error401
1000+ Head-Fier
- Joined
- Oct 11, 2006
- Posts
- 1,244
- Likes
- 11
Quote:
I'm far from an EE, so take my words with a grain of salt. It's mostly what I've been able to glean from publications outside the audio field, since I don't trust anyone's opinion there to be anything more than 'it sounds better to me', which I personally don't put a lot of stock in.
Decoupling capacitors exist to make the power supply impedance as low as possible across as wide a frequency range as the part will be operating at. As you should already know, wires and PCB traces have parasitic inductance which presents a relatively high impedance to fast signals. A high impedance means that an instantaneous current draw on the pin will cause a voltage fluctuation we want to avoid. Placing a decoupling capacitor near the part counters those effects with a small amount of capacitance to stabilize the voltage at the pin when a fast load transient is encountered. DC and slower signals suffer much less from inductive parasitics, and decoupling caps generally have little effect at these frequencies; the power supply or bulk capacitors can respond quickly enough.
As a result, you want the fastest cap possible so that it can respond more quickly and keep the voltage sag during a load transient as short as possible. More uF doesn't help with this. Linearity isn't too big a deal either since the goal is not to pass a signal linearly, but to respond quickly. A ceramic cap, while nonlinear, should still respond fairly well far outside the effective range of a film cap (well into the GHz). A larger value than necessary can only have a detrimental effect, the cap only needs to be large enough to respond to the transient while the power supply catches up. It just needs to counter the frequency response of the power supply lines and the power supply itself, so it's only working in the MHz range, where small values are king. Generally small currents are involved as well, so a large capacitance is not required.
From what I've read about digital decoupling, 0.1uF as a standard value is a holdout from the 70s when not much was known about the physics of complex power supply networks. Most of the papers I've seen suggest that this value is much too large, especially when there are lots of them across the same power supply rails.
At low frequencies like audio, this isn't really important though. 0.1uF caps at each part is a fine solution, but I don't see a reason to go any larger than that. 0.01uF should perform just as well.
Some reading material if you're interested, this is a complex field when we're talking about GHz digital signals:
Electronic Design Welcome
http://download.cypress.com.edgesuit..._an1032_12.pdf
Decoupling
Originally Posted by infinitesymphony /img/forum/go_quote.gif Very interesting--thanks for clearing me up! I hadn't heard that 0.1 uF was the typical maximum limit for decoupling, or maybe I had my values confused with those of the input capacitors (0.1 uF minimum, higher is better to avoid low bass roll-off). I'd heard that film capacitors were better than ceramics for decoupling due to higher linearity in the audio range, whereas ceramic capacitors had better high-frequency linearity, but I might be misremembering what I've read. Either way, it's nice to know that ceramic isn't a bad choice. Just curious... When would it be recommended to use additional decoupling capacitors? |
I'm far from an EE, so take my words with a grain of salt. It's mostly what I've been able to glean from publications outside the audio field, since I don't trust anyone's opinion there to be anything more than 'it sounds better to me', which I personally don't put a lot of stock in.
Decoupling capacitors exist to make the power supply impedance as low as possible across as wide a frequency range as the part will be operating at. As you should already know, wires and PCB traces have parasitic inductance which presents a relatively high impedance to fast signals. A high impedance means that an instantaneous current draw on the pin will cause a voltage fluctuation we want to avoid. Placing a decoupling capacitor near the part counters those effects with a small amount of capacitance to stabilize the voltage at the pin when a fast load transient is encountered. DC and slower signals suffer much less from inductive parasitics, and decoupling caps generally have little effect at these frequencies; the power supply or bulk capacitors can respond quickly enough.
As a result, you want the fastest cap possible so that it can respond more quickly and keep the voltage sag during a load transient as short as possible. More uF doesn't help with this. Linearity isn't too big a deal either since the goal is not to pass a signal linearly, but to respond quickly. A ceramic cap, while nonlinear, should still respond fairly well far outside the effective range of a film cap (well into the GHz). A larger value than necessary can only have a detrimental effect, the cap only needs to be large enough to respond to the transient while the power supply catches up. It just needs to counter the frequency response of the power supply lines and the power supply itself, so it's only working in the MHz range, where small values are king. Generally small currents are involved as well, so a large capacitance is not required.
From what I've read about digital decoupling, 0.1uF as a standard value is a holdout from the 70s when not much was known about the physics of complex power supply networks. Most of the papers I've seen suggest that this value is much too large, especially when there are lots of them across the same power supply rails.
At low frequencies like audio, this isn't really important though. 0.1uF caps at each part is a fine solution, but I don't see a reason to go any larger than that. 0.01uF should perform just as well.
Some reading material if you're interested, this is a complex field when we're talking about GHz digital signals:
Electronic Design Welcome
http://download.cypress.com.edgesuit..._an1032_12.pdf
Decoupling