Bob Cordell's Power amplifier book

[T117]

I think 2 transistor is not enough for good current source. Everyone can understand the simple idea, that 4 of the transistor is two times better than 2 of the transistor. The way it is.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

The output impedance of the current source 4T is better at 40 dB than 2T.

 

[T117]

No. It's too complicated current source. You can do just a little simpler and better.

An externally hosted image should be here but it was not working when we last tested it.

I throw one resistor.

An externally hosted image should be here but it was not working when we last tested it.

And the output impedance is increased even more!
I think to throw away a few more details and I will inform you about the results.

 

[Ranchu32]

I believe it might help a whole lot to point out that a resistor in series with the base of the control transistor, works in conjunction with the control transistor's Collector-Base capacitance, giving classical "pole splitting". The final location of the LF pole depends on the external resistor, the Ccb capacitance, the stage gain (Ceffective = C*(1+gain) i.e. the classical Miller Effect), and other factors. You could suggest, as I occasionally have done, that people select Rbase such that (Rbase * Ccb) > (20 * Tauf) where of course Tauf = 1/(2*pi*fT). It's overkill but it's also very safe.

Mark, is the factor of 20 in your formula something you arrived at empirically or is there more to it than that? By my calcs, Rb ends up in the order of 10k with the commonly used transistors or 10X the biggest value seen in most published schematics.

 

[kgrlee]

Walt Jung's schematic is attached;
....
When I plot open loop gain and phase of the circuit in LTSPICE AC analysis, I find that item (a) above is crucial. Small changes in load resistance produce large changes in stability margin.

Thanks for the clarification Mark & mlloyd1.

I was horrified to read this as I've used this a lot in commercial designs. But closer scrutiny shows in all cases, I've slugged the base of the output transistor with a 1-10u electrolytic.

Going through Jurassic notes reminds me the reason for doing this was oscillation wriggles on overload.

Of course this means the Output Z is then no better than the 1T versions but it looks like this is no bad thing.

I think 2 transistor is not enough for good current source.

Just a reminder that the aim of the exercise is better Power Amps rather than better current sources.

In most cases, the PSR & other good stuff provided by these current sources is swamped by other factors.
________________

Mark Johnson, could you post the *.ASC file that you used for the results in #7618. I hope you show where you place your stability probe.

I'm having difficulty simming oscillation or interpreting my poor LTspice efforts in the files you and T117 posted

 

[Mark Johnson]

I told Excel to build a table showing the calculated values from

Rseries = (20 * Tauf) / Ccb​

for some commonly used transistors. ("Cob" is datasheet speak for Ccb). The table includes several resistors greater than 10K, which, as member Ranchu32 points out, is greater than the 2.7Kohm value mentioned by Douglas Self and quoted in post #7618 here in this DIYA thread.

Why not wait and see what Bob Cordell has to say? Perhaps he has a different rule-of-thumb in mind. I think it is possible that Bob may suggest

• Connect a resistor in series with the control transistor's base, chosen so that (base current) x (resistor value) equals 2-5 millivolts

Recall that (base current) = (collector current) / Beta

Bob may suggest that you use the "min" value of Beta from the datasheet.

 

[kgrlee]

I think it is possible that Bob may suggest

• Connect a resistor in series with the control transistor's base, chosen so that (base current) x (resistor value) equals 2-5 millivolts

Recall that (base current) = (collector current) / Beta

Better would be to put a smaller resistor in the emitter of the control device.

In these cases, it is always the gm of the 'input' which needs reducing. It has the 'same' effect as a base resistor but isn't governed by hfe. (Not quite the 'same' but in fact 'better' cos not affected by Cob bla bla )

And much less noise penalty too.

These methods all reduce the zillion M output Z but what seems to be coming out is that you don't really need this in the usual amp.

The emitter resistor on the input is certainly the way to stabilize a CFP and much more reliable than a base resistor.

 

[Bob Cordell]

Excuse my somewhat blunt question perhaps bottoming in lack of knowledge, but among other CCS circuits, what are the advantages of the 2T CCS recently discussed in this thread, and in particular in audio amplifiers, and even preamplifiers, where voltage supply headroom is plenty?
Has it become used in audio circuits where voltage supply headroom is of no problem mostly out of "monkey see, monkey do" reasons, or are there any historical reasons, did it appear first in ordinary circuits built on discrete devices or did it appear first in IC's?
Although, I can see its usefulness in IC's where among various design limits, the supply voltage can be even very limited.

The feedback current source can be attractive for use as the tail current source for an LTP input stage. This, because of its very high output impedance. However, this application is one wherein there is typically a lot of headroom to spare. I always put at least a 1k resistor in series with the current source in this application to isolate the capacitance of the current source from the tail at high frequencies. This amount of resistance at the output of the current source greatly eases stability concerns. If larger, this resistance can also be increased to drop more voltage to reduce dissipation in the current source transistor. This can be especially helpful when the current source is connected to the negative rail (as opposed to a separate -15V source).

A tail current source with a Vbe, LED or Zener reference will tend to have a lower output impedance. The larger the reference voltage used in such a current source, the higher the output impedance will tend to be.

One can cascode such a current source to obtain a much higher output impedance, at the expense of a few more parts (possibly as little as a diode and a transistor). The added required headroom in this application is no problem.

Cheers,
Bob

 

[Bob Cordell]

I think 2 transistor is not enough for good current source. Everyone can understand the simple idea, that 4 of the transistor is two times better than 2 of the transistor. The way it is.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

The output impedance of the current source 4T is better at 40 dB than 2T.

Note that both of these current sources exhibit the peak in output conductance in the region of 70-100 MHz that is characteristic of the reduced phase margin of these feedback current sources.

In a single-ended VAS application where the current source is used as a VAS load, the desirable output impedance of the current source load is merely that it be at least significantly larger than the output impedance of the VAS itself.

Cheers,
Bob

 

[Bob Cordell]

Bob, I suspect it is possible to write up a few recommendations that will help people avoid trouble with the 2T current source, and I was hoping that other readers of this thread would contribute a few of them for your consideration.

I believe it might help a whole lot to point out that a resistor in series with the base of the control transistor, works in conjunction with the control transistor's Collector-Base capacitance, giving classical "pole splitting". The final location of the LF pole depends on the external resistor, the Ccb capacitance, the stage gain (Ceffective = C*(1+gain) i.e. the classical Miller Effect), and other factors. You could suggest, as I occasionally have done, that people select Rbase such that (Rbase * Ccb) > (20 * Tauf) where of course Tauf = 1/(2*pi*fT). It's overkill but it's also very safe.

Hi Mark,

This is a constructive suggestion that is along the lines of what I have in mind for the second edition. The general theme is that any such arrangement involving feedback may need some form of compensation (but adequate "inherent" compensation may exist in many cases). The challenge is to quantify.

The insertion of a very inexpensive 1k or so resistor in the base circuit is probably an easy solution, but unintended consequences may have to be considered (noise for example?). As touched on earlier, running significant current into the reference transistor can exacerbate the stability issue by increasing the transconductance of the CE stage that the reference transistor forms. However, running too little current into the reference transistor can cause the reference transistor to lose control of the current source if there is a high slew rate at the collector of the current source that creates significant current through Ccb of the current source transistor.

When the current source looks into a very low impedance load, in the limit we have a feedback arrangement with a CE stage with fairly high gain (the reference transistor) followed by effectively an emitter follower (the current source transistor) to feed the signal back. When there is moderate load impedance, there is some Miller effect via Ccb of the current source transistor, and this may help provide adequate phase margin in many cases.

We need to bear in mind that there are other arrangements involving 2 transistors that can have feedback that looks like this at high frequencies. One such example is the use of a Miller compensation capacitor around a 2T VAS.

Cheers,
Bob

 

[Ranchu32]

Better would be to put a smaller resistor in the emitter of the control device.

How much would you degenerate it - about 10:1? Is this the sort of thing you have in mind? NB: 35V rail.

 

[kgrlee]

How much would you degenerate it - about 10:1? Is this the sort of thing you have in mind? NB: 35V rail.

Yes. But the resistor can be a LOT smaller for 'similar' effect. The ratio is hfe.

eg if you find a 3k resistor in Q109 base gives you the desired stability and Q109 has hfe 300, .. an emitter resistor of R113 = 3k / 300 = 10R should do the trick.

But this is just a suggestion on my part. I haven't tried it.

I'm trying to get LTspice running on a 'new' computer to sim all this and its hard work for dis beach bum.

I know it works very well with CFPs in 'real life'.

 

[Ranchu32]

Thanks kgrlee, good to know.

Does anyone have a view on 'how much' output impedance is enough for the input stage and any more doesn't really improve matters? The only real advantage of the 2T CCS in this position that I can see is that it has higher output impedance than the other common alternatives. An LED makes a good, stable, low noise Vref. Maybe the BJT+LED CCS is good enough?

 

[T117]

Note that both of these current sources exhibit the peak in output conductance in the region of 70-100 MHz that is characteristic of the reduced phase margin of these feedback current sources.

These are signs of instability. Mark Johnson spoke about this. This is easy to get rid of. Recent examples of the power sources do not suffer from this drawback.

 

[T117]

Just a reminder that the aim of the exercise is better Power Amps rather than better current sources.

In most cases, the PSR & other good stuff provided by these current sources is swamped by other factors.

I assemble the amplifiers with harmonic distortion less than 1 ppm for a frequency of 20 kHz. For example, some of my amps have a distortion level 0.3 ppm, and some - 0.15 ppm (0.000,15%). Look http://www.diyaudio.com/forums/solid-state/290970-several-schemes.html, http://www.diyaudio.com/forums/headphone-systems/292834-simple-head-amp-high-linearity.html, http://www.diyaudio.com/forums/headphone-systems/292669-ultra-hi-res-headamp.html.

In such amplifiers, many factors can degrade the result. One of these factors - low impedance current source. I'm not talking about the latest CCS cascade VAS, where the high output impedance CCS shunted input and output impedances of three transistors. I mean a good CCS for the input stage (LTP).

A good current source should eliminate the distortion because of the common-mode voltage (peak-to-peak over 4 volts) at the inputs of the LTP non-inverting amplifier. Output current LTP with a good current source is much less dependent on the input common-mode voltage of the amplifier.

In addition, it is important to have a minimum output capacitance CCS. If this capacity is too large, current charge capacity will be more on the positive half-wave of the signal (for NPN transistors LTP) than the discharge currents of this capacity. It can also lead to distortions.

So I need a good current source in the input stage (LTP).

 

[kgrlee]

I assemble the amplifiers with harmonic distortion less than 1 ppm for a frequency of 20 kHz. For example, some of my amps have a distortion level 0.3 ppm, and some - 0.15 ppm (0.000,15%).

That's very commendable.

How do you measure these THD levels? It's not clear to my small brain what the power levels and load are for these amazing measurements and which amps.

Look http://www.diyaudio.com/forums/solid-state/290970-several-schemes.html, http://www.diyaudio.com/forums/headphone-systems/292834-simple-head-amp-high-linearity.html, http://www.diyaudio.com/forums/headphone-systems/292669-ultra-hi-res-headamp.html.

In such amplifiers, many factors can degrade the result. One of these factors - low impedance current source. I'm not talking about the latest CCS cascade VAS, where the high output impedance CCS shunted input and output impedances of three transistors. I mean a good CCS for the input stage (LTP).

A good current source should eliminate the distortion because of the common-mode voltage (peak-to-peak over 4 volts) at the inputs of the LTP non-inverting amplifier. Output current LTP with a good current source is much less dependent on the input common-mode voltage of the amplifier.

So I need a good current source in the input stage (LTP).

Of course an easy way to reduce the input common-mode THD is simply to have more gain. It seems a shame to have such a complex circuit with a gain of 2x.

I'm not convinced you need such complexity to achieve eg 1ppm THD20k

But certainly your circuits show some useful tricks which I'm sure will be useful in my much less complex efforts

BTW, what SPICE programme are you using? Are the ASC files compatible with LTspice?

 

[T117]

Of course an easy way to reduce the input common-mode THD is simply to have more gain. It seems a shame to have such a complex circuit with a gain of 2x.

If you mean head-amp, such amps don't need high gain. This amplifier gives you more volume even with the same gain. After all, you don't plan to become deaf.

I think a big advantage of my circuit design greater depth NFB, more than 100 dB, and thus the stability of the amplifier with a small gain. Despite the apparent complexity, tuning of the amplifier is limited by the setting of the current idle.

I'm not convinced you need such complexity to achieve eg 1ppm THD20k

No, I'm not assembled amplifiers with distortion levels of 1 ppm. It's a shame. I started with amps that have distortion 0.000,3% at 150 Wt / 8 Ohms at 20 kHz http://www.diyaudio.com/forums/attachments/solid-state/548293d1462871663-several-schemes-2011.gif http://www.diyaudio.com/forums/solid-state/290970-several-schemes.html.
So I have assembled amps with less distortion.

But certainly your circuits show some useful tricks which I'm sure will be useful in my much less complex efforts

I'll be very happy if they will be useful to you or someone else. I was under the impression that my circuitry is not needed at all.

BTW, what SPICE programme are you using? Are the ASC files compatible with LTspice?

Yes, my design is very helpful program LTspiceIV.
I often hear the question: how reliable is data from the simulator? I can say: very, very reliable.

 

[dadod]

If you mean head-amp, such amps don't need high gain. This amplifier gives you more volume even with the same gain. After all, you don't plan to become deaf.

I think a big advantage of my circuit design greater depth NFB, more than 100 dB, and thus the stability of the amplifier with a small gain. Despite the apparent complexity, tuning of the amplifier is limited by the setting of the current idle.

No, I'm not assembled amplifiers with distortion levels of 1 ppm. It's a shame. I started with amps that have distortion 0.000,3% at 150 Wt / 8 Ohms at 20 kHz http://www.diyaudio.com/forums/attachments/solid-state/548293d1462871663-several-schemes-2011.gif http://www.diyaudio.com/forums/solid-state/290970-several-schemes.html.
So I have assembled amps with less distortion.


I'll be very happy if they will be useful to you or someone else. I was under the impression that my circuitry is not needed at all.


Yes, my design is very helpful program LTspiceIV.
I often hear the question: how reliable is data from the simulator? I can say: very, very reliable.

I like your circuits very much. LTspice is most use Spice here in this forum, but the models are equally important. What models do you use, can you put them here in one file?

By the way, could you show your real distortion measurements and what tool you use for it, it is not simple to measure 1ppm THD?

 

[T117]

I like your circuits very much.

I also saw your amplifiers, your ideas are very good.

LTspice is most use Spice here in this forum, but the models are equally important. What models do you use, can you put them here in one file?

Library downloaded here: DropMeFiles ? free one-click file sharing service

By the way, could you show your real distortion measurements and what tool you use for it, it is not simple to measure 1ppm THD?

For the measurements use a sound card. Sometimes, for measurements of very small distortions, together with card use an active filter with Op-Amp, which reduces the first harmonic signal.

 

[forr]

The feedback current source can be attractive for use as the tail current source for an LTP input stage. This, because of its very high output impedance. However, this application is one wherein there is typically a lot of headroom to spare. I always put at least a 1k resistor in series with the current source in this application to isolate the capacitance of the current source from the tail at high frequencies.

For years, I suspected that the role of this resistance has something to do with the high frequency behavior of the CCS. But it is the fisrt time I read an explanation for it. Thanks. 1 kOhm is a handy value as the voltage across it gives directly the number of mA delivered by the CCS.

 

[T117]

To separate the tail from the capacity CCS 1 kOhm is not enough. The resistance of the resistor should be chosen so that the voltage at the collector CCS remained at approximately 10 V.