I need block DC in a 50 Ohms terninated system. The passband will go from 100Hz to 100MHz, so the needed capacitance is around hundreds of uFs. For normal chip capacitors, I think I must take the ESL into consideration.
Are there any good methods to achieve a flat passband as wide as this? Or there are some special capacitors can handle this? Any suggestions are appreciated.
Leaving aside whether it's a good idea to attempt this, it should be possible to do it by paralleling multiple identical value high-K ceramics, in series with resistors. The reason I specify those is their low cost, small size and high volumetric capcitance, though watch out for the vicious voltage coefficient of some of the dielectrics.
You have not specified the DC voltage you need, but let's assume it's in single digit or low double digit volts, where MLCC (multi layer ceramic capacitors) are readily available.
The trick is to be able to use them above their SRF.
Consider a 16v 10uF in X5R in an 0805 case by Murata (not recommending the brand, it's just they happen to publish a data sheet). This has an SRF of around 1MHz, which implies a series inductance of about 3nH. At 1MHz, these two components have an impedance of magnitude 16mjohm, so the series resonance would be completely de-Q'd by a series resistance of 1ohm, let alone the 10ohm I've shown.
The voltco of that particular part means it loses about half its capacitance to 5v, so you'd need 20 in parallel for 100uF, at 5v. All those 10ohms in parallel are less than 1ohm, insignificant in series in a 50 ohm system. The 3nHs in parallel at 100MHz come to about 0.1johm as well.
The method you parallel several capacitors to achieve the total value is important. You must avoid exciting resonances between any wire inductance that connects the capacitors. One way to do this is to make a transmission line structure, and place the capacitors along it, much as you'd make a travelling wave amplifier. Your top frequency of 100MHz has a wavelength of 3m, so any structure below 200mm in length is electrically short, and if you make a matched structure, you can go to any length. Like this ...
simulate this circuit – Schematic created using CircuitLab
Physically, microstrip would be a good and easy to use structure. Use a pair of lines side by side over ground plane, which taken together in parallel give you 50 ohms. On FR4, that's a W/h ratio of 2. Go in on one, out on the other.
In case anybody is wondering whether this actually works, yes it does. This is how I designed a DC block for the input of a 1MHz to 4.2GHz spectrum analyser, using 5 single layer 500pF capacitors in a edge-line structure, and that behaved without the resistors.
Once you've established the required voltage, and the maximum permissible series resistive loss, it's worth redoing the sums for other values and dielectrics than the 10uF X5R I've used here. X7R has a much lower voltco, but needs a larger package.
