[DESIGN] Original(?) mike-amp design

[chrissugar]

Ted,

here is the link for the "trafoless microphone preamp wish list" thread:
http://www.groupdiy.com/index.php?topic=7982

chrissugar

 

[adrianh]

PRR;
Yes the noise floor in the thread was very bad.
But there was a few of us lurking in the "grass"
learning something.

 

[PRR]

> Q15 and Q16 are darlington transistors (right?). Are they supposed to be monolithic devices (if so, what part number?) or would you rather build them from two transistors each.

Whatever. Being lazy, I like the PNP version of TIP120. (I do have a mike-amp that is nothing but a TIP120 per channel: hardly lowest-noise but adequate for the mikes and the room, and monster output to overwhelm elevator-motor garbage.)

> Why did you abandon current sources instead of R20 and R21?

I thought this had been covered? The current sources feed noise current into the load and feedback resistors. At high gain, the feedback shunt resistor is low and the gain from the current-source is low. At low gain, the feedback shunt resistor is large, voltage gain is about 10, and current-source noise dominates all other noise sources.

In general, a variable-gain amplifier's noise may be modeled as an input noise and an output noise. At high gain, input noise rules. At some low gain, output noise dominates. I basically overlooked this, thought I had an idea that would be input-noise limited for any useful setting. Wrong.

The best "fix" is to drop huge voltage across larger current-source resistors. Basically you need twice the supply voltage (and power) or accept half the output. You can compromise, with compromised results.

As shown, output noise is always less than the classic $2 input stages, and maybe less than some $200 inputs at some gains. It isn't bad, except it is a PIG for power without any compelling audible advantage. It may be fun to build; I have not seen anybody try.

> PRR started with the 48V single rail option; which must be good.

"Good" in the sense that I am a lazy git, the one rail we "need" today is a good Phantom, and why build any other power voltage when it is possible to make a fine amp with 48V?

In a larger system, you always end up with other voltages for other things, so this mono-mania loses its attraction.

> Is there really a market for a super mic preamp?

There are markets for super-cheap and super-marketed preamps.

> And if there is, what are the parameters?

Price less than a mike cable, or more than one buzz-word per dollar. That SM8(?) with eight $2 preamps (each one has an XLR and a knob) retailing for $99 sells in enough quantity to eke out a volume-profit. The guys selling $2000 preamps have multi-page brochures full of technical flummery. Both classes of product also have serious engineering (production or listening) behind them, but that aint what makes them "super".

> a mic pre that will do anything you can think of, perfectly.

I'm suspecting that it is not possible to produce a simple elegant design that handles ALL needs well. The low-noise high-gain designs overload (or noise-rise) on AKG414 jammed against a Fender Deluxe, and need (ugh) pads to handle heavy soundz. Some excellent medium/low gain preamps can't handle very weak mikes on very weak acts. With conventional level structure, you may need gain from 80dB to -10dB to handle any situation; to do that with low noise and strong output, much less consistent "sound", isn't easy and may be impossible.

> Audio is not a business to make money.... If I wanted to be rich, I would have stuck to civil engineering!

And that is pretty sad: name any rich or famous civil engineers. For every Roebling, there's a guy who works 19 years for modest salary then gets laid-off and has to work at HomeDepot to keep his mortgage out of default. At that, it is more-likely money than most audio work.

You have to be crazy to make a living in audio. Or to DIY when so much really excellent gear is sold at bargan price. Being crazy, you can and should do things your own way.

 

[Rossi]

Thanks for your explanations, PRR!

Ted, if you're looking for a challenge, how about a dedicated ribbon mic preamp? There certainly isn't a huge market for it, but a lot of pros like things that do just one thing but do it excellently. Plus, some otherwise well received preamps are a bit noisy with ribbons or don't offer enough gain to begin with.

 

[TedF]

Thanks Rossi.... you're actually echoing my thoughts!

I have been playing around with higher gain mic amps precisely to come to a better compromise between gain, overload margin and noise.
I have to admit that although the idea of using a 'current mode' mic amp would seem to be ideal for ribbon mics, in practice, because the physics makes them a couple of dB noisier than conventional front ends, with ribbon mics that's vitally important, so I may have to admit defeat; or at least go back to the drawing board.

I have some nice transformers, and a few odd ribbons in the drawer.... I shall potter around with them over the next few days.

 

[Kingston]

sorry to pull up an older topic, but I just have to say I had a great morning studying this thread and learned a great deal of the real world problems faced with mic amp design. I've mostly basic theoretical background and this really helped. :grin: Instead of just building from other people's designs I have a desire to understand these things as well.

Thanks PRR (and everybody else who contributed).


It's been mentioned several times here, but has there been any advances on "the universal pre for ribbons"? I'd be interested in something like that, especially if it followed this "keeping it simple, pro and +48V" topology...

Cheers,
Mike

 

[Samuel Groner]

Are you talking about a "pre-pre", e.g. a phantom powered small box with 20 dB gain to plug ahead of another mic pre or a dedicated mic pre specifically designed for ribbon mics?

Samuel

 

[Kingston]

Agh,

I suppose I narrowed it down to a pre-pre. What I meant was a simple preamp specifically designed for ribbons of known types. (which in fact i'm sort of building based on several recent one bottle discussions)

But a little pre-pre box like that phantom powered thing inside royer 122 wouldn't be a complete misfire of an idea. :idea:

 

[henk]

After some time of waiting my stuff is now on a server and in very good hands i hope.

The ETI stuff from Mr Barry Porter can you found at
http://Headfonz.rutgers.edu/henk/eti1987/
And
http://Headfonz.rutgers.edu/henk/eti1989/

So now everybody can enjoy the design from Mr Barry Porter.

 

[Guest]

The link dosen't work for me.

 

[ulysses]

It's too bad this design idea didn't pan out for PRR, but maybe we can back up a bit and steer this thread a little bit. There's been a lot of great discussion so far, and while reading it I had a couple of ideas that might be fun to discuss...

[quote author="PRR"]

• Transformerless. I like transformers, but good ones are expensive. In effect you buy performance, and that isn't really DIY. (Unless you do considerable set-up to wind-your-own and learn all the little secrets, which is maybe too DIY.) [/quote]
  That's a good argument. On the other side of the equation, an input transformer gives you "free" gain without noise. It also eliminates the need for a differential input amplifier with high CMRR, which can open up some more possibilities for simple DIY amplifiers. There are some trade-offs involved, but the free gain and amplifier freedom combined with the galvanic isolation make a pretty convincing argument for keeping the input transformer. I'm just going to leave it open for the time being.
  
  [quote author="PRR"]
• "Professional" interfaces: differential input, differential 600-ohm output, +4dBm nominal with ample headroom. I still argue that "semi-Pro" interfacing is more appropriate for small studios and cleaner sound, but there is a historical legacy that is hard to fight. [/quote]
  I'm starting to believe that, shy of an output transformer, impedance-balanced outptuts really are the best solution. But I consider the output topology to be pretty much an option that can be tacked onto the end of another circuit.
  
  [quote author="PRR"]
• All modern mike preamps have to have a +48V supply for Phantom. It seems silly to also have various other voltages like +/-15V. You can build a fine preamp with just +48V. [/quote]
  I had just been thinking the same thing before I stumbled upon this thread. Bipolar supplies get you thinking you'll be able to DC couple everything, but when you're done you find yourself sitting on a fraction of a volt at a low-impedance node, which forces you to use electrolytic caps that won't be properly polarized. Bipolar caps don't seem to be available with the same quality as good polarized caps. And for the DIY crowd, a single rail for the whole circuit can make things much more straightforward and achievable.
  
  [quote author="PRR"]
• S-i-m-p-l-e-! Don't make the audio go through a lot of tricky-stuff. Don't design something you can't easily build or debug or tweak.
  
• Wide bandwidth and low distortion, but not obsessively so. A few-dB droop can be fixed in the mix. If the signal path is simple, the distortion will be inoffensive.[/quote]
  I generally try to use the simplest circuit possible. Sometimes that means hand-matching components, which I'm not into. So let's say the "simplest" circuit means the fewest components and no critical component selection.
  
  [quote author="PRR"]
  
  
  
  
  To change gain, you can change either feedback resistor. But for a number of reasons, changing Rshunt is most common in mike preamps.
  
  At 60dB gain, the amplifier sees 200 ohms of mike resistance noise at one input, and 50 ohms of feedback resistor noise at the other input. (Really 50 in parallel with 50K, but that's essentually 50 ohms.) The mike sees a total noise resistance of 250 ohms, of which 200 ohms is "good" (microphone signal) and 50 ohms is "bad" (feedback network dead resistance). Noise figure is 1dB, very good.
  
  At 40dB gain, we have 700 ohms of noise with only 200 ohms contributing signal. Noise figure has gone up to 5dB.
  
  At 20dB gain, we have 5,200 ohms of noise with only 200 ohms contributing signal. Noise figure has gone up to 14dB.
  
  We would like to reduce feedback network impedance. But if we reduce Rseries, it loads the amplifier output, and increases the strain on the amplifier. Maximum output falls (or the output must be beefed-up), distortion rises. [/quote]
  
  This is the main thing I wanted to address. It seems to me you could improve the noise performance without too much fuss if you were to adjust both the series AND the shunt resistors to control gain. True, to get down to unity gain you'll have to make the shunt resistor zero. But we can also lighten the load on this amp by using it to feed a 2nd amp which will have only one job: Current drive to feed the outside world.
  
  Varying both resistors simultaneously sounds like a pain, but if you're using a switch then you can select the resistors to add up and give you exactly what you need. Something like this:
  
  big image
  
  
  
  {large image changed to link -PRR}
  
  If you look at the first gain stage, you can see it uses a 6-position switch to give a gain range of 60dB with 12dB steps. You could just as easily use a 12-position switch with 6dB steps. You'll notice the resistors all add up to 10K ohms, which made the math easy but this could be scaled to any total. I've listed the gain in dB, its corresponding amplification factor, and the ratio of Rf to Rs (Feedback resistance to shunt resistance) for each position. I've seen this arrangement done with a pot, but I don't think that works nearly as well.
  
  The key to this arrangement is that the inverting input sees both resistors in parallel, so that the worst case is an equivalent of about 1875 ohms (the worst case would be 2500 ohms if the gain switch had a 6dB setting).
  
  On an unrelated note, that pot between the stages allows you to hit all the gain levels in-between the switch settings. The resistor sitting below the pot means the attenuator will only go down to -6dB, rather than all the way down to silence. That gives you much better resolution from the pot, so that you can get very good channel matching and repeatability with the usual grease pencil or eyeball method of setting the level. I think this is really the best of both worlds - a stepped gain switch combined with a precise attenuator.
  
  [quote author="PRR"]
  The most obvious 2-stage amp is sketched on the left:
  
  
  
  
  The amp on the right uses complementary devices to get the load resistor going toward the feedback network. In fact it works well if the load resistor IS the feedback resistor. And that gives the lowest feedback network inpedance, lowest noise for any gain setting, highest first-stage gain when needed.
  [/quote]
  
  So the thing I'm interested in is the possibility of using the arrangement I described above in conjunction with the amplifier concept here. Let's stick with the single-ended arrangement for now since it appears that the differential version is what got PRR into trouble with other noise issues.
  
  It seems to me that you could vary the emitter resistor of Q1 without varying the load resistor of Q2, if you switched both resistors together. This would increase the feedback from Q2 to Q1 at the same time that it increased Q1's own self-feedback. It will also muck with Q1's DC operating point, but that's something we can work on.
  
  Any thoughts?

 

[PRR]

No argument on the transformer transformerless argument. You state the situation well.

I would only quibble that "free gain" is sometimes not worth the price: gain is often too cheap to matter. And high secondary impedances (high step-up from conventional impedance) force treble compromise, as well as costly winding technique.

Your switch has the minor problem that if the wiper ever loses contact, gain rises to infinity and throws your speaker cones behind the tape decks. Since in that situation, the inverting input is hi-Z, you can connect a modest resistor from OUT to In-; then an open wiper goes to unity gain, a more polite failure. This added resistor does make the gain calculations ugly.

This is in fact a common plan in older designs, except they often used a pot.

Your plan as drawn has huge DC gain at high gain settings, and might face offset trouble. This of course can be dealt with many ways; you are just sketching. I'm not sure any opamps are really making good audio at 60dB gain (I think that's why twin-990 preamps exist). Arguing for a transformer, you can get gain there too.

The change-both-resistors plan DOES give lower noise for the same output loading, as your analysis shows.

The worst-case output noise is at 6dB gain. With the values you sketched, input noise is like a 2K5 resistor, or 10dB noise figure (ASSuming a 1:1 input tranny, which is probably not the case). This is only ~1.2uV output noise, nearly 120dB below nominal output, far better than most preamps or recorders or nearly any musical source. So good, that IIRC the old designs used a 50K pot instead of your 10K network.

While most designs are naturally unbalanced and the balance-out is an add-on, the design I posted in this or a similar thread is oddly symmetrical and nearly balanced-out by essential nature (only going "off" when the source is unbalanced and gain is low). I thought that was an odd spin-out.

 

[ulysses]

I like a lot of these ideas, and I appreciate the discussion. I haven't thought of everything myself just yet.

I hadn't thought about the open-switch squeal condition, but I have encountered that problem before, on the Telefunken V672. Any circuit that gives higher gain with a higher-value gain resistor is going to suffer this problem. The solution is to use a shorting switch, or "make before break." The good news is almost any switch is available in this configuration. The bad news is it'll most assuredly be a custom order with minimum quantity requirement, which makes it a no-go for DIY projects.

As for the DC gain - that's easy, we'll just stick a capacitor at the bottom of that resistor string and we have unity gain at DC. This is assuming a normal op-amp circuit though - if we want to try and use this gain control arrangement with your complimentary pair, I guess we'll have to have a fixed DC resistance alongside an AC-coupled gain control. I really haven't dug into that circuit at all yet, I just thought it looked neat in the original post (which was like 2 years ago, but I just saw it yesterday).

 

[PRR]

> use a shorting switch

Sure. Except in real-world, I don't trust them. So it is good that there IS a hardwired passive fix: the extra resistor. If nothing else is connected (switch/pot wiper lost contact), it WILL go to unity gain, which is pretty benign in a mike-amp. The extra loading can easily be tolerated. The uglier calculations can be done by hand in under an hour, or in a spreadsheet in 50 minutes (0.002 seconds to calculate, 49 minutes to set it up). Actually, if the ladder is low-Z and the opamp not too leaky, you can just ignore the error: 10K string (2K5 maximum impedance) with a 25K fail-safe resistor gives +5dB at the "+6dB" click, less in all other positions, and only the most anal customers will call you on that.

> assuredly be a custom order with minimum quantity requirement

I think the common $3 2P6T or 1P12T rotary is available shorting; gives you huge range in mono and a workable range as stereo.

 

[clintrubber]

from page 5 of this thread:
[quote author="PRR"]
Beta (Hfe) affects this a little. But a 10:1 change in Hfe changes the optimum current only 3:1. And the optimum is so broad that a 3:1 "error" in current causes only a small change in noise. What is really happening though is that low-Beta transistors have more current noise. Everything else equal, you can usually do better with a high-Beta transistor. The very small (50mA max) "low noise" transistors of the 1970s were really high-Beta transistors: good for high impedance sources but the cost of processing for high Beta means they are small and won't be best for low-Z sources. [/quote]

Just thinking out loud here, is the in bold above directly because of this stuff below here that I'm about to type or are there other things involved ?

It's about the influence of higher hFE for a given a type of transistor (so assuming constant Rbb).


/start/

If I'm not mistaken then the benefits from a sample with higher hFE
are because of a decrease of the In-component: ~1/sqrt(hFE)
At the same time the In-component is also a sqrt-function of Ic though: ~sqrt(Ic),
so one would expect no net effect from a sample with a higher hFE, since for higher hFE-values,
Ic needs to increase as well to remain optimal w.r.t. noise (for a given source-resistance Rs).

When Ic is 'optimally adjusted' for that higher hFE (using Ic_opt = (kT/q)*(sqrt(hFE))/Rs)
there is a reduction of In though:
the denumerator increases more rapidly than the numerator increases.

So a sample with higher hFE can reduce noise, but to notice it (assuming other contributions are not already dominating), the Ic needs to be kept at its optimal value, correct ?

As said, very possibly the resulting reduction in noise may be hardly noticable nor relevant any longer, but it's always nice to know why in which direction and by how much the other things go when something changes.

/end/

All directly resulting from the well-known formulas, but I was wondering if it's indeed because of that or that there are other things involved.
So please correct/shoot...

Regards,

Peter

 

[Samuel Groner]

So a sample with higher hFE can reduce noise, but to notice it, the Ic needs to be kept at its optimal value, correct?

Well, the sentence as written doesn't make sense to me--how can you keep Ic at its optimum value if the optimum value is a function of hFE which has changed? If you mean at the optimum value for the lower-hFE part then the answer is no, as the noise figure at the new optimum Ic will be lower for a given source impedance (both as the noise figure of the lower-hFE part at its optimum collector current and the higher-hFE part at the optimum collector current of the lower-hFE part).

Lets run an example (rbb = 10 Ohm):

Case A:
hFE = 100
Ic = 1 mA
For a simple transistor noise model a voltage noise of 0.61 nV/sqrt(Hz) and a current noise of 1.8 pA/sqrt(Hz) results. OSI is 343 Ohm with a noise figure of 0.54 dB.

Case B:
hFE = 1000
Ic = 1 mA
Gives still 0.61 nV/sqrt(Hz) but 0.57 pA/sqrt(Hz). OSI is now 1.1 kOhm and noise figure for 343 Ohm 0.3 dB.

Case C:
hFE = 1000
Ic = 5.4 mA
Gives 0.45 nV/sqrt(Hz) and 1.3 pA/sqrt(Hz). OSI is again 343 Ohm with a new noise figure of 0.15 dB. We note that both voltage and current noise has improved over case A. And we see that the higher-hFE part gives an improvement at both collector currents but at the higher current even more of it.

I did the calculations in a hurry, hope they're about right...

Of course there's much to say about the practical relevance of this (1/f noise and the difficulty to measure noise to 0.1 dB accuracy is just the start of it) but I think it might answer the question.

Samuel

 

[clintrubber]

[quote author="Samuel Groner"]

So a sample with higher hFE can reduce noise, but to notice it, the Ic needs to be kept at its optimal value, correct?

Well, the sentence as written doesn't make sense to me--how can you keep Ic at its optimum value if the optimum value is a function of hFE which has changed? If you mean at the optimum value for the lower-hFE part then the answer is no, as the noise figure at the new optimum Ic will be lower for a given source impedance (both as the noise figure of the lower-hFE part at its optimum collector current and the higher-hFE part at the optimum collector current of the lower-hFE part).[/quote]
Thanks Samuel for responding. I'll read thru your info & examples in more detail later, but wanted to add or clarify already that I meant that for a changed hFE the Ic is adjusted accordingly of course (so a higher Ic for higher hFE; the optimisation has to remain 'valid' for an unchanged Rs:
Ic_opt = (k*T/q)*(sqrt(hFE))/Rs).

More to follow; I guess I realise just now there seems already a better NF to be had before adjusting the Ic for the increased hFE.

OK, to be continued, have a good weekend,

Peter

 

[Samuel Groner]

I guess I realise just now there seems already a better NF to be had before adjusting the Ic for the increased hFE.

Sure, lower current noise always helps except for zero source impedance (although there is still rbb as hidden impedance to convert current to voltage in a real transistor).

Samuel

 

[clintrubber]

[quote author="Samuel Groner"]

So a sample with higher hFE can reduce noise, but to notice it, the Ic needs to be kept at its optimal value, correct?

Well, the sentence as written doesn't make sense to me--how can you keep Ic at its optimum value if the optimum value is a function of hFE which has changed? [/quote]
'Optimal' as I originally wrote it was meant that Ic was 'adapted/updated' for the changed hFE, so that is was yet again at its optimal value, that might have been not clear enough.

If you mean at the optimum value for the lower-hFE part then the answer is no, as the noise figure at the new optimum Ic will be lower for a given source impedance (both as the noise figure of the lower-hFE part at its optimum collector current and the higher-hFE part at the optimum collector current of the lower-hFE part).

Which is more or less the same what I said, but I went for an increased hFE... which gives already a lower NF (because of the lower in already - which I didn't realize at first). And even a further reduced NF when Ic is made optimal again for the 'actual (new) value' of hFE. Your examples illustrate this.

Of course there's much to say about the practical relevance of this (1/f noise and the difficulty to measure noise to 0.1 dB accuracy is just the start of it) but I think it might answer the question.

Still fully agreed :grin: more something to learn from than something else, see my original 'disclaimer' above:
[quote author="clintrubber"]As said, very possibly the resulting reduction in noise may be hardly noticable nor relevant any longer, but it's always nice to know why in which direction and by how much the other things go when something changes. [/quote]


Regards,

Peter

 

[ruffrecords]

[quote author="PRR"]I've been pondering a somewhat different mike-amp.[/quote]
I have not read this entire huge thread so forgive me if this has been covered before. I think it important to think carefully about how a mic pre is used and not get too hung up with noise figure specs.

At 60dB gain, the amplifier sees 200 ohms of mike resistance noise at one input, and 50 ohms of feedback resistor noise at the other input. (Really 50 in parallel with 50K, but that's essentually 50 ohms.) The mike sees a total noise resistance of 250 ohms, of which 200 ohms is "good" (microphone signal) and 50 ohms is "bad" (feedback network dead resistance). Noise figure is 1dB, very good. If the mike maker claims mike self-noise equal to 20 dB SPL, we will observe noise equal to 21 dB SPL, which is about as good as we can get.

Assuming the 20KHz bandwidth noise of a 200 ohm resistor is -131 dBU then the noise at the output of this 60dB amp is just -70dBu. Assuming FSD is +4dBu then your dynamic range is a mere 74dB - nowhere near CD quality. The self noise figure of the mic alone is insufficient to determine the noise at its output terminals - you need its sensitivity as well. The point is the noise at the amp output will never be better than -70dBU.

At 40dB gain, we have 700 ohms of noise with only 200 ohms contributing signal. Noise figure has gone up to 5dB. If the mike maker claims 20 dB SPL, we will observe noise equal to 25 dB SPL, which is not so good, though in many situations it isn't a problem. If we use a hot dynamic mike, and nominal preamp output level is -10dBV, we would use 40dB gain for sounds of 108 dB SPL. In pauses, the noise floor is 25 dB SPL. The total dynamic range is 83dB, acceptable only because acoustic background noise may be around 25 dB SPL in many studios. (But I'm booked for a 108dBSPL gig where the noise drops below 15 dB SPL after rush-hour.)

This amp has 20dB less gain but 4dB more noise so its output noise is actually 16dB lower than the 60dB one at -84dBU with a dynamic range of 88dB ref +4dBU.. Most of the improvement is due to the reduction in gain. Equally, to reach +4dBU at the output, the imput needs to be 20dB higher which means the sound level to the mic needs to be 20dB higher which means the mic's S/N is 20dB better too.

At 20dB gain, we have 5,200 ohms of noise with only 200 o
hms contributing signal. Noise figure has gone up to 14dB. If the mike maker claims 20 dB SPL, we will observe noise equal to 34 dB SPL, which is pretty poor. If we use a hot dynamic mike, and nominal preamp output level is -10dBV, we would use 20dB gain for sounds of 128 dB SPL. In pauses, the noise floor is 34 dB SPL. The total dynamic range is 94dB, barely 16-bit quality. .

Once again, the gain has reduced by 20dB but the noise has only gone up by 9dB so the noise level is now -93dBU with a dynamic range of 97dB. Again the input signal must be 20dB higher to achieve +4dBU output so the mic's output must be 20dB higher and its S/N will be 20dB better.

The thing is, even if the amp had a 1dB noise figure at 20dB gain, its output noise would still only be -110dBU. In practice this never happens because this analysis has only considered input noise appearing at the amp output. Real amps also have output noise and rarely will you see one that can achieve better than -100dBU at 20dB gain. Even if you set FSD to +20dBU you would still only achieve a 120dB dynamic range (about 20 bits).

The bottom line is that most real world analog sources rarely achieve a S/N greater than 90dB at the input to the mic pre . The only situation where noise factor is a real issue is with really low level sounds and for those you need max gain and a mic with a self noise of a lot better than 20dB SPL.

Cheers

Ian