Cathode resistors and bypass caps.

[Diabolical Artificer]

Did one last test yesterday using this circuit lifted from Valve Wizard - http://www.valvewizard.co.uk/localfeedback.html it had the best performance in terms of F ressponse and distortion as would be expected but the gain was sacrificed. THD = 0.3%, F response flat from 10hz to 15khz, 3dB down at 32khz. Gain less than 10, so you'd need to use the other half of the ECC83 to get the gain up and with local NFB stability might be an issue with global NFB applied.

I'll do a proper write up later.

Andy.

[Radio Wrangler]

Originally Posted by Diabolical Artificer

"Making wideband amplifiers is serious work" I'm finding this, but would you class 20hz ish to 20khz wideband? To be honest am not sure why amp designers bother to go this high, most of us can't hear that high as you know, only youngsters who are all listening to crap pop music anyway and couldn't give a stuff about fidelity.

So your saying a bit iof degeneration is a good thing? Degeneration I presume is loss or attenuation?

Andy.

If you try to get a large gain in a stage the anode load impedance has to go up, this brings down the roll-off point, the Miller effect gets scaled up and the roll-off point comes down seriously. Phase shifts increase to match.

If you're designing an amplifier intended to have overall feedback, you want the feedback to be effective over the range people can hear (or marketing insist on a larger number which looks competitive.impressive) THEN you need your base amplifier (to which the feedback is applied) to have a response to appreciably higher frequency, trying to roll off its gain down to unity loop gain, before the phase lags build up to 1980 degrees, else the thing will oscillate.

So for a standard hifi amp, expected to be pretty flat to 20kHz (for marketing reasons if not audiometry) then the bandwidth from each individual stage needs to be several times that. The output transformer is a much more difficult thing, so do the electronics well enough to keep out of the way.

Degeneration is not loss or attenuation. It is local feedback acting within an individual stage. It reduces the stage gain from what it could have been if degeneration hadn't been applied (or if a decoupler was shorting the signal voltage across a cathode/emitter resistor) Degeneration doesn't have a visible loop on a schematic that people can point to and say "Aha! Feedback" Degeneration works by having an impedance common to both the loop of the input circuit and the loop of the output circuit of a device.

Think of it as crosstalk between the input circuit and the output circuit and the phase of the crosstalk provides negative feedback around the amplifier.

David

[Diabolical Artificer]

Got most of that David, so for an overall response of flat to 15khz say we'd design for 25khz say?

" roll off its gain down to unity loop gain, before the phase lags build up to 1980 degrees, else the thing will oscillate." Presume that's 180 deg; could you explain unity loop gain please?

My approach at present is to try and get each stage as good as poss with as low as possible distortion and good F response, with as many stages as poss directly connected to reduce phase shift but havn't found a good guide as to how to go about designing the whole amplifier as a whole yet, with FB in mind.

I've also tried to squeeze as much gain out of each stage so I have headroom to spare when NFB is applied, but from what your saying more gain = lower roll off of higher frequency's. My little experiment so far seems to back this up.

More later, Andy.

[Herald1360]

Unity loop gain is the point at which gain measured (not necessarily easy to do.....) between the point at which feedback is applied, and the output from the feedback circuit when this is disconnected from its application point, has fallen to unity. The phase shift between this input and output will determine whether the system is stable or not.

[Radio Wrangler]

Originally Posted by Diabolical Artificer

Got most of that David, so for an overall response of flat to 15khz say we'd design for 25khz say?

" roll off its gain down to unity loop gain, before the phase lags build up to 1980 degrees, else the thing will oscillate." Presume that's 180 deg; could you explain unity loop gain please?

My approach at present is to try and get each stage as good as poss with as low as possible distortion and good F response, with as many stages as poss directly connected to reduce phase shift but havn't found a good guide as to how to go about designing the whole amplifier as a whole yet, with FB in mind.

I've also tried to squeeze as much gain out of each stage so I have headroom to spare when NFB is applied, but from what your saying more gain = lower roll off of higher frequency's. My little experiment so far seems to back this up.

More later, Andy.

I think Chris has covered the first bit.

what bandwidth you're looking for in an individual stage depends on the whole of the rest of the amplifier design, if you're going to apply feedback and want it to be stable.

No, no, no, you can't squeeze every bit of gain out of each stage so that you have plenty of gain in anticipation of feedback reducing it.

What you are doing is having a race. You want sufficiently more forward path gain than you want out of the finished (feedback running) amplifier, but you are caught between a rock and a hard place, getting the gain by squeezing the pips increases the phase shift in a stage and you want the surplus gain to roll off in a gentle fashion that creates a controlled amount of phase shift that doesn't build up to more than 180 degrees of lag before you reach the frequency where the loop gain has fallen below unity. So the tendency is to have more stages and for the gain/phase characteristic of each to be carefully controlled.

I'm finding this difficult to explain in words. It's really a case for mathematics. The analysis of the stability of a feedback loop amounts to solving differential equations and messing with complex numbers. It's not hard to do because there are standard approaches, once you understand what you are doing and why. But explaining what and why needs the maths in the first place.

It's possible for a piece of circuitry to write an equation for its transfer function (volts out/volts in) If the impedances of capacitors and inductors are written as complex numbers, the transfer function is a vector, having both magnitude and phase aspects all in the same equation.

Equations of various sections then get multiplied together and you get an equation for the trip round the loop (this is the open loop gain and phase)

Then this stuff can be put into the 'feedback equation' and you now have an equation for the behaviour of the amp with the loop closed.

Some vandal now came along and said "What happens if I put frequency in not as a simple plain old number, but as a complex number having both real and imaginary parts?"

The answer is staggering. At some values of complex F the gain can hit infinity, at others it can hit zero. (At a certain wholly imaginary frequency an R-C lowpass filter can give infinite gain! So that proves imaginary frequency is impossible). But still, the location of the infinite and zero values are useful markers. like flags in golf holes and trig points on hills. Frm them you can deduce the shape of the land. The pattern of their locations give massive clues. You can calculate od see graphically, stability margins.

If this sounds wild, it is. If it sounds crazy, it isn't. It works and it's a tool to make a nasty task routine.

David

[Diabolical Artificer]

"Unity loop gain is the point at which gain measured (not necessarily easy to do.....) between the point at which feedback is applied, and the output from the feedback circuit " Firstly some of the terms used if one is not familiar with them make it harder to understand, like unity, but I'll have to go and look that up.

One other phrase is less than clear "the point at which gain (is) measured" Is this a random point? Or are we talking a theoretical point?

For second there I thought I had it, EG if I measured gain between the first stage of an amplifier, a triode gain stage and OP IE secondary of the OPT with NFB applied (closed loop), then measured again without NFB (open loop) , then I get stuck on unity.

The mists are lifting a bit here, there is a relationship between gain, frequency and phase which for a stable circuit has to be at "state X"... unity?

Don't worry chaps, I know it must be very hard to explain, I'll need to keep slogging away a while before the eureka moment.

An apt metaphor might be instead of setting off from the start line with the throttle wide open we need to apply power (gain) slowly lest the wheels spin (phase lag) sending us crashing into the spectators (instability/oscillation).

Thanks again, Andy.

[Radio Wrangler]

An amplifier with overall feedback applied around it forms a loop. Signal passes along the stages of the amplifier in what's usually called the forwards direction to the output. The signal voltage at the output is then passed through an attenuation network back to oppose the input So the fed-bach signal goes forwards through the amplifier again, and round and round and round the loop.

It's like saying I've got a hoop from an old barrel lying in my garage. I want to measure its circumference, where should I start measuring?
Wherever you pick you get the same result, so it doesn't matter, but practically, some places are easier to dothan others. EG the rivet makes a handy marker for the starting place for the tape measure.

The game is to have the signal looping round and round get progressively smaller. We need to have the gain of the forwards parh of the amplifier multiplied by the attenuation of the feedback path fall below 1 before the frequencies at which the fed-back signal is so delayed that it reinforces itself and becomes self sustaining (this is how an oscillator works!)

Because its a loop, we could think of breaking the loop at any point and driving a test-signal in at that point (facing in the right direction of course) and then seeing what comes out of the other end of our cut. Think of sweeping the test signal and looking not just at the amplitude of what appears at the other end of the cut but at the phase as well.

Actually doing this can be too difficult to bother. Some amplifiers have so much forwards gain and use the overall feedback to stabilise their DC bias conditions that it would be problematical to get the amplifier biassed right to do any open loop measurements. Most valve amplifiers are DC biassed locally and don't have immense gain so you could do this and measure their gain/phase if you wanted to and had the test gear.

Most often, the loop gain is calculated. Any 'interesting' sections of the circuit may get built and measured in isolation from the rest, the results going in to the calculations of the rest.


The classical, grossly simplified explanation of a feedback loop on an amplifier is that the forwards gain is immense. So for any output voltage, the inpuut to the forwards amplifier is negligible. This means that the fed-back voltage must equal and cancel the isignal applied to the whole thing... its the only way you can get a net input of zero to the forwards amplifier. Flipping this on its head, you can say that the gain of the whole thing from the input voltage to the output voltage must be the reciprocal of the transfer function of the feedback attenuator.

Meanwhile, back in the real world, the forwards gain won't be infinite, and the amplifier will need some residual difference between the input voltage and the fed-back voltage to act as its input to drive the output voltage. So some of the forwards amplifier characteristics make it through to the performance of the overall system performance, the feedback network hasn't overruled everything. What it has done is diluted the characteristics of the forwards-path amplifier.

Think of the feedback system as throttling-back the gain etc. of the forwards path amplifier. The more gain the forwards path amplifier has over what it gets throttled back to, the more the feedback dilutes its imperfections. Mr Leak made an amplifier with about 1% distortion and enough gain that he could use feedback to parlay it down to 0.1% AND he got the phase shift in all the stages and the output transformer down so that it hadn't built up to 180 degrees of lag until a frequency high enough that the loop gain had rolled off to less than 1 (=unity)

There are transistor amplifiers around with far in excess of 20% distortion, and they have so much surplus gain over what their feedback system brings them down to, that their distortion is reduced to sub 0.1%


David