RF Filters Q&A

• Low-Pass Filter (LPF): Passes signals below the cutoff frequency, attenuates above. Uses: anti-aliasing before ADC; removing harmonics from a PA output; baseband filtering after downconversion.
• High-Pass Filter (HPF): Passes above cutoff, blocks below. Uses: DC blocking in AC-coupled circuits; removing low-frequency interference or self-interference in a receiver; removing LO feedthrough in a direct-conversion architecture.
• Bandpass Filter (BPF): Passes a band of frequencies, attenuates outside. Uses: selecting a specific RF channel from the antenna; interstage filtering in a superheterodyne chain; duplexer/diplexer channel selection.
• Bandstop / Notch Filter (BSF): Rejects a narrow band, passes all others. Uses: GPS receivers blocking out-of-band interference (e.g. LTE at 1.7 GHz blocking GPS at 1.575 GHz); notch filter to remove a specific interferer from a receiver; WLAN 2.4 GHz notch in a wideband spectrum analyser front-end.

• Butterworth (maximally flat): No ripple in passband or stopband. Monotonically decreasing. Roll-off: 20n dB/decade (n = order). Choose when: flattest passband magnitude response is needed and sharp rejection is not critical. Common in anti-aliasing and audio applications.
• Chebyshev Type I: Equiripple in passband, monotonic in stopband. Steeper roll-off than Butterworth for same order by "spending" some passband flatness. Choose when: sharp transition band is needed and some passband ripple (0.1–3 dB) is acceptable. Used in most RF BPF designs where steep skirts matter.
• Chebyshev Type II (inverse): Flat passband, equiripple stopband. Finite zeros in stopband. Choose when: flat passband and specified minimum stopband rejection are both required.
• Elliptic (Cauer): Equiripple in both passband and stopband. Sharpest transition band for a given order — minimum order for a given spec. Choose when: minimum component count with the most selectivity is needed. Penalty: finite transmission zeros close to the band edge can cause group delay peaking, affecting pulse fidelity.
• Bessel (Thomson): Maximally flat group delay (linear phase). No amplitude selectivity advantage, but the pulse shape is preserved. Choose when: time-domain fidelity is more important than selectivity — pulse radar, digital modulation systems where ISI matters, oscilloscope front-ends.

Insertion loss (IL) in a bandpass filter is the power lost when the filter is inserted in the signal path, measured at the centre of the passband. For a 2-port filter: IL = −S21_min (dB).

Physical causes of insertion loss:

• Conductor loss: Series resistance of coils (inductors), resonator walls, or PCB traces. Dominates at lower frequencies. For an inductor: R_series = ωL/Q_conductor.
• Dielectric loss: The substrate or capacitor dielectric absorbs energy. Characterised by tanδ. Low-loss ceramics (NP0/C0G caps, PTFE substrate) minimise this.
• Radiation loss (cavity filters): If the cavity walls are not perfectly enclosed, RF leaks out. Usually small for well-designed metal cavity filters.

Measurement:

• Calibrate VNA (2-port SOLT) at the cable tips
• Connect filter, sweep S21 over full passband
• The peak of S21 in the passband (least negative dB value) is the minimum IL
• Verify with a thru first: if thru S21 ≠ 0 dB, re-calibrate

Filter order N determines the steepness of the transition from passband to stopband:

Choosing minimum order (Butterworth LPF example):

• Given: passband spec (−3 dB at fc), stopband spec (−As dB at fs)
• Compute normalised stopband frequency: Ωs = fs/fc
• Minimum order: N ≥ log(10^(As/10) − 1) / (2·log(Ωs))
• Round up to next integer

Example: Need LPF with −3 dB at 100 MHz and −40 dB at 200 MHz. Ωs = 2. N ≥ log(9999)/log(4) ÷ 2 = 6.6 → N = 7 for Butterworth. A Chebyshev with 0.5 dB ripple would achieve this with N = 5.

Trade-off: Higher order = sharper roll-off but: (1) more components (cost, board area, assembly), (2) higher IL in the passband, (3) worse group delay variation (Butterworth/Chebyshev), (4) more sensitivity to component value tolerances.

Group delay τ_g is the negative derivative of phase with respect to angular frequency — it measures how long different frequency components of a signal are delayed by the filter.

Why it matters for digital communications:

• ISI (Inter-Symbol Interference): If the group delay is not flat across the channel bandwidth, different frequency components of a symbol arrive at different times. This smears the symbol in time, causing adjacent symbols to interfere — degrading BER.
• Eye diagram closure: A signal passed through a filter with significant group delay variation shows eye closure — the received I/Q constellation is distorted even if the amplitude response is flat.
• OFDM systems (LTE/5G): Each subcarrier occupies a narrow bandwidth, so group delay variation across individual subcarriers is small. But across the full channel bandwidth (e.g. 100 MHz for 5G NR), group delay variation of the channel filter must be <0.5 µs to avoid inter-subcarrier distortion.

Filters and group delay:

• Bessel filter: maximally flat group delay — ideal for pulse signals
• Butterworth: moderate group delay variation (peaking at the band edge)
• Chebyshev: larger group delay ripple corresponding to the amplitude ripple
• Elliptic: severe group delay peaking near the transmission zeros → worst for pulse

• LC filters (lumped element): Discrete inductors and capacitors. Q_u = 30–200. Good to ~2 GHz. Larger size, cheap components. IL = 1–5 dB. Use for: sub-GHz filtering (ISM, IoT), anti-aliasing filters, IF filters in superhet radios, wherever small size is not critical.
• SAW (Surface Acoustic Wave): Piezoelectric substrate (LiTaO₃/LiNbO₃). IDT transducers convert RF to acoustic waves on the surface. Q_u = 300–1000. Good to ~3 GHz. Small (1–2 mm²). IL = 1–3 dB. Use for: handset Rx filters below 3 GHz (Wi-Fi 2.4, LTE sub-3 GHz bands), where small size and moderate IL are required. Limited by acoustic velocity — cannot scale to high frequencies.
• BAW (Bulk Acoustic Wave — FBAR/SMR): Thin-film piezoelectric resonator (AlN). Bulk acoustic resonance in a thin film (0.5–2 µm). Q_u = 1000–3000. Operates 1.5–10 GHz. Very small. IL = 0.5–2 dB. Use for: 5G sub-6 GHz handset filters (n77, n78, n79 bands), Wi-Fi 5 GHz, LTE high bands. Dominant technology for sub-7 GHz handset front-end today.
• Cavity filters: Metal resonant cavities (combline, interdigital). Q_u = 2000–30000. Large size (cm to dm). IL < 0.2 dB. Very high power handling (kW). Use for: base station duplexers, broadcast transmitters, military radar filters, anywhere low IL and high power matter and size is not constrained.

A duplexer allows a single antenna to be shared between a transmitter and receiver operating simultaneously on different frequencies (FDD — Frequency Division Duplex). It replaces two separate antennas with a three-port network: Antenna, TX, and RX ports.

Why BAW enables small duplexers:

• BAW resonators (FBAR) have very high Q_u (1000–3000) and very small size (tens to hundreds of µm per resonator). This allows a complete 5-resonator ladder filter to fit in 1–2 mm².
• A complete duplexer (two BAW ladder filters + balancing network) fits in a 1.5 mm × 1.0 mm package — achievable only because of the tiny resonator size and high Q.
• BAW resonators at LTE Band 7 (TX: 2500–2570 MHz, RX: 2620–2690 MHz) have a duplex gap of 120 MHz. The BAW filter can have a sharp enough transition band (due to Q ≥ 1500) to provide 50 dB TX-RX isolation within this gap.

TX-RX isolation requirement: A 5G handset PA outputs +23 dBm. The LNA sensitivity is −95 dBm. If isolation is only 40 dB, the TX leaks at −17 dBm into the LNA — 78 dB above the sensitivity floor → receiver completely desensitised. Duplexer isolation of ≥50 dB ensures TX leakage at the LNA input is −27 dBm → still 68 dB above sensitivity, but the LNA's IP3 must handle it.

The Q factor (Quality factor) of a resonator measures the ratio of energy stored to energy dissipated per cycle. It determines both the sharpness of the resonance and the insertion loss of the filter built from those resonators.

Q vs. insertion loss: For a bandpass filter built from N resonators of unloaded Q_u:

Practical Q values and IL:

• LC chip inductor at 1 GHz: Q_u ≈ 30–80, 5th-order BPF IL ≈ 3–8 dB
• SAW resonator: Q_u ≈ 500–1000, IL ≈ 1–3 dB
• BAW/FBAR: Q_u ≈ 1500–3000, IL ≈ 0.5–1.5 dB
• Coaxial ceramic resonator: Q_u ≈ 300–800, IL ≈ 0.5–2 dB
• Metal cavity (combline): Q_u ≈ 5000–20000, IL ≈ 0.05–0.3 dB

Spurious responses (spurs) are additional transmission passbands that appear at frequencies above the intended passband in a real RF filter. They occur because practical resonators are not ideal — they have higher-order resonance modes.

Causes of spurs:

• Harmonic resonances of resonator: A λ/2 resonator also resonates at λ, 3λ/2, etc. (harmonics). A combline filter centred at 1 GHz will have spur passbands near 2 GHz, 3 GHz.
• SAW/BAW acoustic harmonics: SAW IDT have overtone responses. BAW has bulk harmonic modes above the fundamental. A BAW filter for 3.5 GHz may have a spurious pass around 7 GHz.
• Package parasitics resonance: The package inductance resonates with the chip capacitance at GHz frequencies, creating a spur in the filter response.
• Propagation of the Chebyshev prototype's stopband ripple into spur passbands: The prototype has infinite attenuation between transmission zeros; a distributed realisation repeats the passband periodically.

Management in receiver design:

• Combline filter: Choose resonator coupling dimensions to push the first spur above the image frequency of the receiver
• LPF after SAW/BAW: Add a simple LC LPF after the BAW duplexer to suppress the harmonic spur at 2×f without adding significant IL in the passband
• Diplexer combiner: Combine a LPF and HPF at their transition frequency — creates a wideband BPF with good stopband rejection above and below
• Receiver image rejection: The image frequency (f_LO − f_IF) must fall in a filter stopband, not in a spur passband

Power handling and PIM are critical specifications for base station filters operating at tens of watts TX power with sensitive receivers on the same antenna.

Power handling: A filter's maximum input power is limited by: (1) peak voltage breakdown in capacitor dielectrics, (2) thermal dissipation in resistive elements (IL × Pin = heat), and (3) current-handling of resonator conductors. Base station duplexer filters typically handle 100–500 W average TX power.

PIM (Passive Intermodulation): When two or more TX carriers (f1, f2) pass through a passive device, nonlinear mechanisms generate IM products (2f1−f2, 2f2−f1) — even in "passive" components. These IM3 products can fall directly in the receive band, raising the noise floor and desensitising the base station's own receiver.

Causes of PIM:

• Contact nonlinearity in loose connectors or oxidised metal surfaces (most common)
• Ferromagnetic materials near high-field regions (ferrite beads, steel screws in cavity walls)
• Dielectric nonlinearity in ceramic capacitors (especially X5R/X7R MLCC vs. NP0/C0G)
• Plating defects or damaged plating on antenna connectors

Reduction techniques:

• Use PIM-rated connectors and cables (7/16 DIN, silver-plated, IPC-PIM-G1 spec)
• Use NP0/C0G capacitors (linear dielectric) not X7R in high-power filter paths
• Avoid ferromagnetic materials in any high-current or high-field region
• Torque connectors to specification and inspect regularly
• Test PIM at installation: IEC 62037 specifies PIM test procedures at +43–46 dBm (20–40 W)

A coupled-resonator filter is built from N resonant cavities (or LC tanks) that are coupled to each other and to the source/load by controlled amounts. The bandwidth, return loss, and ripple are all determined by the coupling coefficients and external Q values.

Physical implementation of coupling:

• Inductive (aperture) coupling: An iris or slot between resonator cavities. Aperture size controls k.
• Capacitive coupling: A small gap between resonator conductors. Gap size controls k.
• Mixed coupling: Both electric and magnetic coupling simultaneously (used to create transmission zeros — finite attenuation poles).

Extracting coupling from measurements:

• To measure k_{12}: weakly couple a VNA to both resonators simultaneously (weak probes). Observe two split peaks in |S21|. k = (f₂² − f₁²) / (f₂² + f₁²) where f₁, f₂ are the two peaks.
• To measure Q_e: weakly couple to one resonator at a time. Measure the 3 dB bandwidth of the resonance → Q_e = f₀/BW_3dB of the isolated resonator when loaded by the port only.
• This "coupling matrix extraction" is the fundamental tool in cavity filter design and tuning.

Tuning procedure for a manufactured filter: Adjust each resonator to the target frequency (via tuning screws), then adjust coupling irises to achieve the design coupling coefficients extracted from the S21 response. Iterate using an optimiser (gradient descent) run against the measured vs. ideal S21/S11.

SAW (Surface Acoustic Wave) filters exploit the piezoelectric effect to convert RF electromagnetic energy into acoustic waves that travel along the surface of a piezoelectric crystal, then back into electrical signals.

Physical structure:

• A piezoelectric substrate: LiTaO₃ (lithium tantalate, favoured for low loss) or LiNbO₃ (wider coupling, higher bandwidth)
• Input IDT (Interdigital Transducer): a comb of metal fingers deposited on the surface. When RF voltage is applied, the alternating fringe fields generate a Rayleigh (surface acoustic) wave. The IDT finger pitch p sets the centre frequency: f₀ = v_acoustic / (2p). For v_acoustic ≈ 3500 m/s, at 2 GHz: p = 0.875 µm — near the photolithography limit.
• Output IDT: converts acoustic wave back to electrical signal. The bandpass response is determined by the IDT frequency selectivity (each finger pair is a resonant transducer at f₀).
• Reflectors (gratings): arrays of metal strips on each side of the transducer reflect acoustic waves to create resonant cavities with high Q.

Frequency limits: SAW is limited by photolithography. The IDT finger width must be λ_acoustic/4 = v/(4f). At 3 GHz, this is 0.29 µm — below standard mass-production lithography limits (~0.5 µm). This fundamentally limits SAW to <~3 GHz in volume production. TC-SAW (temperature-compensated) uses a silicon dioxide overcoat to reduce frequency-temperature coefficient (from −40 ppm/°C to near zero) enabling more stable filters over temperature.

Performance:

• Q_u: 500–1500 (mechanical resonance, much higher than LC)
• IL: 1–3 dB
• Temperature stability: ±50 kHz / °C (uncompensated) → problem for LTE narrow channels; TC-SAW reduces this to ±5 kHz / °C
• Power handling: 1–2 W maximum (limited by SAW amplitude in piezoelectric substrate)
• Size: 1–4 mm² package

A tunable RF filter can change its centre frequency (and sometimes bandwidth) under electronic control, enabling multi-band operation from a single filter or real-time frequency agility.

Tuning technologies:

• Varactor diode: Reverse-biased diode with voltage-controlled capacitance. Adding a varactor in parallel with a resonator tunes its resonant frequency. Tuning range: 1.5:1 to 3:1. Trade-offs: varactor Q is typically 50–200 (lower than fixed capacitor Q) → increases IL. Also adds nonlinearity (IM3 from the p-n junction) → reduces IIP3 by ~10–20 dB vs. fixed filter.
• RF MEMS (Micro-Electro-Mechanical Systems): Mechanically switched capacitors. Very high Q (Q_u > 200 in switched state), low insertion loss, near-linear (high IIP3). Trade-offs: slow switching (µs to ms), requires high actuation voltage (30–80 V), limited reliability (billions of cycles). Used in military and premium handset designs.
• BST (Barium Strontium Titanate) ferroelectric capacitors: Voltage-tunable dielectric material. No moving parts. Q lower than MEMS (~50–100). Moderate tuning range. Used in some antenna tuning networks.
• PIN diode switched resonators: PIN diodes switch additional LC elements in/out, discretely switching the filter centre frequency between N preset values. High Q of switched elements, but multi-pole switching increases loss and complexity.
• YIG (Yttrium Iron Garnet) sphere: Magnetically tunable single-crystal resonator. Q_u = 3000–10000. Excellent tuning linearity. Used in tunable oscillators and filters up to 40 GHz. Trade-off: requires a magnetic coil for tuning (large, slow, consumes DC power).

Key trade-offs vs. fixed filters:

• Higher insertion loss (varactor/BST Q is lower than the best fixed resonators)
• Reduced linearity (IIP3 typically 5–20 dB worse)
• Additional bias circuits and control logic
• Lower reliability for MEMS (switching fatigue)
• Benefit: a single tunable filter replaces 4–8 fixed filters in a multi-band handset, saving board area and cost despite per-unit cost premium

A narrower-than-designed bandwidth is a common first-pass filter problem. It indicates the coupling coefficients or loading are higher than designed, or the resonators are higher Q than expected.

Step 1 — Confirm the measurement is correct:

• Calibrate VNA and check thru. Measure S21 and S11 simultaneously.
• Confirm the centre frequency is correct — if the filter is also shifted in frequency, that is a separate issue (substrate εr, resonator length error).
• Measure group delay — a too-narrow filter will also show higher group delay peak at band edge.

Step 2 — Identify whether coupling or external Q is wrong:

• Too-narrow BW with poor S11 (return loss worse than spec): external Q (Q_e) is too high — the input/output coupling is too weak. The resonator nearest the port is under-coupled. Fix: increase the input/output coupling (larger coupling aperture, stronger probe, larger gap coupling).
• Too-narrow BW but S11 good (well-matched): the inter-resonator coupling coefficients are too small. Fix: increase coupling between resonators (wider iris, stronger coupling element).

Step 3 — For LC PCB filters:

• Measure each resonator's actual resonant frequency (using weak coupling from a VNA probe). Compare to design values — if individual resonators are off by >1%, the coupling calculation was accurate but the LC values were not.
• Check inductor self-resonance: if the inductor's SRF is close to the filter passband, the effective inductance is higher than the DC-measured value, causing the filter to be narrower and the resonant frequency to shift.
• Swap to the next-larger coupling capacitor (or adjust the inductor turn count) and re-measure.

Step 4 — For cavity filters:

• Use the coupling extraction measurement (weak-probe split-mode technique) to measure actual k_ij values at each coupling iris.
• Compare to design target. If k_ij < k_target, the iris is too small — enlarge it (typically by milling slightly wider or repositioning the coupling screw).
• Iterate: each adjustment affects neighbouring resonators. Use an iterative computerised optimisation after initial manual adjustment.

Base station antennas combine TX and RX sharing a physical tower, often co-located with other cellular and broadcast transmitters. This creates a high-risk environment for PIM.

Scenario: A 4G base station transmits Band 1: TX at 2110–2170 MHz (up to 40 W per carrier), receives on RX 1920–1980 MHz. A co-located TV broadcast transmitter uses frequencies in the 600–800 MHz range. The antenna feed system is shared.

PIM generation and path to receiver:

• Two TX carriers (f1 = 2125 MHz, f2 = 2140 MHz) mix in a nonlinear junction
• IM3 products: 2×2125 − 2140 = 2110 MHz (falls at the lower edge of the TX band but could combine with other products to land in RX)
• 5th-order products (3f1 − 2f2): 3×2125 − 2×2140 = 1095 MHz — outside the band but shows the density of products
• Cross-band PIM between the TV transmitter (700 MHz) and the 4G TX (2.1 GHz): IM products at 2×2100 − 700 = 3500 MHz (outside LTE) and 2×700 − 2100 = −700 MHz (not physical) — less of a concern for cross-band scenarios

How it desensitises the receiver:

• The base station RX sensitivity is typically −120 dBm (with LNA noise figure of 2 dB and noise bandwidth of 10 MHz)
• If PIM products are at −100 dBm at the RX input, they raise the noise floor by 10×, degrading sensitivity by 10 dB → effective range reduced by ~70%
• PIM products also spike when multiple TX power levels change simultaneously (transmit power control events) — causing transient desense that is extremely difficult to diagnose without a PIM analyser

Field mitigation:

• Replace all connectors and jumper cables on the antenna system with PIM-rated 7/16 DIN hardware
• Check all screws and mounting hardware near RF — replace any ferromagnetic steel screws with brass or aluminium in high-current regions
• Use a PIM test set (e.g. Anritsu Site Master with PIM option) to measure and locate the PIM source before climbing the tower
• Apply two-carrier PIM test signal at +43 dBm each (3GPP standard) and identify the PIM source by location correlation

Step 1 — Define the specification:

Step 2 — Technology selection: At 3.3–4.2 GHz, SAW cannot be used (limited to ~3 GHz). LC is too lossy (Q ~60 at 4 GHz → IL ≈ 3–5 dB for 900 MHz BW — fails the 1.5 dB spec). BAW (FBAR): Q_u = 1500–3000 at 4 GHz, provides IL ~0.8–1.2 dB for a ladder filter with 25% FBW. Fits in <1.5 mm². Correct choice.

Step 3 — Topology selection: BAW ladder filter (series-shunt alternating resonators). For 25% fractional bandwidth, a 5-resonator ladder (3 series + 2 shunt, or 5-resonator balanced) provides the correct transition steepness.

Step 4 — Prototype design: Normalise to 3rd-order Chebyshev with 0.1 dB ripple. Convert BPF prototype to bandpass using frequency transformation with ω₀ = 2π × 3.75 GHz, BW = 2π × 900 MHz. Extract L and C values for each resonator position. Convert to BAW resonators: series resonators (series arm, fs governs passband upper edge), shunt resonators (shunt arm, fp governs passband lower edge).

Step 5 — Simulation: Build FBAR model in ADS (use mason/BVD model with Q_m = 1500). Simulate IL, S11, stopband attenuation. Iterate resonator area (scales capacitance and coupling coefficient k_t²) to meet transition band spec.

Step 6 — Verification targets:

• S21 at 3.75 GHz: >−1.2 dB
• S21 ripple 3.3–4.2 GHz: <0.5 dB
• S21 at 3.0 GHz: <−40 dB
• S11 across passband: <−10 dB
• No spur below −10 dB at the LTE Band 7 (2.6 GHz) or Wi-Fi 5 GHz frequency

Both interdigital and combline are families of bandpass filters built from coupled transmission line resonators, used in microstrip, stripline, or metal cavity implementations.

Interdigital filter:

• Resonators are λ/4 transmission lines alternately short-circuited and open-circuited at opposite ends
• Adjacent resonators are oriented in opposite directions (interleaved)
• Coupling is primarily capacitive between adjacent lines (parallel lines facing each other)
• Natural balanced topology — input and output ports on opposite sides
• First spurious passband: at 3f₀ (due to 3λ/4 resonance)
• Use when: broadband BPF (10–30% FBW), compact layout in microstrip, frequencies 1–18 GHz

Combline filter:

• Resonators are less than λ/4 long (typically λ/8 at centre frequency) with a lumped capacitor to ground at the open end, making the combined element resonate at f₀
• All resonators shorted at the same end (ground plane) → looks like a "comb"
• Smaller than interdigital for the same frequency (because λ/8 < λ/4)
• First spurious passband: farther from f₀ (depends on line length — can be pushed to >4f₀)
• Tuning screws can be added at the open (capacitive) end to mechanically adjust resonant frequency
• Use when: narrow-to-moderate bandwidth (1–10% FBW), high selectivity, tunable versions needed, low first spur requirement

Cavity version: Both topologies are implemented as enclosed metal cavities (aluminium or copper) for base station applications, using cylindrical or rectangular resonators with coupling irises between them. Cavity Q_u of 5000–20000 is achievable.

In a standard direct-coupled filter (resonators coupled only to adjacent neighbours), the stopband attenuation increases monotonically with distance from the passband. Cross-coupled filters add coupling paths between non-adjacent resonators, creating alternative signal paths that can interfere destructively at specific frequencies — creating transmission zeros (finite attenuation poles).

Types of cross-coupling:

• Synchronously tuned cross-coupling (resonators 1-4 in a 4-pole): Adds one TZ above or below the passband. Placing TZ on the upper edge gives a very steep upper skirt — useful for suppressing an adjacent TX band while keeping lower skirt gradual.
• Quadruplet cross-coupling (symmetric): Adds two TZs symmetrically, one above and one below the passband. Creates an "extracted pole" response that is steeper on both sides — equivalent to a higher-order filter but with only N resonators.
• Cascade triplet / quadruplet: Multiple groups of cross-couplings enabling multiple TZs at specified frequencies for maximum out-of-band rejection.

Practical advantage: Adding one cross-coupling to a 4-pole Chebyshev cavity filter can improve stopband rejection by 20–30 dB at a specific frequency without adding a 5th resonator. This saves size, weight, and cost in base station duplexers where every gram and cubic centimetre is significant.

Design challenge: Cross-coupling must be precisely controlled (small iris or coupling screw). The wrong coupling phase (capacitive vs. inductive) places the TZ on the wrong side of the passband. Complex simulation is required — the coupling matrix must be extracted precisely and realised with tight dimensional tolerances.

A hairpin filter is a compact version of the parallel-coupled half-wavelength resonator filter. The λ/2 resonators are folded (bent) into a U-shape (hairpin) to halve the physical length, making it practical on PCBs at RF frequencies.

Design procedure:

• Step 1 — Prototype selection: Choose order N and ripple level (Chebyshev). Obtain g-values (g₀, g₁, ... g_N) from table.
• Step 2 — Coupling gap calculation: For each pair of coupled resonators, calculate the even and odd mode impedances: Z₀e and Z₀o from the coupling coefficient J_i,i+1. These define the gap between adjacent hairpins.
• Step 3 — Resonator physical dimensions: Each hairpin is λ/2 at f₀ in the substrate. Resonator length L = λ_g/2 = c/(2f₀√ε_eff). Width W is chosen for 50 Ω input/output and calculated for the resonator's characteristic impedance (typically 50–120 Ω for the resonators).
• Step 4 — Gap dimensions: Convert the required Z₀e, Z₀o values to physical gap width s and conductor width W using closed-form equations (Hammerstad-Jensen for microstrip coupled lines) or EM simulator lookup.
• Step 5 — EM simulation: Import the layout into an EM simulator (Keysight Momentum, Sonnet, HFSS). Simulate S21 and S11. Adjust gap widths and resonator lengths iteratively to centre the passband and achieve the target bandwidth.

Key layout considerations:

• Coupling gap accuracy: For frequencies >3 GHz, gaps of 50–100 µm are required — at the edge of standard PCB process capability. Use a board house with ±25 µm etching tolerance or switch to ceramic substrate.
• Ground vias: Each hairpin's open end must have a clear reference to the ground plane below. Insufficient via density causes parasitic resonances in the ground plane.
• Input/output tapping: Tap the 50 Ω line into the hairpin at a point that gives the correct Q_e (usually at 1/3 to 1/4 from the shorted end of the hairpin).
• Substrate selection: Do not use FR4 above 3 GHz (tanδ = 0.02 → excessive IL). Use Rogers RO4003C (tanδ = 0.0027) or PTFE-based laminates.
• Symmetry: Layout must be exactly symmetric about the filter's axis. Any asymmetry introduces even/odd mode imbalance and splits the passband.

Key distinction:

• Diplexer: A three-port passive device that separates signals by frequency band. Connects one "common" port to two "branch" ports, each passing a different frequency band. Both branches may be transmit or both receive — there is no TX/RX constraint. The two channels do not overlap in frequency.
• Duplexer: A three-port device specifically for simultaneous TX and RX on different frequencies (FDD). The TX and RX frequencies are close (same band) and the key spec is TX-to-RX isolation. A duplexer is a special case of a diplexer with very demanding TX/RX isolation.

Diplexer example — dual-band Wi-Fi (2.4 GHz and 5 GHz):

• Common port connects to the antenna (or broadband RF chain)
• Branch 1: LPF or BPF passing 2.4–2.5 GHz, rejecting 5.1–5.9 GHz by ≥40 dB
• Branch 2: HPF or BPF passing 5.1–5.9 GHz, rejecting 2.4–2.5 GHz by ≥40 dB

Design challenge — port impedance matching: The key issue is that each filter, looking from the common port, must present a matched 50 Ω in its passband but a high impedance (not a short) in the other branch's passband — otherwise the two branches load each other and degrade both passbands.

Implementation options:

• LC diplexer: suitable up to 3 GHz, compact, cheap, IL ~1–2 dB
• SAW diplexer: two SAW filters in a single package — used in 2.4/5 GHz Wi-Fi modules
• Microstrip coupled-line diplexer: junction design with branch-line stubs; used at frequencies where PCB fabrication is practical

The quality factor Q of a filter resonator measures the ratio of energy stored to energy dissipated per cycle. It determines both the insertion loss and the achievable selectivity of the filter.

Unloaded Q (Q_u) of different resonator technologies:

• SMD chip inductor at 1 GHz: Q ≈ 30–60 — high loss, avoid for narrow BPF
• Microstrip hairpin resonator at 2.4 GHz: Q ≈ 80–150
• SAW resonator at 1.9 GHz: Q ≈ 500–2000 — enables narrow cellular BPFs
• BAW resonator (FBAR): Q ≈ 1000–3000 — best Q for RF front-end filters
• Cavity resonator at 1 GHz: Q ≈ 5000–50000 — very low IL, used in base stations
• Dielectric resonator at 5 GHz: Q ≈ 5000–20000

Why Q limits selectivity: A higher-order filter with low-Q resonators suffers high insertion loss in the passband. The minimum achievable IL = 8.686×n×f_0/(Q_u×BW) dB. For a 5th-order BPF at 900 MHz with 30 MHz BW using Q=80 resonators: IL = 8.686×5×900/(80×30) = 16.3 dB — unusable. With Q=2000 SAW resonators: IL = 0.65 dB — acceptable.

Both SAW (Surface Acoustic Wave) and BAW (Bulk Acoustic Wave) filters use piezoelectric materials to create acoustic wave resonators with extremely high Q. They are the enabling technology for narrow-band RF filters in smartphones.

SAW filters:

• Use interdigitated transducer (IDT) electrodes on a piezoelectric crystal (LiTaO₃, LiNbO₃, quartz) to launch acoustic waves along the crystal surface
• Resonant frequency determined by IDT finger pitch: fâ‚€ = v_acoustic/λ_acoustic
• Frequency range: 100 MHz – 3 GHz. Above 3 GHz, the IDT fingers become too small for reliable manufacturing (<0.5 μm)
• Insertion loss: 1.5–3 dB. Temperature stability: moderate (TCF ≈ −40 ppm/°C for LiTaO₃)
• Cost: low — mature manufacturing process, wafer-scale production

BAW filters (FBAR or SMR type):

• Use a thin-film piezoelectric layer (AlN, ZnO) suspended over a cavity. Resonance is a bulk thickness vibration: fâ‚€ = v_acoustic/(2t) where t is film thickness
• Frequency range: 1–20 GHz. The thin film can be made thin enough for mmWave frequencies.
• Insertion loss: 0.5–1.5 dB — lower than SAW due to higher Q. Temperature stability: very good (TCF ≈ −20 ppm/°C, can be compensated to <3 ppm/°C)
• Power handling: excellent (up to 4 W in PA output duplexers) — critical for transmit path
• Cost: higher than SAW — requires precise thin-film deposition and MEMS processing

When to choose each:

• SAW: receive path filters below 3 GHz, low-cost applications, wideband filters
• BAW: TX path filters (PA output duplexers), frequencies above 2.5 GHz (LTE Band 7, n41), temperature-critical applications, 5G sub-6 GHz filters

A transmission zero is a frequency at which the filter completely blocks signal transmission (S21=−∞ dB, or in practice, limited by the filter's finite Q to some finite but very large attenuation). They appear as sharp notches in the stopband.

Why they matter: Butterworth and Chebyshev filters have no transmission zeros (all their poles are at s=∞, meaning infinite-frequency stopband) — their stopband attenuation increases at only 20·n dB/decade. Transmission zeros at finite frequencies create much steeper skirts right outside the passband, achieving the required stopband rejection with fewer filter poles (lower insertion loss, smaller size).

Elliptic (Cauer) filter:

• Places transmission zeros at finite frequencies in both the stopband above AND below the passband (for a BPF)
• Result: equiripple in both passband AND stopband, with steepest possible transition for a given filter order
• Tradeoff: non-monotonic stopband (the attenuation level between TZs may not be the minimum required), and non-linear group delay (poor phase response)

Quasi-elliptic filter:

• Places one or a few transmission zeros near the passband edge but not a full equiripple stopband design
• Common in SAW/BAW filters — the ladder topology naturally creates TZs at the series and shunt resonator anti-resonances
• More practical to implement than full elliptic, with most of the selectivity benefit

Physical implementation of TZs:

• Cross-coupling between non-adjacent resonators creates signal paths that cancel at specific frequencies — the "source-load coupling" technique
• Bypassing a resonator with a capacitor or shunt stub creates a notch at the resonator's frequency
• In BAW ladder filters, the ratio of series to shunt resonator frequencies is set by electrode geometry to place TZs at desired frequencies

This is the classic filter design procedure: LP prototype → frequency scaling → bandpass transformation → normalised element values → final component values.

Step 1 — LP prototype element values (0.5 dB ripple Chebyshev, n=3):

Step 2 — Bandpass transformation:

• Centre frequency: fâ‚€=2400 MHz, bandwidth: BW=100 MHz, fractional BW: Δ=BW/fâ‚€=0.0417
• Each LP series inductor L_i → series LC bandpass resonator: L_si=g_i/(ω₀Δ), C_si=Δ/(ω₀g_i)
• Each LP shunt capacitor C_i → shunt LC bandpass resonator: C_pi=g_i/(ω₀Δ), L_pi=Δ/(ω₀g_i)

Step 3 — Denormalise (Z₀=50 Ω, ω₀=2π×2.4×10⁹):

• Series resonator 1 (from g₁=1.5963): L_s1=Z₀·g₁/(ω₀Δ)=50×1.5963/(2π×2.4e9×0.0417)=255 pH, C_s1=Δ/(ω₀Zâ‚€g₁)=0.0417/(2π×2.4e9×50×1.5963)=17.3 fF
• Shunt resonator 2 (from gâ‚‚=1.0967): C_p2=gâ‚‚/(ω₀Z₀Δ)=1.0967/(2π×2.4e9×50×0.0417)=174 fF, L_p2=Z₀Δ/(ω₀gâ‚‚)=50×0.0417/(2π×2.4e9×1.0967)=127 pH
• Series resonator 3 (same as 1 by symmetry): L_s3=255 pH, C_s3=17.3 fF

Step 4 — Practical realisation:

• 255 pH at 2.4 GHz: achievable with a bondwire or short microstrip stub
• 17 fF at 2.4 GHz: extremely small — a 0.1 mm × 0.1 mm gap between microstrip lines
• For microwave implementation: use coupled-line or inter-digital filter sections that realise the same response in distributed form

A cavity filter uses enclosed metal cavities as the resonating elements. Each cavity resonates when its dimensions equal half a wavelength (or a quarter-wave when shorted at one end). The EM field is entirely contained inside the metal enclosure, dramatically reducing radiation losses and enabling extremely high Q factors.

How cavity filters achieve high Q:

• The resonant energy is stored in the electric and magnetic fields inside the cavity — there is no radiation loss (fully enclosed)
• The only loss is conduction loss in the cavity walls (R_wall ∝ √f due to skin depth)
• Q_u ∝ V/S_wall ratio — larger cavities (lower frequency or larger physical size) have higher Q because the volume-to-surface-area ratio increases
• Silver plating the interior reduces R_wall → Q increases. Typical Q: 5000–50000
• TE₀₁₁ mode in cylindrical cavities has no wall currents along the z-axis — coupling screws don't degrade Q. This is why cylindrical TE₀₁₁ cavities achieve Q >100000 for filter applications

Where cavity filters are used:

• Cellular base station duplexers (TX +43–53 dBm, isolation >80 dB needed, IL <0.5 dB)
• Amateur radio repeater duplexers (separate TX and RX on same antenna tower)
• Satellite uplink/downlink multiplexers (narrow channel spacing, very low IL)
• Radar receiver protectors (high-Q bandpass filter before the LNA)
• EMC testing: cavity filters block unwanted harmonics from test signals

Tuning: A metal screw inserted into the cavity changes the effective cavity volume, shifting the resonant frequency. This allows field tuning after assembly to correct manufacturing tolerances.

The image parameter method (1920s–1940s) was the first systematic approach to filter design. It characterises the filter by the impedance a semi-infinite ladder network of identical sections would present — the "image impedance." When the source and load are matched to the image impedance, the transfer function can be calculated.

Image parameter method:

• Filter is a cascade of identical LC ladder sections (constant-k or m-derived sections)
• Image impedance Z_i = √(Z_series × Z_shunt) where Z_series and Z_shunt are the series and shunt arms
• Propagation constant: γ = arccosh(1 + Z_series/(2Z_shunt))
• Design is intuitive and graphical — but gives imprecise results when the actual source/load are 50 Ω (not the image impedance)
• Terminated sections have ripple in the passband and rounded stopband transitions — worse than the design predicts

Modern network synthesis (Darlington, 1950s onwards):

• Start with a mathematically specified transfer function H(s) (Butterworth, Chebyshev, etc.) with exact passband ripple and stopband attenuation
• Factor H(s) into a realizable network topology using Cauer synthesis or ladder synthesis algorithms
• The resulting component values (g₁, gâ‚‚, ...) guarantee exact frequency response when terminated in Zâ‚€
• Extends straightforwardly to bandpass, highpass, bandstop transformations via frequency transformations
• Optimal: minimum number of elements for a given specification

When image parameters are still used:

• Transmission line design at microwave frequencies where distributed elements don't map cleanly to lumped-element synthesis
• Intuitive insight into filter behaviour (why adding m-derived end sections improves termination match)

Both combline and interdigital filters are microstrip or coaxial bandpass filters using coupled resonator rods, but they differ in the grounding of the rods and the resulting frequency response characteristics.

Combline filter:

• All resonator rods are shorted to ground at the same end (hence "comb" — all teeth on the same rail)
• Each rod is electrically shorter than λ/4 (typically 0.1λ to 0.25λ), loaded with a lumped capacitor to ground at the open end to bring it to resonance
• Adjacent rods couple through the fringing electric field between them (TEM-mode coupling)
• Advantages: compact size, spurious-free region extends to 5–6× the passband frequency, easy tuning (adjust capacitor screws)
• Used in: VHF/UHF base station filters, satellite uplink filters

Interdigital filter:

• Alternate rods are shorted at opposite ends — one rod grounded at top, next at bottom, next at top... (hence "interdigital")
• Each rod is exactly λ/4 long at the centre frequency (no capacitor loading needed)
• Coupling is through both electric and magnetic fields between adjacent rods
• Advantages: very wide stopband (spurious-free to 2× passband), no tuning elements needed (fewer manufacturing steps)
• Disadvantages: longer than combline (λ/4 rods vs shorter combline rods), harder to tune after assembly
• Used in: microwave filters where spurious suppression is critical, mmWave bandpass filters

Group delay τ_g(ω) = −dφ/dω where φ(ω) is the phase of the filter's transfer function H(jω). It represents the delay experienced by the envelope of a narrowband signal at frequency ω.

Linear phase filter: If φ(ω)=−ω·τ₀ (perfectly linear phase), then τ_g=τ₀ = constant — all frequency components of the signal experience the same delay. The signal is delayed but its waveform shape is perfectly preserved.

Why linear phase matters:

• If group delay varies across the filter passband (non-constant), different frequency components of the signal arrive at the filter output at different times
• For a digital signal with bandwidth B, group delay variation ΔGD across B causes inter-symbol interference (ISI): each symbol spreads into adjacent symbols by ΔGD seconds
• Specification: ΔGD should be much less than one symbol period T_s. For 1 Gbps: T_s=1 ns → ΔGD should be <0.1 ns

Filter types and phase response:

• Butterworth: moderate group delay variation — acceptable for many applications
• Chebyshev: worse group delay (more ripple in GD near the band edges due to the equiripple amplitude response)
• Elliptic: worst group delay — large peaks near the transmission zeros. Only usable with a GD equaliser
• Bessel-Thomson: maximally flat group delay — all frequencies delayed by the same amount. Poorest amplitude selectivity but best phase response
• Linear phase FIR (digital): exact linear phase by design, at the cost of computational complexity

A bandstop filter (BSF), also called a notch filter or band-reject filter, passes all frequencies except those in the stop band, which is attenuated. The attenuation reaches a maximum at the notch frequency fâ‚€.

Design procedure (LP prototype transformation):

• Select a LP prototype (Butterworth, Chebyshev) of order n with the stopband frequency equal to the prototype's cutoff
• Apply the bandstop frequency transformation: s → ω₀/(s·BW_frac) where BW_frac is the fractional bandwidth
• Each LP shunt capacitor becomes a shunt LC parallel resonator (high impedance at fâ‚€ → removes signal)
• Each LP series inductor becomes a series LC series resonator (low impedance at fâ‚€ → shorts signal to ground)

Practical BSF example — notching 2.4 GHz WiFi to protect a 433 MHz IoT receiver:

• Design a 3rd-order Chebyshev BSF centred at 2.4 GHz with 200 MHz notch bandwidth
• LP prototype g values → BSF shunt parallel resonators: C_p=g_i/(ω₀·Z₀·Δ), L_p=Z₀·Δ/(ω₀·g_i)
• Series resonators: C_s=Δ/(ω₀·Z₀·g_i), L_s=Z₀·g_i/(ω₀·Δ)
• At 2.4 GHz: component values are in the range of 1–10 pF and 1–5 nH — achievable with SMD components below 5 GHz

Microwave implementation: At 5+ GHz, a λ/4 shorted stub presents an open circuit at f₀ (the notch frequency) — inserted in series with the transmission line, it blocks signal at exactly that frequency. Multiple stubs create multiple notches.

Common applications:

• Blocking an interfering carrier (e.g. cellular blocking an adjacent IoT band receiver)
• Harmonic suppression in transmitter output stages
• Radar: blocking transmit bleedthrough at the receiver (implemented as a duplexer with a BSF characteristic)
• EMC: notching internal clock harmonics before they reach an antenna

Component tolerance causes the actual filter response to deviate from the design — passband edges shift, ripple increases, stopband attenuation degrades, and the centre frequency drifts. The sensitivity of the response to component variation depends critically on the filter type and order.

Sensitivity analysis:

• Sensitivity S^H_C = (dH/H)/(dC/C) = fractional change in response per fractional change in component value
• High-order Chebyshev and elliptic filters have very high sensitivity — a 2% capacitor error can cause 3 dB of ripple variation in a 0.1 dB design
• Butterworth filters have lower sensitivity — more forgiving to component errors
• Bessel filters have the lowest sensitivity — the gradually sloping response means component errors cause only gentle deviations

Designing for tolerance robustness:

• Use wider specification margins: Design for 0.5 dB ripple if your spec says 1 dB — the tolerance spread won't exceed the spec
• Use lower-order filters: Every extra pole increases sensitivity. A 3rd-order design is far more robust than a 7th-order design for the same specification if that specification can be met.
• Use 1% tolerance components: At the cost of ~3× price, 1% components reduce sensitivity contribution by 5× compared to 5% components. For microwave filters this is non-negotiable.
• Use C0G/NP0 capacitors: X5R and X7R capacitors have ±15% capacitance change with voltage (piezoelectric effect) — disastrous for a filter. C0G is ±30 ppm/°C with no voltage coefficient — essential for precision RF filters.
• Monte Carlo simulation: Before building, run 1000 simulations with component values randomly varied within ±5%. If >99% of simulations meet the spec, the design is tolerant. If not, re-design with lower order or wider margin.
• Tunable filters: Add trim capacitors or variable inductors that can be adjusted post-manufacture to centre the response. Used in VHF/UHF cavity filters.

The LP prototype is a normalised (Z₀=1 Ω, cutoff frequency ω_c=1 rad/s) filter with specific g-values that define the Butterworth, Chebyshev, or other response. All practical filters are derived from this prototype via mathematical transformations.

LP to LP (frequency scaling): Multiply all reactive component values by 1/ω_c. An inductor becomes L_new=L_prototype/ω_c; a capacitor becomes C_new=C_prototype/ω_c.

LP to HP: Replace s with 1/s in the transfer function. Each inductor L → capacitor 1/L; each capacitor C → inductor 1/C. The prototype LPF inductor becomes an HPF capacitor.

LP to BP (bandpass transformation):

Practical procedure:

• Each LP prototype series element → series LC resonator (resonates at fâ‚€)
• Each LP prototype shunt element → shunt LC resonator (resonates at fâ‚€)
• Number of components doubles (from n LP elements to 2n BP elements)
• Then scale for impedance: multiply inductors by Zâ‚€, divide capacitors by Zâ‚€

A multiplexer connects N bandpass filters to a common port (antenna port) — a triplexer for 3 bands, quadraplexer for 4, etc. Each filter passes one band and rejects all others, allowing multiple radios to share a single antenna.

Design challenge 1 — Inter-channel interference:

• The stopband of each filter must adequately reject all other channels. If filter 1 (2.4 GHz) passes −40 dB of 5 GHz signal, and the 5 GHz radio transmits +20 dBm, the 5 GHz signal reaching the 2.4 GHz LNA is −20 dBm — potentially saturating it.
• Solution: each filter must have sufficient stopband rejection at ALL other channel frequencies, not just adjacent channels.

Design challenge 2 — Port impedance mismatch loading:

• At the common port, all N filters are in parallel. At 2.4 GHz, the 2.4 GHz filter presents a low (matched) impedance; the 5 GHz filter presents its stopband impedance.
• If the 5 GHz filter's stopband impedance is not a high impedance (open circuit) at 2.4 GHz, it loads the 2.4 GHz filter — degrading its passband response.
• Ideal filter: presents its characteristic impedance in the passband AND a high (open circuit) impedance in the stopband. Real filters approach but don't perfectly achieve this.

Design challenge 3 — Junction design:

• The physical junction where all filter inputs connect must be modelled as a transmission line junction — its electrical length and parasitic inductance (PCB via, SMD pad) affects all channels.
• Solution: EM simulation of the full junction + filter assembly, not just each filter in isolation.
• Manifold multiplexer: uses a transmission line manifold with filter stubs — the line length between filter connection points is part of the design and must be optimised.

Co-design approach:

• Each filter's termination impedance at the junction depends on what the other filters present — they are not independent
• Modern approach: co-synthesise all filters simultaneously with the junction model, optimising the entire multiplexer response as a system

A helical resonator is a coil of conductor wound inside a shielded metallic can. The coil acts as a distributed LC resonator — the inductance of the helix and the capacitance between the helix and the can wall resonate at the design frequency. One end of the helix is open (unconnected), the other is shorted to the can wall.

Advantages over standard LC filters at VHF/UHF:

• High Q: Typical Q_u = 500–2000 at 100–500 MHz — far higher than a wound chip inductor (Q=50–100 at these frequencies). Lower insertion loss for the same filter order.
• Compact vs cavity: A cavity resonator at 150 MHz would need to be approximately 50 cm long (λ/4 at 150 MHz). A helical resonator achieves the same Q in a can only 3–5 cm tall by using the loading capacitance to "slow" the resonance.
• Better shielding: The metal can completely encloses the resonator, preventing radiation and coupling to nearby circuits.
• Tunable: A small tuning screw through the top of the can adjusts the capacitance — field tuning after assembly corrects manufacturing tolerances.

Disadvantages:

• Hand-wound coils have tight manufacturing tolerances — relatively expensive
• Not suitable above ~500 MHz where the helix self-resonance creates spurious modes
• Not miniaturisable — always at least several centimetres

Applications:

• VHF/UHF amateur radio repeater input filters (146 MHz, 440 MHz)
• Land mobile radio (LMR) base station bandpass filters
• First IF filter in high-performance HF receivers (455 kHz to 10.7 MHz helical resonators exist)

Return loss (RL) of a filter — the S11 measured in the passband — indicates how well the filter input is matched to the source impedance (usually 50 Ω). For a well-designed filter, the passband return loss should be high (good match). Poor passband RL means a significant fraction of the signal is reflected rather than transmitted.

What poor passband return loss indicates:

• Design error: The filter was designed for a different source/load impedance than 50 Ω. A Chebyshev filter designed for gâ‚€=1.0 but terminated by a source that includes cable loss will show RL degradation.
• Calibration plane error: The calibration reference plane is not at the filter input — transmission line between calibration plane and filter transforms the impedance, changing the apparent S11.
• Component value errors: Particularly the first series inductor or first shunt capacitor — these components control S11 most directly. A 5% error in the first element can drop RL from 20 dB to 10 dB.
• Resonator detuning: A resonator whose resonant frequency is off (due to manufacturing tolerance) mismatches the filter at the frequencies where that resonator contributes most.
• Connection losses: A lossy connector or poorly soldered joint between the source and filter creates a "padding resistor" that absorbs reflected power — artificially improving apparent RL while hiding real filter problems.

Passband RL and its relationship to filter type:

• Butterworth N=3: passband RL ≥ 17 dB at band centre, degrades to ≥ 0 dB at band edge (monotonic)
• Chebyshev 0.5 dB ripple: passband RL oscillates between 9.6 dB (at ripple maxima) and ∞ dB (at ripple nulls where perfect match occurs)
• These values assume perfectly matched terminations — any deviation degrades them proportionally

A dielectric resonator filter uses a high-permittivity ceramic disc or puck (typically barium titanate, alumina, or titanate-based ceramics with εr=20–100) as the resonating element, placed inside a metallic enclosure. The high εr causes the EM fields to be mostly concentrated inside the ceramic, with only fringing fields outside — dramatically reducing radiation losses and achieving very high Q.

How it achieves high Q:

• Most of the resonant energy is stored in the high-εr ceramic where the dielectric loss tangent (tanδ=10⁻⁵ to 10⁻⁴) is extremely low for high-purity ceramics
• Radiation losses are small because the high εr "traps" the EM field inside — like total internal reflection in fibre optics
• Conductor losses from the enclosure walls are reduced because the metallic walls see only weak fringing fields
• Achievable Q_u: 5000–50000 at 1–30 GHz
• Size: a factor of √εr smaller than the equivalent air-filled cavity → compact at microwave frequencies

Why DRFs dominate satellite communications:

• Satellite output multiplexers (OMUX) must combine 24–60 transponder channels (each 36–72 MHz wide) into a single output. Each channel filter must have insertion loss <0.3 dB — any more and the satellite's DC-to-RF efficiency drops unacceptably (affecting battery sizing and launch mass).
• DRFs achieve this with Q_u >30000 — no competing technology (LC filters, SAW/BAW, waveguide cavities) achieves this combination of size, mass, and insertion loss at 3–30 GHz.
• Temperature stability: satellite components must operate from −40°C to +80°C in orbit. High-quality ceramic formulations have TCF (temperature coefficient of frequency) below ±2 ppm/°C — critical for maintaining channel alignment across the orbital thermal cycle.

Coupling mechanisms: Adjacent DRs are coupled through the fringing field between them. Coupling coefficient is adjusted by changing the spacing or by adding a metal iris (aperture) between cavities. Input/output coupling uses a small coupling loop or probe inserted near the DR.