There is a better choice: thin air. As is shown in this thread, amongst others is that paralleling capacitors results most of the times in degradation.
If paralleling is actually needed (doubtful), it is a complex job, and requires multiple elements with scaled damping resistors.
If you're not prepared to go to such lengths, the simplest option is to bypass it (!)
No but there are two in series ==> double the inductance.
If a low-impedance path (one or more capacitors) exists between two nodes, adding a capacitor will close the loop and create resonances.
Thank you for sharing the link. It´s a very interesting thread with scientific data to back it up!
But now I'm confused about this: In PSU I use to put two 100nF just after the rectifier (1 per rail) with the purpose of filtering noise and 2 more after the capacitor bank. I'm I correct to assume it's better to get rid of all the 4 caps?
Thank you for sharing the link. It´s a very interesting thread with scientific data to back it up!
But now I'm confused about this: In PSU I use to put two 100nF just after the rectifier (1 per rail) with the purpose of filtering noise and 2 more after the capacitor bank. I'm I correct to assume it's better to get rid of all the 4 caps?
No. There is enough loss in the PC traces to spread out the potential resonances in all but badly designed cases.
I have seen recommendations to use the higher esr MKS rather than the lower esr of MKP.
I have also read that the lossy non C0G caps can better be used due to the damping of the lossy element.
Certainly use of resistor damping is fundamental to Hagerman's method.
Who's the digital/switching expert on the Forum? He did a long and detailed posting of test results for oscillation due to paralleled capacitances.
But absolutely not teflons alone.
I have also read that the lossy non C0G caps can better be used due to the damping of the lossy element.
Certainly use of resistor damping is fundamental to Hagerman's method.
Who's the digital/switching expert on the Forum? He did a long and detailed posting of test results for oscillation due to paralleled capacitances.
But absolutely not teflons alone.
The devil is in the details, and very much depends on the specifics of the case.Thank you for sharing the link. It´s a very interesting thread with scientific data to back it up!
But now I'm confused about this: In PSU I use to put two 100nF just after the rectifier (1 per rail) with the purpose of filtering noise and 2 more after the capacitor bank. I'm I correct to assume it's better to get rid of all the 4 caps?
An important factor is the inductance to resistance ratio of the wiring or tracks linking the capacitors: if the inductance is low compared to the resistance (hot and return wires twisted together f.e.), the damping will dominate and it will be harmless.
You have to describe your layout to go further.
Yikes that was obvious once you pointed it out!
I am reading Small Signal Audio Design by Self at the moment, and he recommends 0.1uF rail to rail bypass for opamps instead of the commonly seen rails to ground caps as this avoids injecting currents into ground, but realizing the electrolytics are in series, this can't be good?
I've been using 0.22uF Polyester films on top of opamps, while MKT generally has higher ESR than "better" film types and should be less resonant, I am now thinking about cutting traces and giving each opamp supply pin isolation resistors.
--
>AndrewT
Are you perhaps referring to Eva?
http://www.diyaudio.com/forums/power-supplies/76229-power-supply-decoupling.html
I first came across this thread and learned that the common "bypassing" can be a bad practice, albeit the pictures in this thread happened to be all dead.
I'd really love to be able to see those plots.
It depends where the lytics are located: if they are far enough, the track resistance/impedance will be high, making them not only harmless but necessary: what you want to achieve is a low impedance, wideband, seen from the opamp. If the supply alone cannot provide for that, a local decoupling is necessary. But you always want to avoid resonances.
Note that most of the times, there will be no obvious noticeable effects, even when the configuration is rather catastrophic: maybe some susceptibilty to some kind of interference, increased distortions at spot frequencies, etc. Only in rare instances will it cause gross malfunctioning, such as oscillations.
And to be honest, most decoupling configurations tend to look apparently neutral, even when they are excellent or atrocious. But when you accumulate problems, there is a chance you arrive at some tipping point, and you begin to see odd, unexplained effects.
ESR is somewhat misleading, for a variety of reasons. For MKT, it is measured at a much lower frequency than any possible resonance, and at the MHz or tens of MHz where the resonance occurs, the Q of the dielectric is much higher than the global Q including the whole circuit. This means that all "decent" dielectrics are more or less equivalent on this. Only with the worst ceramics is there a small difference.
The apparent capacitance at 10MHz might be reduced, but the remainder has a high enough Q to be troublesome.
When you want damping, the best option is to add it explicitly, with a small series resistor. And if, as a result the VHF decoupling properties become insufficient, you then add a small additional parallel capacitor.
A small SMD tantalum capacitor can do all of that in a single component, as is shown in the study.
PS:
Isolation resistors are excellent, because they "break" the Q of any loop, local or global.
If you use them as standard practice, you will encounter very few problems.
Last edited:
Yes, modern small aluminium capacitors have excellent HF properties, and their inherent damping avoids any possibility of severe resonances.
Would you be able to use Clarity Cap from their ESA range?
ClarityCap ESA range of capacitors home page
ClarityCap ESA range of capacitors home page
No, it's to big. It has to be very small but now I'm leaning to not use any bypass.
The PSU in question is made with these boards: (this is another question - not the board for the small 100nf)
HackerCAP and HackerNAP boards - Blue vs Green Back | Flickr - Photo Sharing!
Should I add 100nF caps at the rails?
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.