Note that a 440pF cap with less than 6nH ESL is hard to come by (an 0805 ceramic chip is already about three), and more to the point, you won't be able to tap onto the "top" and "bottom" of that 6nH as suggested by the schematic!

The series branch is a bit easier to deal with, as 260nH inductors with much less than 10pF stray (end-to-end and to ground) are common enough.

What effect does this have? Basically you'd be connecting to a tapped inductor, which means impedance transformation -- whether you meant to or not.

Consider transforming the entire network, so that for example it's three LCs chained, with a bit of mutual inductance linking them, and the two end tanks using tapped feedpoints (tapped L, or C divider).

Or just raising the impedance, so that the end tanks are reasonable (say, 200 ohms), but then the middle tank will be unreasonable (you won't find a ~uH inductor with <1pF strays), so again, consider a series-parallel transformation.

Or use the strays to your advantage as impedance transformation. This requires a good model of them, but it is solvable.

I made this 100-130MHz bandpass,

with this [simulated] response,

This can be described as a:

- One port unterminated filter (this is a perfectly normal type -- at least one port must be terminated, and the other can be whatever impedance, including open or short, it just uses different coefficients; see tables in Zverev)

- 2nd order highpass, from the bias choke L1 and coupling cap C2

- 5th order lowpass, approx. Chebyshev 0.5dB or so I think

- Impedance transformer from the capacitor divider (C2, C5, C6)

- Two zeroes from the inductor stray capacitance

- Terminated into a complex load (Q2's base input impedance is around 500 ohms and a few pF, adjustable by R13)

- Capacitors adjusted with trimmers and gimmicks (bits of wire/PCB tacked on)

Overall, the impedance matching gains several dB over a more cookbook approach, and more than 6dB over a naive double terminated filter.

So, it's a neat curiosity, but hardly necessary with gain-bandwidth being as affordable as it is!

This is more to say -- you can pull off an oddball circuit, which behaves like a certain amalgam of prototypical filters and well understood transformations, and which is motivated by real parasitics that are missing from the standard ladder (alternating parallel-series) prototype. But choose your transformations wisely, because some are easier than others, and an optimal (lead component count?) approach can incur a lot more design and tuning time.

So, impedance transformations are easy, and series-parallel transformations are modest. Keep things symmetrical, so the same thing happens at both ends (and you maintain 50 ohm system impedance).

Series-parallel transformation is where you transform one branch into another. The transition band response is preserved, but the asymptotes may shift, depending on what reactances are used to perform the transformation. The most common example is a quartz crystal, which is usually series resonant between two pins (actually it's both, due to internal capacitance), but a parallel resonant response is required. By making a pi network with two capacitors and the crystal between them, a transformation is performed and the desired response is obtained.

Similarly, we can transform a parallel-series-parallel LC bandpass into an all-parallel bandpass, where the parallel tanks are identical and linked with series capacitors.

Or more generally, any coupling will do, not just series capacitors -- series inductors could be used, or mutual inductance between the tank inductors, or many other topologies as well.

If series caps are used, the asymptotic response tends towards high-pass (the HF asymptote is flat, a constant -whatever dB, corresponding to the C ratios; while the LF is the full -20*(filter order) dB); if inductors, low-pass; if mutual, then symmetrical band-pass.

Tim