S-Parameters Q&A

S-parameters (scattering parameters) describe a network in terms of travelling power waves — incident, reflected, and transmitted — at each port. The S-matrix relates the outgoing wave amplitudes to the incoming wave amplitudes.

Why not Z/Y at RF? Z and Y parameters require open-circuit or short-circuit terminations while measuring. At high frequencies this is nearly impossible — a short circuit becomes an inductor, an open circuit radiates, and probe parasitics dominate. The measurement is meaningless.

S-parameters are measured with the ports terminated in the reference impedance Z₀ (usually 50 Ω), which is easy to realise with coaxial lines and matched loads. A VNA does exactly this: it sources a wave, measures reflections and transmissions, and the device is always in its normal operating condition.

For a 2-port device (DUT connected to Port 1 = input, Port 2 = output):

• S11 — Input reflection coefficient: How much of the signal applied at Port 1 bounces back. Bad S11 (close to 0 dB) means you're wasting power fighting reflections. Target: S11 < −10 dB (VSWR < 2:1) for most designs.
• S21 — Forward transmission (gain/loss): How much power gets through from Port 1 to Port 2. For an amplifier S21 > 0 dB (gain). For a filter, S21 in the passband is the insertion loss (negative, ideally close to 0).
• S12 — Reverse transmission (isolation): Signal leaking from output back to input. High S12 causes feedback and instability in amplifiers. For a filter, S12 ≈ S21 (reciprocal). For an amplifier you want S12 as negative as possible.
• S22 — Output reflection coefficient: Mismatch at the output port. Affects how well the device drives the next stage.

Practical example — measuring a bandpass filter at 2.4 GHz:

• S21 in passband should be > −1 dB (low insertion loss)
• S21 in stopband should be < −40 dB (good rejection)
• S11 in passband should be < −15 dB (matched input)

Return Loss (RL) is a positive dB number expressing how much less the reflected power is compared to incident power. The higher the RL, the better the match.

Key values to memorise:

• RL = 6 dB → VSWR 3.0:1 → 25% reflected — barely usable
• RL = 10 dB → VSWR 1.93:1 → 10% reflected — minimum for most RF
• RL = 14 dB → VSWR 1.5:1 → 4% reflected — typical antenna spec
• RL = 20 dB → VSWR 1.22:1 → 1% reflected — very good
• RL = 40 dB → VSWR 1.02:1 → 0.01% reflected — excellent (lab standard)

Common pitfall: A VNA displays S11 in dB as a negative number (e.g. −14 dB). Return loss is the positive version of this. Don't confuse them — a device with S11 = −14 dB has a return loss of 14 dB.

For an air-filled coaxial cable, the optimum impedance depends on what you want to optimise:

50 Ω was standardised by the US military in the 1940s for radar equipment and became universal in RF/microwave. All VNAs, signal generators, spectrum analysers, and most RF components are 50 Ω.

75 Ω in cable TV: Cable TV systems optimise for minimum loss over long cable runs (the attenuation minimum is nearer 77 Ω). Since the cables can be hundreds of metres long, a few tenths of a dB per metre saved adds up significantly. Power handling is not a concern at the milliwatt levels used in broadcasting.

Insertion loss is the reduction in transmitted power caused by inserting a device into a signal path. For a passive device it equals −S21 in dB (since S21 is a negative number for a lossy device).

Correct VNA measurement procedure:

• Step 1 — Calibrate: Perform a full 2-port calibration (SOLT or equivalent) at the cable ends where the DUT will connect. This removes cable and connector effects.
• Step 2 — Thru verification: Connect a thru (zero-length connector) and verify S21 reads 0.0 dB ± 0.1 dB. If not, re-calibrate.
• Step 3 — Connect DUT: Insert the device under test between the calibrated ports.
• Step 4 — Read S21: The displayed S21 magnitude is the insertion loss.

Common mistake: Measuring insertion loss without calibration includes cable attenuation (~0.5 dB/m at 1 GHz for RG58). A 2 m setup has 1 dB of cable error — enough to completely mischaracterise a low-loss filter.

VNA calibration moves the measurement reference plane to the DUT connector tips and removes systematic errors (source match, load match, directivity, reflection tracking, transmission tracking, isolation). SOLT uses 4 known standards:

• S — Short: |Γ| = 1, phase ≈ 180°. Defines the reference plane for reflection. Even a "short" has a small inductance (offset L) — the cal kit data sheet specifies it as a polynomial vs. frequency.
• O — Open: |Γ| = 1, phase ≈ 0°. The fringing capacitance of an open connector is characterised by the cal kit (C0, C1, C2, C3 polynomial). Without this correction, open vs. short are indistinguishable in phase.
• L — Load (Match): |Γ| ≈ 0 (typically −40 dB or better). Sets the system impedance reference. A poor load standard (e.g. a worn SMA terminator) directly limits your return loss measurement accuracy.
• T — Thru: Measures the transmission path. Zero-length thru (ports mated directly) is ideal. If using an insertable thru (e.g. male-male), the cal kit must include its electrical length.

Practical notes:

• Always torque SMA connectors to 0.9 N·m (8 in-lb) — under/over torque shifts the reference plane
• Calibrate at the temperature you will measure — connectors expand with heat and shift electrical length
• Re-calibrate if you change cables, add adapters, or move connectors
• SOLR (Unknown Thru) is preferred when you cannot mate ports directly — it self-characterises the thru

The Smith Chart is a polar plot of the reflection coefficient Γ, with overlaid constant-resistance (R) and constant-reactance (X) circles. It maps complex impedance onto the unit circle of |Γ| ≤ 1.

Reading impedance: Find the intersection of the R-circle and X-arc. The normalised value reads directly (multiply by Z₀ = 50 Ω). A point at (r=1, x=0) = 50+j0 Ω = perfect match.

Moving along a transmission line: Rotate clockwise (toward generator) or counter-clockwise (toward load) around a constant-|Γ| circle. One full rotation = λ/2 of line length. The outer scale "wavelengths toward generator" tracks this.

Adding shunt/series elements:

• Series inductor L: move along constant-R circle, upward (increasing +jX)
• Series capacitor C: move along constant-R circle, downward (decreasing −jX)
• Shunt inductor: move along constant-G circle on the admittance chart
• Shunt capacitor: move along constant-G circle upward on admittance chart

This is one of the most important practical questions. Sim vs. measured discrepancies are almost universal and have specific root causes:

• Substrate parameters (εr, tanδ): PCB manufacturers quote nominal εr (e.g. 4.4 for FR4) but actual εr varies ±0.2 between board lots and is frequency-dependent. At 5 GHz, FR4 εr may be 4.1, not 4.4. Trace widths must be re-tuned.
• Conductor losses: Most simulators use ideal smooth conductors. Real PCB copper has surface roughness of 0.5–2 μm Rz, which at GHz frequencies (skin depth <2 μm) adds 20–50% extra loss.
• Component parasitics: A 10 nH inductor in simulation is ideal. The real part has self-resonance at maybe 3 GHz and looks capacitive above that. Always use vendor S-parameter models, not ideal values.
• Via and pad models: PCB vias have parasitic inductance (0.5–1 nH) and capacitance — often not modelled. At 5 GHz a 1 nH via inductance is 31 Ω of reactance — not negligible.
• Calibration plane mismatch: If calibration was done at cable ends but the simulation model starts at the SMA pad, the electrical length difference (few mm) shifts S11 phase significantly at high frequencies.
• Connector models: SMA connectors are not ideal transmission lines. They have discontinuities, especially at transitions between pin and PCB.
• Temperature and humidity: FR4 εr shifts with moisture absorption — outdoor/humid environments can shift resonant filters by 10–50 MHz.

A VNA's ports are designed to present and receive signals at exactly Z₀ = 50 Ω (or 75 Ω on some instruments). The S-parameter definition requires this: S11 is measured with port 2 terminated in 50 Ω and the source impedance at port 1 is 50 Ω.

What happens if the DUT is not 50 Ω: You still get a perfectly valid S11 — it just tells you how much the DUT reflects relative to 50 Ω. If the DUT is designed for 75 Ω, you'll measure a large S11 even though the device works perfectly in its intended 75 Ω environment.

Port impedance renormalisation: Modern VNAs can renormalise S-parameters to any impedance after measurement. If your DUT works in a 75 Ω system, tell the VNA to use Z₀ = 75 Ω and it converts the 50 Ω raw data mathematically.

Practical example: You are characterising a 100 Ω balanced differential trace on a PCB. Measuring with 50 Ω probes gives misleading S11. You need either: (a) a 50-to-100 Ω balun, (b) a differential VNA measurement, or (c) post-measurement renormalisation to 100 Ω.

Isolation is the attenuation of an unwanted signal between two ports that should be independent. For a switch in the OFF state, it's −S21 in dB. For a diplexer, it's the rejection between the two output ports.

How to measure isolation: It's simply −S21 or −S12. But measuring high isolation (e.g. >80 dB) hits the dynamic range limit of the VNA.

What limits isolation measurement:

• VNA dynamic range: A typical VNA at 1 GHz has 100–130 dB dynamic range. Above this limit, you're measuring VNA noise, not DUT isolation.
• Direct electromagnetic coupling between cables: If the two VNA cables run parallel and close together, they couple electromagnetically, adding a "leakage floor" of −60 to −90 dB regardless of what's connected. Separate and shield the cables.
• Ground loops and radiated coupling: Even without cables, RF radiates from the cable shields and couples into nearby conductors.
• IFBW (IF bandwidth): Narrowing the IF bandwidth reduces noise floor and extends measurable isolation. At IFBW = 10 Hz, you can typically measure 120+ dB isolation, but each sweep takes minutes.

Practical technique for high isolation: Use a coaxial shield cage around the DUT, route cables in opposite directions, and reduce IFBW to 100 Hz or less. Also perform an isolation calibration (measure isolation with ports terminated in matched loads before connecting DUT).

S12 is the reverse transmission coefficient — signal leaking from the amplifier output back to the input. Even a tiny S12 can cause feedback if the phase is wrong, leading to oscillation.

Rollett stability factor K:

Why it matters in practice:

• K < 1 means the device is potentially unstable — it can oscillate for some source/load impedances. The amplifier may work fine with 50 Ω but oscillate when connected to a reactive antenna.
• Even unconditionally stable amplifiers (K > 1) can oscillate at frequencies outside the measurement range — always check stability over a wide frequency span (e.g. 1 MHz to 10× operating frequency).
• Stabilisation techniques: series resistor at input/output (reduces gain but improves K), feedback resistor, or lossy matching network.

Practical measurement tip: When you connect your amplifier and it oscillates, check: (1) Is K < 1 at any frequency? (2) Is your source or load impedance reactive? (3) Does adding a ferrite bead or 10 Ω series resistor at the input stop it? If yes, the instability is caused by reactive loading together with insufficient reverse isolation.

De-embedding is the mathematical process of removing the effects of fixtures, probes, connectors, or PCB launch structures from raw VNA measurements, to obtain the S-parameters of just the DUT itself.

Why you need it: When measuring a chip or bare die, you cannot connect a VNA directly — you use probes, a probe station, and a PCB test fixture. These add length, impedance discontinuities, and loss. If you don't remove them, you're measuring the fixture, not the chip.

Methods:

• ABCD matrix method: Convert the total measurement to ABCD matrix, subtract (divide) the fixture ABCD matrices on both sides: [DUT_ABCD] = [Fixture_left]⁻¹ · [Total_ABCD] · [Fixture_right]⁻¹. Then convert back to S-parameters.
• Two-port SOLT at the DUT pads: The gold standard — physically calibrate at the DUT's exact reference plane using on-board calibration structures (short, open, load printed on the PCB). Removes everything up to the pad.
• Open-short de-embedding (semiconductor): Measure an "open" structure (same fixture, no DUT) and a "short" structure, use Y and Z matrix subtraction. Common for transistor characterisation on-wafer.

Practical example: You are measuring a 0402 chip capacitor on a PCB. The SMA connector, 5 mm of microstrip, and the pad structure all affect the measurement. Print a "thru" structure (same launch — no capacitor, just the pads shorted) on the same PCB, measure it, and de-embed it from your capacitor measurement to get the true chip S-parameters.

S-parameters cannot be directly cascaded by matrix multiplication because they relate scattered waves at all ports simultaneously. To cascade two networks (e.g. amplifier + filter), convert to ABCD matrices, multiply, then convert back.

Why useful in RF design:

• System-level simulation: chain amplifier + filter + coupler + cable by multiplying ABCD matrices at each frequency point
• Antenna feed analysis: cascade balun + matching network + transmission line + antenna
• De-embedding: remove known fixture by multiplying by its inverse ABCD matrix
• Used by all RF simulation tools (ADS, Microwave Office) internally

Other parameter conversions: Y-parameters are useful for shunt elements (add Y matrices for parallel networks). Z-parameters useful for series elements (add Z matrices for series networks). T-parameters are similar to ABCD but use wave quantities — useful for coaxial system analysis.

Connectors are the most common source of S-parameter measurement errors. They cause: additional insertion loss, impedance discontinuities, phase shift, and if damaged — large ripple across the frequency range.

How a bad connector manifests in measurements:

• Ripple in S21 and S11: A damaged or poorly mated connector creates a small reflection. This reflection bounces back and forth between two discontinuities, creating sinusoidal ripple in S21. The period of ripple tells you the round-trip electrical length (Δf = c / (2·l·√εr)).
• Elevated insertion loss: Oxidised or worn contacts add 0.1–0.5 dB per connector at GHz frequencies.
• High S11 floor: A damaged center pin doesn't mate tightly — you get a floating contact and unpredictable reflection.
• Non-repeatable measurements: If S21 changes by >0.1 dB each time you re-connect, the connector is suspect.

Diagnostic procedure:

• Connect a thru (short cable). S21 should be <0.2 dB loss and flat. Any tilt or ripple is cable/connector error.
• Wiggle the cable while watching the VNA — if trace moves, there's a bad connection.
• Inspect visually: bent pin, debris in connector, gold worn off contact.
• Try phase of S11: it should rotate uniformly with frequency. Abrupt phase jumps indicate multiple reflections from a damaged connector.

Connector lifetime: SMA connectors are rated for 500–1000 matings. 2.4 mm connectors rate ~2000 matings. Keep a logbook on frequently used cal kit standards — after 500 matings, re-verify them on a reference VNA.

Standard single-ended S-parameters describe each port as one conductor referenced to ground. Modern RF ICs and high-speed digital interfaces use differential (balanced) ports where the signal is the difference between two conductors. Mixed-mode S-parameters describe how differential and common-mode signals interact.

Mixed-mode S-matrix structure:

Why mode conversion matters: If Scd21 is high, your differential amplifier is converting differential signal into common-mode noise, which can interfere with other circuits. A good differential LNA should have Scd21 < −40 dB.

Measurement: A 4-port VNA can measure the full 4-port single-ended S-matrix of a differential device (2 differential ports = 4 single-ended ports). The VNA software then calculates mixed-mode parameters via a linear transformation (the "mode transformation matrix").

A VNA with time-domain (TD) capability performs an Inverse Fourier Transform of the frequency-domain S-parameters to produce a time-domain impulse or step response. This effectively gives you TDR (time-domain reflectometry) from S11, and TDT (time-domain transmission) from S21.

What you can see in time domain:

• Impedance discontinuities: Any reflection in S11 shows up as a peak in TDR. The sign tells you whether it's capacitive (negative spike) or inductive (positive spike). The time position tells you the distance: d = (t × vp) / 2.
• Connector locations: Each connector shows as a small blip at the expected electrical length.
• PCB impedance profile: A 50 cm PCB trace shows as a flat region at Z₀ = 50 Ω; if there's a pad or via, you see a deviation.
• Open and short faults: An open cable end gives a large positive reflection at the exact length of the cable. A short gives a large negative reflection.

Key parameters affecting TDR resolution:

• Rise time (10–90%) ≈ 0.35 / BW_max. At 20 GHz measurement bandwidth, risetime ≈ 17.5 ps, resolving features >1.75 mm (in air). Narrower features are smeared.
• Time span = λ_span / 2 (one-way propagation over the sweep span)
• Window function: Kaiser-Bessel reduces sidelobe ripple but broadens features slightly

Small-signal S-parameters are measured with a very low signal level (typically −20 to −30 dBm at the DUT input) where the device behaves linearly. They fully describe the device only in this linear regime.

When small-signal S-params fail:

• Gain compression: An amplifier's S21 (gain) decreases as input power increases. S21 at −30 dBm may be 20 dB, but at −5 dBm it's only 17 dB (3 dB compressed = P1dB). If you use small-signal S21 to predict output power at P1dB, you get a 3 dB error.
• Harmonic and intermodulation distortion: S-parameters are single-frequency quantities (superposition principle applies). In a nonlinear device, two tones at f1 and f2 produce products at 2f1−f2, 3f1−2f2 etc. — not captured by S-params.
• AM/AM and AM/PM conversion: In power amplifiers, the gain and phase shift depend on signal amplitude. S-parameters assume constant gain and phase — false for large signals.

Large-signal solutions:

• Load pull / source pull: Measure S21 (or output power) vs. source/load impedance at the actual operating power — finds optimum matching for maximum power or PAE.
• Large-signal S-parameters (LSSP): Some VNAs with power sweeps can extract "describing functions" — S-parameters measured at each power level.
• X-parameters / poly-harmonic distortion: Keysight's X-parameter measurement characterises nonlinear devices over power and frequency including harmonics.

This is a classic practical debugging scenario. A passband with ±0.3 dB ripple and poor S11 (meaning significant reflected power) suggests one or more of these root causes:

1. Calibration or fixture error (most common first suspect):

• Re-calibrate. Connect a thru — does S21 = 0 dB ± 0.05 dB? If not, your calibration is off.
• Check connector torque. Re-mate all connectors and re-measure — if ripple disappears, it was a loose connector creating a "resonant cavity" between two reflections.

2. Impedance mismatch at filter ports:

• A filter designed for 50 Ω but presented with a slightly different source/load impedance will show ripple and S11 degradation. The ripple spacing Δf = vp/(2·L) where L is the resonant path length.
• Check: is the filter datasheet specifying 50 Ω terminations? Are your test cables and connector adapters all 50 Ω? A 75→50 Ω mismatch would explain exactly this behaviour.

3. Resonance between filter and PCB/fixture:

• The transmission line sections between connectors and filter ports act as impedance transformers. At certain frequencies they transform the 50 Ω to something else, causing the ripple.
• Try connecting the filter with different cable lengths — if ripple shifts frequency, this is the cause. Solution: move the calibration plane to the exact filter input pads.

4. Component value tolerance or layout error:

• If the ripple is repeatable and calibration is verified: compare simulated vs. measured. Shift component values ±5% in simulation until the ripple pattern matches. This tells you which component is out of spec.
• For PCB filters: check trace widths with a microscope. A 50 μm under-etching on a 0.5 mm trace at 50 Ω changes Z₀ by ~3 Ω, causing ~12 dB return loss — consistent with your measurement.

S-parameters are defined relative to a reference impedance Z₀ at each port — almost always 50 Ω. The definition normalises voltages and currents to travelling waves: a₁ = (V₁ + Z₀I₁)/(2√Z₀) (incident wave), b₁ = (V₁ − Z₀I₁)/(2√Z₀) (reflected wave). S₁₁ = b₁/a₁ with all other ports terminated in Z₀.

Why the reference matters:

• If you measure a device with Zâ‚€ = 50 Ω but it operates between 75 Ω source and load, the S11 you measured is not the reflection the device sees in-circuit. Renormalisation is required.
• Touchstone files embed Zâ‚€ in the header. Always check it matches your system impedance before using the data.
• Z-parameters (open-circuit) and Y-parameters (short-circuit) are independent of reference impedance — they are physical quantities. Conversion S→Z requires Zâ‚€: [Z] = Zâ‚€(I+S)(I−S)⁻¹.

Renormalisation: To convert S-parameters from Z₀=50 Ω to Z₀_new=75 Ω, first convert to Z-matrix, then back to S with new Z₀. Modern VNA software does this automatically.

Transmission loss (TL): Simply |S21|² — the fraction of power transmitted from port 1 to port 2, regardless of matching. It can be less than 1 due to reflections, absorption, or both.

Insertion loss (IL): The additional loss caused by inserting the DUT into a system compared to a direct through connection. IL = 10·log₁₀(P_through/P_with_DUT). If the system is matched (Z_S=Z_L=Z₀), then IL = −|S21|²_dB and IL=TL.

When they differ:

• In a mismatched system, the through connection itself has reflections. Inserting a DUT that acts as an impedance transformer improving the match can reduce IL below TL.
• An attenuator has IL=attenuation value and TL=attenuation value — equal because it improves the match and the comparison is to a matched through.
• A filter in a mismatched system may have TL=3 dB but IL=1 dB if the source sees better coupling through the filter than through a direct connection.

A VNA achieves high dynamic range through narrowband IF detection — it is essentially a very selective superheterodyne receiver.

Measurement chain: The VNA source outputs +5 to +10 dBm at the test port. After passing through the DUT (80 dB attenuation), the received signal may be at −70 dBm. The VNA downconverts this to an IF, then applies a very narrow digital IF filter (IFBW).

How IFBW sets dynamic range:

• Noise floor: N_floor = −174 dBm + NF_receiver + 10·log₁₀(IFBW)
• At IFBW=1 kHz: N_floor = −174+10+30 = −134 dBm → dynamic range ≈ 144 dB (theoretical)
• Narrowing IFBW by 10× improves dynamic range by 10 dB but slows measurement 10×

Practical limits on dynamic range:

• Directivity: Finite-directivity coupler in the VNA test port allows a small portion of incident signal to leak into the reflected port — limits S11 accuracy to 40–50 dB before calibration, 60–80 dB after.
• Port crosstalk: Isolation between test ports limits S21 measurement at high insertion loss. Corrected by isolation calibration.
• Source phase noise: Noise sidebands of the VNA source set a noise floor independent of IFBW below a certain offset.

Group delay is defined as GD(f) = −dφ/dω where φ is the transmission phase (∠S21) and ω=2πf. It represents the time delay experienced by the envelope of a narrowband signal at frequency f.

Why flat group delay matters:

• A signal with bandwidth B occupies f_c−B/2 to f_c+B/2. If group delay varies across this band, different frequency components arrive at different times — this is group delay variation (GDV).
• GDV causes inter-symbol interference (ISI) — each symbol bleeds into adjacent symbols, degrading BER.
• Specification: GDV should be much less than one symbol period. For 100 Msps: T_s=10 ns → GDV should be <1–2 ns.

Measurement: VNA measures S21 phase vs. frequency, then numerically differentiates: GD = −ΔΦ/Δω over a small frequency aperture.

Equalisation:

• Analogue: all-pass filter networks that add phase without affecting magnitude — designed to cancel GDV of the primary filter.
• Digital: FIR/IIR equaliser at baseband — corrects both amplitude and phase errors. Used in all modern digital receivers (WiFi, LTE, 5G).

De-embedding removes parasitic effects of the test fixture from measured S-parameters to reveal the true DUT characteristics. The measurement always includes: [S_measured] = [fixture_in] × [DUT] × [fixture_out].

Technique 1 — SOLT calibration to the DUT reference plane:

• Build calibration standards (Short, Open, Load, Thru) replicating the DUT's connector interface.
• Perform full 2-port SOLT cal — the reference plane is now at the DUT interface. Fixture effects removed inside the VNA.
• Best accuracy but requires accurate physical standards for the DUT's connector type.

Technique 2 — Time-domain gating:

• Measure S11 over wide frequency range, IFFT to time domain.
• Connector/fixture reflection appears as early-time; DUT reflection appears later.
• Apply a time-domain window (gate) passing only the DUT region, zero the fixture. Transform back to frequency domain.
• Effective but reduces frequency resolution due to gate windowing.

Technique 3 — T-matrix cascade de-embedding:

• Measure a dummy fixture (fixture with no DUT, just a thru) separately.
• [T_DUT] = [T_fixture_in]⁻¹ × [T_measured] × [T_fixture_out]⁻¹, then convert back to S.

The Smith chart is a polar plot of the reflection coefficient Γ=(Z−Z₀)/(Z+Z₀). Circles appear because the transformation from Z to Γ maps lines and circles in the Z-plane to circles in the Γ-plane (Möbius transformation).

Three key circle families:

• Constant resistance circles (r-circles): All points with the same normalised resistance r=R/Zâ‚€. Centred at (r/(r+1),0) with radius 1/(r+1). The r=0 circle is the outer rim; r=1 passes through centre.
• Constant reactance circles (x-circles): All points with the same normalised reactance x=X/Zâ‚€. Centred at (1,1/x) with radius 1/|x|. Upper half=inductive; lower half=capacitive.
• Constant |Γ| circles (VSWR circles): All points with the same reflection coefficient magnitude. Centred at the origin. Moving around this circle represents adding lossless transmission line — the impedance rotates at constant VSWR.

Also important: Stability circles — loci of source or load impedances causing |S₁₁_eff|=1 or |S₂₂_eff|=1, marking the boundary between stable and unstable regions.

A reciprocal network satisfies Sᵢⱼ=Sⱼᵢ for all i,j — the transmission from port j to port i equals the transmission from port i to port j. Physically, swapping the source and detector leaves transmitted power unchanged.

Conditions for reciprocity: The device must contain only linear, time-invariant materials with symmetric permittivity and permeability tensors. Isotropic media (vacuum, silicon, FR4) are always reciprocal.

Non-reciprocal device — the circulator:

• A ferrite circulator placed in a DC magnetic field has an asymmetric permeability tensor μ. The magnetic bias breaks time-reversal symmetry.
• Signal from Port 1 exits Port 2 (S21≠0), but signal from Port 2 exits Port 3, NOT Port 1 (S12≈0).
• Numerically: S21=0 dB (or small IL), S12=−20 to −35 dB (isolation).
• Other non-reciprocal devices: isolators, active amplifiers (S21=gain, S12≈0), electro-optic modulators.

Why amplifiers appear non-reciprocal: They contain active elements with DC bias that inject energy, violating Lorentz reciprocity (which requires passive linear media only).

The output stability circle is the locus of load impedances Γ_L that cause |Γ_in|=1 (input reflection coefficient has unity magnitude — verge of oscillation).

Derivation:

• Γ_in = S11 + S12·S21·Γ_L / (1−S22·Γ_L). Setting |Γ_in|=1 gives a circle in the Γ_L plane.
• Load stability circle centre: C_L = (S22−Δ·S11*)* / (|S22|²−|Δ|²) where Δ=S11·S22−S12·S21
• Load stability circle radius: R_L = |S12·S21| / (|S22|²−|Δ|²)

Interpretation:

• Points inside the stability circle on the Smith chart are unstable load impedances.
• If |S11|<1, Γ_L=0 (matched load) is stable, so the region NOT containing the centre is unstable.

When the stability circle passes through the chart centre (Γ_L=0):

• Γ_L=0 corresponds to a 50 Ω load. If this falls exactly on the boundary, |Γ_in|=1 — the device is marginally stable even with matched terminations.
• This means |S11|≈1 — the device is right at the edge of instability. Any inductance from a PCB trace on the load will push into the unstable region. Virtually guaranteed to oscillate in practice.

Three gain definitions appear in amplifier design, each useful for a different design step:

• Transducer power gain G_T: P_delivered_to_load / P_available_from_source. Accounts for both input and output mismatches. G_T=|S21|² when source and load are both Zâ‚€.
• Available power gain G_A: P_available_from_two-port / P_available_from_source. Depends only on source impedance, not load. Used during input matching design and in Friis noise cascade formula.
• Operating power gain G_P: P_delivered_to_load / P_input_to_two-port. Depends only on load impedance, not source. Used during output matching design.

Special case — matched source and load (Γ_S=Γ_L=0, i.e. Z₀ everywhere):

• All three gains become equal: G_T=G_A=G_P=|S21|²
• The VNA measures |S21|² with all ports at Zâ‚€ — it measures transducer gain in this matched condition.

Mixed-mode S-parameters characterise differential and common-mode signal behaviour of a two-port differential network, decomposing into: Sdd (differential-mode), Scc (common-mode), Sdc (common-to-differential conversion), Scd (differential-to-common conversion).

Four mixed-mode sub-matrices:

• Sdd21: Differential gain/insertion loss — the primary design parameter.
• Scc21: Common-mode transmission. Ideally very small (differential amplifier rejects common mode).
• Sdc21: Mode conversion — common-mode input produces differential output. High Sdc21 means the device converts noise into the signal — bad for EMI immunity.
• Scd21: Differential signal generates common-mode output — radiates as EMI. High Scd21 fails EMC tests.

When to use mixed-mode S-parameters:

• Characterising differential amplifiers, ADC input stages, LVDS links, USB/PCIe/HDMI differential pairs
• CMRR characterisation: CMRR=20·log₁₀(|Sdd21/Scd21|)
• EMC analysis — how much differential signal radiates (Scd) or how much common-mode noise enters as differential (Sdc)

Conversion from single-ended: Sdd21=(S31−S41−S32+S42)/2. A 4-port VNA measures all 16 single-ended parameters and computes the full 4×4 mixed-mode matrix.

Touchstone (.snp, n=number of ports) is the universal industry-standard format for S-parameter data exchange. A .s2p file contains 2-port S-parameter data.

File structure:

• Option line: Starts with "#", e.g. "# GHz S MA R 50" — specifies: frequency unit, parameter type (S), data format (MA=Magnitude/Angle, RI=Real/Imaginary, DB=dB/Angle), reference impedance.
• Comment lines (!): Contain manufacturer info, date, temperature, bias conditions. Always read these — they tell you at what bias and temperature the data was measured.
• Data lines: freq S11_mag S11_ang S21_mag S21_ang S12_mag S12_ang S22_mag S22_ang

Limitations of .s2p:

• No temperature, bias, or power level in the data format (must be in comments)
• No noise parameter data — noise figure requires separate .nsp or inline noise parameter columns (Touchstone 2.0 extension)
• No large-signal or harmonic data — S-parameters are linear small-signal only
• Extrapolating outside the measured frequency range is dangerous and meaningless

Mason's unilateral gain U is the maximum power gain achievable from a transistor after: (a) losslessly embedding in any passive network, and (b) making the device unilateral (S12=0) by feedback cancellation. It is invariant to lossless reciprocal embedding.

Formula: U = |S21/S12−1|² / (2·K·|S21/S12|−2·Re(S21/S12))

Why U is the ultimate figure of merit:

• U is invariant to all lossless, reciprocal network transformations (matching networks, transmission lines, lossless feedback). You cannot improve it by clever circuit design — it is a property of the transistor alone.
• f_max (maximum oscillation frequency) is the frequency at which U=1 (0 dB). Above f_max, the transistor cannot provide net power gain even with optimal matching.
• U rolls off at 20 dB/decade. Plotting U vs frequency on log-log gives a straight line; extrapolating to 0 dB gives f_max.
• Unlike f_T (current gain unity frequency), f_max captures the effect of gate resistance R_g and drain-gate capacitance C_gd. Modern InP HEMTs achieve f_max >1 THz.

VSWR=(1+|Γ|)/(1−|Γ|). The angle of Γ does not affect VSWR — only the magnitude matters.

Port 1 (S11=0.5∠90°):

• |S11|=0.5
• VSWR₁=(1+0.5)/(1−0.5)=1.5/0.5=3.0
• Return loss = −20·log₁₀(0.5) = 6.02 dB. Reflected power = 25%.

Port 2 (S22=0.3∠−30°):

• |S22|=0.3
• VSWRâ‚‚=(1+0.3)/(1−0.3)=1.3/0.7=1.86
• Return loss = −20·log₁₀(0.3) = 10.5 dB. Reflected power = 9%.

Physical interpretation: Port 1 (VSWR=3) has significant mismatch. Port 2 (VSWR=1.86) is better but still not ideal. For most RF systems, VSWR<1.5 (|Γ|<0.2, RL>14 dB) is considered acceptable.

A signal flow graph (SFG) represents a network as nodes (wave quantities a₁, b₁, a₂, b₂...) connected by directed branches. Mason's rule gives gain between any two nodes: G=Σ(Pₖ·Δₖ)/Δ

• Pâ‚– = gain of the k-th forward path from source to sink node
• Δ = 1 − Σ(loop gains) + Σ(products of non-touching loop gains) − ... (graph determinant)
• Δₖ = value of Δ with all loops touching path k removed

Two-stage amplifier (unilateral, S12a=S12b=0):

• Forward path: S21a·S21b
• Loop between stages: S22a·S11b (inter-stage mismatch loop)
• Δ = 1−S22a·S11b, Δ₁=1
• G = S21a·S21b / (1−S22a·S11b)

The denominator represents the inter-stage mismatch — this is why the inter-stage impedance matters and why S22a=S11b* (conjugate match) maximises transducer gain.

An ideal 90° hybrid coupler splits input power equally between two output ports with exactly 90° phase difference. In an IQ mixer, this hybrid provides the quadrature LO signals needed for I and Q downconversion.

Ideal 4-port hybrid S-matrix (port 1=input):

• S11=S22=S33=S44=0 (all ports matched)
• S21=−j/√2 (−3 dB, −90° from port 1 to 2)
• S31=−1/√2 (−3 dB, −180° from port 1 to 3)
• S41=0 (isolation)

Amplitude imbalance: |S21|≠|S31| — two outputs deliver different power levels. ±0.5 dB amplitude imbalance in an IQ mixer creates an elliptical constellation, adding ~0.5 dB EVM degradation for 64-QAM.

Phase imbalance: ∠S21−∠S31≠90°. A 2° phase error causes the IQ constellation to rotate and compress. Image rejection: IRR=20·log₁₀(|1+r·e^jΔφ|/|1−r·e^jΔφ|). At 0 dB amplitude error and 2° phase error: IRR≈35 dB.

TDR is obtained by inverse-Fourier-transforming VNA frequency-domain S11 data. The result shows reflected signal amplitude vs. time — since the signal travels at v_p=c/√εr, time converts to distance: d=v_p×t/2 (factor of 2 for round trip).

What each feature means in TDR:

• Positive step: Impedance increases — open circuit, too-narrow trace, or gap in ground plane.
• Negative step: Impedance decreases — via stub, solder bridge, too-wide trace, or connector capacitance.
• Oscillating response: A resonant structure — a stub, bondwire+pad capacitance resonating, or a PCB via resonance.

Locating a fault:

• Measure S11 from DC to f_max. Apply Kaiser-Bessel window, then IFFT.
• Read time position of fault reflection: t_fault. Distance=v_p×t_fault/2.
• For microstrip on FR4 (εr_eff≈3.5): v_p=1.6e8 m/s. Fault at t=2 ns → d=1.6e8×2e-9/2=16 cm from the probe.

The Bode-Fano limit is a fundamental theorem constraining the achievable reflection coefficient bandwidth product for any passive matching network connected to a load containing a reactive element. No matter how complex the matching network, the integral of ln(1/|Γ|) over all frequencies is bounded.

For a parallel RC load: ∫₀^∞ ln(1/|Γ(f)|) df ≤ π/(R·C)

Physical meaning:

• The total area under the "return loss" curve is bounded — you cannot make |Γ| small over an arbitrarily wide bandwidth without it being large outside that band.
• For simple RC load: maximum achievable bandwidth for a given reflection level is BW_max ≈ 1/(π·R·C·|ln|Γ_max||)

Why it matters for real design:

• A chip antenna with 2 pF input capacitance at 50 Ω has RC=0.1 ns. Bode-Fano limit at |Γ|<0.1: BW_max≈1/(π×0.1e-9×2.3)=1.4 GHz. No passive matching network can beat this.
• This is why electrically small antennas have fundamental bandwidth limitations — the Chu-Harrington limit is derived from the same principle.
• Active matching networks can in principle exceed Bode-Fano (by adding power) but at the cost of noise and stability.