Lemme see...1.27mV error gives a current ratio error of 5 percent which is not too hot,to say the least. If one wanted one percent worst case current ratio error, one would need to match Vbe to 260uV or better...
But, there IS a dual source chip: LM13700, just strap a program current resistor to it and put three resistors in a string to bias both amplifier sections inputs with 1V.
You're reading the intro, not the actual data. Look at the FIRST item on page 2, under the column for the _LM334_.
Cut-and-pasted from that datasheet: "Set Current Error, V+=3D2.5V, 10=C2=B5A =E2=89=A4 ISET =E2=89=A4 1mA = 6 %"
0.02% is typical. The _guaranteed_ spec is 0.1% (max) change in set current per 1 volt change in output voltage.
At 1mA output that's 1uA/V, or R.out=3D1meg.
LTSpice says to a first order that R.out for my circuit's fed-back output is about 80meg, or 12.5nA/V =3D 0.0013% regulation on a 1mA set current.
That simulation isn't realistic for various reasons, esp. for the non- fed-back transistor--that output's regulation would degrade drastically with differential heating. I guesstimate an error of ~.
2% / volt. That's what Phil was pointing to up-thread.
I suspect your present LM334 circuit is worse under that condition-- the IC will heat up, and its external sensing diode will not promptly or accurately follow.
The fed-back output on my ckt, OTOH, should greatly out-perform the LM334's regulation under all conditions.
It all depends what you want.
I made it up. I don't have specs--you'd have to calculate them.
If super accuracy and performance are important, I'd use something else--I was just trying to suggest something ultra-simple, yet slightly better than what you already have.
More to the point, I don't see a simple, as-accurate way to transform my topology into a current source. That would take a more complicated circuit, which you've already ruled out.
Excuse me, that's bullshit. You have misunderstood the data, and decided that I haven't read it.
Sheesh!!!!!!!! Do you not understand that +/- 3% is a total of 6%, not +/- 6%?
The datasheet PLAINLY says +/- 3%. The datasheet PLAINLY does NOT say +/- 6% in the line you pasted, yet you said you infer that to mean +/- 6% ??????????? Quoting you: "I infer they mean +/- 6%, not 0 >accuracy". The datasheet provides numbers - there are no numbers
You misunderstand what the datasheet says, again. You seem to be referring to the "Average Change in Set Current with Input Voltage" portion of the datasheet. It says _Input Voltage_, not _output_ voltage. Huge difference.
That seems like a comparison of apples to pine trees. The 334 circuit is not driving a mirror - the "non-fed-back-transistor" condition does not exist for it.
The 334 circuit measured way better than your 2% estimate, and way worse than your .0013% figure from LTSpice. So, if the 2% estimate is valid for the mirror sink circuit, it is totally off the map for consideration based on performance. It would be far far better to use the 334, based on performance alone, even if the source/sink issue didn't apply. Maybe the estimate is off by a decimal point? .2 percent would put it back on the map.
What's the modification you have in mind for feed back with your mirror circuit? You can speculate all day long about what some unspecified circuit should do. I haven't seen anything that supports the contention that the 431 circuit should greatly outperform the 334 circuit. I do not know, one way or the other, yet. My comparison will be of the
334 datasheet tempco source circuit trimmed to 1.000 mA vs the 431 circuit I mention below, with no mirror. If you have a different
431 source circuit to offer for the test, I might be able to include that.
Super accuracy has no bearing on it. I'm just trying to understand what you have in mind when you mention numberless things to indicate one circuit performs better than the other.
We're back to the simple vs simple comparison. (If your 2% estimate is correct, we're not even back to that.) Personally, I find
2 LM334's, 4 resistors and 2 diodes in 2 separate circuits simpler than your mirror consisting of 1 431, 3 transistors and 2 resistors in a dual circuit, ignoring the sink vs source issue. I suppose that's just a matter of different approaches.
Well, it's simple (3 components: add R, D & NPN) to modify the 431 datasheet source circuit. The circuit on the datasheet yields poor performance, where "poor" means at least 20% variation in Iout when Vin varies by +/- 30%. (Based on breadboard measurements.) Worse, with Vin fixed, Iout varies with change in Rload. The technical term for regulation with that circuit is "ugly". :-(
The modification yields far better circuit performance: Less than 1uA/V Iout change over a Vin range of 7.38 to 14.48 volts. (Total measured change was 6 uA) Less than 5 uA change with Rload change from 0 to 2700 ohms. It's in the same ballpark with the LM334 circuit, and might be a tiny bit better. That determination will have to wait until I build a test jig to host both circuits simultaneously. That will be on a PCB and will eliminate breadboard errors that you have to assume are embedded in the above measurements.
While still of interest, it proves hands down that the LM334 circuit is preferable due to lower parts count. There is simply not enough measured performance gain to warrant using 2 transistors, an IC, 3 resistors and a diode to replace one IC, 2 resistors and a diode. If it was driving a mirror, as your original does, it would perhaps be worth considering. That would, of course, depend on the performance delivered to the load, implying matched xsistors.
It'd be a) dumb to spec 0-6% when you could split the difference and spec +/- 3%, and b) the datasheet covers several parts, and National traditionally gives better-grade (or "typical") part specs in the cover-page blurb.
So, I take "+/-3%" in the 1rst page promo as most likely meaning the LM234.
Umm, it's a two-port network, right?
~.
No, the comparison exactly applies. See page 5 of the datasheet under the heading "Thermal Effects."
0.2% per volt, for one of the dual outputs.
is
se
ent
Yes, 0.2%/V. The text wrapped.
It already has feedback via the LMV431 input, no modification needed.
tion
be
sheet
your
it,
I missed that comment before. I see they have a cheesy source at the end of the LMV431 datasheet, and my basic sink idea too. That source is horrible--the reference's idle current bypasses the sense resistor, wrecking the regulation.
I meant to suggest a topology, not design the circuit and specify the performance. Those are easily done, but I don't know what the requirements or environment are, and you haven't given them.
Mine was 1 LMV431, 3 resistors, 1 dual transistor. 5 parts. Yours is
8.
ated
m
That's not surprising if you're using the LMV431 "source" circuit on the last page, which it kind of sounds like you are. What circuit are you using?
It's not clear exactly what you're looking for i.e. a single package, tempco, and/or current source matching etc...so until you state /what/ the currents are supposed to do, will be unlikely a viable suggestion is made.
I'm talking math, not assumption. If I may put words in your mouth, I think you are saying that you do not assume that the
6% figure on the datasheet represents a total of +/- 3%.
I've learned something here, thanks to you. I _am_ making an assumption, where I previously thought I was stating a fact. I am assuming that the +/- 3%, which mathematically is 6% total, applies to the 334 at 1mA. But seeing your perspective on it, proves to me that I am assuming, and it very well could be that the 6% is +/-6%.
Sentence # 3, page 1, and the last bullet under features. Here's sentences 1, 2 and 3: "The LM134/LM234/LM334 are 3-terminal adjustable current sources featuring 10,000:1 range in operating current, excellent current regulation and a wide dynamic voltage range of
1V to 40V. Current is established with one external resistor and no other parts are required. Initial current accuracy is ±3%."
Ok, thanks for hanging in there - I finally (I think) understand your point of view.
Umm, does the datasheet say _output_ voltage?
See (far) below for discussion, with numbers, of why it makes a huge difference. I have no idea what you have in mind with your question about a two-port network.
Your circuit drives mirror transistors and you refer to the "non-fed-back transistor".
You say you suspect the LM334 circuit is worse under that condition.
It can't be "under that condition" - there are no transistors in the circuit, and it is temperature compensated - there is feed back provided by the diode & R used for that. So I'm missing the comparison you're making.
I'm glad to hear the estimate was really .2% as I suspected, instead of 2%.
Earlier your referred to the "non-fed-back transistor", which is Q2 on your schematic. Without Q2 there is no mirror. My question asks about the mirror. I thought you had something in mind to get feedback from Q2 (and possibly Q3...Qn).
Right. My initial question required no more than what I posted. The mirrored sink topology is fine, just not applicable to my question. The ensuing conversation whet my appetite to investigate your idea, when modified as a source, and your discussion of accuracy, stability etc. is germane to that.
There have been good breadboard results so far, but the thing has got to be soldered on a PCB to make reliable measurements of single digit uV and uA changes. What I see on the breadboard is enough of a go/no go test to warrant the better testing environment.
Right. But I get 2 separate temperature compensated sources. There's no compensation in the mirror. The dual transistor might be good enough, but it is simpler to me to include the tempco than to worry about whether or not the dual transistor is good enough. And to me, it is simpler to have two independently trimmable sources, rather than a single dual source that cannot be independently trimmed. Still, that's just different preferences. I can see someone preferring to trim one source instead of two.
I used that (the "source" from the datasheet) and commented on it - it's a dog. Then I added a PNP, cap, diode & Rs like this:
A 10 turn (or more) pot is the key to trimming it to exactly
1.000 mA in the test circuit. Once set, the thing held regulation from Vin 7.38 to Vin 14.48, and from Rload 0 to Rload 2700. (The datasheet source didn't hold regulation at all with Vin and Rload variation.) Total measured change while varying Vin over the above range was 6 uA. That's +/- 3uA, not +/- 6 uA. Total measured change when going from Rload = 0 to Rload = 2700 was 5 uV. Not +/- 5 uV. Down to 995 uA with Rl = 0; 1000 uA at Rload = 2700.
Those numbers are all based on the circuit built on the breadboard. They really have to be based on a soldered PCB circuit before I'll be satisfied.
Getting back to the mistake of treating "Average Change in Set Current with Input Voltage" as _output_ voltage: in the example above, with Rl = 2700: _input_ voltage changes from 7.38 to 14.48 or 7.1 volts _output_ voltage changes from 2.6919 to 2.7081 or 16.2 mV
There is a tremendous difference between delta Iset/Vx if you use Vout instead of Vin for Vx.
Yes. Wouldn't it be nice if it was all rolled into a single chip containing 2 separate sources!
(Note to self: No good deed goes unpunished, especially trying to help people on SED.)
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058
email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net
It does seem ludicrous that they show the "plot" of peak output current at 1000 uA when they earlier show 650 uA as the max. I'll believe the 650 uA over the supposed "plot" for 2 strong reasons: 1) I can read the 650 figure easily. The "plot" is very difficult to read at the 10^3 line with Iadc = 1000 uA.
2) If we assume the graph is right, and the peak output current figure of 650 uA on page 2 is wrong, we're still at the very end of the range - and that's for _peak_ output current. It is not prudent to run parts at their max, nor to use a maximum peak current figure as the current that can be supplied continuously.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.