I remember a discussion you had years ago about a technique to generate a small negative voltage from a transistor. It had to do with something about reverse-biasing one of the junctions and photons were generated and reabsorbed, generating a negative voltage on one of the pins. Do you have information on this?
Yeah, but there ainâ€™t much current gainâ€”like about 0.03%. You might do a bit better with a 4N28. If you feed 6 mA through 1k into the emitter of an NPN, which is zenering into the base, which is GROUNDED, the collector will go to â€“0.3 V or so. But with not much current, maybe a couple microamps? A 4N28 could do better.
I would like to feed two channels of an analog-to-digital converter (ADC) with the same signal, but with one channel, say, 20 dB hotter. The input stage should be very quiet and still able to handle line-level signals so I can use low-level microphone signals or line-level signals as source (in other words, a gain-ranging analog-to-digital stage). The problem is that the input stage has to perform a gain, but at the same time must not overload the ADC.
(Oh? It seems to me that you donâ€™t care if the larger signal distorts. You just want to ignore it and not look at it. Thatâ€™s much easier than doing analog or digital selectors. Or limiters or clamps or clippers. Some of the statements I have read indicate that you want to look at the larger of two signals, but if it gets too big, you want to look at the smaller one. But itâ€™s not nearly as simple as that. /rap)
This has been done before, as there have been quite a few gain-ranging designs on the market for a while. But I think that those designs have always been crossfading between the converter channels, i.e., they have allowed some time for the overdriven stage to recover. Iâ€™m wondering whether it is possible to reduce the recovery time to well below a sample (at 192 kHz, that would be some 5 Âµs), so I could think about other ways to use the two converter channels.
Hello to Stephan,
Okay, if I understood this, I think this would turn into a huge multilevel awareness problem. Letâ€™s presume that you have one signal, which may be small or big. It can be buffered and sent to one low-gain preamp and one high-gain preamp.
If the signal is large, the output of the high-gain preamp will get too big and will distort. You donâ€™t give a darn if it distorts. You just want to disconnect it. You donâ€™t want to look at it. That is a fine theory. But thereâ€™s still nothing easy about it.
If the signal â€śgets bigger,â€ť you would like to detect that and switch over to the smaller signal. That sounds easy. But youâ€™re going to have to invent some criteria to decide what is big.
Letâ€™s say a nominal audio signal (200 Hz to 5 kHz) is fed to two preamps: one with G = 2 and one with G = 20. If the volume gradually grows until distortion will soon start, you can shift quickly to the low-gain preamp. But you have to decide what is too big.
Continue on next page
What if the signal is a 30-Hz or 100-Hz sine wave? For several milliseconds, there seems to be no signal, and then a cycle comes and it becomes a big signal. And what if you have music, where thereâ€™s a clash of sound, and then a low level of music? How in the name of God are you going to decide what is real?
This is a classical problem that every AVC amplifier in the world, for 90 years, has tried to solve, and everything is a compromise. AVC amplifiers usually have a â€śfast attack, slow decay.â€ť The circuit has to look at typical signals and gradually crank the gain up (or quickly turn the gain down) to avoid excess distortion.
You want to do this quickly and nimbly as you switch between the low-gain preamp and the high-gain one, and you want the ADCâ€™s output to be re-scaled ~ very quickly to compensate for the change of analog gain. And no glitches or clicks, eh? Donâ€™t bet on it.
So letâ€™s say you invent 10 or 100 test signals. You want the detector/gain system to do the right thing, and I can assure you it wonâ€™t be easy to make a system that does the right thing.
What if one of these test signals is just a family of slowly changing sines? Of various sizes and frequencies? That part is probably easy. What if itâ€™s a family of sines whose amplitudes change suddenly? Much harder. What if you have low-frequency sines, where the spaces between half-cycles is like dead time?
Now, turn on your radio. Record samples of 50 kinds of music and 50 kinds of plain speech, like news reporters and talk shows. Theyâ€™re notoriously hard to handle because the dead space between phrases varies all over the place.
Can you find any system that will do what you want for most of these samples? Make the controller as easy as possible for you to manipulate. Youâ€™ll need all the help you can get. I think that even starting to define â€śdo what you wantâ€ť is going to be brutally difficult.
What if you had 10 modes of operation? One mode could be symphonic music, where the volume usually changes in believable ways. But there will still be cymbal clashes and loud transients.
Sound engineers have not gone out of work and starved to death because AVC systems do so well. The good ones are in demand because thereâ€™s no such thing as a really good AVC system. Am I wrong? Good sound engineers are wise enough to anticipate things coming.
If youâ€™re thinking of doing audio recording, you can record both channels and decide later what is â€śbig.â€ť But if you want to do this in real time, itâ€™s not easy at all.
One of the modes could be â€śtalk-show modeâ€ť so the spaces between phrases can be tolerated without the gain growing or leaping to high values. Thatâ€™s easy to wish for, but not easy to do well. Can you imagine a â€śrecord-the-rock-bandâ€ť mode? A mode that would be agreeable to more than half the rock bands in the world? And would sound ~ good? I canâ€™t.
If I understand your wish, then changing the analog gain and simultaneously changing the digital gain is the easiest part of problem. Deciding what you wish, and detecting signals that can be detected to see if they fit â€śwhat you wishâ€ť sounds like, is at least a half a lifetimeâ€™s workâ€”or many years of frustration until you give up! Which will you do?
Continue on next page
I have been following your Electronic Design columns for decades now and hope you have some information on an obsolete device. Recently, I found several 1N430 and 1N430A devices: temperature-compensated zener diodes in a large metal package (comparable to a 6AK5 tube in size).
The only data I can find online is the basic spec: 8.4 V at 10 mA, with different temperature coefficients per suffix. Can you suggest a source for further information on these quaint devices?
I believe they were developed for critical military applications, but I vaguely remember a column of yours about Dr. Julie where he had suspicions about the stability of references used in inertial navigation computers.
Thanking you in advance for any information,
Timothy R. Fox
Hello, Tim Fox,
One company wanted to sell â€™em for $77 to $90. 4-STAR Electronics claims it has 1100 of them, so maybe they arenâ€™t quite as rare and valuable as $77 would indicate. I am sure they were medium-crappy for things like hysteresis and long-term stability. You could aim for Minsk and hit Pinsk. I agree, the datasheet info is very thin. I canâ€™t find a decent one, either. Sorry, they were before my time. I have no source of info.
Â Dear Bob,
Thanks for your quick response. In graduate school (mid-1970â€™s), I scrounged a Julie zener-diode module (1-inch ice cube with nine-pin tube base) with matching resistor that we put in a small temperature chamber and had excellent results (monitored by a Fluke differential voltmeter). (Well, I know a lot of circuits that give excellent results if you put them in a (constant) temp chamber! /rap) I only paid a few dollars each from Alltronics, who are now out of stock.
Yeah, thatâ€™s how it goes. I pulled a good ref-grade zener out of a Minuteman II, something like a selected 1N825. I ran it for months and it never held 50 ppm worth CRAP. There goes Pinsk.