This thread is a sequel of this one:
http://www.diyaudio.com/forums/equipment-tools/266421-power-oriented-cap-meter.html
I have finalized and documented my second-choice cap-meter.
It operates on completely different principles from the other one, being of the counting variety.
It is also less sophisticated, being only able to display one "generic" type of capacitance.
All in all, not very original, apart from one or two subtleties and peculiarities, the most notable one being the peak measuring current, reaching 500mA on the three highest ranges. This is about 100x what is generally found on this kind of instrument.
It is however insufficient to exercise but the most damaged samples it is presented with: another 100x factor would be required for that, which is not really practical in this format: the supply needs to actually provide the test current, for periods of 100's of ms at a time.
As I didn't want to build a monster of an instrument, I shied away from the manly option and contented myself with a miserable 500mA....
It works nicely though, and allows interesting comparisons with its less informal elder brother. I'll come back on that later.
The component choices were again dictated by my stock contents, hence the vintage TIL311's and slightly insane 5-digit resolution: since it came for free as I had half a counter spare, I implemented it, but it is certainly not the most reasonable option.
It does save one range though, which is not negligible since it allows for a very simple range selector.
If someone decides to build his own variant, other counting/display methods should be chosen, either based on a microcontroller or a LSI counter for example
http://www.diyaudio.com/forums/equipment-tools/266421-power-oriented-cap-meter.html
I have finalized and documented my second-choice cap-meter.
It operates on completely different principles from the other one, being of the counting variety.
It is also less sophisticated, being only able to display one "generic" type of capacitance.
All in all, not very original, apart from one or two subtleties and peculiarities, the most notable one being the peak measuring current, reaching 500mA on the three highest ranges. This is about 100x what is generally found on this kind of instrument.
It is however insufficient to exercise but the most damaged samples it is presented with: another 100x factor would be required for that, which is not really practical in this format: the supply needs to actually provide the test current, for periods of 100's of ms at a time.
As I didn't want to build a monster of an instrument, I shied away from the manly option and contented myself with a miserable 500mA....
It works nicely though, and allows interesting comparisons with its less informal elder brother. I'll come back on that later.
The component choices were again dictated by my stock contents, hence the vintage TIL311's and slightly insane 5-digit resolution: since it came for free as I had half a counter spare, I implemented it, but it is certainly not the most reasonable option.
It does save one range though, which is not negligible since it allows for a very simple range selector.
If someone decides to build his own variant, other counting/display methods should be chosen, either based on a microcontroller or a LSI counter for example
Attachments
A bit of explanation on how it works:
All the timing signals are derived from a common timebase, clocked by a 455KHz ceramic resonator.
The total cycle period is the clock divided by 131072 or 1310720, depending on the range.
This results in a ~0.28s cycle time for most of the ranges (and 2.8s for the rest), which seems (to me at least) optimum for ergonomy: slower or faster is equally annoying, for opposite reasons.
When a cycle starts, the CUT begins charging through calibrated resistors, controlled a by discrete MOS inverter.
The lowest value resistors have slightly different values to account for the difference in Rdson of P and N devices.
The charge starts from zero, but the counting gate opens only when the low threshold of Vcc/4 is crossed (detection by U18). This allows for a better accuracy, since the initial conditions are positively defined.
When the upper threshold (3*Vcc/4) is reached, the circuit switches to discharge, and the counting window closes when the capacitor voltage passes below the lower threshold.
The result accumulated in the counters is then latched and displayed in quite standard manner.
In principle, the CMOS 4518's should not be capable of driving the TTL TIL311; I made a check on the TIL311 of my stock, and in fact they do not conform to the datasheet: the input current they take is much lower than a standard unit load, even lower than a LS input, more like a 74L.
Probably a later series, made to be compatible with CMOS.
Rather handy (otherwise I would have used 74HC390's), but there is a downside: they are also as slow as 74L's, which caused an unreliable display.
The hold time which is normally 40ns, should have posed no problem with the theoretically slower CMOS glue logic, but here it was much longer, hence the delay introduced by C7/R15.
The cycle time is therefore increased to 2.8s, which is unpleasant but bearable.
For the highest range of 1F, I had to make a difficult choice: without decreasing the resistance, or increasing the cycle time further to an eternity of 28s, the only left parameter was the thresholds.
It is a very poor method of action, because the time is only weakly and non-linearly influenced by the thresholds, but I had no other choice.
M4 has the job of switching the thresholds, to ~350mV and ~425mV.
They are alarmingly close and small, but there is no other way: making them closer to Vcc/2 would have been healthier, but with a prohibitive inconvenient: for the first acquisition, the capacitor would have had to be charged at Vcc/2, and this would have required 28s/2, or half an eternity: clearly too much.
Initially, I had the ambition of making this tester completely adjustment-free, but the reality decided otherwise: without adjustment, the error on this range was intolerable and I had to add AJ1 to tweak the thresholds.
Once adjusted, it is stable enough to be usable.
Two accessory indicators are included: one is the overflow, the other shows the activity.
With this design, there is about 30% overflow margin (131,000/100,000), meaning the displayed digits remain usable up to 30% in an overflow condition. This extends even further the already unusual 5 digit resolution.
The activity indicator helps figuring what's going on: since this tester is going to be used with possibly faulty capacitors, it is good to know exactly why an indication is not present, and if it is worth waiting longer.
The red LED is a pulse signal, flashing very briefly at the beginning of each cycle. It is normally not visible, unless no capacitor is connected. It lets you know that the tester is working, and that nothing is connected (capacitor open, or range much too high).
The charge part of the cycle lights the green LED. When it is present, it makes the red flash invisible. If the green doesn't blink, or does so occasionally, this means that the capacitor is shorted or too large for the range chosen.
- Control:
All the timing signals are derived from a common timebase, clocked by a 455KHz ceramic resonator.
The total cycle period is the clock divided by 131072 or 1310720, depending on the range.
This results in a ~0.28s cycle time for most of the ranges (and 2.8s for the rest), which seems (to me at least) optimum for ergonomy: slower or faster is equally annoying, for opposite reasons.
When a cycle starts, the CUT begins charging through calibrated resistors, controlled a by discrete MOS inverter.
The lowest value resistors have slightly different values to account for the difference in Rdson of P and N devices.
The charge starts from zero, but the counting gate opens only when the low threshold of Vcc/4 is crossed (detection by U18). This allows for a better accuracy, since the initial conditions are positively defined.
When the upper threshold (3*Vcc/4) is reached, the circuit switches to discharge, and the counting window closes when the capacitor voltage passes below the lower threshold.
The result accumulated in the counters is then latched and displayed in quite standard manner.
In principle, the CMOS 4518's should not be capable of driving the TTL TIL311; I made a check on the TIL311 of my stock, and in fact they do not conform to the datasheet: the input current they take is much lower than a standard unit load, even lower than a LS input, more like a 74L.
Probably a later series, made to be compatible with CMOS.
Rather handy (otherwise I would have used 74HC390's), but there is a downside: they are also as slow as 74L's, which caused an unreliable display.
The hold time which is normally 40ns, should have posed no problem with the theoretically slower CMOS glue logic, but here it was much longer, hence the delay introduced by C7/R15.
- Range switching
The cycle time is therefore increased to 2.8s, which is unpleasant but bearable.
For the highest range of 1F, I had to make a difficult choice: without decreasing the resistance, or increasing the cycle time further to an eternity of 28s, the only left parameter was the thresholds.
It is a very poor method of action, because the time is only weakly and non-linearly influenced by the thresholds, but I had no other choice.
M4 has the job of switching the thresholds, to ~350mV and ~425mV.
They are alarmingly close and small, but there is no other way: making them closer to Vcc/2 would have been healthier, but with a prohibitive inconvenient: for the first acquisition, the capacitor would have had to be charged at Vcc/2, and this would have required 28s/2, or half an eternity: clearly too much.
Initially, I had the ambition of making this tester completely adjustment-free, but the reality decided otherwise: without adjustment, the error on this range was intolerable and I had to add AJ1 to tweak the thresholds.
Once adjusted, it is stable enough to be usable.
Two accessory indicators are included: one is the overflow, the other shows the activity.
With this design, there is about 30% overflow margin (131,000/100,000), meaning the displayed digits remain usable up to 30% in an overflow condition. This extends even further the already unusual 5 digit resolution.
The activity indicator helps figuring what's going on: since this tester is going to be used with possibly faulty capacitors, it is good to know exactly why an indication is not present, and if it is worth waiting longer.
The red LED is a pulse signal, flashing very briefly at the beginning of each cycle. It is normally not visible, unless no capacitor is connected. It lets you know that the tester is working, and that nothing is connected (capacitor open, or range much too high).
The charge part of the cycle lights the green LED. When it is present, it makes the red flash invisible. If the green doesn't blink, or does so occasionally, this means that the capacitor is shorted or too large for the range chosen.
Attachments
Results, comparisons, comments
It is immediately obvious that the 500mA maximum test current is widely insufficient to bring additional information compared to a "normal", low power meter.
Even in the purely linear domain, it under-performs compared to a state-of the-art vectorial meter (like modern hand-held RLC meters f.e.): the values indicated for capacitors in a good condition are larger than even the theoretical series equivalent.
How can this be?
The measurement process is slow, and involves charging resistances that can be high for the lower ranges. This means that the "ghost capacitance", hidden in the dielectric absorption is accounted for in the total capacitance.
This hypothesis is easily confirmed, thanks to the unreasonable resolution: it is possible to measure the same capacitors on several ranges with a sufficient accuracy.
On the lower ranges, the capacitance appears larger, and decreases on higher ranges.
On the lower ranges, the value displayed is > than Ct, but on the highest range, it becomes intermediate between Ct and Cb, and even sometimes Ce.
This variation is only observed with E-caps: with plastic or even paper types, the value remains consistent for all ranges, with only negligible variations.
This apparent capacitance variation with the range is an informal, indirect way of estimating the losses: it takes into account the DA, but also the series resistance: if such a resistance is present, it will cause an early upper threshold crossing, when the capacitor's voltage is still lagging, and the discharge phase will be shorter, because the capacitor was not fully charged, and because of the early threshold crossing caused by the I*R error.
The displayed value will be reduced accordingly
The losses information is thus theoretically accessible, but indirectly and not in a clean and practical format, unlike a vector-meter, which dematrixes exactly the parameter you need and nothing else.
Real power testing with this method is essentially ruled out, because of the huge power supply that would be required.
To summarize, this is a nice, general-purpose tester, but it is unable to detect faulty capacitors (except grossly faulty ones), let alone marginal ones.
Since it applies a bias voltage, it is also unsuitable for in-circuit applications.
One could say that the capacitance measured by this meter is the "timing capacitance", one more fancy definition...
Erratum:
On the initial schematic, I had omitted some bypass capacitors. Here is the corrected version:
It is immediately obvious that the 500mA maximum test current is widely insufficient to bring additional information compared to a "normal", low power meter.
Even in the purely linear domain, it under-performs compared to a state-of the-art vectorial meter (like modern hand-held RLC meters f.e.): the values indicated for capacitors in a good condition are larger than even the theoretical series equivalent.
How can this be?
The measurement process is slow, and involves charging resistances that can be high for the lower ranges. This means that the "ghost capacitance", hidden in the dielectric absorption is accounted for in the total capacitance.
This hypothesis is easily confirmed, thanks to the unreasonable resolution: it is possible to measure the same capacitors on several ranges with a sufficient accuracy.
On the lower ranges, the capacitance appears larger, and decreases on higher ranges.
On the lower ranges, the value displayed is > than Ct, but on the highest range, it becomes intermediate between Ct and Cb, and even sometimes Ce.
This variation is only observed with E-caps: with plastic or even paper types, the value remains consistent for all ranges, with only negligible variations.
This apparent capacitance variation with the range is an informal, indirect way of estimating the losses: it takes into account the DA, but also the series resistance: if such a resistance is present, it will cause an early upper threshold crossing, when the capacitor's voltage is still lagging, and the discharge phase will be shorter, because the capacitor was not fully charged, and because of the early threshold crossing caused by the I*R error.
The displayed value will be reduced accordingly
The losses information is thus theoretically accessible, but indirectly and not in a clean and practical format, unlike a vector-meter, which dematrixes exactly the parameter you need and nothing else.
Real power testing with this method is essentially ruled out, because of the huge power supply that would be required.
To summarize, this is a nice, general-purpose tester, but it is unable to detect faulty capacitors (except grossly faulty ones), let alone marginal ones.
Since it applies a bias voltage, it is also unsuitable for in-circuit applications.
One could say that the capacitance measured by this meter is the "timing capacitance", one more fancy definition...
Erratum:
On the initial schematic, I had omitted some bypass capacitors. Here is the corrected version:
Attachments
- Status
- Not open for further replies.