Analysis of Crosstalk Between Cables — MTL Theory and Practical Simulation
Theory and Physics
Overview of Multiconductor Transmission Line (MTL) Theory
Professor, is crosstalk analysis for wire harnesses done with 3D FEM? Automotive cable wiring is long, do you mesh the entire length?
Good question. With 3D FEM, meshing the entire cable length becomes enormous. For example, an automotive harness can be over 5km in total length, and the conductor cross-section is on the order of 0.5mm in diameter. The aspect ratio would exceed 10,000, making proper meshing impossible.
Then how is it solved in practice?
In practice, we use a 1D solver based on Multiconductor Transmission Line (MTL) theory. The concept is this:
- Obtain only the cross-sectional parameters (L, C, R, G matrices) using 2D FEM
- Solve for the length direction using MTL equations (a system of 1D ordinary differential equations)
Since automotive harnesses have over 500 wires running in parallel, this efficiency is essential. 3D FEM would require millions of elements even for a single wiring run, but the MTL approach only needs 2D FEM for the cross-section (thousands of elements) + 1D matrix exponential calculations. The computation time differs by at least three orders of magnitude.
I see! So the idea is to "extract the cross-sectional information and apply it to the entire line." Smart...!
Exactly. The prerequisite for MTL theory to hold is that the Transverse Electromagnetic (TEM) mode approximation is valid—meaning the cable cross-sectional dimensions are sufficiently small compared to the wavelength. For automotive EMC target frequencies of around 150kHz to 200MHz, the wavelength is 1.5m or more, which is sufficiently valid for a harness bundle a few cm in diameter. It's a different story for GHz-band SI (Signal Integrity) problems, but cable crosstalk in EMC is fundamentally within the scope of MTL.
NEXT (Near-End Crosstalk) and FEXT (Far-End Crosstalk)
I've heard there are NEXT and FEXT types of crosstalk. What's the difference?
First, let's set up the scenario. Assume there are two parallel conductors (aggressor line and victim line).
- NEXT (Near-End Crosstalk): Crosstalk observed at the victim line end on the same side as the signal input. Appears as the sum of capacitive coupling $C_m$ and inductive coupling $L_m$.
- FEXT (Far-End Crosstalk): Crosstalk observed at the victim line end on the opposite side from the signal input. Appears as the difference between capacitive and inductive coupling.
For example, Ethernet cable (LAN cable) specification tests measure both, but generally NEXT levels are larger. This is because capacitive and inductive coupling add together.
Wait, FEXT is determined by the difference? So theoretically, it could become zero under certain conditions?
Sharp observation! Theoretically, if $L_m / L = C_m / C$ (the so-called homogeneous medium condition) holds, FEXT becomes zero. This approximately holds for coaxial cables or stripline structures in homogeneous dielectrics. However, automotive harnesses are a mixed medium of air and insulation, so the homogeneous condition rarely holds. FEXT also needs to be properly evaluated.
MTL Governing Equations
Please teach me the governing equations for MTL!
For an MTL system with $N$ conductors (+ a return conductor = ground), the frequency-domain governing equations are:
Where:
- $\mathbf{V}(z)$, $\mathbf{I}(z)$: $N \times 1$ voltage and current vectors
- $\mathbf{Z} = \mathbf{R} + j\omega\mathbf{L}$: $N \times N$ per-unit-length impedance matrix
- $\mathbf{Y} = \mathbf{G} + j\omega\mathbf{C}$: $N \times N$ per-unit-length admittance matrix
In essence, it's an extension of the telegrapher's equations to multiple conductors, right?
Exactly. Combining the two MTL equations yields a second-order wave equation:
$\boldsymbol{\gamma}^2 = \mathbf{Z}\mathbf{Y}$ is the propagation constant matrix, whose eigenvalues give the propagation constant for each mode. The general solution can be written using the matrix exponential function as follows:
Matrix exponential...! I understand it's an extension of the scalar $e^{-\gamma z}$, but it seems tough to compute.
That's why in actual computation, we perform eigenvalue decomposition of $\boldsymbol{\gamma}^2$ and compute it as a scalar exponential function for each mode. If $\boldsymbol{\gamma}^2 = \mathbf{T}\boldsymbol{\Lambda}\mathbf{T}^{-1}$ can be diagonalized, then $e^{\boldsymbol{\gamma}z} = \mathbf{T} \, \text{diag}(e^{\sqrt{\lambda_i}z}) \, \mathbf{T}^{-1}$, allowing each mode to be solved independently. This is the core algorithm of MTL solvers.
Per-Unit-Length L/C Matrices
How do you obtain the L and C matrices that are inputs to the MTL equations?
This is where 2D FEM first comes into play. Model the cable cross-section (conductor arrangement, insulation, shield) in 2D and obtain the matrices from electrostatic and magnetostatic field analyses.
How to obtain the C matrix: Apply a unit charge to conductor $i$, ground all other conductors, and solve for the potential distribution. $C_{ij}$ is obtained from the charge induced on conductor $j$.
How to obtain the L matrix: Pass a unit current through conductor $i$, set currents in all other conductors to zero, and solve for the magnetic field distribution. $L_{ij}$ is obtained from the magnetic flux linking conductor $j$.
I see, so you solve the 2D cross-sectional problem $N$ times (for the number of conductors). That's incredibly light compared to 3D FEM!
For example, with 10 conductors, you need 10 electrostatic analyses + 10 magnetostatic analyses = 20 total analyses with 2D FEM. Each analysis is a 2D problem with a few thousand elements, so the whole thing finishes in seconds.
Since the matrices are symmetric, even fewer calculations are needed in practice. Both $\mathbf{L}$ and $\mathbf{C}$ are symmetric positive-definite matrices:
The diagonal elements $L_{ii}$ represent the self-inductance of conductor $i$, and the off-diagonal elements $L_{ij}$ represent the mutual inductance between conductors $i$ and $j$. These off-diagonal elements are precisely what determine crosstalk.
Crosstalk Voltage Prediction via Weak-Coupling Approximation
Is there a more intuitive formula for how much crosstalk voltage actually occurs?
Under weak-coupling conditions (when mutual coupling is small compared to self-coupling), approximate formulas can be used. For the two-wire case, the NEXT voltage $V_\text{NE}$ and FEXT voltage $V_\text{FE}$ are:
Here, $V_s$ is the aggressor line signal voltage, $\tau = \ell / v$ is the line propagation delay, $\ell$ is the parallel run length, and $v$ is the propagation velocity.
Wow, with this you can roughly estimate by hand calculation! It worsens proportionally with frequency...
Yes. Crosstalk increases proportionally with frequency—that's why, with the rise of high-frequency switching (hundreds of kHz operation of SiC/GaN inverters) due to electrification, crosstalk problems are becoming increasingly severe. Another point to note: while NEXT does not depend on parallel run length (it saturates), FEXT increases proportionally with parallel run length $\ell$. There are cases where FEXT becomes dominant when there are long parallel sections.
Automotive Harness Design—How to Bundle Hundreds of Wires
Modern passenger cars are equipped with wire harnesses exceeding 5km in total length and weighing 30–50kg. In EVs/HEVs, 400V–800V high-voltage battery cables and sensor signal lines on the order of a few mV run within the same vehicle body. This voltage difference is over 1 million times. Shielding all cables would solve crosstalk, but it would increase vehicle weight by tens of kg and skyrocket costs. Therefore, quantitatively evaluating induced voltage with MTL simulation and sorting out "wires that truly need shielding" from "wires that only require separation distance" is key to balancing cost, weight, and performance.
Physical Meaning of Each Parameter
- Self-inductance $L_{ii}$: The proportion of magnetic flux linking conductor $i$ itself when current flows through it. Becomes larger as the conductor cross-sectional area decreases and as the distance from the return conductor increases. Typical value: 0.3–1.0 $\mu$H/m for automotive harnesses.
- Mutual inductance $L_{ij}$: Magnetic flux from conductor $i$'s current that links conductor $j$. Larger when the two conductors are closer. Source of inductive coupling in crosstalk. Decreases inversely proportional to the logarithm of distance.
- Self-capacitance $C_{ii}$: Electrostatic capacitance between conductor $i$ and the return conductor. Determined by insulation permittivity and thickness. Typical value: 50–100 pF/m.
- Mutual capacitance $C_{ij}$: Coupling capacitance between conductors $i$ and $j$. Depends on inter-conductor distance and insulation conditions. Source of capacitive coupling in crosstalk.
- Per-unit-length resistance $R_{ii}$: Conductor DC resistance + increment due to skin effect at high frequencies. About 33 m$\Omega$/m for AWG20 Cu wire (DC). Increases as $R \propto \sqrt{f}$ with frequency $f$.
Application Conditions and Limitations of MTL Theory
- TEM approximation: Cross-sectional dimension $d$ must be sufficiently small compared to wavelength $\lambda$ (guideline: $d < \lambda/10$). At 200MHz, $\lambda = 1.5$m, so it holds for harness bundles under 15cm in diameter.
- Cross-sectional uniformity: Conductor arrangement must be constant along the length direction. At bends, branches, and connector junctions, the MTL assumption breaks down, requiring modeling as lumped-parameter circuits.
- Neglect of radiation loss: MTL assumes a closed TEM system, so it cannot handle power loss due to radiation. Full-wave analysis is needed when antenna effects become problematic at high frequencies.
- Linear system: L, C, R, G parameters must not depend on signal amplitude. Nonlinear materials like ferrite cores require separate consideration.
Numerical Methods and Implementation
Cross-Sectional Parameter Extraction via 2D FEM
Please teach me the specific procedure for obtaining L/C matrices with 2D FEM.
Let's start with C matrix extraction (electrostatic field analysis). Solve Laplace's equation:
Set the potential of conductor $i$ to $\phi_i = 1$ V and all other conductors and the return conductor to $\phi = 0$ V as boundary conditions, then integrate the charge $Q_j = \oint \varepsilon \frac{\partial \phi}{\partial n} dS$ on each conductor surface. This yields column $i$ of the $C$ matrix:
The diagonal element is the sum of coupling capacitances to all other conductors, right? Is that the Maxwell capacitance matrix?
To be precise, what we obtain here is the short-circuit capacitance matrix (the circuit-style C matrix). The Maxwell capacitance matrix $\mathbf{C}_M$ has the opposite sign convention, with the relation $C_{ij} = -C_{M,ij}$ ($i \neq j$). Note that output formats differ by software.
L matrix extraction (magnetostatic field analysis) is similar: give conductor $i$ a unit current $I_i = 1$A, set other conductors to $I = 0$, and solve for the magnetic vector potential $\mathbf{A}$:
Obtain $L_{ji} = \Phi_{ji} / I_i$ from the magnetic flux $\Phi_{ji} = \oint \mathbf{A} \cdot d\mathbf{l}$.
There are several practical points:
- Mesh: Place at least 3 or more element layers in gaps between conductors. Also include 2–3 layers of boundary layer mesh near conductor surfaces.
- External boundary: When there is no clear return conductor (e.g., vehicle body ground), set an electric wall ($\phi=0$) at a sufficient distance from the cross-section. A distance 5–10 times the outer diameter of the conductor group is a guideline.
- Skin effect: The R matrix at high frequencies is obtained from 2D eddy current analysis considering skin depth $\delta = \sqrt{2/(\omega\mu\sigma)}$.
Numerical Solution of MTL Equations
Once you have the L/C matrices, how do you solve the MTL equations?
There are three main approaches:
1. Mode Decomposition Method (Eigenvalue Decomposition)
This method performs eigenvalue decomposition of $\boldsymbol{\gamma}^2 = \mathbf{ZY}$ to separate it into $N$ independent scalar transmission line problems. For uniform lines (parameters independent of $z$), an exact solution is obtained. It's the fastest and the standard method for frequency-domain analysis.
2. Chain Parameter Matrix Method
This method divides the line into small segments $\Delta z$ and multiplies the transfer matrix (multiconductor extension of the ABCD matrix) for each segment in a chain:
The advantage is applicability to non-uniform lines (where cross-section varies along $z$). It shines in cases like automotive harnesses where branching and bundle configuration change.
3. FDTD Method (Time-Domain Direct Solution)
This method discretizes the MTL equations in both space and time. Its characteristic is the natural handling of nonlinear terminations (like diode clamps). Suitable for analyzing wideband transient signals such as ESD (electrostatic discharge) pulses. However, a time step satisfying the stability condition (CFL condition) is required.
I see... So if it's uniform and linear, method 1 is fastest; for complex wiring topology, method 2; and for transient analysis including nonlinearities, method 3. Is that the usage?
Perfect summary. Actual tools (CST Cable Studio, Ansys EMA3D Cable, etc.) internally switch between multiple methods.
| Method | Domain | Non-uniform Line | Nonlinear Termination | Computation Speed |
|---|---|---|---|---|
| Mode Decomposition | Frequency | No | No | Fastest |