Power Amplifier for 100kHz.

E.M.Cherry's PA circuitry and their clones can form rugged designs, but they were never intended for data transmission. What's the format?

Thermal compensation of quiescent biasing was a good trick, then, but you'd have to reconsider options over an extended range. Claiming to have done such a design without doing so is silly.

You may have to satisfy yourself with something that doesn't work very well (or at all) at room temperature, if you stick with the limitations of Self's variation (intended to address inaudible distortion).

Self-heating to within the range isn't out of the question, though inconvenient on the bench.

RL

So, first of all you tell me that I haven't done what I have done (and what many others could have done), then you tell me that it won't work very well, if at all.

Who's going to break the news to the users? They'll be understandably upset that their systems work by some magic other than the electronics they paid for.

It performs well enough to pass the acceptance tests without issue. What I'm looking towards is the next iteration - can I tweak what I have or should I start afresh?

The latter is always preferable to the designer, the former to their paymasters. High temperature work always takes longer and costs more, mostly because of the testing and the restricted range of components.

<snip>

Cardboard is good for a short time, I often use similar with a heat gun blowing in.

A proper lab oven for long-term testing of course. First time I did this, I thought I'd inspect after 1000 hours. Too enthusiastic - cooled down, reached in and zap, static, dead board.

I often use a fairly wide-mouthed stainless steel Thermos (Dewar) flask (my boards are long and thin) with an aquarium pump on the bench pumping air through an insulated high power resistor (the type with the water cooling tube down the middle) and put the hot silicone pipe to the bottom of the flask with wadding in the top.

Cheap Chinese temperature controller and you have something compact and relatively safe using minimal power.

Right. I take it all back.

My reading and comprehension skills seem to be taking a hit today.

RL

** Only time I've seen that sort of temp happen involved an audio power stage employing a small, fan blown heatsink. When the fan failed, output BJTs got so hot the solder connecting them to the PCB melted. Needless to say, collectors were all shorted to emitters by the time I saw it.

The OP seems intent on breaking the temperature limits of silicon devices. Maybe he should try vacuum tubes instead, where plates can run a dull red colour for long periods. Likely need a bit more than a 60V supply and not use any electros.

.... Phil

Yes, the idea appealled to me too. I built the circuit from Quad 405 in:

It was very robust amp worked very well for me.

piglet

That may use the same bias method, but I was thinking of a discrete transistor design, posted on his web site and possibly still available somewhere:

From: Jim Thompson snipped-for-privacy@On-My-Web-Site.com Newsgroups: sci.electronics.design Subject: Unusual Bias Method Date: Sun, 24 Nov 2013 08:53:30 -0700 Message-ID: <sl749997np4gq2giqk9d8k77orheh6qt3d@4ax

Here's half of the full H-bridge amplifiers that I built for my 1977

280Z... Image scanned in quarters and pieced together for easier understanding...

No one has commented on the unusual bias scheme in this amplifier since I originally posted it.

No actual circuit designers in our midst ?:-} ...Jim Thompson

From old LTSpice trash here;

RL

Thanks!

Looks like the top output Darlington is AC coupled and when the comparator detects a quiescent current through the output resistors transitioning to less than some value, it pumps the upper Darlington base voltage up a bit, otherwise, the upper Darlington base voltage drifts down.

Is that about right?

Not sure it would work in my application as my signal isn't continuous - it spends some proportion of the time idling at half supply. Still, I could probably arrange a clock to force a comparator sample somehow.

Or maybe make the adjustment non-volatile (digipot?) and clock it both up and down. The signal comes from a DAC, so I do have access to timing signals.

A lot of the bumph is dedicated only to biasing and it would take some doing to get it to work over temperature given those polarized cap sizes. Integrated darlingtons are also best avoided. By 'wide range', the author was talking standard industrial temperatures.

You'd also have to do some thin'in around the gain-setting regime. Doubt this was a consideration in this drawing ( . . . 'or' . . .), nor was 100KHz ( hence zobel network ).

I don't see quiescent conditions being an issue, but start-up and shutdown could be surprising. Not sure that was Thompson's strong point.

RL

There was a lot of discussion of this circuit when it was posted, and Jim posted some models and simulations possibly still available on Phil's archive. I don't have time to actually think about it right now, but here are some post snips with comments and model links, sorry about the length:

------------ Here's half of the full H-bridge amplifiers that I built for my 1977

280Z... Image scanned in quarters and pieced together for easier understanding...

Here is a link to Mr Thompson's file (JimThompsons_A-B- Bias_Amplifier.zip) should anyone want to simulate the circuit.

I'm not sure that we're looking at the same schematic. In the one I have, "Half_Bridge_for_77_280Z_thompson.pdf", the LM311 is a weird sort of switching bias supervisor, not a linear amp. It's running as a normal open-collector comparator, with its output wire-ORed with the shutdown transistor Q7.

When it fires, or the shutdown line is high, it steals Q3's base bias. That causes Darlington Q5/Q6 to turn on more, with a TC of about 2 seconds. That reduces the quiescent bias.

In small-signal conditions, that'll just oscillate irregularly and keep the class-A bias current of very roughly 60 mA. (*) In large-signal conditions, the comparator will be pulling low most of the time, which reduces the quiescent bias progressively. (If the gain of the bias loop is high enough, it may not drift that far, but I'd probably need to use SPICE to find that out.)

When the shutdown pin is active, the class-A bias will gradually go to

0, turning the output totem pole into a really bad class B. (One gathers that the shutdown turns off the audio input as well.)

One aspect that I don't understand well is Q3. With a 3k/100 ohm voltage divider in the Q4 leg, ISTM that the base voltage of Q3 will be nearly the same as that of Q2, which is driven from a 620 ohm / 22 ohm divider (plus various V_BEs). That makes the current through the 220 ohm hard to estimate by eyeball. (The average current is obviously going to be small, on account of that 100k resistor.)

(I sort of gather that it's pretty tweaky, due to that scribbled-in 10k to ground from the bases of Q3 and Q4.)

Cheers

Phil Hobbs

(*) That 60 mA number is based on the ~20 mA emitter current of Q1. That'll drop about 450 mV across the 22 ohms, which translates to about

1.5 mA collector current in Q8. That puts roughly 40 mV across the 27 ohms, which divided by 0.66 ohms gets you roughly 60 mA of Class-A bias current.

Phil,

The 'bumph' controlling the biasing is a sub-audible current switch into fairly large capacitors, so simulation would have to take this into account.

In my internet/bugs/ampjt folder there are a number of simulations posted by Jim around 2013. They used the .op spice directive to establish DC operating point values only.

I couldn't get it to operate as an amplifier (starts with static latched-off biasing) using his TL081/071 subcircuit. Using the basic LTSpice single or doublepole OA would allow it to demonstrate a signal path in a .tran simulation.

His sims left out the electrolytic cap on the base of Q2 and the gain was set to simple unity (inverting) using 10K resistors and a cap-coupled source.

Crossover distortion was easily visible. As I couldn't see any slow-moving voltages or currents in the biasing section to correct this, while a signal was being processed, or any reason why they should change (with the LM311 inputs overloaded by normal operating current), I set the thing aside.

RL

In practise, of the LM311 sees zero current, it allows the bias to increase.

RL

In 1977 it wasn't that easy to do a fully-differential measurement of a

40-mV signal sitting on 12 Vpp of audio.

Re-framing the problem as preventing the measured voltage from falling much below 40 mV, and letting it gradually decrease otherwise, is an interesting approach.

JT was a smart guy, even if he was a tiny bit too aware of that. ;)

May God hold him in memory eternal.

Cheers

Phil Hobbs

<snip>

Yes, it's a good idea. I'd go for a current monitor - eg INA169 - to get things down to the 5V domain (I have a 60V supply) and take it from there.

The concept is sound, but not actually demonstrated here.

In the presence of signal, the biasing reverts to bad classB, as you've suggested. The op amp is stressed to supply the ~ 20mA in Q2, while slewing a full darlington Veb at crossover, even for 40mV signal amplitudes.

All subckts in the sim are questionable in function, even the simple darlingtons (from Modpex 2004?). There are no thermal spice parameters included.

High temperature 100KHz is probably out of the question.

RL

RL

Something like an LM10 will operate at low power within a few Vebs of local supply voltage, in a bootstrapped circuit.

Not sure if it was there in 1977; definitely 1979. Possible that wafer size will prevent miniturization in SMD, but hermetic packages may make more sense anyways, at high temperature.

A range of integrated circuits will intentionally shut down above a certain temperature (~ overload protection circuitry), so 'simplicity' or low junction/device count is probably better.

RL