Semi parts). Even Self was surprised at the level of performance these parts brought to the party, compared with the older more conventional 4-5 MHz Ft transistors with gain droop above 1-2A.

Also, as Edward Cherry discussed in his AES paper in the late 70's or early 80's (sorry, can't remember the title off the top of my head, it's at home somewhere) layout and power supply issues are also quite important. Layout issues resulted in a difference of as much as 0.02% IM versus as little as 0.002% IM in an amp I worked on in the late 70's. Self discusses this, too, though not with the specific recommendations Chrry had, but still his current book may be useful for more than a few topics.

With regards to an NFB type amplifier, with no interstage feedback, getting the output stage clean is only half the battle. Perhaps the easier half.

Best regards,

Jon

 

[Banfi T.]

Traderbam,

thanks for the comments.

Does it means, that one is able to build a working open loop output stage just by using simple emitter followers with a high enough bias?

Then why everybody uses feedback topologies? Because cooling is expensive? I mean there are many-many really expensive amps with lots of negative multistage feedback...

Somehow I didn't even thought about the possiblity that such simple configuration can work. A bridged, push-pull, heavliy biased amplifier was the simpliest I could think of, but then you have to make a phase splitter which is a simple task if you don't go into the details too much....

Shame on me, but I have to confess that I want to build an amplifier without multistage feedback because:
1, I really do not understand how feedback actually can work. It works, I know, but I simply can not understand. (How can a signal modifiy itself when it has already left the point where the modification should happen?) I know this is my limitation, but I don't want to have an amplifier built by mself if I don't understand it.
2, I don't have appropriate measuring equiipment around to check high frequency oscillations if they appear for whatever reasons...

 

[peranders]

Re: The caps.

Peranders,

The caps are wearing mini skirts !
Did they come that way or did you cut away the rest of the plastic sleeve? If so ,why ?

Ashok.

They look like this. I'm not the "peel-cover-off-guy"

 

[traderbam]

Does it means, that one is able to build a working open loop output stage just by using simple emitter followers with a high enough bias?

Sure. I'm not saying it will be distortion-free...but it is certainly a good approach if you choose the transistors carefully.

Then why everybody uses feedback topologies? Because cooling is expensive? I mean there are many-many really expensive amps with lots of negative multistage feedback...

Well, cooling and the size of he psu are a high price to pay for avoiding global feedback. Just look at Pass Labs gear. IMO using feedback is the best approach if you know what you are doing but it is very useful experience to make the best sounding open-loop you can.

The two reasons you have for avoiding global feedback are quite fair. I really support your desire to understand the fundamentals before you make use of something. A good lesson for all.

 

[traderbam]

Well, let's just consider what Doug Self found- paralleling sets of extended beta devices brought the open loop distortion down as well as the open loop impedance.

I'm afraid I have little time for D. Self. I have not ready his book. I have read his website. He appears to design using an oscilloscope and a slide rule rather than his ears.

By your own calculations, achieving a low open loop output impedance will require a combination of operating bias and paralled devices, so your last comment seems a little at odds with the rest. I'd say lowering output distortion and output impedance, as well as increasing SOA are fairly "compelling reasons", no?

I'm sure a single BJT can be found that will have a Gm of 5 or higher at high enough Ic. If paralleling makes the output sound better then I agree. But what do you mean by output distortion?

Besides the use of a properly biased EF triple, the right output transistors are mandatory(the extended beta Toshiba or On Semi parts). Even Self was surprised at the level of performance these parts brought to the party, compared with the older more conventional 4-5 MHz Ft transistors with gain droop above 1-2A.

I beg to differ. A push-pull triple darlington is a disaster for sound quality. Very, very hard to make it sound good.

...layout and power supply issues are also quite important. Layout issues resulted in a difference of as much as 0.02% IM versus as little as 0.002% IM in an amp I worked on in the late 70's.

Agreed. Especially psu wiring and grounding.

 

[azira]

2. The output impedance of the amplifier (or should I say "resistance"?) is decreased by using negative feedback. This is - of course - essential to reach good damping.

The 2. point is where my knowledge seems to give up. I read many times that connecting several power BJTs in parallel should solve this problem, but I do not see how this practise is capable to deliver the goods.

What it isn't clear for me:
- Can the equation I used above applied to calculate both AC output impedance and DC resistance?
- Which of the two parameters (AC and DC output impedance/resistance) is really important?

Ok, I cut some of what you said to save space 'cause I'm going to repeat it anyway. First in BJTs Gm = Ic/Vt. Re (the output resistance of an EF) is 1/Gm. So as you have noticed, if you increase Ic, your bias current, your output resistance goes down. Firstly this is only for "small signal" or AC situations. In DC, these parameters hold no weight, partly because VCC no longer looks the same as GND.

Now, you're right about doubling the output current in a single device would seeminly lead to the same as having 2 devices in parallel. Except since you have two devices (the PNP/NPN) facing eachother, you have to put in some resistance to protect the devices from trying to drive eachother and eat up minor DC offsets. These resistances actually dominate the output resistance. So if you put another device in parallel, the emitter resistors will also be in parallel.

Shame on me, but I have to confess that I want to build an amplifier without multistage feedback because:
1, I really do not understand how feedback actually can work. It works, I know, but I simply can not understand. (How can a signal modifiy itself when it has already left the point where the modification should happen?) I know this is my limitation, but I don't want to have an amplifier built by mself if I don't understand it.

Feedback sometimes feels like doing a convolution But you're on the right track. Just add the time component into it. Assuming your time delay in your feedback network isn't very significant, it's just 1 resistor after all, your feedback represents what is exactly on the output at a particular instance. Your signal input, instead, is now the direction you'd like the output to go with respect to what is currently on the output. The bigger the delay from input to output with respect to the speed the input changes, the less "accurate" the feedback is of the actual output (it ends up representing an "old" value). Start by pretending that the signal changes slowly compared to the delay and you'll see that the feedback signal seems more "accurate".

Or... just do the math and see the terms cancel
--
Danny

 

[Banfi T.]

azira

"Assuming your time delay in your feedback network isn't very significant, it's just 1 resistor after all, your feedback represents what is exactly on the output at a particular instance."

The feedback loop - as I see - does contain the whole circuitry (first the signal has to travel through the whole circuit) and the feedback network (which is sometimes just a resistor). Therefore sometimes I think that having a fast amplifier - or with an other word a high bandwith amplifier - is an advantage becasue the faster the circuit the faster the feedback network is. Of course it is only true if the amplifier has a high open loop bandwith...


"Your signal input, instead, is now the direction you'd like the output to go with respect to what is currently on the output. "

If this is process is something like adding the two signal together, then the output will also have its share from the steering...or not?
I know how most of the students book do the maths about feedback but they don't calculate with the delay. I never saw any calculation which included it as a matter of fact.

And thanks for the info about paralleling and the emitter resistors. That makes sense.

 

[jan.didden]

[snip]I really do not understand how feedback actually can work. It works, I know, but I simply can not understand. (How can a signal modifiy itself when it has already left the point where the modification should happen?) I know this is my limitation, but I don't want to have an amplifier built by mself if I don't understand it.[snip].

Fair enough. But maybe it may help to put the feedback thing in perspective if you look at it in the right, err, perspective. Look at driving a car. You could say, I see that I go off the right track, and then I try to correct the steering to stay right. How can that work? After all, you can only decide that you need to correct it if you are already in the wrong direction! The clue is that the loop of detection-decision-correction is much faster than the occurence of the error.

It is similar in amps (you may have noted that the above example is just another case of neg feedback anyway). As long as the feedback loop, and the amp itself, which is part of the loop as noted above, are fast enough, it works. Your gut feeling is right: when the loop slows down (or, when the frequency rises, same thing) it breaks down. In effect, at high frequencies, the feedback comes so late back to the input that the signal is already changed. When the delay is so large (or the signal change so fast, same effect) you get ringing: the correction overshoots because it is trying to correct a signal that already has moved on. If you get to a delay of half the signal period (180 degr shift) you get an oscillator: The feedback tries to raise the signal to correct for a too low output signal, and at the same time the signal is also rising.

The key thing is to understand that feedback works *good enough*, not perfect, and only within specific, strict circumstances, and breaks down outside those. Part of the art of amp design is in judging where and how to make compromises.

Does this make sense to you?

Jan Didden

 

[Banfi T.]

Jan,

yes it makes sense. I just hard for me to *really* understand the connection between delay time and frequency. But anyway, thanks.

 

[jan.didden]

Jan,

yes it makes sense. I just hard for me to *really* understand the connection between delay time and frequency. But anyway, thanks.

Oh, that's easy. If the delay through the loop gets in the neighbourhood of the period of the signal, the loop is no longer fast with respect to chnages of the signal and you get problems like ringing and oscillations.

Say your loop has a delay of 1uS, meaning that it takes a signal 1uS to travel from input, to output, through the feedback circuitry, back to the input. Now, if the signal changes appreciably within that 1uS, you're in trouble. When does a signal chnage appreciably in 1uSec? When it has a frequency of, say 200kHz (which is a period time of 5uS), it "moves on" 1/5 of a cycle (assume sine wave just for the sake of discussion). But if the freq is 500kHz, it moves on 1/2 a cycle. That means that if you have a rising signal to begin with, by the time the correction arrives back at the input, the signal has moved on and is now also rising. The feedback is no longer negative (working against the origianl signal) but positive, reinforcing the original signal.

And oscillators start easy: thers's always some noise that get amplified and grows and grows, until it reaches the limits of the amp or power supply and then remains there. So how do we prevent every amp becoming an oscillator? Just make sure that for those frequencies, that may be too fast for the loop, the amp gain is less than 1, because then the oscillations can not be maintained, they die out. (Actually, they never start).

You get the point: we want to make the amp fast, with wide bandwidth, but as soon as we add feedback we must roll of the gain for higher frequencies, which means the feedback is less effective, which meams the distortion rises with frequency. Which is of course what we normally see in audio amps.

You know, feedback and control systems are really, really very interesting....

Jan Didden