I am working on a project were it is useful to have accurate matching of resistors throughout. The project I am building is SY's Equal Opportunity MM Preamp. This design is a fully balanced design and, since noise rejection is key in phono preamps, the better matching of all components, the better the noise. My goal was to have 0.1% match for each resistor pair in the signal path.
I found an approach online for building a bridge out of non-precision resistors. The idea was to start with random resistors and measure voltage across the bridge, while changing one resistor position through multiple resistors. When you find two resistors that have equal voltage across the bridge, then those resistors are equal. Using those two resistors as either the top or bottom of a new bridge, you can match resistors in the other half of the bridge by looking for zero volts across the bridge.
I tried this approach, and although it seemed to work, I did not have high confidence in my results. Even though I used a regulated power supply and timed the point of taking a measurement, the bridge tended to drift so much that I wasn't sure I had a good result.
I just picked up a Keithley 2015 Multimeter and have not yet installed the resistor pairs that I measured with the bridge. I thought I would compare the bridge results with the Keithley. For those of you unfamiliar, the Keithley is a 6.5 digit benchtop multimeter with enough precision and accuracy to easily test 0.1%. It also has Kelvin resistance measurement to further improve accuracy and stability.
The resistors that I had matched with the bridge were 8.2k Vishay RLR Series and I used Kelvin leads to measure on the Keithley. On measuring the two pairs with the Keithley, the difference within a pair was 0.015 % for the first pair and 0.0002% for the second. At that level, the readings from the Keithley are at the edge of it's accuracy, but you can clearly see that the bridge method works, even though it is time consuming and subject to drift. I would say that the bridge method is very dependent on procedure, but those results were achieved in my first attempt at bridge measurement.
Naturally, I'm happy to have the Keithley because it takes so much less time and gives me confidence in the results. But a good 6.5 or higher digit multimeter is expensive, even used, so I hope this gives you confidence to give the bridge a try.
I found an approach online for building a bridge out of non-precision resistors. The idea was to start with random resistors and measure voltage across the bridge, while changing one resistor position through multiple resistors. When you find two resistors that have equal voltage across the bridge, then those resistors are equal. Using those two resistors as either the top or bottom of a new bridge, you can match resistors in the other half of the bridge by looking for zero volts across the bridge.
I tried this approach, and although it seemed to work, I did not have high confidence in my results. Even though I used a regulated power supply and timed the point of taking a measurement, the bridge tended to drift so much that I wasn't sure I had a good result.
I just picked up a Keithley 2015 Multimeter and have not yet installed the resistor pairs that I measured with the bridge. I thought I would compare the bridge results with the Keithley. For those of you unfamiliar, the Keithley is a 6.5 digit benchtop multimeter with enough precision and accuracy to easily test 0.1%. It also has Kelvin resistance measurement to further improve accuracy and stability.
The resistors that I had matched with the bridge were 8.2k Vishay RLR Series and I used Kelvin leads to measure on the Keithley. On measuring the two pairs with the Keithley, the difference within a pair was 0.015 % for the first pair and 0.0002% for the second. At that level, the readings from the Keithley are at the edge of it's accuracy, but you can clearly see that the bridge method works, even though it is time consuming and subject to drift. I would say that the bridge method is very dependent on procedure, but those results were achieved in my first attempt at bridge measurement.
Naturally, I'm happy to have the Keithley because it takes so much less time and gives me confidence in the results. But a good 6.5 or higher digit multimeter is expensive, even used, so I hope this gives you confidence to give the bridge a try.
Last edited:
You might be shooting yourself in the foot, here... Resistors lacking excellent stability specs may be matched by selection, but you can't trust them to stay that way. I found out the hard way, by designing a product that depended on 0.1% resistor matching to meet its specs and then sorting 1% metal film resistors into 0.1% bins prior to product assembly. The mere act of soldering those resistors into the PCBs caused enough drift to force many boards to be rejected at test time. It's usually better to just pay up for excellent resistors when you need 'em.
BTW a much cheaper meter with a delta or relative measurement function is extremely useful in these matching cases. what you pay for is absolute accuracy and is not typically needed for audio amps using feedback ratios. even a select in test part can speed up the whole process after all this DIY not a precision 'hands free' production run. as usual limited budgets and ingenuity can take you to the same places indeed.
Last edited:
Mike,
Very interesting point. I had never heard of resistors changing value during soldering. I would think that how much they change is less of a factor of their sorted tolerance (0.1% or 1%) than it is the quality of the design and the specs that confirm that. Does that fit your experience?
For example, the Vishay RLR Series that I mentioned are Mil spec designed for high frequency and low noise, but are only available in 1%. I'm not sure what spec would refer to the change that you mentioned. They do have a MIL spec for solderability.
Jac
I agree in spirit. That's also the reason I am proposing the bridge approach. It is essentially a way to improve matching with a simple 3.5 or 4.5 digit multimeter. Just more time consuming.
analyze the finished product for a null procedure instead time consuming nulling all the parts individually before assembly.
The 1% resistors in this case were a 1/4W axial-lead commodity type from Mouser. Not sure if they were 100ppm or 50ppm. Since the switch to 0.1% resistors, there were zero rejects at production test. No changes were made in assembly procedures. I suspect that allowing extra lead length and/or heatsinking the leads during soldering would have made improvements, but I now consider my original design philosophy to be deeply flawed. Resistor manufacturers know what they're doing (surprise!) and they don't make the product any better than it absolutely has to be. Anything else would be suicidal in such a competitive business environment.
solder a string of nominally identical resistors together.
Pass a current that roughly matches the operating current the circuit will use. A CCS is good for this, if you allow a bit of time for the CCS to stabilise to it's working temperature.
Then just measure the voltage drop across each resistor.
A 2000 count DMM can give you 0.05% resolution
A 20000 count DMM gives 0.005% resolution.
Repeating the measurements two or three times helps remove/average the variations.
If the resistors are very low value you will find that the lead out can be a significant proportion of the total resistance. You must measure each resistor with the test probes at the same distance apart, to remove this source of error.
I reckon I can match 0r22 power resistors using a 50000 count bench DMM to better than ±0.01% using this method.
Pass a current that roughly matches the operating current the circuit will use. A CCS is good for this, if you allow a bit of time for the CCS to stabilise to it's working temperature.
Then just measure the voltage drop across each resistor.
A 2000 count DMM can give you 0.05% resolution
A 20000 count DMM gives 0.005% resolution.
Repeating the measurements two or three times helps remove/average the variations.
If the resistors are very low value you will find that the lead out can be a significant proportion of the total resistance. You must measure each resistor with the test probes at the same distance apart, to remove this source of error.
I reckon I can match 0r22 power resistors using a 50000 count bench DMM to better than ±0.01% using this method.
Andrew,
A very interesting method. One advantage that I see over the bridge is that there may be less drift because every resistor sees the same current and you can let everything reach equilibrium before starting measurement. A potential drawback would be the need for a high voltage power supply or a low current requirement for a long string of high resistance, but then we don't usually push a lot of current through high value resistors.
I learn something new everyday.
Jac
I went looking, after Andrew's comment, at some of the datasheets for resistors in my kit. As Andrew points out, there is sometimes/often a spec in the datasheet.
KOA Speer, MF series, has a spec of +/- 0.5% +/- 0.05 Ohm for max change in resistance due to soldering. Interestingly, there is no different spec between 1/4 and 1/2 watt versions, nor for 1% and 5% tolerance resistors.
The military standard seems to specify how they test, but not the limit. Interesting that the mil-spec Vishay RLR doesn't quote a spec, only that they meet the MIL spec.
Vishay MBA/MBB Series is a "professional" industrial series. They have 3 stability classes with different solder resistance change. It's not completely clear, but these stability classes appear to coincide with their resistance tolerance.
Resistance tolerance of 0.5% gets stability of +/-0.1% + 0.01 Ohm. A tolerance of 1% gets you stability of +/- 0.25% +0.05 Ohm. It appears that a tolerance of 2 or 5% gives you a stability of +/- 0.5%. There are similar stability specs for vibration, overload, etc. What is interesting is that the stability class isn't correlated to temperature coefficient, at least in this series.
I guess this proves two things. Although it seems to vary by resistor series, BinaryMike is right that soldering can shift your resistance by a significant amount, enough to screw up your careful pre-solder matching. The second thing is that I need to read the datasheets even closer.
Jac
If you tack on a nominally same value high accuracy resistor to the string, you can do fairly accurate measuring of the DUTs.
I have a small stock of 0.1% resistors over a fairly wide range of common values and an even smaller stock of 0.3% capacitors.
what is a significant amount? most reputable resistance products hold to the stated tolerance, if its going off spec then yer either soldering it wrong or got the wrong parts for the job. you seem to want to achieve best matching outside the limits so your testing matters and if the parts change after that then you just entered the "DIY parts biz " building super unspec'd parts. I still say one can avoid a lot of grief (time and money) by nulling ALL the parts after assembly with a smart test procedure and product qualification over the expected temperature range.
Last edited:
I'm not trying to re-invent the wheel here. I'm just a DIY guy building a project and learning along the way. In my previous projects, close matching was not critical and every time that I checked a resistor before and after soldering, it was the same within my ability to measure it. BinaryMike brought up the issue of resistors changing value after soldering. I am just learning the implications of that through this discussion and further reading of the datasheets.
To answer your question. I'm sure all the parts are within their spec. I understand that in most DIY audio, the design wouldn't be sensitive to the small change in resistance during soldering. In this project, the designer requested that the input resistors of the balanced circuit be matched to 0.1%, I assume to meeting his noise goals with balanced input impedance.
I got carried away. I figured, I might as well match all the resistors balanced signal chain, along with the other active and passive components. So, in this case, a change of up to 0.5% during soldering, if both resistors in the pair don't change the same amount and direction, would take me out of the specified match. I did mean to imply that 0.5% change during soldering is significant in all situations.
When you say "nulling ALL the parts after assembly with a smart test procedure", I think of a design that has a parallel potentiometer in one leg and balancing the two legs after soldering. Is that what you had in mind? If not, please explain it a bit further.
Jac
look at the balance sensistivity from the ratios of initial tolerance then either use a small value pot and / or select in test resistor. check the sensitivity due to temperature. IDK ask the designer for a good check out test. perhaps short one input measure the output voltage repeat on the other input then proceed to nulling as needed. a procedure may need to be built in the PCB for ease of final check out. again since this is DIY so the 'touch labor' isn't a big deal on a small run. this touch labor is typically undesirable for large production runs where the accuracy is normally pushed down to individual parts vendors (what yer doing) or live with the reduced specs of the final product is acceptable. also super high precision means a custom thick film assy where the ratio depends on the resistors geometry only and process variations is nulled by design.
Balanced impedance connections requires the source to be balanced impedance and the receiver to be balanced impedance.
Impedance means capacitance as well as resistance. Treat 0.1% as the minimum required for effective common mode attenuation. If you can get better than 0.01%, then go for that.
Select your capacitance to be matched as well. This becomes critical at the upper end of the audio frequency range. Here I would relax the cap matching tolerances by a factor of ten.
Have you read W.Jung?
He likens the source/receiver to a balanced bridge where the interference attenuation increases as the ratio of receiver input impedance to source output impedance increases. 10ohm source to 1M receiver is better than 100ohm source to 100k receiver.
I have since seen the same argument adopted by another Balanced Impedance Design author.
Last edited:
Infinia,
I get what you are saying. It is clear that you are coming from a more knowledgeable and more professional viewpoint than my meager experience. Thanks for the explanation.
Andrew,
I haven't read Yung, but I have read others with a similar approach. Of course, that gets tricky with a moving magnet phono preamp, such as this project. MM cartridges typically have output impedance on the order of 1k and the accepted norm for preamp input impedance is 47k.
As for the approach you outlined for impedance matching, it describes what I am trying to do. Also include matching actives to that the signal gain remains balanced. I look forward to seeing how well I do.
Jac
I get what you are saying. It is clear that you are coming from a more knowledgeable and more professional viewpoint than my meager experience. Thanks for the explanation.
Andrew,
I haven't read Yung, but I have read others with a similar approach. Of course, that gets tricky with a moving magnet phono preamp, such as this project. MM cartridges typically have output impedance on the order of 1k and the accepted norm for preamp input impedance is 47k.
As for the approach you outlined for impedance matching, it describes what I am trying to do. Also include matching actives to that the signal gain remains balanced. I look forward to seeing how well I do.
Jac
Signal matching has nothing to do with Balanced Impedance connections.
The 10r : 1M would not be applicable to cartridge level signals.
The 10r : 1M would not be applicable to cartridge level signals.
Resistors change with voltage temperature humidity and time. What starts as a .1% resistor can be as much as 1% from its original value a year later. Mechanical stress changes them. Humidity can change them a lot. The worst part is that they won't coorperate and change in the same direction. Even ultra close tolareance parts like Vishay foils can change over time. In fact the NBS's standard resistors (made in 1933) have a known drift value now that they can be measured to the necessary accuracy.
Match the parts using close tolerance parts and add a trim to balance them. Typically a trimmer resistor of 5% of the value and reduce the resistor by 2%. Add a trim cap in the right place and you can get 60 dB or better CMRR. If you want to get better study the circuitry of a Tek 7A13. 7A13 - TekWiki It gets 80 dB CMRR and its all discrete.
Match the parts using close tolerance parts and add a trim to balance them. Typically a trimmer resistor of 5% of the value and reduce the resistor by 2%. Add a trim cap in the right place and you can get 60 dB or better CMRR. If you want to get better study the circuitry of a Tek 7A13. 7A13 - TekWiki It gets 80 dB CMRR and its all discrete.
Thank you, Demian. You put it into a nice overall perspective. I can clearly see the importance of designing for change over time and why production designs focus on minimizing component sensitivity.
I started out this thread exploring my ability to accurately match resistors and learned a lot about how resistors change over time and conditions. Thanks.
Jac
I started out this thread exploring my ability to accurately match resistors and learned a lot about how resistors change over time and conditions. Thanks.
Jac
- Status
- Not open for further replies.