Transistor CCS resistor value calculation.

[Diabolical Artificer]

"I ask again, how much does the Zout matter in this application?" AFAIK the higher Z out the better the frequency response, so it a matters a lot. We are trying to get better balance from the two OP's of a LTP, a more linear Op and more gain by using a CCS instead of a resistor. We want low DC Z and high AC Z.

Your right there are a lot of topologies Jeffrey, but I'm after something simple, high voltage and cheap that doesn't use exotic parts like Vref chips etc.

"Personally, it's far easier to measure Zout than to try to calculate it" The only way I know how to measure Z out is to use a variable R in parallel with the OP and twiddle until OP = 1/2. I suspect there's more to it than that. Using datasheet stated hFE is going to give you an approximate Z out yes? As the hFE of a Q as per datasheet is an approximation; could be lower than 100 or could be 250. As my circuits arn't critical and I'm still feeling my way an ish figure is good enough in most cases.

Compliance, heard of it, will do some reading up ont subject. How does one calculate it, or do you measure it?

Temp stability - the CCS is near valves, albeit somewhat shielded. Got you re using a 1N4148 as well as zener.

The ring of two sounds favourite Mark, I'll knock one up. The CCS I've been using has a 220r fixed R and either a 1k or 4k7 preset as Re. I've found that my calculated figures are often off. For instance say I've drawn a load line and chosen -8v as a good bias point but when the circuit is running while scoping the two OP's of an LTP I've found a better bias point by adjusting the CCS Re.

Thanks again for all your input, i'll do some building and testing and report back.

A.

[ppppenguin]

I can't see that increasing Z out will make much difference to frequency response. In any case the Z out at higher frequencies will differ from that at low. If only due to stray C. But we're only talking about audio here where the performance of the CCS is going to be pretty good.

BTW, you can't have low Z at DC and high Z for AC using a CCS. The definition of an ideal CCS is that it has infinite Z at all frequencies. What that means at DC is that if you change the applied voltage slightly then the current hardly varies.

Compliance is the range of voltages over which a CCS will work. For your sort of design the upper limit is set by blowing up the transistor, the lower limit is set by taking the collector voltage below the base voltage.

Trying to measure Z out by paralleling with a resistor is bound to fail for a CCS like this. Conceptually you can do it by applying a variable voltage to the output of the CCS and measuring the current. In a working circuit you can 'scope the emitter as the collector current is close to the emittter current. This will also give you some idea of the dynamic performance of the CCS.

[Craig]

Excellent articles by Walt Jung on current regulators

http://waltjung.org/PDFs/Sources_101_P1.pdf
http://waltjung.org/PDFs/Sources_101_P2.pdf

Note particularly the one in part 2 with an LM317 voltage regulator and a heatsunk MOSFET that goes up to 450V

[Mark Hennessy]

Hi Andy,

The reason I asked about this application is simple: you might be worrying about the performance of the current source for no good reason.

I'm guessing that this is being used to supply the tail current to a valve-based LTP at the front end of an audio amplifier?

I've no real experience of using valves as a LTP, so what follows comes from the perspective of transistor circuits. However, I don't think I'll be that wide of the mark.

Before we can design the right current source for a circuit, we need to know what purpose the current source is fulfilling. Note that "right" does not automatically equal "best" - over-engineering part of a circuit can be as much of a crime as under-engineering might be.

In a conventional amp, using a current source on the LTP brings 2 benefits:

1. Greatly improved common-mode rejection
2. Better rejection of noise on the negative PSU rail (assuming NPN or valves).

These come about because of this simple fact: if you vary the tail current, then the currents in the collector/anode load resistors vary - and of course, you're taking an output from at least one of these resistors.

Let's think about point 1 - CMRR - first of all.

This is important in a general-purpose op-amp, where the two input voltages could be placed anywhere within the range of allowable values (which is close to the entire +/- supply rails for many op-amps). With just a tail resistor, the tail current is obviously directly related to the common-mode input voltage so the common-mode rejection is terrible, but a current source fixes that.

But for an audio amp, there is very little common-mode signal. In fact, the common-mode signal is the audio input itself - perhaps only a volt or 2. It would be different if you were running the power amp at a lower overall gain - and indeed, a unity-gain buffer is the toughest challenge for an op-amp in terms of CMRR (and common-mode distortion, which FET op-amps often suffer from).

In short, the current source only needs to keep the current constant (hence, Zout high) over a very small range of voltages. With that in mind, you actually have to think pretty hard about including one in the first place. If your negative rail is in the order of -80V, then a 15k resistor might do the job as well as a current source would in the overall scheme of things. Actually, 5mA is a pretty high tail current for a transistor amp - 1 or 2 mA is more typical, so that's a 75k or 39k resistor.

How about point 2?

This is simple: if V- varies, then the voltage across the tail resistor also varies, which means the current in it must also be varying. And we already know what that means...

A current source obviously fixes this problem. How good does our current source need to be for this problem? Well, the noise on the -V rail will be 100Hz and harmonics thereof. Even the most basic current source will cope with low-frequency stuff like this. If you're really going for broke, you might look at ways to improve it further, but the sad truth is this: it's actually far easier to decouple the -V rail before it hits the current source. A resistor and capacitor is far cheaper and dependable in mass-production than complicated current sources.

In fact, let's say that you didn't want a current source at all. Above, I said that a simple tail resistor might be adequate when the supply rails are high (high rails means large tail resistors, and large resistors connected to large supply rails look awfully like current sources in my book). If you did this but found the PSU rejection was poor on the -ve rail, a 1k resistor and a big cap ahead of the tail resistor would fix it - cheaply and elegantly.

The right way to do this is to first establish what sort of performance you need from your current source. So I'd build up the whole amplifier, but using just a tail resistor, and then inject a signal into the tail at different levels and frequencies, and see what sort of levels causes that source to appear at the output above the noise floor. Bear in mind you've got negative feedback on your side! Having done that, you now have a spec for the current source, and you can target your efforts into achieving this. As there's every chance that a simple "ring or two" would be well in excess of the actual requirements, that saves getting stuck in the rabbit hole.

Looking at the articles in post #13, the ring-of-two does pretty well. But given the high voltage/valve nature of this amplifier, I think I'd be tempted to use a MOS-FET as the second transistor

[ppppenguin]

Mark has given a far more eloquent and comprehensive answer than I. His experience as an educator definitely shows

The real questions are about what you are trying to achieve with a CCS. As Mark said, if a resistor is good enough then why bother with a CCS at all. If you do find you need a CCS then how good does it need to be? Any plausible sort of CCS will give much higher output Z than a resistor.

[Diabolical Artificer]

Thanks chaps, lots to think about. I'll breadboard some circuits today and report back soon as.

A.

[Diabolical Artificer]

Have read parts of Craig's articles and your comments Mark, what follows is me thinking out loud trying to assimilate all the info, so if I've missed some of the points you've made, I apologise. I'll need to re read everything a few ( a lot of) times before it coalesses.

The first PDF that Craig linked goes on about "power supply related noise" ripple yes? One thing I have noticed about the CCS I've been using is the large ripple. Have knocked up a test circuit - a phase splitter, but this is unsuitable if i want to measure Z out, as it's tricky to vary two loads at the same time. So thought about a simple common cathode gain stage instead. What I am uncertain about as regards a test circuit is whether I need a high voltage gain stage as part of the set up.

As I understand it at present (this might be a bit sketchy ) I need to test a CCS for 1) compliance - IE it stays at XmA over a differing voltage range. 2) Z out 3) frequency response 4) good ripple rejection? 5) not go bang : )

Therefore to test 1) I'd need a variable negative DC voltage supply ( though over not much of a range, say 10v? , to test 2) I need to measure V AC out at no load, then either compare to a load at 90% or measure V out with no load load then attach and lower load till Vout is half. Finally to measure 3) I need to measure V AC out from 20hz to 30khz say. The only way I can do this with a certainty of the result is to do a manual sweep but in steps of 1khz say. As regards 4) simples, scope it. 5) switch on and duck.

Been looking online at mosfets and Darlington's (have some TIP122's). Mosfets are tricky as they have no hFE, but transconductance, no worries there, it's like a valve, just more mental gymnastics.

Andy.

[Mark Hennessy]

MOS-FETs don't draw gate current*, so the hFE is arguably infinite Yes, they have transconductance, but much less than a typical bipolar transistor. If you can live with the base current - and that's the way to regard base current (an inconvenience, rather than something that can be depended on) - then bipolar transistors will often do a better job. However, FETs are more marketable in some circles for whatever reason... And easier to understand than valves IMHO.

*Of course, that's DC current. At AC the gate will draw current because it looks like a capacitor.

Any PSU can be negative or positive, assuming one end hasn't been connected to earth. That's certainly the case with bench power supplies, where the outputs deliberately float. However, in lieu of that, then use a variable resistance in series with a milliampmeter. As you vary the resistance, the voltage across it varies in sympathy. Note the points where the milliamp reading changes, and measure the voltage across the current source... Watch for power dissipation in the resistance, of course. A decade box is nice to use if you have one, but a pot will do.

Again, Zout is almost certainly irrelevant for your application, but to measure it, you apply an AC voltage to the current source (technically a "sink" here) via a series resistor, and sweep the frequency. You can use a DMM to measure that voltage if you know that it accurate at the frequencies of interest; otherwise use a pair of 'scope probes to perform a differential measurement. I'll leave it as an exercise to work out how to crunch the numbers.

Ripple rejection is easy. Let the current source/sink establish a flow of current in something like a 1k resistor. Monitor the AC voltage across the resistor (hopefully it'll stay at 0V). Then, look to see how much ripple you have on the negative rail feeding your current sink. From there, you can do 20*log(Vr/Vs) (where Vr is the AC voltage across the resistor, and Vs is the AC voltage at the supply rail). I'll leave it to you to decide if you need to consider the value of the load resistor in those calculations (hint: you might want to turn that AC voltage into an AC current). If you find that the AC voltage across the 1k resistor is hard to measure on a 'scope (and it ought to be), then try increasing the resistor... You can also increase the ripple on the supply rail - either by reducing the value of the smoothing capacitor(s), or by deliberately drawing a large current from the rail with a power resistor.

[Diabolical Artificer]

Been trying to test CCS's but I'm struggling. For the life of me I can't figure out where to put a signal in, to frequency sweep for instance, or where to take an  OP off. See att. Whether it's the chilly temperature in the workshop or I'm over thinking things or just being dense, I don't know.

A.

[Mark Hennessy]

You connect the signal source between the rail and the 1k resistor. Between points E and D, if you prefer.

But you have to think hard about grounded test gear.

The scope will be grounded - can't do much about that apart from taking a differential measurement - so let's hope that the signal source and power supply are both floating. The latter has to be, but if the oscillator is also grounded, then define point E as ground.