Svart
Well-known member
anyone?
[DESIGN] Original(?) mike-amp design
- Thread starter PRR
- Start date
Help Support GroupDIY Audio Forum:
I'm real reluctant to get started on low-noise design lessons. A little bit of knowledge WILL lead you to wrong conclusions. It should probably be a full-semester hands-on lab course and seminar. And for most folks (even with EE papers) it should be preceeded by a boot-camp in basic electricity and amplifier understanding.
You might browse Leach's class-notes. Don't think, though, that reading the notes (or even attending his class) really means you understand noise.
You might browse Leach's class-notes. Don't think, though, that reading the notes (or even attending his class) really means you understand noise.
Learner
Well-known member
Hey PRR,
Thanks for the link!!! :thumb:
:guinness:
Thanks for the link!!! :thumb:
:guinness:
The first page or two of National's AN-222 has an OK essay on BJT noise.
Ignore the first part where they toot their own horn. Paralleling identical devices will reduce voltage noise compared to a single device, true. However one BIG device works the same as many small devices. Since we can buy hefty switching transistors much larger than anything on an IC, for cheap, that argument won't hold water.
Their idea was to build a dual device with excellent matching by laying out 100 transistors like this:
ABABA
BABAB
ABABA
BABAB
ABABA
Note that there is a "defect", a thin-spot or spec of dust on the wafer. If they put two big transistors on one wafer, this would usually screw-up one side of the "matched" pair. By interleaving many-many A and B pieces, any defect will probably screw-up both sides equally, so the match stays good.
Their noise essay does not explain why there is voltage noise and current noise (the LM194 isn't especially "special" for current noise"). But their theory is plain and correct: for silicon transistors made in the last 30 years, noise is simply predicted by theory plus "Rbb" which is related to base contact area, which is related to overall size, which frankly is a lot like the Max Current spec. And for all normal transistors for the impedances we would call "medium" or "high", you can calculate the noise without knowing much about the actual transistor.
With any silicon transistor, the optimum emitter current in Amps for lowest noise from a specified source resistance Rs is about 0.6/Rs. Or for convenience: milliAmps = 600/Rs. Or: for 600Ω, use 1mA; for 60KΩ, use 0.01mA or 10uA.
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.
All this is fine from fairly low impedances up to quite high impedances. But when we get down to 150Ω, as when trying to get signal from a recording microphone without a transformer, we can not ignore the base spreading resistance Rbb. In many small transistors it can be over 100Ω, which must be added to our source resistance, but adds only noise not signal. The LM194 has Rbb below 40Ω, so even two of these (differential) has only marginal effect with a 150Ω source. It is still the largest "unnecessary" noise source, so we could wish for something better, but there are only a few options. The MAT duals used to be a little better, but not enough to notice, and it seems they may now be made on LM194-type dies. THATcorp has some transistor arrays. And then there are BIG transistors, like the TIP series. These may have low base-spread resistance. But they are pretty fat, and when run at the 0.6/Rs current level (about 8mA for a 150Ω source into a differential input) their Beta may be extremely low, causing not only noise but bias difficulties. They may also have leakage current, which is not a problem for clean silicon working anywhere near a "reasonable" current for the device size, but when running FAR below rating we may drown in Iceo noise. Hence we see a lot of 2N4401-like devices: 500mA parts that don't stink too bad at 1ma-5mA.
Ignore the first part where they toot their own horn. Paralleling identical devices will reduce voltage noise compared to a single device, true. However one BIG device works the same as many small devices. Since we can buy hefty switching transistors much larger than anything on an IC, for cheap, that argument won't hold water.
Their idea was to build a dual device with excellent matching by laying out 100 transistors like this:
ABABA
BABAB
ABABA
BABAB
ABABA
Note that there is a "defect", a thin-spot or spec of dust on the wafer. If they put two big transistors on one wafer, this would usually screw-up one side of the "matched" pair. By interleaving many-many A and B pieces, any defect will probably screw-up both sides equally, so the match stays good.
Their noise essay does not explain why there is voltage noise and current noise (the LM194 isn't especially "special" for current noise"). But their theory is plain and correct: for silicon transistors made in the last 30 years, noise is simply predicted by theory plus "Rbb" which is related to base contact area, which is related to overall size, which frankly is a lot like the Max Current spec. And for all normal transistors for the impedances we would call "medium" or "high", you can calculate the noise without knowing much about the actual transistor.
With any silicon transistor, the optimum emitter current in Amps for lowest noise from a specified source resistance Rs is about 0.6/Rs. Or for convenience: milliAmps = 600/Rs. Or: for 600Ω, use 1mA; for 60KΩ, use 0.01mA or 10uA.
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.
All this is fine from fairly low impedances up to quite high impedances. But when we get down to 150Ω, as when trying to get signal from a recording microphone without a transformer, we can not ignore the base spreading resistance Rbb. In many small transistors it can be over 100Ω, which must be added to our source resistance, but adds only noise not signal. The LM194 has Rbb below 40Ω, so even two of these (differential) has only marginal effect with a 150Ω source. It is still the largest "unnecessary" noise source, so we could wish for something better, but there are only a few options. The MAT duals used to be a little better, but not enough to notice, and it seems they may now be made on LM194-type dies. THATcorp has some transistor arrays. And then there are BIG transistors, like the TIP series. These may have low base-spread resistance. But they are pretty fat, and when run at the 0.6/Rs current level (about 8mA for a 150Ω source into a differential input) their Beta may be extremely low, causing not only noise but bias difficulties. They may also have leakage current, which is not a problem for clean silicon working anywhere near a "reasonable" current for the device size, but when running FAR below rating we may drown in Iceo noise. Hence we see a lot of 2N4401-like devices: 500mA parts that don't stink too bad at 1ma-5mA.
bluebird
Well-known member
So is this going to be possible with an LM-194?
G
Guest
Guest
[quote author="PRR"]I'm real reluctant to get started on low-noise design lessons. [/quote]
No everyone can have chance to go to the Low noise lessons.
At my school every professor capable to teach it is emeritus.
I can go to consultations only, but no lessons. :-(((((
PRR, if you have chance to go to low noise lessons, go there.
I used it for exploring Motchenbacher and Fitchen s bible.
Motchenbacher s Low Noise Electronic System Design is other
candidate for scan to PDF. Nice and easy, easy readable.
No everyone can have chance to go to the Low noise lessons.
At my school every professor capable to teach it is emeritus.
I can go to consultations only, but no lessons. :-(((((
PRR, if you have chance to go to low noise lessons, go there.
Thank you, professor Leach. For Lecture notes, and bibliography.
I used it for exploring Motchenbacher and Fitchen s bible.
Motchenbacher s Low Noise Electronic System Design is other
candidate for scan to PDF. Nice and easy, easy readable.
Svart
Well-known member
Thanks PRR, that was as educational as always! I know it's a lot of work but can you explain some of the various ways that people/companies have gotten around some of the limitations you explained, and maybe some experimental situations as well?
:guinness:
:guinness:
henk
Well-known member
Has anyone looked at the "old" thinks of Mr Barry Porter as disscribe in ETI february 1987?
It uses a LM394/194 perhaps these days we should use MAT04,That...
The only dislike it because you only can use +/-17V max (max 35V) , then power must by very good in filter to LM394
This could bring us to some new ideas
It uses a LM394/194 perhaps these days we should use MAT04,That...
The only dislike it because you only can use +/-17V max (max 35V) , then power must by very good in filter to LM394
This could bring us to some new ideas
Svart
Well-known member
still looking for the teacher... :guinness: :green:
> It uses a LM394/194 perhaps these days we should use MAT04
"these days"? I think the specification on the MAT parts is older than the 194, and that current production actually IS an LM194 die in a MAT can.
Either one, or the THAT, will give essentially the same results. Slightly lower noise than a pair of switch transistors, much lower rate of excessively noisy parts, and much easier balancing due to high Beta and great Vbe match.
"these days"? I think the specification on the MAT parts is older than the 194, and that current production actually IS an LM194 die in a MAT can.
Either one, or the THAT, will give essentially the same results. Slightly lower noise than a pair of switch transistors, much lower rate of excessively noisy parts, and much easier balancing due to high Beta and great Vbe match.
dale116dot7
Well-known member
There is a problem with getting a big transistor to work as a preamp... you can get a big die but with a whopping beta of maybe 10. Not ideal from that perspective.
Matching is likely most critical in a transformerless application where you depend on matching for good CMRR, at least I've found that to be the case.
Matching is likely most critical in a transformerless application where you depend on matching for good CMRR, at least I've found that to be the case.
henk
Well-known member
I am in for a design with lm194/394 but de voltage should be higher , so it could fit to +/- 24V.
The other possiblillty would be puting more transistors parallel to them like the preamp of the soundcraft 2400.
Because ssm 2017/2019 is shit , we should design something that is in silicon , no tube (sorry) , and must sound great.
Didn't had the time to try a ina 103/163 , I would like to try the ina 103 because it can have +/-24V and fit to opa2604 in that way.
But if you read the datasheet of ina 103 , it will stop production as statet on the last page. it is a pitty!(not for new designs)
So what is the replacment of the ina103 , not the ina163 , because the figures are not so good as them from the ina103.
Further i would like to put some stuff on a server/ftp site , who knows a site where i can put things for this forum and other forums?
The other possiblillty would be puting more transistors parallel to them like the preamp of the soundcraft 2400.
Because ssm 2017/2019 is shit , we should design something that is in silicon , no tube (sorry) , and must sound great.
Didn't had the time to try a ina 103/163 , I would like to try the ina 103 because it can have +/-24V and fit to opa2604 in that way.
But if you read the datasheet of ina 103 , it will stop production as statet on the last page. it is a pitty!(not for new designs)
So what is the replacment of the ina103 , not the ina163 , because the figures are not so good as them from the ina103.
Further i would like to put some stuff on a server/ftp site , who knows a site where i can put things for this forum and other forums?
[quote author="henk"]Because ssm 2017/2019 is shit ...[/quote]
Only in inapt hands.
Only in inapt hands.
henk
Well-known member
not only in my hand the ssm2017/2019 are shit , they are also shit in many tables that are built in this world.
I had to update many of these tables and none of them could get a nice sound , tables with other input do sound better and do have more headroom. Perhaps you should try to repair an Interface DDA or so.
There are many tables that sound better!!!
I had to update many of these tables and none of them could get a nice sound , tables with other input do sound better and do have more headroom. Perhaps you should try to repair an Interface DDA or so.
There are many tables that sound better!!!
NewYorkDave
Well-known member
Just slap a label on it that says "Malt Liquor" and they might let it through :grin:
(Yes, that's a little-known loophole in the regulations: over-proof beer can be sold here as long as it's not labelled as "beer").
(Yes, that's a little-known loophole in the regulations: over-proof beer can be sold here as long as it's not labelled as "beer").
Svart
Well-known member
i don't think the ssm2019 is shit on it's own. i think it's more ofthe total design than anything.
we all know that most consoles that would be using this IC to lessen cost will be cost cutting in other areas like using 5% carbon resistors and cheap caps along with using marginal values that might be on the verge of being too low because we know that the larger the cap, the more it costs.
I've been playing with a pair of them lately and yes, they aren't as *nice n' colorful* as other preamps but I've discovered that they like certain combinations of supporting components, mainly good caps. I don't have hard measurements but nichion VX caps have a soft high end and used as coupling before/after the ssm2019 tend to muffle the IC beyond what they would normally sound like.. I'm not sure what caused this but using panasonic FCs fixed this and opened the sound considerably more than using the FCs on the regular circuit.. maybe an ESR relation?
:guinness:
we all know that most consoles that would be using this IC to lessen cost will be cost cutting in other areas like using 5% carbon resistors and cheap caps along with using marginal values that might be on the verge of being too low because we know that the larger the cap, the more it costs.
I've been playing with a pair of them lately and yes, they aren't as *nice n' colorful* as other preamps but I've discovered that they like certain combinations of supporting components, mainly good caps. I don't have hard measurements but nichion VX caps have a soft high end and used as coupling before/after the ssm2019 tend to muffle the IC beyond what they would normally sound like.. I'm not sure what caused this but using panasonic FCs fixed this and opened the sound considerably more than using the FCs on the regular circuit.. maybe an ESR relation?
:guinness:
I am on the opinion as well that chips like the SSM2019 and INA103 get their bad reputation due to poor circuit design and poor components of choice. Cost is already a major issue if a design uses these opamps so further savings are squeezed out by using cheap caps. Spending a little more money on Elna Cerafine, Nichicon Muse or even Panasonic FC caps pays off. If you can spend about $15 for a pair of Solen Polypropylene caps on the input you will be surprised how much nicer the entire range comes out.
Can circuitry on the front of the preamp limit bandwidth (zeners + capacitors + relays + 2 miles of PCB trace + more crap)?
Then there are cheap potentiometers ususally the $1 Alpha kind that does not help a bit.
A 6dB driver/buffer after the SSM2019 also helps the user from having to crank it up all the time near max gain.
Can circuitry on the front of the preamp limit bandwidth (zeners + capacitors + relays + 2 miles of PCB trace + more crap)?
Then there are cheap potentiometers ususally the $1 Alpha kind that does not help a bit.
A 6dB driver/buffer after the SSM2019 also helps the user from having to crank it up all the time near max gain.
henk
Well-known member
I have done a few tables with ssm 2017 like DDA interface and midas xl200 , the work great in studio app. but if you would use it at live and have some dB's then the 2017 will do some strangs this with it and you dont know what it is , later you findout that with those tables the sound different is than with soundcraft 8000 or so.
The sound isn't nice to here it when it is in overdrive , distorsion sounds different then when you drive a soundcraft 8000 in distorsion.
Therefor it can work good in studio but i dont like it.
The sound isn't nice to here it when it is in overdrive , distorsion sounds different then when you drive a soundcraft 8000 in distorsion.
Therefor it can work good in studio but i dont like it.
