On Tuesday 26 October 2004 05:03 am, Olin Lathrop wrote: > Jose Da Silva wrote: --snip-- > The general method is to pick the first resistor to be the smallest > standard value guaranteed to not be smaller than the desired value. Fo= r > each subsequent resistor but the last one, you measure the resistance s= o > far, then add the minimum parallel value to guarantee the result will n= ot > be less than the desired value. For the last resistor, you pick the va= lue > that results in the least overall error. Only two 1% resistors are nee= ded > to make a .1% resistor within the standard value range. 2 resistors will do, but the granularity of your result will be larger pl= us=20 less likely that you strike identical values. Suppose you put 2 resistors in parallel. 27ohms + 27ohms (+/-1 ohm) Your known value is now 13.5 +/- 0.5 Suppose you put 10 resistors in parallel. 135+...+135 ohms (+/-1 ohm) Your known value is now 13.5 +/- 0.1 I am more likely "able" to reproduce 2 identical resistors of 13.5+/-0.1=20 using 10 versus using only 2 resistors. Let's suppose my ohmmeter is accurate to 3 digits.... I don't like dealing with fractions of ohms here, but just for the sake o= f=20 using the above 13.5 ohm result. Say we get 27.4 || 27.4 =3D 13.7 and we are aiming to be at 13.5+/-0.1 so= to=20 get within the same window, I need to put another resistor in parallel of= =20 924ohms (ideal) or 1863 just to get within the window. Now we build a 2nd 13.5 resistor and find, 26.7, so we need to put someth= ing=20 at 27.3 to be accurate... let's use 39.1 || 90.5=20 Can you reasonably say that the two 13.5 resistors are identical once I s= tart=20 pumping power thru them? Once the temperature starts going up due to the power, I don't think that= the=20 27.4 || 27.4 || 924 has the same resistance as the 27.3 || 39.1 || 90.5=20 because they are going to have different loads of power, therefore temp=20 difference, therefore resistance differences. I would think that the "complicated" 10 resistors stands a better chance = of=20 being identical if comparing 1 versus ther 2nd, plus the bonus is reduced= =20 temperature generated due to reduced power across each resistor.... You are closer to ambient temperature with 10 versus 2. Likewise, you are closer to ambient temperature with 20 versus 10. > Note however that this assumes you have the ability to measure a > resistance somewhat better than your desired end tolerance. Of course. Your numbers are only as good as the tools you use to measure them with. So, if your meter is only good enough to 2 digits, then 2 digits is all y= ou=20 can trust, but again, the precision is more trustworthy. You are closer t= o=20 the target value if you spread it over 10 versus 2. > Also note that this discussion is not relevant to the original post. If you recall (see ###1), someone else asked a question of where to find=20 0.1% resistors, so if you can't find them, then you can build them. > The OP wanted to measure very small drift in a current value over seve= ral > hours. He was interested in precision, not accuracy. To get back "On Topic"... ;-) After reading the solution you suggested in another thread.... I fully agree with you that resistor accuracy is not relevant considering= you=20 have your eye set on trying to propose a time-based weight-drop solution. Let's put the resistor stuff aside.... ;-) However, a weight-drop apparatus is only able to measure one instance at = a=20 time. you got to reset the apparatus, and therefore you have less guarant= ee=20 that you set it up identical to the 1st condition. You can't measure=20 steadily over hours of time, therefore you may collect some numbers, but=20 those may or may not be enough to correlate/prove/disprove something. You can't build the apparatus as part of the vacuum chamber... it has to = be=20 free-standing within the chamber, due to air crushing of the chamber. You can't discount for vibrations or random events which can be filtered = out=20 because it is a one-shot type measurement. Finally, you are trading one set of drift problems with another set of dr= ift=20 problems because you are building a 1-of-a-kind apparatus that you can't=20 reference against an identical 1-of-a-kind apparatus. You can't discount=20 voltage, temp, air pressure,or other drifts because you got nothing to kn= ock=20 your drift values against. If you were able to build a 2nd "reference" to discount drifts, then a=20 weight-drop solution may be the way to go, but how do you propose that? you can't build a 2nd weight-drop apparatus in the same room as the 1st=20 because they are both affected by gravity. > The problem for the=20 > OP is dealing with various causes of drift, not so much initial accurac= y. For myself, I prefer the 1st solution posted (see ###2) and I would consi= der=20 that a better suggestion since it can be measured over hours, you can=20 filter-out noises like vibrations, plus you can build 2 identical apparat= us. One is your "no load" for reference purposes to discount for various drif= ts,=20 the second is your "under load" device measured over time. You build both with common things, such as common power supply input, com= mon=20 A/D, common......etc. so that you can discount drift. ...whatever affects= the=20 refence affects the measured. Some things won't be common between both=20 circuits, but you will aim for identical in function. Taking the example of the 13.5 ohm resistor above.... Is [ 27.4 || 27.4 || 924 ] identical to [ 27.3 || 39.1 || 90.5 ] ...ummm, no, because power is going to go thru the various resistors at=20 different rates, therefore putting them at different temps, therefore giv= ing=20 different resistances. You'll probably have a better chance of identical with 10 resistors sprea= d=20 across.... even better, perhaps 20 or more in parallel. .....perhaps the OT resistor stuff may not be so OT after all. just my 2 cents. >###1------------------------------- >At 04:54 PM 10/25/2004 -0400, you wrote: >=A0 ... >>You might look at using a 24 bit ADC with some ultra-stable resistors. >>If you need to create a virtual ground, you'll need to add some additio= nal >>complexity- a precision high-gain op-amp and a buffer, most likely. > >Spehro, where can a hobbyist get high stability high precision (e.g. 0.1= %=20 >or better) resistors? I've had a couple of projects in which I needed 0.= 1%,=20 >but haven't been able to find a source. One of the surplus joints=20 >advertised 0.1% stuff, but when it arrived, it was just 1%. I complained= ,=20 >they cancelled the bill, but they are still advertising the same stuff=20 >erroneously. >------------------------------- >###2------------------------------- >Re: [EE:] Sensing very small current changes > Date: Mon Oct 25 13:54:04 2004 > From: Spehro Pefhany > To: "Microcontroller discussion list - Public." > Reply to: "Microcontroller discussion list - Public." > >At 03:54 PM 10/25/2004 -0400, you wrote: >>I'm looking for recomendations on how to measure a current with >>extremely fine precision - nanoamps. The signal I'm interested in will >>be at about 50 ma all the time, but I'm interested in sub microamp >>drift over a period of several hours. Expect dynamic range to be less >>than 50ua. >> >>Sugestions ? - search keywords? > >You might look at using a 24 bit ADC with some ultra-stable resistors. >If you need to create a virtual ground, you'll need to add some addition= al >complexity- a precision high-gain op-amp and a buffer, most likely. > >1nA represents about 0.02ppm, so any kind of temperature or other drift = in >your reference or measurement circuit will quickly swamp that. At 100nA >you have a fighting chance without getting into really exotic stuff, but >you're still essentially talking about building a 6 digit DVM. > >Best regards, >------------------------------- _______________________________________________ http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist