RF Filter Theory
A complete, example-driven guide to RF filter design — from approximation theory to component synthesis, bandpass/bandstop transformations, and practical microwave filter implementations. Every design is verified with numerical component values.
Filter Fundamentals
An RF filter is a two-port network that passes signals in a desired passband while attenuating signals in the stopband. Filters appear everywhere in RF systems: after the antenna to reject out-of-band blockers, between stages to suppress harmonics, in diplexers to separate TX and RX bands, and as channel-select filters in receivers.
Ap — max passband ripple (dB) · As — min stopband attenuation (dB)
Ωs = fs/fc — normalised stopband frequency (selectivity factor)
Steeper rolloff or larger As−Ap requires higher order n — more components.
Filter Approximations
Butterworth — Maximally Flat
Rolloff: −20n dB/decade · No passband ripple — maximally flat magnitude
Poles: equally spaced on circle of radius ωc in left half-plane
Chebyshev — Equiripple Passband
ε² = 10^(Ap/10)−1 · Ap=0.5 dB: ε=0.3493 · Ap=3 dB: ε=0.9976
Advantage: Steeper rolloff than Butterworth for same n · Trade-off: Passband ripple
Bessel — Maximally Flat Group Delay
Step response: zero overshoot · Rolloff: slower than Butterworth
Use for: radar pulse shaping, digital IF filters, wideband modulations
Comparison — n=5, same cutoff
| Parameter | Butterworth n=5 | Chebyshev 0.5 dB n=5 | Bessel n=5 |
|---|---|---|---|
| Passband ripple | 0 dB (flat) | 0.5 dB | 0 dB (flat) |
| Atten. at 2×fc | −31 dB | −43 dB | −14 dB |
| Atten. at 3×fc | −50 dB | −67 dB | −25 dB |
| Group delay | Good | Poor | Excellent |
| Best for | General purpose | Sharp selectivity | Linear phase / pulse |
Butterworth n=5 at 2fc: 10·log₁₀(1+2¹⁰) = 10·log₁₀(1025) = 30.1 dB ≈ −31 dB ✓
Determining Filter Order
Chebyshev: n ≥ arccosh(√[(10^(As/10)−1)/(10^(Ap/10)−1)]) / arccosh(Ωs)
Worked Example
LP Prototype & Denormalization
All filter designs start from a normalized lowpass prototype (ωc=1 rad/s, R=1 Ω). Denormalization scales to any cutoff frequency and impedance.
Prototype g-values
Butterworth (−3 dB at ω=1)
| n | g₁ | g₂ | g₃ | g₄ | g₅ | gn+1 |
|---|---|---|---|---|---|---|
| 1 | 2.0000 | — | — | — | — | 1.0000 |
| 2 | 1.4142 | 1.4142 | — | — | — | 1.0000 |
| 3 | 1.0000 | 2.0000 | 1.0000 | — | — | 1.0000 |
| 4 | 0.7654 | 1.8478 | 1.8478 | 0.7654 | — | 1.0000 |
| 5 | 0.6180 | 1.6180 | 2.0000 | 1.6180 | 0.6180 | 1.0000 |
Chebyshev — 0.5 dB ripple
| n | g₁ | g₂ | g₃ | g₄ | g₅ | gn+1 |
|---|---|---|---|---|---|---|
| 1 | 0.6986 | — | — | — | — | 1.0000 |
| 2 | 1.4029 | 0.7071 | — | — | — | 1.9841 |
| 3 | 1.5963 | 1.0967 | 1.5963 | — | — | 1.0000 |
| 4 | 1.6703 | 1.1926 | 2.3661 | 0.8419 | — | 1.9841 |
| 5 | 1.7058 | 1.2296 | 2.5408 | 1.2296 | 1.7058 | 1.0000 |
Topology: C–L–C–L… (shunt-first) or L–C–L–C… (series-first, dual network, same response)
Worked Example — 3rd-order Butterworth LP at 100 MHz, Z₀=50 Ω
g-values: g₁=1.0, g₂=2.0, g₃=1.0. Topology: C–L–C shunt-first. ωc=2π×10⁸=6.283×10⁸ rad/s
LP → HP / BP / BS Transformations
Lowpass → Highpass
LP shunt C → HP shunt L: L = Z₀/(gk·ωc)
Example n=3 Butterworth HP at 100 MHz, Z₀=50 Ω:
C₁=C₃ = 1/(1.0×50×6.283×10⁸) = 31.83 pF · L₂ = 50/(2.0×6.283×10⁸) = 39.79 nH
Lowpass → Bandpass
LP series Lk → series LC: Ls=gk·Z₀/BW · Cs=BW/(gk·Z₀·ω₀²)
LP shunt Ck → parallel LC: Cp=gk/(Z₀·BW) · Lp=Z₀·BW/(gk·ω₀²)
Bandpass Design Example
g₁=1.5963, g₂=1.0967, g₃=1.5963. ω₀=5.655×10⁹ rad/s, BW=3.142×10⁸ rad/s, Q=18.0
Lowpass → Bandstop
LP shunt C → series LC (resonates at ω₀, short at notch frequency)
Single-section notch at 2.4 GHz: L=1 nH, C=1/((2π×2.4G)²×1n) = 4.39 pF
Notch depth ≈ 20–40 dB in practice (limited by component Q)
Microwave Filter Realisation
Above ~1 GHz, lumped LC components become impractical due to parasitics. Distributed transmission line structures replace them. Same prototype g-values — only the physical realisation changes.
Edge-Coupled BPF — Most Common Microwave BPF
Z0e,k = Z₀·[1 + Jk/Y₀ + (Jk/Y₀)²] · Z0o,k = Z₀·[1 − Jk/Y₀ + (Jk/Y₀)²]
Practical Considerations
Component Q and Insertion Loss
5th-order Butterworth: Σgk=6.472. QL=50 (cheap SMD): IL=0.562 dB. QL=200 (good SMD): IL=0.140 dB
| Technology | QL | f range | IL (n=3) | Best for |
|---|---|---|---|---|
| SMD LC | 30–150 | DC–3 GHz | 0.3–2 dB | PCB, cost-sensitive |
| Microstrip coupled | 100–200 | 0.5–20 GHz | 0.5–2 dB | Microwave PCB standard |
| SAW | 500–2000 | 10 MHz–3 GHz | 0.5–3 dB | Mobile band-select |
| BAW / FBAR | 1000–3000 | 0.5–6 GHz | 0.5–2 dB | 5G, WiFi duplexers |
| Cavity / coaxial | 2000–10000 | 100 MHz–10 GHz | 0.1–0.5 dB | Base stations, instruments |
| Waveguide | 5000–50000 | 2–100 GHz | 0.05–0.3 dB | Satellite, radar, mmWave |
· Using inductor Q at 1 MHz to design a 900 MHz filter — Q drops dramatically with frequency
· Ignoring PCB parasitic capacitance across series inductors — shifts cutoff upward
· Not accounting for end-loading in microstrip edge-coupled filters — resonators appear too long
· Too-narrow BPF bandwidth — insertion loss increases as 1/BW²
Put This Theory Into Practice
Use these RFLab tools to design, compute component values, and verify your filters — linking the prototype tables and transformations on this page directly to physical designs.