S-Parameter Analysis
Theory and Physics
What are S-Parameters?
Professor, what quantity do S-parameters represent?
They represent the input/output characteristics of high-frequency circuits using the ratio of reflected and transmitted waves. For a 2-port case:
$a_i$: incident wave, $b_i$: reflected wave. $S_{11}$: reflection coefficient, $S_{21}$: transmission coefficient.
So a smaller $|S_{11}|$ indicates better impedance matching, right?
Correct. $S_{11} = -20$ dB means reflected power is 1%. $S_{21} = -3$ dB means transmitted power is halved (3 dB loss). S-parameters are functions of frequency and are measured with a VNA (Vector Network Analyzer).
Summary
- $S_{11}$: Reflection coefficient — Indicator of impedance matching
- $S_{21}$: Transmission coefficient — Indicator of insertion loss
- Function of frequency — Measured with VNA, calculated with FEM
The Birth of S-Parameters——How the Scattering Matrix Changed "Inter-Port Relationships"
The concept of S-parameters (scattering parameters) was formalized by K. Kuroki and D. M. Pozar, among others. They provided a unified description of the relationships between incident, reflected, and transmitted waves in microwave circuits, which were difficult to handle with Z-parameters or Y-parameters. In particular, the "reflection coefficient Γ (=S₁₁)" and "transmission coefficient S₂₁," which are measurable on transmission lines, directly connected experiment and theory, becoming the common language of high-frequency design alongside the spread of network analyzers. S-parameter calculation in CAE is realized through the process of eigenmode expansion → port mode normalization.
Physical Meaning of Each Term
- Electric field term $\nabla \times \mathbf{E} = -\partial \mathbf{B}/\partial t$: Faraday's law of electromagnetic induction. A time-varying magnetic flux density generates an electromotive force. 【Everyday Example】A bicycle dynamo (generator) generates voltage in a nearby coil by rotating a magnet—a direct application of this law where a changing magnetic field induces an electric field. An IH (induction heating) cooking heater also uses the same principle, where high-frequency magnetic field changes induce eddy currents in the pot bottom, heating it via Joule heat.
- Magnetic field term $\nabla \times \mathbf{H} = \mathbf{J} + \partial \mathbf{D}/\partial t$: Ampère-Maxwell's law. Current and displacement current generate a magnetic field. 【Everyday Example】When current flows through a wire, a magnetic field is created around it—this is Ampère's law. An electromagnet operates on this principle, passing current through a coil to create a strong magnetic field. A smartphone speaker also applies this law: current → magnetic field → force on the diaphragm. At high frequencies (e.g., GHz-band antennas), the displacement current $\partial D/\partial t$ cannot be ignored, describing electromagnetic wave radiation.
- Gauss's Law $\nabla \cdot \mathbf{D} = \rho_v$: Indicates that electric charge is the divergence source of electric flux. 【Everyday Example】Rubbing hair with a plastic sheet creates static electricity, making hair stand up—electric field lines radiate from the charged sheet (electric charge), exerting force on the light hair. In capacitor design, the electric field distribution between electrodes is calculated using this law. ESD (electrostatic discharge) countermeasures are also based on electric field analysis derived from Gauss's law.
- Magnetic flux conservation $\nabla \cdot \mathbf{B} = 0$: Expresses that magnetic monopoles do not exist. 【Everyday Example】Even if you cut a bar magnet in half, you cannot create a magnet with only an N pole or only an S pole—N and S poles always exist as a pair. This means magnetic field lines form "closed loops with no start or end points." In numerical analysis, to satisfy this condition, the formulation using vector potential $\mathbf{B} = \nabla \times \mathbf{A}$ is used, automatically guaranteeing magnetic flux conservation.
Assumptions and Applicability Limits
- Linear material assumption: Permeability and permittivity do not depend on magnetic/electric field strength (nonlinear B-H curve needed in saturation region)
- Quasi-static approximation (low frequency): Displacement current term can be ignored ($\omega \varepsilon \ll \sigma$). Common in eddy current analysis
- 2D assumption (cross-section analysis): Effective when current direction is uniform and end effects can be ignored
- Isotropic assumption: For anisotropic materials (e.g., silicon steel rolling direction), direction-specific property definitions are needed
- Non-applicable cases: Additional constitutive laws are needed for plasma (ionized gas), superconductors, nonlinear optical materials
Dimensional Analysis and Unit Systems
| Variable | SI Unit | Notes / Conversion Memo |
|---|---|---|
| Magnetic Flux Density $B$ | T (Tesla) | 1T = 1 Wb/m². Permanent magnet: 0.2〜1.4T |
| Magnetic Field Strength $H$ | A/m | Horizontal axis of B-H curve. Conversion with CGS unit Oe (Oersted): 1 Oe = 79.577 A/m |
| Current Density $J$ | A/m² | Calculated from conductor cross-section and total current. Note non-uniform distribution due to skin effect |
| Permeability $\mu$ | H/m | $\mu = \mu_0 \mu_r$. In vacuum $\mu_0 = 4\pi \times 10^{-7}$ H/m |
| Electrical Conductivity $\sigma$ | S/m | Copper: approx. 5.96×10⁷ S/m. Decreases with temperature rise |
Numerical Methods and Implementation
S-Parameter Extraction in FEM
How do you extract S-parameters from FEM?
1. Set mode patterns (e.g., TE10) on ports
2. Excite an incident wave from one port
3. Calculate reflected/transmitted waves at each port
4. Calculate via $S_{ij} = b_i/a_j$
HFSS's adaptive mesh uses $\Delta S$ (change in S-parameters) for convergence judgment.
What about multi-port cases?
For $N$ ports, it's an $N \times N$ S-matrix. Solve sequentially by exciting from each port ($N$ times, Direct Solver). Or solve all ports simultaneously and extract $S$ via matrix operations (Fast Frequency Sweep).
Summary
- Port Mode Setting — Definition of incident wave
- $\Delta S$ Convergence Criterion — Indicator for adaptive mesh
- Fast Frequency Sweep — Fast acquisition of S-parameters across the entire band
De-embedding——The Technology to "Remove Port Influence from Measured Values"
When actually measuring S-parameters with a VNA, the influence of connectors, cables, and fixtures gets mixed in. "De-embedding" removes this by measuring known reference structures (Open/Short/Thru) and removing the port model via inverse matrix operations. A similar de-embedding concept is used in CAE, extracting actual device characteristics from port waveforms within the analysis domain. TRL (Thru-Reflect-Line) calibration is the standard method for VNAs, and the port settings in CST/HFSS automate this procedure.
Edge Elements (Nedelec Elements)
Elements specialized for electromagnetic field analysis. Automatically guarantee continuity of tangential components and eliminate spurious modes. Standard for 3D high-frequency analysis.
Nodal Elements
Used for scalar potential formulations. Effective for scalar potential methods in magnetostatics and electrostatic analysis.
FEM vs BEM (Boundary Element Method)
FEM: Handles nonlinear materials and inhomogeneous media. BEM: Naturally handles infinite domains (open region problems). Hybrid FEM-BEM is also effective.
Nonlinear Convergence (Magnetic Saturation)
Nonlinearity of B-H curve handled by Newton-Raphson method. Residual criterion: $||R||/||R_0|| < 10^{-4}$ is typical.
Frequency Domain Analysis
Reduced to a steady-state problem via time-harmonic assumption. Requires complex number operations, but wideband characteristics are obtained via time-domain analysis.
Time Domain Time Step
Time step less than 1/20 of the highest frequency component is required. Implicit time integration allows larger steps but requires attention to accuracy.
Choosing Between Frequency Domain and Time Domain
Frequency domain analysis is like "tuning a radio to a specific frequency"—it can efficiently calculate the response at a single frequency. Time domain analysis is like "recording all channels simultaneously"—it can reproduce transient phenomena containing all frequency components, but computational cost is high.
Related Topics
なった
詳しく
報告