This subject is detached from another thread.
Here is where we are now:
First task: find a fair way of comparing different setups: simply saying, we apply a constant voltage across the composite cap is certainly not one: you are going to change the corner frequency, and the composite arrangements will be bulkier than a single cap.
The value of the caps could be changed depending on the configuration, but that is certainly not a good solution, because we want to use the same caps throughout (different caps could have different behaviors).
Here is what I propose to (sensibly) normalize the setups and conditions with identical caps:
-The single cap is the reference: its C*V (4µ7*35V) value is one unit. The normalized corner frequency is also one unit.
-The series composite has two CV units and two frequency units. To make a fair comparison (same volume for the same corner frequency), the voltage across the composite has to be corrected by a 2*2 factor = 4
-The parallel composite also has two CV units, but 0.5 frequency unit, resulting in a 1 correcting factor wrt the single cap. This configuration is disadvantageous though, because it has zero additional DC blocking capability.
Maybe this necessitates some kind of penalty?
I don't know, this has to be discussed, in the mean time, I will make new measurements taking into account the above weighting system
Here is where we are now:
How many times will I need to repeat that:
This choice is debatable, but we have to find a fair way of making comparisons: if for example you are allowed to connect in series an indefinite series of capacitors, you will probably achieve a very low THD, but it will require a large volume if want to keep the same corner frequency
I have used spice only to illustrate the setup and indicate the voltages applied to the capacitors, not for the actual measurements
See above: I physically measured real parts, but used LTspice to present the results
Unlikely: people make errors, if you see what I mean....
No sim, pure measurements in my case
First task: find a fair way of comparing different setups: simply saying, we apply a constant voltage across the composite cap is certainly not one: you are going to change the corner frequency, and the composite arrangements will be bulkier than a single cap.
The value of the caps could be changed depending on the configuration, but that is certainly not a good solution, because we want to use the same caps throughout (different caps could have different behaviors).
Here is what I propose to (sensibly) normalize the setups and conditions with identical caps:
-The single cap is the reference: its C*V (4µ7*35V) value is one unit. The normalized corner frequency is also one unit.
-The series composite has two CV units and two frequency units. To make a fair comparison (same volume for the same corner frequency), the voltage across the composite has to be corrected by a 2*2 factor = 4
-The parallel composite also has two CV units, but 0.5 frequency unit, resulting in a 1 correcting factor wrt the single cap. This configuration is disadvantageous though, because it has zero additional DC blocking capability.
Maybe this necessitates some kind of penalty?
I don't know, this has to be discussed, in the mean time, I will make new measurements taking into account the above weighting system
why rehash known also ran "solutions"
Bateman's Capacitor Sound articles should have put this to rest
his data supports his explanation that polar Electrolytics have a thin spontaneous oxide layer on one foil - the distortion problem is the huge signal V/M in this dielectric layer
not that it is somehow rectifying and we just have to bias it - it is the AC signal and inherent Vcoeff
his recommendation is non/bipolar Al electrolytics with symmetric full depth oxide layers grown on both foils - and even then to reduce V/Mm further by using higher V rating than required, even to series 2 of these electros - both options measured significantly better than back to back polar Al electros w/wo bias V
Bateman's Capacitor Sound articles should have put this to rest
his data supports his explanation that polar Electrolytics have a thin spontaneous oxide layer on one foil - the distortion problem is the huge signal V/M in this dielectric layer
not that it is somehow rectifying and we just have to bias it - it is the AC signal and inherent Vcoeff
his recommendation is non/bipolar Al electrolytics with symmetric full depth oxide layers grown on both foils - and even then to reduce V/Mm further by using higher V rating than required, even to series 2 of these electros - both options measured significantly better than back to back polar Al electros w/wo bias V
Last edited:
I now modified the setup according to the above. Note that every detail is of importance, in particular the low (50Ω) output impedance of the generator, and the full DC-coupling (apart from the CUT of course).
The frequency is 100Hz; to make a measurement, the voltage across the cap or composite is measured, and the generator is adjusted to read 2.5Vrms or 10Vrms (I had to reduce from 3V to 2.5V, because the generator was not capable of delivering 12V).
The actual measurement of THD is then made across the 100Ω resistor.
I remeasured the metal case caps, and also drop-types.
I have also included the 2.5V measurement for the series arrangement, although it is "unfair"
The frequency is 100Hz; to make a measurement, the voltage across the cap or composite is measured, and the generator is adjusted to read 2.5Vrms or 10Vrms (I had to reduce from 3V to 2.5V, because the generator was not capable of delivering 12V).
The actual measurement of THD is then made across the 100Ω resistor.
I remeasured the metal case caps, and also drop-types.
I have also included the 2.5V measurement for the series arrangement, although it is "unfair"
Code:
[U]Configuration/Vrms Hermetic Epoxy drop[/U]
Single, 2.5V 0.08% 1%
Series, 10V 0.105% 0.45%
Series, 2.5V 0.025% 0.15%
Shunt, 2.5V 0.02% 0.14%Attachments
I could do that (and I will, later), but there are good reasons for the choice I made:
-It is a good representation of the real-life situation: that is how coupling caps work
-It is more sensitive, because you use the setup as a high-pass filter, thus maximizing the harmonics towards the output.
In order to normalize completely, we would need to apply a voltage (or current) directly to the cap, and measure the THD of the resulting current (or voltage).
Here, with various resistances here and there, the THD becomes "diluted" in a larger amount of fundamental.
If the generator's output is not stiff enough (or worst, not DC-coupled), the resulting THD will appear lower
No problem there
What strikes me at this stage is the huge difference between two capacitor samples (theoretically the same technology, just a different encapsulation), and the fact that the shunt combination still manages to remain the winner, even with the gross handicap of having the same voltage across as the series connection: this means that it is more than four times more advantageous, if you take all elements into account.
Quite puzzling.
Quite puzzling.
At the end of the day, it strikes me that the use of the shunt antiparallel caps is limited to very low d.c. bias, typically in the a.c. coupling of the bottom of a feedback divider to common to limit the d.c. gain and hence limit output d.c. offset. So interesting as it is, and potentially useful, it's not applicable to traditional a.c. coupling applications that sustain a significant d.c. offset across the capacitor(s).
But, for those cases, especially with solid-state, it's usually as or more convenient to manage d.c. coupling anyway. With hollow state it becomes a good deal more painful.
Yes, I also mentioned it; perhaps of even more concern is the long term behavior of such an arrangement: the oxide layer in tantalum caps has no self-healing properties, quite the opposite in fact, and with a shunt arrangement, one of the caps will always experience reverse-bias. Not exactly healthy IMHO
DC coupling is of course the gold standard, but it isn't always practicable or even desirable.
In those cases, adapting the impedance level and relying on a film cap is probably a better option. I do not picture myself connecting antiparallel tantalums, even after the measurements I just performed....
Perhaps it is sign I am getting old.... or wise? or both?
Although I used plenty of them in scientific instrumentation for years, I'm tantalum-shy these days. They do tend to fail shorted, and I had one such where I managed to get the polarity wrong. Fortunately the power supply was current-limited and at a level that did not cause a fire.
I've watched a few solid tantalum caps go up in flames
who knew mixing finely powdered metal and MnO2 oxidizer together could be a problem?
who knew mixing finely powdered metal and MnO2 oxidizer together could be a problem?
+1
I had replaced chemicals in a mixing desk by tantalum. After careful listening to ensure the sound quality was not affected.
In one month, we had some parts shorted (i knew that). But, after this period no failure any more. 20 Years later, all tantalum are OK. I believe chemical should be aged and need a recap ?
Elvee , thanks for the measurements.
Tant seem to last forever if conservatively used. HP and TEK and many other brands used them almost exclusively and those instruments seem to just go on for ever.
Great testing and discussion over here, BTW !
THx-RNMarsh
Great testing and discussion over here, BTW !
THx-RNMarsh
I know I'll never use the anti-parallel combination, yet I would very much like to understand why it works so well: it is completely counter-intuitive.
In both cases, there is a symmetry that should cancel non-linearities in a comparable way.
The difference is that the series combination works by voltage cancellation, and the shunt one by current cancellation.
Another bizarre anomaly are SY and Pavel's results: they are significantly lower than mine, but that could be explained by the different setups (gentler), and the wide variation in behavior for caps from different origins: the two samples I took randomly are already ~20dB apart, and they are probably not extreme examples
In both cases, there is a symmetry that should cancel non-linearities in a comparable way.
The difference is that the series combination works by voltage cancellation, and the shunt one by current cancellation.
Another bizarre anomaly are SY and Pavel's results: they are significantly lower than mine, but that could be explained by the different setups (gentler), and the wide variation in behavior for caps from different origins: the two samples I took randomly are already ~20dB apart, and they are probably not extreme examples
Your measurement conditions are much gentler than mine: 1.55V across the whole RC against 2.5V across the cap, and total circuit resistance at least 1.5K against 150 ohm, this explains the much lower figures
I have tested two more samples: another hermetic, made by Kemet IIRC and another drop, made by STC (all 4µ7/35V)
Code:
[U]Configuration/Vrms Hermetic Epoxy drop Kemet STC[/U]
Single, 2.5V 0.08% 1% 0.125% 0.215%
Series, 10V 0.105% 0.45% 0.087% 0.15%
Series, 2.5V 0.025% 0.15% 0.015% 0.046%
Shunt, 2.5V 0.02% 0.14% 0.021% 0.045%No clear pattern seems to emerge: the type/manufacturer of the cap has an overwhelming effect.
In general though, the shunt arrangement remains better, even at 2.5V which favours the series connection
Attachments
how sensitive is the anti-parallel to DC? - you know, the actual purpose of a DC blocking C?
really can't recommend even a few 100 mV reverse V on a electro
really can't recommend even a few 100 mV reverse V on a electro
What will happen with prolonged reverse voltage on an Al electrolytic: if it doesn't go into a runaway condition and explode, it will reform to some extent and remain functional. This saved some agony on a consumer audio product, where it was discovered late in the game that a silkscreen was incorrect. The power available was limited, and iirc it was across a shunt regulator. No recall required, thank goodness. Of course the error was corrected asap.
However, clearly not a recommended practice.
However, clearly not a recommended practice.
Yes, and that is the reason why I said I would never use this configuration. I made a quick check though, and I have to say I was somewhat surprised by the results: I applied an increasing voltage while measuring the leakage current, and noted the value for each decade:
Reaching 1µA required 2V, 10µA-3V, and 100µA-5.3V. I attempted 1mA, but the cap went dead short before I could note anything 😀 ...there are limits.
To summarize, I would never use this where any actual DC blocking is required, but for example in the feedback of an amplifier where the cap would only see the offset voltage, it is probably acceptable.
Al caps are more tolerant than tantalums, and solid Al caps even have both voltages specified and marked on the case
The dc polarity across the feedback capacitor is often predictable although small in many designs
Keep that system without power for two to three months and then say so.
Gajanan Phadte
- Status
- Not open for further replies.