How do X-parameters extend S-parameters?

S-parameters describe linear behavior: B_p = S_pq × A_q (scattered wave B at port p is a linear function of incident wave A at port q). This breaks down for nonlinear devices (amplifiers near compression, mixers, oscillators) where harmonics and intermodulation products are generated. X-parameters extend this to: B_p,k = X^(F)_p,k(|A_1,1|) + Σ X^(S)_p,k,q,l × A_q,l + Σ X^(T)_p,k,q,l × conj(A_q,l). Where k,l are harmonic indices (fundamental, 2nd, 3rd...), X^(F) is the large-signal operating point (function of drive level |A_1,1|), X^(S) captures forward-scattering sensitivity to small perturbations, and X^(T) captures conjugate-scattering (mixing products). X-parameters are measured on a nonlinear VNA (NVNA) or large-signal network analyzer (LSNA) and stored as frequency/power-indexed lookup tables. They enable cascading nonlinear blocks in simulation (ADS, Microwave Office) without exposing proprietary transistor-level designs. Accuracy: typically within 0.5 dB of gain compression and ±2° of AM-PM distortion versus measured data.

When should you use a behavioral model versus a circuit model?

The choice depends on the design phase and IP protection needs: Use behavioral models when: 1) The component is from a third party (vendor-supplied MMIC, packaged PA) and internal topology is proprietary. The vendor provides X-parameters or S2P files instead of a transistor-level schematic. 2) System-level simulation is needed: cascading 5-10 blocks in a receiver chain. Behavioral models simulate in seconds versus hours for circuit-level. 3) Optimization requires many iterations: neural network surrogates can replace full EM simulation (HFSS, CST) for antenna/filter optimization, providing 100-1000x speedup. 4) The device operates in a well-defined regime (fixed bias, limited bandwidth). Use circuit models when: 1) Designing the component itself (transistor biasing, matching network synthesis). 2) Predicting behavior outside the characterized range (temperature extremes, frequencies beyond measurement bandwidth). 3) Understanding failure mechanisms (breakdown, thermal runaway). 4) Nonlinear interactions between components require co-simulation at the circuit level. Hybrid approach: use circuit models for the critical block under design and behavioral models for all surrounding blocks.

What is the accuracy limitation of behavioral models?

Behavioral models are interpolation-based: they are accurate within the characterized measurement space but unreliable for extrapolation. Specific limitations: 1) Power range: X-parameters are measured at discrete power levels (e.g., -20 to +20 dBm in 2 dB steps). Between points, cubic interpolation introduces ±0.1-0.3 dB error. Beyond the measured range, errors grow rapidly (no physical constraints prevent unphysical predictions). 2) Frequency coverage: measured at discrete frequencies (e.g., 0.5-6 GHz in 100 MHz steps). Inter-frequency interpolation works well for smooth responses but fails near resonances or band edges. 3) Load sensitivity: standard X-parameters assume a 50Ω load. Under mismatch (VSWR > 2:1), X^(S) and X^(T) terms may be insufficient. The Cardiff model handles this better by including load-pull data across the Smith chart. 4) Memory effects: X-parameters capture memoryless nonlinearity. For devices with significant memory effects (bias-dependent thermal, trapping in GaN HEMTs), envelope-domain behavioral models (e.g., X-parameters with memory extension) or Volterra series are needed. 5) Temperature: most behavioral models are characterized at 25°C. Temperature coefficient data must be added separately or a family of models provided.

RF Simulation

Behavioral Model

/bih-HAY-vyor-ul MOD-ul/

A black-box representation of an RF component capturing input-output behavior (gain, compression, harmonics, intermodulation) without internal circuit topology. Types: S-parameters (linear), X-parameters (nonlinear, PHD-based, harmonics indexed by power), Cardiff Model (load-pull contours), Volterra series (weakly nonlinear kernels), and neural network surrogates (100 to 1,000x speedup over EM simulation). Accuracy: ±0.5 dB gain compression, ±2° AM-PM versus measured data within characterized range.

Linear: S-parameters

Nonlinear: X-parameters

Speedup: 100–1000x

Understanding Behavioral Models

Behavioral models exist because RF/microwave design requires simulating components whose internal details are either proprietary (vendor-supplied MMICs), too complex for circuit-level simulation (EM structures), or both. The model captures what the component does (its transfer function) without modeling how it does it (transistor physics, EM field distributions).

The progression from S-parameters to X-parameters mirrors the industry's shift from linear to nonlinear system design. S-parameters (introduced 1960s) describe linear two-port behavior. X-parameters (introduced 2008 by Agilent, now Keysight) extend this framework to large-signal operation where harmonics and intermodulation products are generated, capturing the full poly-harmonic distortion behavior in a format suitable for commercial simulators.

X-Parameter Framework

S-Parameter (linear):
B_p = ∑ S_pq × A_q

X-Parameter (nonlinear):
B_p,k = X^(F)_p,k(|A_1,1|)
+ ∑ X^(S)_p,k,q,l × A_q,l
+ ∑ X^(T)_p,k,q,l × A*_q,l

Where:
k,l = harmonic index (1,2,3...)
X^(F) = large-signal operating point
X^(S) = forward scattering sensitivity
X^(T) = conjugate scattering (mixing)

Neural Network Surrogate:
y = f_NN(freq, P_in, Z_load, bias, T)
Training: 500–5,000 EM simulations
Inference: 100–1,000x faster

Behavioral Model Type Comparison

Model Type	Linearity	Harmonics	Load-Pull	Speed
S-parameters	Linear only	No	No	Fastest
X-parameters	Nonlinear	Yes (indexed)	Limited	Fast
Cardiff Model	Nonlinear	Yes	Full Smith chart	Fast
Volterra	Weakly NL	Yes (kernels)	No	Medium
Neural network	Any	Trained	Trained	Very fast

Common Questions

Frequently Asked Questions

X-parameters vs. S-parameters?

S-parameters: linear, B = S×A. X-parameters: nonlinear, include harmonic-indexed X^(F), X^(S), X^(T) terms. Measured on NVNA/LSNA. Cascadable in ADS/MWO. Accuracy: ±0.5 dB compression, ±2° AM-PM.

When to use behavioral vs. circuit?

Behavioral: third-party IP (vendor MMIC), system-level cascade (5 to 10 blocks), optimization (NN surrogates for EM). Circuit: designing the component, predicting out-of-range behavior, failure analysis. Hybrid approach is common.

Accuracy limits?

Interpolation-only: accurate within characterized power/freq/load. Power: ±0.1 to 0.3 dB between measured levels. Load mismatch: X-params limited above VSWR 2:1. Memory effects (GaN): require envelope-domain extensions. Temperature: usually 25°C only.

Related Terms

← Bed of Nails BeiDou B1 →

RF Simulation

Precision RF Components

RF Essentials provides precision terminations and custom waveguide assemblies for load-pull measurement systems, NVNA calibration, and behavioral model extraction equipment.

Request a Quote