Every time I see a schematic for a simple power supply, like something based on a 7805 or 7812 for example, there’s invariably two capacitors on the output - the first much larger than the second. Usually it’ll be something like a 10uf followed by a 0.01uf. I get that they’re filter capacitors, but they’re in parallel, so their capacitance sums. What’s the purpose of this? Since the 10uf cap is likely an electrolytic anyway, with a 20% or more tolerance, what the heck difference would it make to have an extra 0.01uf in parallel with it?
The only thing I can think of is that the 0.01uf is there to absorb little reverse-polarity spikes that might otherwise damage a polarized electrolytic over time.
Electrolytic caps don’t work well at high frequencies. The electrolytic is there to do the bulk of the filtering, and the small ceramic disk is there to get the higher frequency stuff that it misses.
And if anyone’s unfamiliar with how capacitors filter, the small capacitance will do an excellent job at the filtering high frequencies, so it doesn’t need to be as large as the other one.
They do make large value ceramic caps (I’ve seen 22uF (multilayer ceramic chip) 1206 size), but they’re a bit more expensive.
There is another specification of capacitors which is almost never published on a schematic, which is the Equivalent Series Resistance, or ESR. It appears in the equivalent circuit as a resistor in series with the capacitor. There is also an equivalent series inductor, but that is not as relevant to power supply capacitors. Anyway, the ESR directly effects how fast charge can be stored and removed in a capacitor and how the impedance looks at high frequencies. Two differant values are used because the ESR of the lower cap is low and it still looks like a capacitor at high frequency. The big capacitor looks like a resistance at high frequency.
I have no idea of the source, but many years ago I saw published a “top 10” list of things they don’t teach Electrical engineers in engineering school. Two that stick in my mind are:
-There are a least 10,000 kinds of capacitors. The best one for your application may be two or more. The kinds that will work in your application all have 15 week lead times.
-“Dilbert” is not a comic strip. It is a documentary.
And if they’re milspec, multiply all costs by a factor of 3 and delivery times by 5, even when (or especially when) the spec is identical to a COTS (commercial off the shelf) item.
A 78xx/79xx voltage regulator shouldn’t need a 10uF decoupling capacitor anywhere. The recommended values vary, so refer to the relevant manufacturer’s data sheet when in doubt. For these particular regulators most manufacturers specify between 330nF - 470nF on the input, and 100nF on the output. One manufacturer of a 79xx device specs 2.2uF and 1uF tantalums respectively, so it’s always best to check.
It’s common practice to slap a small ceramic cap (say 10nF - 100nF) across an electrolytic to improve the HF characteristics, but the technique isn’t a cure-all, and should be used with thought and discrimination. Capacitors are self-resonant circuits in themselves due to their inherent inductance, and adding badly matched self-resonant circuits in parallel can sometimes make matters worse. For instance, decoupling a reservoir electrolytic with a ceramic capacitor might reduce the amplitude of voltage spikes, but the subsequent resonance may contain more energy than the original noise spike.
If you’re using several decoupling capacitors in an electronic design, then it’s best to make them all identical, not just in value, but in type too. The impedance versus frequency graph for a capacitor looks like a “V”, with the capacitive impedance dominating at low frequencies, and the inductive impedance dominating at high frequencies. Put 2 different types of capacitor in parallel, and there is a danger that the two characteristics will combine to form a big spike in the frequency response, making the whole circuit more susceptible to noise at those frequencies.
Everything has a tolerance associated with it, and chances are they add instead of cancel when more than one part is in parallel or series. Even circuit board traces (wires) have tolerances.
The part you want to use isn’t available. You will have to use the one that takes more work, usually after you have already designed the unavailable part into the system.
You will blame problems on software, and software will blame problems on you. In the end you will both blame marketing for writing the specs too tight.
Then there’s the heat in the chassis and the emi issues that you have to solve to be able to sell it.