LM394 National Semiconductor 'supermatch' NPN pair

[flippineck]

LM394 datasheet

https://www.farnell.com/datasheets/25507.pdf

I have a simplish circuit which seems to be a long tailed pair made from an LM394 driving a galvanometer, to give an indication of circuit power.

Although the occasional LM394 does appear on eBay it seems to be obsolete and expensive if available.

I was wondering if, two BC54x discrete transistors could be used instead, albeit with a less accurate result, if one hand-matched things like Hfe and Vbe, and maybe mounted them both in close proximity on a heatsink?

BC54 6,7,8 A,B,C datasheet

https://www.farnell.com/datasheets/55009.pdf

A bag of 50 BC547's is a third of the price of the one secondhand LM394 I can see at the moment.

Maybe there's a better alternative for the LM394?

Would BCM847 work in this circuit?

https://www.mouser.com/datasheet/2/916/BCM847DS-3081843.pdf

https://modwiggler.com/forum/viewtopic.php?t=55387

 

[Delta Prime]

I was wondering if, two BC54x discrete transistors could be used instead, albeit with a less accurate result, if one hand-matched things like Hfe and Vbe, and maybe mounted them both in close proximity on a heatsink?

That seems to be the consensus.
Even if you match up a pair of discretes on a curve tracer, the thermal differences will still cause gain drift. You will need to either use a monolithic solution or maybe try to enclose the pair with silicon thermal grease and a copper clip to attach them face to face. dbx used to use some adhesive backed aluminum foil

https://gearspace.com/board/geekzone/451413-lm394-replacement.html

 

[crutschow]

There are dual transistors available in a single package that should work better than two discrete transistors.

 

[flippineck]

There are dual transistors available in a single package that should work better than two discrete transistors.

I've ordered a BCM847BS & a board to solder it to.. will give it a try!

 

[crutschow]

I've ordered a BCM847BS & a board to solder it to.. will give it a try!

Do you know how to connect to such a small, surface-mount device?

 

[flippineck]

No problems on the practical side, it's the theory and maths side I struggle with more.

 

[Delta Prime]

the theory and maths side I struggle with more

Here you go buddy.
A little snippet.
It has become clear in the past few years that ultimate performance in monolithic transistor pairs was
being limited by statistical fluctuations in the material itself and in the processing environment.
This led to a matched transistor pair fabricated from many different individual transistors physically located in a manner which tended to average out any residual process or material gradients. At the same time, the large number of
parallel devices would reduce random fluctuations by the square root of the number of devices.

 

[flippineck]

Thanks Delta Prime. Figures 12 & 13 & associated discussion looks very useful! It will take me a while to digest.

My circuit is intended for 0-40V DC @ 0-8A, to measure Watts produced by Trinasolar TSM-230C05 panels which have a rated Pmax of 230W. I'm hoping to try using the BCM847BS and a 0-500uA galvanometer for this.

I've played about trial-and-error style in a simulator and gotten somewhere near sensible operation, but my lack of deep understanding of the operation of the multiplier is preventing me calculating (rather than guessing) really appropriate resistor values.

Falstead simulation link: https://tinyurl.com/yu2ept2r

I'll go over that document until hopefully something clicks!

Tried Googling a SEM picture of a bare LM394/194 chip, hoping to catch sight of the many individual transistors, was quite intrigued! No joy though. So that's how they did it, anyway.. cool!

 

[danadak]

See if you can find sem in reliability handbook, NSC. Its a long shot,
bitsavers one place to look for book, archive.org another.

I found this - https://zeptobars.com/en/read/National-LM394CH-super-matched-bjt


Regards, Dana.

 

[Alec_t]

In your simulation why are you supplying the transistor collectors from a 47k/8.2k voltage divider rather than from the 50V rail? The base-collector junctions are forward-biased.

 

[flippineck]

Hmm.. ok I can see that now..

It's because, that's how I interpreted fig. 12 in Delta Prime's link

I thought I was extrapolating from a circuit intended for mains AC, to a DC version. I figured if 'neutral' (0V) was the top rail, and 'hot' was the bottom rail, and "the circuit measures power on negative cycles only", then when the circuit is measuring power, in that instant the top rail must be positive wrt the bottom rail. & I essentially just copied the circuit 'stuck in a negative half cycle' in an attempt to come up with DC version.

And then, the simulation seemed to work in a basic sense (if with bad scaling/linearity) - so I gave it no further thought.

Tried to make the junctions biased the other way, seems to work, but I still need to calculate more appropriate resistor values I think, to get the scaling and linearity right and keep the power draw down: https://tinyurl.com/ywp92jgq

danadak - that's a beautiful pic!​

 

[flippineck]

Tried again to look at the original National Semiconductor circuit diagram and convert it to a 'snapshot' at the most negative point of the AC cycle:

https://tinyurl.com/yocn73bh

It really would seem that those base-collector junctions are forward biased.

I've had this circuit behave a bunch of times in the simulator but there's always a gotcha. Really wish I could understand what's going on with it.

 

[danadak]

I think V of Rsense is not producing enough G when using a 1 mOhm Rsense (R6) to generate
needed V to run the diff pair. When I hit 1 Ohm for Rsense I get a poor but distinctive
somewhat linear relationship of Pload versus Rload.



Here is same but running much high load current. The 1 Ohm Rsense (R6) is
as before the better choice.



I dont think design is or was fully tested. Note using the 2N3904 models
w/o alteration created a matched pair.

A solution https://www.analog.com/en/technical...simplifies-design-of-power-control-loops.html

Of course with a micro and simple sampling this becomes an fairly easy problem to solve.


Regards, Dana.

 

[flippineck]

Noticed I've got my power supply connected the other way round to yours, I'll have another look at the way I've been thinking the transposition from AC operation to a DC circuit. Wonder if this is the root of my incorrect transistor biasing too.

LT2940 - Common-mode range 4-80V is pretty suitable! Looks like the part can be had quite cheaply too. I don't think I need to look at power much under 4V, with it being solar panels I'm looking at I don't suppose there will be many amps there below such a voltage. It's a definite contender, thanks! (and to me, it seems easier than setting up and programming some kind of microcontroller)

 

[danadak]

The power is the ac line, so only matters where Rsense is and Vbase to the diff pair within its CM
range.


Regards, Dana.

 

[flippineck]

I might try using some form of op-amp to handle the sense resistor, and maybe use a dedicated multiplier chip. I found something along those lines in an old ETI edition..

https://www.worldradiohistory.com/U.../90s/Electronics-Today-1998-No-13.pdf#page=26

Page 26