" precision " ballasting resistors.

[mrpackethead]

Im considering a design idea at the moment, which requires a pair of closely matched resistors.   The pair could fall in the range 1.45 - 1.55 Ohms, will need to be rated at 0.5W,  Ideally there will be less then 0.1% difference in them.

I've had a hunt for such resistors and while they do exisit on a datasheet they have very long lead times, and are basically product unobtainable.

If i bought a reel of 1% parts, would it be insanity to consider measuring them, sorting them into 'range' bins, and then getting closer pairs?      They would still need to be placed on the pcb as pairs, and that is quite a lot of work, since they wont' be on a tape any longer ( though they coudl be loaded into custom trays and picked up, its still quite a lot of work.   And yes, i potentially have to place thousands and thousands of them.

[Miyuki]

If you buy reel they will be very similar
Basically in same batch they are almost same, but wary batch to batch most times

[T3sl4co1l]

You can get 0.1%'s (and .2 and .5) in larger values, and connect them in parallel.

Since you're doing that anyway, it seems the more effective (better guaranteed) method?

The metrologists call it a "statistical array" -- if the resistors are independent variables, then resistors in series add arithmetically while their errors add vectorially (i.e., RMS).  The expected tolerance of the resulting chain is T / sqrt(N), so you'd need 100 resistors to get the next-decade improvement from 1% resistors.

Obviously, such a strategy is susceptible to systematic errors, like if reels/batches have a bias.

The relationship for resistors in parallel is slightly different, because smaller values dominate.  This is obvious when you have 1 ohm in parallel with 100, but less pronounced when you have 1.01 in parallel with 0.99.  You should get comparable results this way, in any case (i.e., a hundred 100R 1%'s in parallel should be close to 1R 0.1%).

Tim

[mrpackethead]

100 Resistors in parrallel might be a bit problematic for other more pracgtical reasons.

[soldar]

If i bought a reel of 1% parts, would it be insanity to consider measuring them, sorting them into 'range' bins, and then getting closer pairs?

I have done this in the past. The question is how many you need to sort and whether it is time/cost efficient for you.

[Kleinstein]

Even if one measures the actual resistance of the new parts, the resistance after soldering can be a different thing.

At 1.5 Ohms and 0.1% tolerance one really has to look at the wires / traces, especially if having resistors in parallel.

One could also consider trimming after soldering the main parts in place.

[mrpackethead]

If i bought a reel of 1% parts, would it be insanity to consider measuring them, sorting them into 'range' bins, and then getting closer pairs?

I have done this in the past. The question is how many you need to sort and whether it is time/cost efficient for you.

Quite a lot. In fact 96 resistors per board, and probaly making 50 boards in a go.

[rstofer]

In a batch of 1% resistors, all of the 0.1% resistors will be sorted out and sold for more money.  You will likely find a bimodal distribution with a gap where the 0.1% resistors should fall.

I wonder if it wouldn't be easier to use PCB traces given the low value desired.

https://www.allaboutcircuits.com/tools/trace-resistance-calculator/

Or, get close with resistors and tune with trace length.

[coppercone2]

dave jones has videos about this and he actually does a test, older videos.

the trap for what you are doing is labor and drift. You can do a monte carlo approximation of the drift effects.

I.e. the 100.1 and 100.09 parts might be close, but if you choose 100ppm resistors, one might drift 1ppm and the other 100ppm... or they might add up and one will drift +100 other -100.

if you use them in a design make sure the drifts don't cause problems, unless you measure them too, but it will get crazy unless you make elaborate automatic thermal test equipment. Don't gamble with the manufacturer giving you something better then they specify, even if its usually the case (and design for something worse actually).

[T3sl4co1l]

In a batch of 1% resistors, all of the 0.1% resistors will be sorted out and sold for more money.  You will likely find a bimodal distribution with a gap where the 0.1% resistors should fall.

I wonder if it wouldn't be easier to use PCB traces given the low value desired.

https://www.allaboutcircuits.com/tools/trace-resistance-calculator/

Or, get close with resistors and tune with trace length.

Bimodal distributions were all the rage back in the carbon comp days.

Laser etched resistors (carbon film, metal film, foil..) take longer (= more expensive) to make more precise, there's no money in making billions of super-rough values and sorting them.  The value is measured once, while it's being trimmed, then it's coated and packaged.

Tim

[mrpackethead]

In a batch of 1% resistors, all of the 0.1% resistors will be sorted out and sold for more money.  You will likely find a bimodal distribution with a gap where the 0.1% resistors should fall.

I wonder if it wouldn't be easier to use PCB traces given the low value desired.

https://www.allaboutcircuits.com/tools/trace-resistance-calculator/

Or, get close with resistors and tune with trace length.

Bimodal distributions were all the rage back in the carbon comp days.

Laser etched resistors (carbon film, metal film, foil..) take longer (= more expensive) to make more precise, there's no money in making billions of super-rough values and sorting them.  The value is measured once, while it's being trimmed, then it's coated and packaged.

Tim

In this design the exact value of the resistors is not critical,  what is important that the pair are as close as possible.

[The Soulman]

Did you take in account the temperature stability of the resistors?
I'd use large resistors (normal tolerance) for minimum temperature rise and find a way of trimming after assembly.
Possibly with trimmer pot ( +couple resistors to define the range) or parallel mounted high value resistors chosen
strategically so that one or more can be clipped away after assembly and measuring.
Either way measuring and manual labor after assembly is mandatory.

Poor design? 

[mrpackethead]

Did you take in account the temperature stability of the resistors?
I'd use large resistors (normal tolerance) for minimum temperature rise and find a way of trimming after assembly.
Possibly with trimmer pot ( +couple resistors to define the range) or parallel mounted high value resistors chosen
strategically so that one or more can be clipped away after assembly and measuring.
Either way measuring and manual labor after assembly is mandatory.

Poor design? 

Multiple resistors is not really an option.   Each side of the pair will be part of a differential signal path that is carrying a signal clocking at 125Mhz.   I'd hate to think what having 10's of resistors in parrallel/series might do to that signal path.      Provided the resistors in the pair behave in a similar way, having them shift in value a bit over temp is not super critical.

The idea of using PCB traces was innovative. :-) though not practical.

[soldar]

At 125 MHz you want super low inductance resistors which rules out wire wound. I have seen carbon resistors drift in value over time so you could install two resistors matched to the max only to have them drift off over time. Make sure to take these two things into account.

[mrpackethead]

[Zero999]

Did you take in account the temperature stability of the resistors?
I'd use large resistors (normal tolerance) for minimum temperature rise and find a way of trimming after assembly.
Possibly with trimmer pot ( +couple resistors to define the range) or parallel mounted high value resistors chosen
strategically so that one or more can be clipped away after assembly and measuring.
Either way measuring and manual labor after assembly is mandatory.

Poor design? 

Multiple resistors is not really an option.   Each side of the pair will be part of a differential signal path that is carrying a signal clocking at 125Mhz.   I'd hate to think what having 10's of resistors in parrallel/series might do to that signal path.      Provided the resistors in the pair behave in a similar way, having them shift in value a bit over temp is not super critical.

The idea of using PCB traces was innovative. :-) though not practical.

Parallel will increase the parasitic capacitance, but reduce the inductance. Series will increase the inductance, but reduce the capacitance.

Series and parallel should theoretically keep the parasitic inductance and capacitance the same as one resistor, but in reality it won't work out because the circuit will be physically larger it will have more L and C.

125MHz is not that difficult, as long as you don't use wire-wound resistors, which will be highly inductive.

[mrpackethead]

watching daves videos was interesting;

Lets assume that; ( all to be prooved, but based on Daves info )

(a)  The distribution of resistors follows a normal distribution.
(b)  Resistors of -/+1% tolerance   have a standard deviation of 0.125 

The mean value is 1.5R

What i'd like to know is the probability that two given resistors are more than 0.2% apart in value.   my statistics is too rusty to do this.   

[spec]

Im considering a design idea at the moment, which requires a pair of closely matched resistors.   The pair could fall in the range 1.45 - 1.55 Ohms, will need to be rated at 0.5W,  Ideally there will be less then 0.1% difference in them.

I've had a hunt for such resistors and while they do exisit on a datasheet they have very long lead times, and are basically product unobtainable.

If i bought a reel of 1% parts, would it be insanity to consider measuring them, sorting them into 'range' bins, and then getting closer pairs?      They would still need to be placed on the pcb as pairs, and that is quite a lot of work, since they wont' be on a tape any longer ( though they coudl be loaded into custom trays and picked up, its still quite a lot of work.   And yes, i potentially have to place thousands and thousands of them.

Hi mrpackethead

In view of the high frequency, and quantities involved, yes it would be madness. Just buy the correct components.

[mvs]

watching daves videos was interesting;

Lets assume that; ( all to be prooved, but based on Daves info )

(a)  The distribution of resistors follows a normal distribution.
(b)  Resistors of -/+1% tolerance   have a standard deviation of 0.125 

The mean value is 1.5R

What i'd like to know is the probability that two given resistors are more than 0.2% apart in value.   my statistics is too rusty to do this.

I have played a bit with Dave's data (400 1% resistors) in excel. Only around 63% possible resistor pairs are within +/- 0.2% tolerance. This might be acceptable if only one resistor pair is used per board, but not in case of 48 pairs.
You need 0.5^(1/48)=98,6% pairs in spec to produce half of the boards w/o rework.

[mrpackethead]

watching daves videos was interesting;

Lets assume that; ( all to be prooved, but based on Daves info )

(a)  The distribution of resistors follows a normal distribution.
(b)  Resistors of -/+1% tolerance   have a standard deviation of 0.125 

The mean value is 1.5R

What i'd like to know is the probability that two given resistors are more than 0.2% apart in value.   my statistics is too rusty to do this.

I have played a bit with Dave's data (400 1% resistors) in excel. Only around 63% possible resistor pairs are within +/- 0.2% tolerance. This might be acceptable if only one resistor pair is used per board, but not in case of 48 pairs.
You need 0.5^(1/48)=98,6% pairs in spec to produce half of the boards w/o rework.

I only need each pair ( not all the 48 pairs ) to be matched.    ( sorry if that was not clear ).   

I wrote a bit of python to programatically work out the statistical likelyhood of picking two resistors at random that were closely enough matched.   on the reel of 1% parts, there is about 55% chance that the two parts would be within a match.  Thats not really good enough, as nearly half of the random pairs would not match. 

Given that obtaining the .1% parts wont' really be possible, sorting them is the last option.  Grrh.

Parts will arrive on a 8mm reel.    I'm thinking i'll use a PNP machine to 'unload' the reel onto a custom made tray, that has cut outs the size of the resistor..  It should take about 10 minutes to unload it into the tray.
I'll make a little test tool, that has a pair of pogo pins, and with the help of a couple of index pins in the tray will line up nicely. 

In stead of connecting to a multimeter, i'll make a small test circuit, that will simplfy it down, and will just beep two tones,  indicating if its in the most common central bins, or if its outside. If its outside, we will just use a small vacum pick up tool to discard them.    Probably will end up with keeping about 60% of the parts, and chucking the rest.    Given the very low cost of the the parts, compared to the labour.

The tray can then go back on teh pnp machine.. I wont' need to tell it when there is missing parts in teh tray, as it will figure that out when it can't pick them up.    and will retry in the next spot.

[mrpackethead]

In view of the high frequency, and quantities involved, yes it would be madness. Just buy the correct components.

unforunatly the 'correct' components are largely unobtanium,  it would be great to buy low value, high precision metal film resistors but they are not stocked, and have lead times of >30weeks.   Thats not something i can decision around.

[Nauris]

0.1% of 1.5 ohms is 1.5 milliohms and at 125 MHz that kind of matching is not possible to maintain with normal smd resistors. It is too low resistance

Back to the drawing board