MUFF WIGGLER Forum Index
 FAQ & Terms of UseFAQ & Terms Of Use   Wiggler RadioMW Radio   Muff Wiggler TwitterTwitter   Support the site @ PatreonPatreon 
 SearchSearch   RegisterSign up   Log inLog in 
WIGGLING 'LITE' IN GUEST MODE

Explain the LM3900 to me
MUFF WIGGLER Forum Index -> Music Tech DIY  
Author Explain the LM3900 to me
pugix
The LM3900 is a quad op amp. I have the data sheet, which says:

"Operation of this amplifier can best be understood by noticing that input currents are differenced at the inverting input terminal and this difference flows through the external feedback resistor to produce the output voltage."

I am building a new modular that is planned to use 9 CGS Serge SSG boards. I'm building some of them to be used as triangle VCOs and others as sample and holds. I need six VCOs and 12 sample and holds, and each board has two circuits.

http://www.cgs.synth.net/modules/cgsssg_ssg.html

I'm trying to understand this amazing circuit, invented by Serge in the 70s. One of my questions is how CV mixing works. Notice on the schematic how a CV input with attenuator is mixed with a current from a trimpot and from an initial rate pot. These mix into the negative input of an LM3900 amp. But this node is clearly not at virtual ground, like a normal op amp. I tried adding a second CV pot in parallel with the first (through another 2M7 resistor) and it didn't work. Just adding another resistor to ground from that node results in more current flow.

Also, take a look at how comparators are made with a large hysteresis for the Cycle outputs. When a Cycle output is patched to the Input, the result is a triangle wave that cycles between about -0.5V and +4.5V. The frequency range is amazing, from off (holding) up to 1-8 Khz, depending on the value of the integrating capacitor.

I'm looking a simple way to reason about the behavior of these amps.
daverj
Have you read this document?

http://www.ti.com/general/docs/lit/getliterature.tsp?baseLiteratureNum ber=SNOA653&fileType=pdf

The 3900 is a unique part. A normal op amp uses a differential amplifier on it's input and amplifies a voltage. The gain is so high, and the normal feedback schemes used make this act like virtual grounds with current inputs. But in reality they are differential voltage inputs with extremely high gain.

The 3900 is very different. It is basically a pair of NPN transistors with their emitters grounded and the collector of one tied to the base of the other. The two bases are the plus and minus inputs. The result is true current inputs one diode junction above the negative rail of the chip. The current into the positive input is subtracted from the current into the negative input, and the resulting current is converted to a voltage and fed out.

So unlike normal op amps where the inputs can be biased to virtually any voltage, the inputs to the 3900 are always at the same voltage (one junction above the minus/ground rail). And you can't use them as voltage followers because you can't directly apply a voltage to their inputs. Only currents.
daverj
Also, with normal op amps you typically bias the inputs to be roughly in the center of the power rails, and the output signal is also roughly centered between the power rails.

With a 3900 the output is still roughly in the center of the power rails, but the input is always in the same spot. One junction above the lower rail (V minus/ground). So you always have to think in terms of sourcing current to those pins, which sit down at a low voltage.
pugix
Dave, thanks for that document and the explanation. I understood that the inputs were current based, but I hadn't realized that the input current is sourced to approximately -V. To make it more complicated the Serge design powers the chip off +/-12 or +/-15V (the latter in my case).

I want to understand the transform of input current to output voltage, if that makes sense. I want to be able to see why the oscillation goes between about -0.5V and roughly +5V. The amplitude of the output when oscillating also increases at the high end of frequency, maybe as much as a volt. These make interesting oscillators, because they have such a wide range. For this application I'm not concerned about stability. I've observed that they wander around until they've warmed up for a while.
daverj
OK, lets take a look at your earlier question about CV mixing. Lets look at the section of that circuit you mentioned which contains U2 pins 2, 3, and 4.

The 2M7 from the VC attenuator to pin 2 is always supplying current, since pin 2 is at -11.3v and point "D" is always above -11.3v. This is balanced out by the current supplied from the Rate pot and the trimpot at pin 3.

If you add another attenuator and 2M7, you have increased the current to pin 2. If both attenuators are at ground, the current is doubled. At other settings the current is the sum of currents through the two 2M7 resistors, created by the voltages through the attenuators.

To balance things out you need to apply a similar current to pin 3. You could do this by adding a 2M7 from ground to pin 3. Or you should be able to adjust the trimpot to get the extra current (though you might need to change the 1M5 to a smaller value if the trimpot won't adjust far enough).
pugix
Dave, here's what I would like to do with the VC.

I want a second manual rate pot, not a second VC input. I tried duplicating the rate pot, with another 2M7 resistor going to pin 3. I was hoping to get CV addition, so that turning either rate pot 'up', i.e. towards ground, will increase the frequency. The reason I want two is so I can have one master rate pot that controls six of these together, with each also having its own individual pot. The lowest frequency/longest rate should be when both pots are CCW. How could this be done?
CLee
Nothing to add here, but thanks for having this discussion. I've always wanted to "figure out" this chip. I've never seen it anywhere other than in Serge designs, and he used it a lot.
Nordcore
It's in the ARP-Filter
http://www.marksmart.net/gearhack/arpomni/FilterMod/Schem4075.JPG
andrewF
The Paia Gnome uses a single LM3900 to get a VCO, VCF, EG & VCA, not a bad effort at all.
Serge's LM3900 based circuits still make wonder- how did he ever come up with those designs?
daverj
Well with one Rate pot it is supplying between -0.26uA and +4.185uA to pin 3. With it in the center it is supplying 1.96uA. If you have two of them, if they are turned together you get double those currents. So if they are both set to the center you get 3.92uA. To get it back to where you were with one pot you need to remove 1.96uA. If the trimpot is set to supply that much or more, you could turn it down. Or you could add a resistor from ground to pin 2 to supply the current there (which gets subtracted from the current at pin 3).

So a 5M7 from ground to pin 2 would compensate for adding a second Rate pot at pin 3. The trimpot at pin 3 would stay set the same, so could still be used to tweak the center point for the Rate pots.

Of course the problem with adding a second pot is the same as with any circuit where you add a second pot for a given function. Either pot works as the original did only when the second one is in the center. If you turn either one up or down the second one can't cover the whole range any more because the other one is biasing it's center point.
dmitri
anybody found a good source for non TI brand LM3900 ?
Or for MC3401? or for any other equivalent ?

For now the TI LM3900s sourced from Jameco & Tayda are saving me a cable when making chords with my Dual Positive Slew. Which actually works really good ! though not intended by design.
J3RK
dmitri wrote:
anybody found a good source for non TI brand LM3900 ?
Or for MC3401? or for any other equivalent ?

For now the TI LM3900s sourced from Jameco & Tayda are saving me a cable when making chords with my Dual Positive Slew. Which actually works really good ! though not intended by design.


You might contact someone at Mammoth. They sell them, and don't specify what the brand is. Odds are it is TI, but might be worth checking.
daverj
I gave up on Jameco years ago when they sent me used and mismarked chips. I've never bought from Tayda.

Are you saying that the TI 3900s are not working as intended? If so, you might double check the wiring. I'd suspect faulty wiring before faulty chips, unless the chip is actually blown (in which case it's not an issue with brand anyways). They do blow easily if fed the wrong voltages on the inputs.

I believe TI is the only one left making those parts. National (now owned by TI) and Motorola (split up and sold off) both discontinued them quite a while ago. You might find some 20 year old stock somewhere, but not sure where. Since they're still made by TI, and relatively cheap, the demand for old stock would be small. They're 19 cents each at Mouser (or $10/100)
dmitri
"Are you saying that the TI 3900s are not working as intended?"
No.

im assuming that the Serge Dual Positive Slew was designed for MC3401.
Using two TI LM3900 in it creates "cross talk" between the two slews.

The cross talk is in the chip./ cant possibly be wiring nor PCB error.

it actually doesn't suck though. (at least i dont mind) its an accidental chord generator that way. hihi CV one , detune the other... seriously makes chord patches easy!
CJ Miller
Aren't the MC3401 just spec for a lower supply voltage? I assumed that Serge used these for trigger subcircuits because he only needed +6V and they were cheaper. Still, I guess swapping out chips won't hurt.
daverj
The 3900 and 3401 are virtually identical at power supplies less than 18v. Neither one should have much crosstalk.

If you used standard resin core solder, be sure that you cleaned it off pretty well. It can cause crosstalk at high impedances.
Tim Stinchcombe
pugix wrote:
I want to be able to see why the oscillation goes between about -0.5V and roughly +5V.
These thresholds derive from the 'cycle' section which acts as a comparator. Looking at U2 section 1/5/6, when the output is high, current into pin 1 will be approx 30V/1meg + (15+13 say)/2meg2 = approx 43uA; whilst current into pin 6 is below this, the o/p stays high, but when above the o/p will flip low, and this will happen when the 'smooth out' is about 470k x 43u = 20V above neg rail, so around +5V. Once the output is low, there is only 30uA into pin 1 (i.e. assume o/p close enough to neg rail so no current in the 2meg2), so voltage across 470k needs to be about 470k x 30u = 14V, minus 15 as referenced from neg rail, so around -1V. (If you work with better figures for the o/p saturation voltage, include the 10k in series with the 470k, and allow for the diode drops at the inputs, the figures are likely to more closely match reality!)

Quote:
The amplitude of the output when oscillating also increases at the high end of frequency, maybe as much as a volt.
Whilst you don't indicate at what sort of frequencies the output swing starts to get bigger, I'm guessing this is related to the output slew rate, which is quite asymmetrical - the high to low value is 40x the other way - and so I suspect it is simply taking proportionately longer to switch at higher frequencies (the LM3900 can hardly be considered a 'high spec' device!)

Tim
pugix
@Tim Stinchcombe Thanks for that explanation. Isn't it amazing that Serge came up with this! screaming goo yo

I'm happy with using the LM3900 in this application, because it's not a 'high spec' synthesizer that I'm building. I am perfectly happy to use circuits with imperfections, since these introduce irregularities which act as stimulants in the complex feedback loops that form the basis of the design.
thebot
I'm really interested to see / hear this thing you're building Pugix. I don't have the foggiest at the moment what you're shooting for here, and that's usually a good sign.
Tim Stinchcombe
pugix wrote:
Isn't it amazing that Serge came up with this!
Well I don't know how it would have appeared back in its day (the 70s, presumably), but by today's reckoning, apart from the use of the LM3900s, the circuit is (mostly) reasonably run-of-the-mill - both the Doepfer A-141 ADSR and A-142 VC Slew, for example, are based on similar circuitry. But it does seem (to me at least) that Serge had a very good understanding of the LM3900, and how to get the most out of it.

The OTA/cap/JFET plus non-inverting LM3900 section gives a variable slew, controlled by the 'Iabc' current feeding the OTA. Adding the comparator at the 'cycle' out gives the basic integrator/comparator oscillator, outputting both a tri and square wave, which appears in many an op amp datasheet/app note. The 'CV control' part of the circuit (based around U2 2/3/4) will likely be the hardest part to fully get ones head round - it is very similar to the set-up in the Plan B M10, which in turn is mostly a direct copy of an early Serge envelope generator. I've had several attempts to come up with an analytic expression describing how it works, but never really succeeded - the interactions between the diodes at the inputs, and (in this instance) that at the OTA Iabc input, make the algebra pretty horrid. I suspect however that the current driving the OTA will be some sort of typical 'hyperbolic tan' shape, which will be exponential-like on the initial bit of the curve, which is presumably the main reason for the set-up (if I get the chance I'll review my simulations of the M10, but even that sometimes struggles with all the interactions).

Tim Stinchcombe wrote:
Whilst you don't indicate at what sort of frequencies the output swing starts to get bigger, I'm guessing this is related to the output slew rate, which is quite asymmetrical
Having run some calculations, and backed up by simulation (once I got it to run at all - even simulations need the power applying to op amps with the correct polarity...), I reckon the max frequency should be around 2kHz, and as such, the slew rates will be nowhere near the maximums of the LM3900, so it seems the source of the increased swing must be caused by something else.

Tim
Fernando
What's the best way of wrapping unused amps on a LM3900?

The only example I find is on the DUSG, but it does not look like actually wrapped : 1M to - input coming from end out, + input to ground, output not connected. As if some function was planned for it but was not used in the end.
(I always think about possible uses of this two unused amps on each DUSG, btw)
Fernando
in case it helps someone: it seems that the consensus is to leave unused amps unconnected, forgot to check SDIY before asking!


It's also good to review the LM3900 AN-72 (application notes)
There is a logic OR gate made with 1/4 of 3900
and a comparator with 2/4 of 3900 (each DUSG have 2 x 3900 amps unused)

cheers
pftjschute
CLee wrote:
Nothing to add here, but thanks for having this discussion. I've always wanted to "figure out" this chip. I've never seen it anywhere other than in Serge designs, and he used it a lot.


i just happened upon this thread because i was curious about the chip while working on my first serge build (dusg!)
MUFF WIGGLER Forum Index -> Music Tech DIY  
Page 1 of 1
Powered by phpBB © phpBB Group