John Curl's Blowtorch preamplifier part II

Status
Not open for further replies.
Administrator
Joined 2007
Paid Member
A question for Scott that might perhaps interest others.

Many IC opamps, such as AD844, is specified at +/-18V max., and +/-15V recommended.
Is there any reason why I should not use them continuously at +/-18V (regulated with soft start) ?
Do I pay for it with reliability, etc. ?


Many thanks in advance,
Patrick

It seemed worth sacrificing some junkbox parts as a bit of fun. I'm typing this while looking at a TL081C deliver around 60 volts peak to peak into its 22k feedback network. Its running on 63 volts DC and is as cool as a cucumber. Been running for several minutes now.

An ancient RCA741 (marked CA741CG) is good to around 54 volts. Above that and the output suddenly jumps to the positive rail. Its not destructive though, bring the voltage down and all is good.

Next !

An old TI 741 from the late 70's. This is good to 63 volts like the TL081.

I need a higher supply :eek: More voltage needed.
 
I was talking about a 16 bit file aligned to 0. Going all 24 bit on it would force lowering the "gain" of the file to get the over 0 levels back to 0.

John

I am not a specialist in DAW-Software but it will most likely not work that way. FS means usually that all "bits are ones" regardless if 16 bits or 24 bits.

I assume most people would not like an automatic regaining that destroys their carefully adjusted clipping (for artistic reasons) .....
 
Disabled Account
Joined 2012
Outside the box - testing

It seemed worth sacrificing some junkbox parts as a bit of fun. I'm typing this while looking at a TL081C deliver around 60 volts peak to peak into its 22k feedback network. Its running on 63 volts DC and is as cool as a cucumber. Been running for several minutes now.

An ancient RCA741 (marked CA741CG) is good to around 54 volts. Above that and the output suddenly jumps to the positive rail. Its not destructive though, bring the voltage down and all is good.

Next !

An old TI 741 from the late 70's. This is good to 63 volts like the TL081.

I need a higher supply :eek: More voltage needed.


You actually tried it. Pretty interesting, isnt it? The so-called Absolute Max ratings (at least for the max supply voltage) is not very absolute on many many devices.

So what's that all about?

:cool: :)

-RNM
 
Last edited:
I am not a specialist in DAW-Software but it will most likely not work that way. FS means usually that all "bits are ones" regardless if 16 bits or 24 bits.

I assume most people would not like an automatic regaining that destroys their carefully adjusted clipping (for artistic reasons) .....

The discussion was centered around the fact that a reconstructed waveform can exceed 0dB. If I have a 16 bit file that does that, and was normalized to 0 by software (such that the peak digital value in the file is all "ones"), then by reconstructing it 24 bit I expose the over 0dB waveform the reconstruction creates...then I push the "normalize button such that the peak value of the 24 bit file is all "ones" again.

John
 
Disabled Account
Joined 2012
I do not recommend using transistors beyond their design voltages. I've had many devices punch through as a result of that. In many cases, a mask defect at the diffusion level creates a local punch through region where gain will go up very fast as CE voltage rises. Think of it as an extreme early effect, where a small spot under an emitter finger gains up faster.


Working a transistor beyond it's design voltage brings the danger of a punch through caused current crowding failure under an emitter finger if the device is allowed to voltage slew it's CE voltage with any current being passed.

John

Some time ago I was using my curve tracer on devices and ran the voltage dial up above the max to see where the max actually was (current limited) ..... much higher than was spec'ed. It makes one wonder how that max rating is derived.... it is a very conservative spec.


Next -- what do the mfr have to say about how they determine max voltage rating......http://ieeexplore.ieee.org/xpl/logi...re.ieee.org/xpls/abs_all.jsp?arnumber=5244143

Might be fun to run a good opamp on +/- 24vdc rails for Pro apps for line stage. Most are internally current limited so that just leaves packaging dissipation to check. Some may increase input current or noise.... maybe.... there might be some good IC for audio found which safely and performance-wise are good for higher than 15/18v.. I havent done any long term tests to see about reliability.... but I would bet a dollar several such opamps exist.




THx-RNMarsh
 
Last edited:
Some time ago I was using my curve tracer on devices and ran the voltage dial up above the max to see where the max actually was..... much higher than was spec'ed. It makes one wonder how that max rating is derived.... it is a very conservative spec.

Next -- what do the mfr have to say....


THx-RNMarsh

The max rating was typically the designed level required for the part. To make all the specs, like bvceo and Icbo, you can't just have the diffusion run at the max voltage required, headroom is needed.

I probe tested hundreds of thousands of 2n2222 and 2n2907 at the wafer level, and they typically ran to 75 or 80 volts. But when a primary mask defect is in the mix, they would datalog test ok at exactly 60 volts, the toast at the next incremental voltage.

Because the test system is designed P/F on all tests, it reports the last passed test as the actual value of the breakdown. When I got a readout with 30 consecutive transistors showing exactly 60.000 volts breakdown, sure enough, I would find 30 blown transistors. And, believe it or not, I would set up one of the operators on a 576 to test the next sample batch running just Bvce(sus). The tracer's slew rate was much lower than the test system, the hotspot had time to thermally diffuse horizontally to spread out the high gain area. I would tell the production guys not to use them at high slew rates, they would select parts based on that.

Running over voltage, sure it can be done. But the manu says it's your risk.

Newer generation parts, historically from my experience, are closer to pushing the envelope parameter wise. I'm confident they have less headroom to play with.

John
 
Last edited:
The discussion was centered around the fact that a reconstructed waveform can exceed 0dB. If I have a 16 bit file that does that, and was normalized to 0 by software (such that the peak digital value in the file is all "ones"), then by reconstructing it 24 bit I expose the over 0dB waveform the reconstruction creates...then I push the "normalize button such that the peak value of the 24 bit file is all "ones" again.

John

That´s what i tried to describe in my post to George.
The problem occurs because people/software normalize the _samples_ to 0 dB (aka FS) because they don´t realize that the actually resulting waveform , which is represented by the samples, will inevitably exceed 0 dB (during DA-Conversion).
But some of them might do it conciously because they want a distorsion effect as part of the artistic impression. If the 24 bit reconstruction process shifts the gain down by 8 bits and normalizes afterwards this effect would be lost although the unwanted intersample overs would have been removed.
 
Last edited:
Administrator
Joined 2007
Paid Member
You actually tried it. Pretty interesting, isnt it? The so-called Absolute Max ratings (at least for the max supply voltage) is not very absolute on many many devices.

So what's that all about?

:cool: :)

-RNM

I sure did, and yes, I'm surprised they survived. Quite remarkable really when you think about it. I'd have thought maybe 45 to 50 volts and things might start going pop but no. The quiescent current is amazingly well controlled on the TL081. The old TI 741 was running quite warm (getting toward hot) but it hung in there.

If you think of a typical modern well designed discrete amp then its pretty stable over a massive supply range... but a fabricated chip with its literally wafer thin semiconductors... quite amazing. It could be that modern high performance parts are less tolerant of such abuse. Smaller chip parts and all.

Edit... which is what jneutron is saying :)
 
diyAudio Member RIP
Joined 2005
On overvoltage: increasing numbers of internal parts are DMOS (I was told that one boutique analog IC company has bipolars only where needed and the rest DMOS now). DMOS has a problem at high voltages and temperatures of electromigration, resulting in long-term threshold shifts before overt gate oxide puncturing. So a short-term test will not expose this, necessarily.
 
Administrator
Joined 2007
Paid Member
That makes perfect sense although the test I did was intended to be destructive. Over several minutes at voltages up to 63 volts that just didn't happen.

I can well believe characteristics would change long term and these parts as they stand now are destined for the bin.
 
sigh, away for a week and 3 topics that I have previously expanded on, contributed substantial links for resurface

the discussion not reaching even some of the other past participants previous level - some of them in this very thread



235586d1313458216-john-curls-blowtorch-preamplifier-part-ii-fig6.gif


in the digital domain you see intersample overs with upsampling to a higher rate, having to interpolate the intersample points

the digital reconstruction filters in most common upsampling audio DAC are steep ~brickwall with substantial Gibbs pre and post ringing

you can reduce Gibbs somewhat with slower slope roll off, and/or move the ringing to post transient as in Apodizing filters

I believe it is a red herring - "pure math" results do not have to reflect real world hardware, signal limits - no digital music recording will come close to the conditions, and you may note his formulation doesn't include amplitude limitation of the sum of infinite arbitrary bandwidth limited sines

a known issue that can produce "higher than expected" peaks in a amplitude, bandwidth limited digital waveform is illustrated by the "over 0 dB fs problem"

illustration from: Over The Top?
shows what you can get by deliberate construction, applying the realistic limit that real DAC hardware will amplitude limit so the "run up" to the "over" has to be both amplitude and bandwidth limited

by "eyeball" that isn't an "infinite" slew rate - 2x maybe?

I believe I heard Bob Stuart claim better oversampling audio DACs do use saturation arithmetic in the digital filters and include 3 dB over range for the more common hard clipping of simpler music signals

and some DAW sw is "aware" of the problem and provide tools to view reconstructed signal to avoid the problem in the mastering

Mastering audio: the art and the science - Google Books
 
Last edited:
Your example with a sinus of 11025 Hz (Fs = 44.1 kHz) will work but you have to move the sampling events on the time scale. If you start sampling exactly synchronized with the zero crossing of the sine wave, the sample hit every maximum and the zero crossings in between.

Start sampling not at zero crossing but a 1/8th of a cycle later and you will get the effect.
Usually intersample overs will occur during gain manipulation in the DAW, if the level meters will only rely on the samples compared to FS . Therefore the ~3dB "overshoot" is the worst case for this example.

This example of course is fabricated, 3dB only occurs if the entire passage is an exact 11025Hz sine wave. I hope we're actually worried about recorded music. The problem I assume was normalizing a CD to exactly the maximum data point and then running it through a reconstruction filter.
 
Member
Joined 2011
Paid Member
My boss designed metal gate PMOS chips in the early era of MOS LSI, when everybody was trying to reduce the number of different power supplies needed and trying to make it slightly less painful to use these weird new things, that were fantastically higher level of integration and fantastically lower power per gate, than bipolar.

Apparently one day The Enemy -- the marketing department -- requested that a certain new chip be able to support a 9 volt power supply spec, AND a 12 volt power supply spec, AND a 16 volt power supply spec, AND a 22 volt power supply spec. This would make them easier to sell. (oops, I mean, this would make them easier to use in a wire variety of applications).

So the boss said, farg it, and wrapped a depletion mode MOSFET all the way around the entire perimeter of the chip, giving it a shape factor (W/L) of ~ infinity. This was placed in series with the power supply pin, thereby forming an on-chip voltage regulator whose vout equals Vth_depletion. Now the entire chip ran on a constant, regulated voltage, independent of the external power supply pin voltage.

First wafers arrived and worked fine on the tester. Boss calls the marketing puke over and says: See that knob there? That's the power supply. This digital voltmeter shows the voltage. Right now it's 6.1 volts. Do you see that? (Yes). And the tester says it's passing, do you see the tester green light (Yes).
Crank up the power supply knob to 9 volts please. See the green light (Yes). It works at 9 volts.
Crank it up to 12 volts please. Do you see it working (Yes).
Crank it up to 16 volts please. Do you see it working (Yes).
Crank it up to 22 volts. Is it working (Yes).
Now crank it up to 30 volts. Does it work (Yes).
Now crank it up to 40 volts. Does it work (Yes)
Now crank it up to 50 volts. Does it work (Yes)
Do you think you can sell that? (Yes). Now get the hell out of my lab.​
 
I named AD844 as an example.
It is not the latest generation, and it is all BJT as far as I know.

Hopefully Scott can enlighten us.

Patrick

36V operation should be fine, a manufacturable process needs lots of headroom for reliability. This does vary from process to process and there is a phenomena called time dependent breakdown but I don't think it comes into play here. I ran the original RCA CMOS (~1971) up to almost 100V before it broke. Permanently in this case.
 
Status
Not open for further replies.