Questions re: Design of Neve BA283

[Ian MacGregor]

Ok,
Now that I have passed my first real transistor design class (with a grade of A, btw; thanks especially to PRR and those who helped on my labs), I have been looking at the class A Neve BA283 cards and have been trying to figure them out. (Schematics here)

My first questions is this (283AM):
What is the function of the 56k resistor and 4.7k trim pot? I know from experience that the pot can be adjusted to change the amount of aysmmetrical clipping on the amplifer. I can see how this works, as changing the pot will change the DC level at the input of this stage, but is there an AC function as well?? It appears to be some sort of feedback??

I'm sure I will come up with more questions later...

Ian

 

[bcarso]

Ian, congrats on getting the A.

At the risk of heresy (No one expects the Spanish Inquisition!) this is a fairly messy design to analyze. But your first supposition is I believe correct: these are primarily d.c. feedback adjustment parts---notice that the 56k R is the only d.c. connection to the input Q base. The main a.c. feedback setting the gain at the collector of the 3055 is via the 3.3k/80uF feedback components to the emitter of the input Q, and the feedback divider series R/C between "K" and "J" (in parallel with the 1.2k that's there all the time).

You still have a little a.c. feedback due to the 56k at low frequencies since the 3055 emitter moves a little, restrained by the 80uF from "D" to "M", but I don't believe that's particularly significant.

Brad

 

[PRR]

> What is the function of the 56k resistor and 4.7k trim pot? I know from experience that the pot can be adjusted to change the amount of aysmmetrical clipping on the amplifer. I can see how this works, as changing the pot will change the DC level at the input of this stage

That's a side-effect.

What defines "symmetrical clipping" in this stage? And under what conditions?

If you run it un-loaded, it is never symmetrical, and the trim pot gives no real optimization.

It has to be adjusted (and probably operated) with rated load, probably 600Ω. Now what is the condition for symmetrical clipping?

Assuming the 47R emitter resistor (strapped to B-) is low compared to transformer primary impedance when loaded (roughly true, about 47R to 300R), the 3055 transistor can pull down any current or voltage (to about 24V*47/(47+300)= 3.2V, or 20.7V swing. To pull up, the 3055 transistor can only turn-off. The transformer, working as a choke, must be flowing enough current to kick the load the other way. Assuming negative swings are 20.7V and the transformer is 300Ω, we better be flowing 20.7V/300Ω= 0.069A. That bias current causes some bias drop in the 47R, so the negative swing is a little less, maybe only 6V drop or 18V peak. Now bias current should be 18V/300Ω= 0.060A. Now the bias drop in the 47R is less... We can iterate to an answer, but we have neglected several small factors. Since the "right" answer is "enough bias to maximize output, but no more", the best set-up is to build it and trim.

What we are really doing, of course, is setting Q1 base-voltage so Q1 collector sets Q2 and Q3 so the 47R has the right current. We are not sure what that voltage is, but the 47R will have only a few volts, so Q2 Base is 1.2V higher, still a few volts. The 68K will have most of the 24V supply. Q1 flows maybe 0.25mA, the 1K2 has maybe 0.3V, Q1 Base must be around 1V. Base current in the 56K makes very small voltage drop: since it is comparable to the 68K which has ~20V, but flows 100-200 times less current, a few tenths of a volt. So the trim pot wiper is something like 1.0 to 1.5V, set mostly by Q1 Vbe, some by supply voltage and 68K/1K2, some by Q1 Base current and 56K.

Meanwhile the 47R is dropping something like 0.060*47= 2.5V to 3.0V. The trim-pot ends up somewhere in the center of rotation.

Look also at the extremes. Trim-pot wiper grounded, 47R will rise almost to the supply voltage, eat 10 Watts and die. Pot wiper at top, current is something like ~1.5V/47Ω or 32mA.

AC feedback is a mess. I'm sure you can figure out why diddling the trimpot has almost no effect on AC gain. But there is an ugly sneak path through the source, which may be why that 2K2 resistor is there. This is a lot like a once-common phono preamp. With a capacitor on the output emitter, certain values of source impedance make a dandy phase-shift oscillator.

 

[bcarso]

...like I say, messy....

But yes, the issue is setting up the optimal quiescent current in Q3. That might not be the optimal sound you are after, but for a given load you may at least have symmetrical clipping, if that's what you want (although as PRR said in another thread, "Who clips?").

Note also (after pondering and grasping PRR's exegesis), as far as how well the whole thing works, that the temperature coefficient of the whole thing is a factor too.

"The sound is in the iron." Well, yes; but this may well be a design where the interaction with the solid-state portion, temperature, and even signal history is where the sound is as well.

 

[PRR]

> although as PRR said in another thread, "Who clips?"

You aren't going to forget that, are you?

> after pondering and grasping PRR's exegesis)

I did meander a bit.

The pot sets the current. What current do we want?

The maximum negative swing is "most of the supply voltage". We paid good money for that supply, and that transformer to give the required output voltage from the specified (probably battery-backed) supply voltage.

The maximum positive swing, OTOH, is the stored current in the transformer times the load, which is actually about 300Ω at the primary. The stored current is basically the idle current.

Test tones are symmetrical. Speech/music never is, but we never know which side the peaks will land on. So the maximum output is the symmetrical output.

So we need to set current so positive swing is at least as much as negative swing. We could go wild and set it higher "just in case", but that increases heat and electric-bills!

Setting high current does improve clipping in those moments when speech/music peaks land on the positive side, but do not help (slightly hurt) negative peaks, and does not help pass the Acceptance Test which will look at maximum output before clipping (either side).

The optimum bias is (most of B+)/(primary impedance) or around 60mA. +/-10% variation will give either 10% extra heat or 1dB less output: +/-20% variations are probably quite acceptable.

The design and the bias-set process assume a specific load. With maybe some leeway: if you dedicate a module to drive a 1K load, you could fudge the trim to a lower current and save a little heat/power (but lose interchangeability). If you had 600Ω||Ω load, 150Ω at the transformer primary, you could increase current to either double (for symmetrical clipping) or until the 2N3055 or 47R sizzled.

And of course the load is often +/-20%, so +/-20% tolerance on bias-set better be acceptable.

And "symmetrical clipping" isn't going to happen. On one side the transistor is fat and happy, on the other side it starves. You can trim so the flat-tops are about equal, but it is going to hit bottom more sharply than it goes over the top.

Today however these beasts may be seeing 10KΩ loads. The Zobel on the secondary damps the main top-ring in the transformer but the load impedance is rising as frequency falls. Without a slip-stick, about 1KΩ at 8KHz but 10KΩ at 800Hz and heading for ~1K at 80Hz due to inductance. That changes distortion production, gain/feedback, and HF loss in various ways: it may have a quite different "sound" when not loaded in 600Ω as intended. I wonder which way the kids today like it.

I'm overlooking the fact that high idle current will soak the core in DC, first phattening the bass, then reducing bass output, and finally going all to heck. There may be enough air-gap so it fails 20Hz flatness without actually getting real-sick.

> is there an AC function as well?? It appears to be some sort of feedback??

When is that prof going to have a lesson on feedback?

Break one loop and rough-estimate the gain implied by the other loop.

The 2K2 and 56K imply a gain of about 25. We estimate the trimmer is set halfway, so gain to the emitter is 50, assuming the emitter cap isn't there (it isn't doing enough in the bass). Gain from emitter to collector is about 300/47 or 6. So this loop holds gain down to about 300.

The default 3K3 1K2 feedback values set gain to the collector at ~4. So the lower network's gain of 300 makes only 1% difference to the gain=~4 loop.

The table says that pin K can see 110 ohms, giving gain of about 33. (The table quotes gain including ~3dB gain in the iron.) Still the gain=300 loop has only about 10% effect on the gain=~33 loop. Even less when the emitter cap is working.

I neglected that emitter cap. If you see Q1 as a high-gain voltage amp, and Q2 Q3 as a big emitter follower, then the emitter cap gives one pole and the 56K resistor with the input cap gives another pole. Q1 can have a LOT of gain, so if the two poles are anywhere near the same frequency you have a 2-pole response that is liable to ring. Many phono preamps exhibit this, and sometimes at the 0.5Hz record-warp frequency. The cheap fixes include undersizing the input cap and adding some damping resistance in series (the 2K2). Both are bad ideas in phono amps: they raise input noise. When I worked this out, I resorted to hypersizing the emitter cap: 10,000uFd on a 2mA transistor's emitter. (Hey, surplus caps are cheap.) I see some of that in the BA first stage: 125uFd with 470R||1K5 is pretty beefy.

I don't understand putting 15K from S to U. That appears to give a very low input impedance with a high noise resistance, but it varies with gain-setting? I suspect a typo.

 

[bcarso]

Is the output tranny 1:1.5 as connected? That's the only way I get a gain of 15dB or thereabouts with nothing across K and J except the 1.2k.

 

[PRR]

> Is the output tranny 1:1.5 as connected?

Something like that. It does have voltage-gain, I know. And 3.5dB or 1.5 falls out of the gain-chart, as you say. And a 2N3055 in a TO-3 on a small finned sink would be a little overkill for 600Ω with 20V-24V swings.

This gives clipping level just at +28dBm minus losses. I would not rate it +28, but it is up in that ballpark. It isn't just a mike-amp: when the bombs fall (WWII was not that long in the past) and you have to patch a radio broadcast around a tattered network, this will give +8VU cleanly even over long lines.

 

[Ian MacGregor]

Wow... Some really cool stuff here. I'm slowly following along, but these discussions exceed my skill level quite quickly. When I was analyzing the amplifier, I overlooked the 47 ohm resistor on the emitter of the 2N3055 :shock:

Is there a good reference that covers feedback theory in general (not just in particular to op amps??)

Ian

 

[bcarso]

A good reference---hmmm. There are so many, each with a particular flavor. I just reviewed the material in Chapter 4 of Horowitz and Hill's Art of Electronics 2nd Ed., and it isn't too bad: it starts with smple op amp circuits, gives some motivating examples, then gets to the effects of finite open-loop gain, more examples, and finally tackles gain/phase issues. It's not how I learned about it but it's not too shabby---a decent place to start.

Avoid books with a specific control theory emphasis for now, as there are some significant differences in certain definitions of terms compared to amplifier-oriented accounts.

 

[PRR]

> I overlooked the 47 ohm resistor on the emitter of the 2N3055

So where is the output stage current flowing down through? The 4K7 trim-pot? Aside from bad form, at max it would flow 24V/5K= 5mA, or for "best power" 12V/5K= 2.5mA, allowing only 12V*2.5mA for the output stage. Even if a miracle allowed 100% efficiency, 30mW is not much when mike-amps can generally make +18dBm or 50+mW.

At some point in your career, (after you learn to find emitter resistors) you should figure out why Rupert didn't wire the 47R emitter resistor or its bypass cap to the ground bus. Hint: it might not matter for a single amp, but becomes a problem in a large system. And it might matter less in music-mixdown than in some radio situations.

> a good reference that covers feedback

Start with much less messy designs than this one, and work up.

This is a good example of complexity through forced simplicity. Transistors cost money then. This has very few transistors compared to modern designs, so it had to be very clever to make the most from the least. Use a 99-transistor chip, and the feedback action can be very simple.

But the idea here is: Q3 emitter to Q1 base is an inverting amplifier, Q3 collector to Q1 emitter is a non-inverting amplifier. Each has different AC and DC gain. The main DC reference is Q1 Vbe, though other parts have lesser roles to play.

 

[peter purpose]

I don't know if this helps or hinders.
From Geoff Tanner.....
"The B283 (or 183) is designed to drive an inductive load in the collector of the 2N3055. If you don't use a LO1166 transformer for the load (the easiest solution) you have to use a T1310 inductor as the load and wire a LO2567 transformer across the capacitor de-coupled unbalanced output."

I've just measured some LO1166s and they range from 450 to 500mH and a T1310 measures 900mH.

peter

 

[PRR]

> designed to drive an inductive load in the collector

Yes, but it will run fine (at much lower maximum output) with 150Ω R in the collector.

> measured some LO1166s and they range from 450 to 500mH

?? Seems low? I'd expect 1H to 10H for a winding supporting a 300Ω load.

Also you have to measure the way it is used: with ~60mA DC flowing in it.

And you should measure at 20Hz-50Hz; a 1KHz test on an iron-core will give a different number.

Iron-core is also sensitive to signal level, though with heavy DC flowing the effect is small enough to neglect for first-order approximation and cloning.

 

[PRR]

> these discussions exceed my skill level quite quickly.

Yes, but also you are too young to have even a nodding acquaintance with some of the old-time classic circuits.

As a glance I class the 283 as a super phono amp. If you consider the power darlington as a single transistor, and resistor-couple the output, this was a VERY popular plan for phono (huge black CDs) preamps.

I've re-drawn the 283 in simpler form. We don't need the monster output to understand the basic action. I've omitted AC paths and feedback: the DC operation is fascinating.

I know this circuit is "somewhat stable" against supply and temperature variation, but never really looked at it. Note the blue curve, output stage current for supply from 0V to 40V and temperatures of 0, 30, and 60 deg C. Over a 4:1 range of supply and a large range of temperature, the output current varies only 2:1 from 0.85mA to 1.6mA and in the "good" direction (when voltage is ample, current is ample). Considering the prime reference is the drifty Vbe, temperature has hardly-no effect. Less than the effect of using 10% resistors.

Replacing R3 with a short or a choke does not affect the DC bias enough to notice.

If you are used to high-gain balanced circuits where everything is stable to point-oh accuracy, this plan is sloppy. Everything has a little effect and they all interact in little ways. The value of R5 has more effect than I thought, even with SPICE's optimistic Beta values.

Another little trick: if you want more constant output current, you reduce the value of R1. This is opposite to simple 1-transistor stages where you need a large emitter resistor for bias stability. And small R1 means small voltage-noise.

We could spend a week on the AC action. Like: as-shown it has a very low AC input impedance. And when we fix that, the value of R2 has almost no effect on AC gain (so we can use odd-lot resistors). It can be an amplifier, or a phase-shift oscillator, or both. All good jolly fun, but you may see why when chips got inexpensive enough to use more complex and less interacting topologies, cheap-HiFi designers quit playing with this plan.

 

[peter purpose]

[quote author="PRR"]you have to measure the way it is used: with ~60mA DC flowing in it. [/quote]

Yes sorry, measured out of circuit with a $50 LCR meter.

 

[PRR]

> Is the output tranny 1:1.5 as connected?

See CJ's dissection

Turns ratio is 624:1044 or 1:1.673. Z ratio is 2.8. 600Ω on the secondary gives about 214Ω on the primary. Or more exactly: sec DCR is ~40Ω, so we have 640Ω over there, 229Ω inside the primary, plus ~15Ω pri DCR is 244 ohms looking into the primary, shunted by reactances and eddy losses.

DCR referred to the secondary is ~82Ω implying something over 1dB loss. In fact the loss nearly equals the difference between nice-round 1.5 and the actual 1.673. They fudged the ratio to hide the loss.

Since all feedback is from the primary and gain is high, true output impedance is ~80-100Ω.

Total primary inductance (according to CJ, who may be testing without DC, and maybe at 1KHz) is 1.22H. This crosses the 244Ω primary resistance around 32Hz. There is enough feedback to extend the -3dB point to a few Hz, but (assuming just 1.22H) the power response will suffer (or "color" will improve) below 32Hz.

 

[echorec]

Just found this great and informative thread. Lots of it exceeds my level though...

I have a cople of 1272 amps and one of my 283 cards has a problem. The sound is a bit more "midrangey" and honky and noise bursts out on some positions on the gain control potentiometer on the front.

All the electrolytics have been replaced and the gain potentiometer is new and when I swap the card for another 283, everything works great so it must be the 283, not anything else in the 1272 housing.

Could it be that the trim pot is set wrong? The 3055 gets a bit hot. Could I damage anything by experimenting with different settings on the trim pot?

Thanks!

 

[Tekay]

measure the current with a working 283 then check the current draw for the "bad" card. If the current ain't the same try adjust the trimmer so it reads like a working 283.

 

[echorec]

Thanks!

Stupid (?) question: Where should I measure, at 0V and 24V?

 

[Greg]

The +24 rail, but remember to measure current you just don't probe. You have make the ammeter part of circuit.

 

[PRR]

> Where should I measure, at 0V and 24V?

Measure the voltage drop across the 47 ohm resistor in the 2N3055's emitter.