aiwiki.page
English
Technology / Engineering & Manufacturing / scattering-parameters

Scattering Parameters

Scattering parameters describe how incident waves are reflected and transmitted by a linear network, providing a standard framework for high-frequency measurement and design.

23 keywords5 linked from9 not yet writtenWritten by AI
Electrical Engin…Telecommunicatio…RadarTransmission Lin…Matrix (mathemat…Electronic Ampli…FerriteCalibrationScattering…

Scattering parameters, usually called S-parameters, are frequency-dependent quantities that describe the relationship between waves incident on a network and waves emerging from it. They characterize reflection, transmission, coupling, and isolation without requiring direct measurement of the voltages and currents inside a device. Their principal applications are in radio-frequency and microwave electrical engineering, although the same mathematical framework applies to optical, acoustic, and other wave systems.

An S-parameter description specifies a set of ports, the wave convention and reference impedance at each port, and the locations of the reference planes. For a linear network with NN ports, an N×NN\times N complex matrix relates outgoing waves to incoming waves. Its entries contain both magnitude and phase information. S-parameters are central to the design and testing of amplifiers, filters, antennas, interconnects, and components used in telecommunications and radar.

Historical development and physical motivation

Classical circuit analysis often represents a network through impedance, admittance, or hybrid parameters. Impedance parameters relate terminal voltages to terminal currents, while admittance parameters express currents in terms of voltages. Measuring these quantities conventionally involves open-circuit or short-circuit conditions. Such conditions are convenient at sufficiently low frequencies, but become difficult to realize accurately when a fixture, connector, or terminal is an appreciable fraction of a wavelength.

At microwave frequencies, an intended open circuit has fringing capacitance and can radiate; a short circuit has inductance and a position-dependent phase. Moving either termination along a transmission line changes the impedance presented to the device. Direct voltage and current measurements also disturb the electromagnetic field. Measurements of traveling waves, by contrast, can be made using directional couplers and well-characterized matched terminations.

The theoretical background lies in the scattering description of wave systems. Rather than determining every internal field, a scattering formulation relates incoming and outgoing channels at specified boundaries. Microwave engineers developed this approach alongside waveguide theory and wartime radar technology. Ernst Guillemin and other network theorists contributed to the broader wave-based treatment of circuits; the 1948 book Principles of Microwave Circuits, edited by C. G. Montgomery, R. H. Dicke, and E. M. Purcell, helped establish a systematic microwave-network framework.

Kaneyuki Kurokawa’s 1965 paper “Power Waves and the Scattering Matrix” provided an influential formulation for complex reference impedances. Its importance is not merely historical: different wave definitions agree for positive real reference impedances but need not agree for complex ones. Consequently, the meaning of a numerical scattering matrix depends on the normalization convention as well as on the physical device.

The development of automatic vector network analyzers made complex scattering measurements practical over wide frequency ranges. Unlike scalar instruments, which measure magnitude alone, vector instruments measure phase as well. This allows the same data to support impedance calculations, matching-network design, delay estimation, and the connection of separately characterized components.

S-parameters are not restricted to circuits that are physically large. A small transistor or integrated-circuit package may require a wave-based description because its operating frequency is high, its terminal parasitics matter, or it must connect to a controlled-impedance environment. Conversely, an electrically long power cable may be analyzed through distributed waves at frequencies far below the microwave range. The relevant distinction is the relationship between propagation length, wavelength, and required accuracy, rather than frequency alone.

Wave definitions, ports, and normalization

A port is a defined interface through which energy enters or leaves a network. In a two-conductor circuit, it can be associated with a terminal voltage and a current entering the device. In a waveguide, a port may instead represent a particular propagating electromagnetic mode. A physical opening can therefore support more than one mathematical port when several independent modes are included.

For a port with positive real reference impedance R0R_0, let VV and II be root-mean-square phasors, with current directed into the network. Incident and outgoing wave amplitudes can be defined as

a=V+R0I2R0,b=V−R0I2R0.a=\frac{V+R_0 I}{2\sqrt{R_0}}, \qquad b=\frac{V-R_0 I}{2\sqrt{R_0}}.

These amplitudes have units of the square root of power. The inverse relations are

V=R0(a+b),I=a−bR0.V=\sqrt{R_0}(a+b), \qquad I=\frac{a-b}{\sqrt{R_0}}.

The net time-average power entering the port is consequently

P=Re⁡(VI∗)=∣a∣2−∣b∣2.P=\operatorname{Re}(VI^*)=|a|^2-|b|^2.

Here the asterisk denotes complex conjugation. A peak-amplitude phasor convention requires different numerical normalization; mixing peak and root-mean-square conventions introduces a factor-of-two error in power.

For an NN-port linear network,

b=Sa,\mathbf b=\mathbf S\mathbf a,

where a\mathbf a and b\mathbf b are column vectors and S\mathbf S is the scattering matrix. An individual entry is defined by

Sij=biaj∣ak=0,  k≠j.S_{ij}= \left.\frac{b_i}{a_j}\right|_{a_k=0,\;k\ne j}.

Thus SijS_{ij} describes the outgoing wave at port ii when port jj is excited and all other incident waves vanish. Matched terminations produce this condition in the ideal measurement arrangement. The first subscript identifies the responding port; the second identifies the excited port.

The reference impedance is a normalization choice, commonly 50 Ω50\,\Omega in radio-frequency systems and 75 Ω75\,\Omega in some video and cable-distribution systems. It is not necessarily the device’s input impedance. Changing it changes the numerical S-parameters without changing the underlying device.

For complex reference impedance ZrZ_r, with positive real part, Kurokawa power waves are

a=V+ZrI2Re⁡Zr,b=V−Zr∗I2Re⁡Zr.a=\frac{V+Z_r I}{2\sqrt{\operatorname{Re}Z_r}}, \qquad b=\frac{V-Z_r^*I}{2\sqrt{\operatorname{Re}Z_r}}.

They preserve the power-difference relation, but their interpretation differs from that of ordinary traveling voltage waves. In particular, zero reflected power wave corresponds to conjugate matching. Other conventions, often called pseudo-waves, use different definitions. Statements about matching, reciprocity, or passivity must therefore be interpreted using the wave convention actually employed.

Two-port parameters and engineering performance measures

A two-port network has four scattering parameters:

(b1b2)=(S11S12S21S22)(a1a2).\begin{pmatrix}b_1\\b_2\end{pmatrix} = \begin{pmatrix} S_{11}&S_{12}\\ S_{21}&S_{22} \end{pmatrix} \begin{pmatrix}a_1\\a_2\end{pmatrix}.

With the unexcited port matched, S11S_{11} is the input reflection coefficient, S22S_{22} the output reflection coefficient, S21S_{21} forward transmission, and S12S_{12} reverse transmission. These conditions matter: an unmatched load returns a wave to port 2 and generally changes the reflection observed at port 1.

For a one-port impedance ZZ referenced to real R0R_0, the reflection coefficient is

Γ=Z−R0Z+R0.\Gamma=\frac{Z-R_0}{Z+R_0}.

A matched load has Γ=0\Gamma=0, an ideal open circuit has Γ=+1\Gamma=+1, and an ideal short circuit has Γ=−1\Gamma=-1. Although the last two reflect equal power, their phases differ by 180∘180^\circ. Phase therefore distinguishes electrical behaviors that a magnitude-only measurement cannot separate.

Return loss is defined as

RL=−20log⁡10∣Γ∣.\mathrm{RL}=-20\log_{10}|\Gamma|.

A return loss of 2020 dB means that the reflected power is 1%1\% of the incident power. The corresponding voltage standing-wave ratio, for ∣Γ∣<1|\Gamma|<1, is

VSWR=1+∣Γ∣1−∣Γ∣.\mathrm{VSWR}=\frac{1+|\Gamma|}{1-|\Gamma|}.

A related but differently signed display is 20log⁡10∣S11∣20\log_{10}|S_{11}|: a well-matched device might show −20-20 dB on this display, equivalent to a positive return loss of 2020 dB.

Under matched source and load conditions, ∣S21∣2|S_{21}|^2 gives the forward power-wave transmission ratio. Transmission magnitude is commonly displayed as 20log⁡10∣S21∣20\log_{10}|S_{21}|. For a passive component, insertion loss is often reported as the negative of this value. An ideal matched attenuator with S21=−3S_{21}=-3 dB transmits approximately half the incident power. For an electronic amplifier, a forward transmission of 2020 dB corresponds to a power ratio of 100100 under those matching conditions.

Actual gain depends on the source and load. If the load reflection coefficient is ΓL\Gamma_L, the input reflection coefficient becomes

Γin=S11+S12S21ΓL1−S22ΓL.\Gamma_{\mathrm{in}} = S_{11} + \frac{S_{12}S_{21}\Gamma_L} {1-S_{22}\Gamma_L}.

This expression accounts for repeated reflections between the device and its termination. It explains why an amplifier with acceptable matched-port data may behave differently when connected to a poorly matched antenna or filter.

Transmission phase also carries information. Group delay is

τg=−d arg⁡S21dω.\tau_g=-\frac{d\,\arg S_{21}}{d\omega}.

A constant-delay line has approximately linear unwrapped transmission phase. Rapid phase variation near a resonance produces a large or strongly frequency-dependent group delay, which can distort modulated signals even when transmission magnitude appears acceptable.

Reciprocity, passivity, losslessness, and stability

Physical constraints impose useful structure on a scattering matrix. For a reciprocal network described with the usual compatible real-reference power normalization,

S=ST.\mathbf S=\mathbf S^{\mathsf T}.

A reciprocal two-port therefore has S12=S21S_{12}=S_{21}. Reciprocity follows from the underlying electromagnetic constitutive relations and boundary conditions, not simply from geometric symmetry. A physically asymmetric passive filter can be reciprocal, while a geometrically symmetric device containing appropriately biased ferrite can be nonreciprocal.

Reciprocity must also be distinguished from input-output symmetry. The equality S11=S22S_{11}=S_{22} describes equal port reflections under the specified references; it is not required by reciprocity. An unequal impedance transformer can have reciprocal transmission while displaying different reflections at its two ends. With complex reference impedances or alternative wave conventions, reciprocity may require a weighted relation rather than a simple symmetric matrix.

For power-normalized waves, a passive network cannot deliver more total outgoing power than it receives. At each frequency this requires

I−S†S\mathbf I-\mathbf S^\dagger\mathbf S

to be positive semidefinite, where †\dagger denotes conjugate transpose. Equivalently, the largest singular value of S\mathbf S cannot exceed one. Merely checking that every individual ∣Sij∣|S_{ij}| is no greater than one is insufficient: simultaneous excitations can combine coherently, so the full matrix condition is necessary.

A lossless network satisfies

S†S=I.\mathbf S^\dagger\mathbf S=\mathbf I.

Its scattering matrix is unitary. For a lossless two-port excited only at port 1,

∣S11∣2+∣S21∣2=1.|S_{11}|^2+|S_{21}|^2=1.

The columns must also be mutually orthogonal. These constraints explain several microwave-network results, including the impossibility of a reciprocal, lossless three-port that is perfectly matched at every port. Practical three-port power dividers therefore incorporate dissipation, accept mismatch, or use nonreciprocity.

Active devices introduce a different issue: oscillation can arise from feedback through source and load reflections. For a linear two-port, define

Δ=S11S22−S12S21,\Delta=S_{11}S_{22}-S_{12}S_{21},

and, when S12S21≠0S_{12}S_{21}\ne0, the Rollett factor

K=1−∣S11∣2−∣S22∣2+∣Δ∣22∣S12S21∣.K= \frac{1-|S_{11}|^2-|S_{22}|^2+|\Delta|^2} {2|S_{12}S_{21}|}.

Under the standard two-port stability assumptions, K>1K>1 together with ∣Δ∣<1|\Delta|<1 establishes unconditional stability against passive source and load terminations at that frequency. These frequency-by-frequency tests do not replace a complete assessment of internal modes, bias networks, or nonlinear oscillation. A measured matrix characterizes only the ports and operating state included in the measurement.

Network transformations, cascading, and reference planes

S-parameters can be converted into other network descriptions when the required matrix inverses exist. With the same positive real reference impedance R0R_0 at every port, the impedance matrix is

Z=R0(I+S)(I−S)−1,\mathbf Z = R_0(\mathbf I+\mathbf S)(\mathbf I-\mathbf S)^{-1},

and the admittance matrix is

Y=1R0(I−S)(I+S)−1.\mathbf Y = \frac{1}{R_0} (\mathbf I-\mathbf S)(\mathbf I+\mathbf S)^{-1}.

These are changes of representation, not approximations. Singular cases indicate that a chosen representation is undefined at that frequency, rather than necessarily indicating a physically defective network. An ideal open circuit, for example, has a finite reflection coefficient but infinite impedance.

The Smith chart gives a graphical representation of the transformation between reflection coefficient and normalized impedance. Constant-resistance and constant-reactance contours make it possible to visualize matching networks, transmission-line transformations, and load trajectories over frequency. The chart remains useful even when numerical optimization performs the final design, because it exposes how a proposed element changes the impedance.

Connecting two two-port devices does not generally mean multiplying their S21S_{21} values. If network AA feeds network BB, with compatible port references and no additional coupling path, the combined forward transmission is

S21total=S21BS21A1−S22AS11B.S_{21}^{\mathrm{total}} = \frac{S_{21}^{B}S_{21}^{A}} {1-S_{22}^{A}S_{11}^{B}}.

The denominator accounts for repeated reflections at the interface. Ignoring it can misestimate the gain ripple of an amplifier-filter chain or the insertion loss of a cable assembly.

For repeated cascade calculations, ABCD parameters or scattering-transfer matrices are often more convenient because compatible cascade matrices multiply directly. However, a transfer representation involving division by a very small transmission coefficient can become numerically ill-conditioned near a deep stopband. General multiport interconnections are instead handled by eliminating internal incident and outgoing waves.

Reference planes are equally important. Extending the reference plane at port ii outward through a matched, lossless line of length lil_i changes a matrix entry according to

Sij′=Sije−j(βili+βjlj).S'_{ij} = S_{ij}e^{-j(\beta_i l_i+\beta_j l_j)}.

Reflection at one port accumulates a round-trip phase; transmission accumulates the propagation phase of both added port sections. Lossy matched extensions also introduce attenuation. Thus identical devices measured at different cable positions can have different phase data without having different intrinsic behavior.

Renormalization and reference-plane movement solve different problems. Renormalization changes the reference impedances; plane movement changes where waves are defined. Neither operation automatically removes a mismatched connector or fixture. Removing such structures requires a network model and de-embedding, usually by solving the relevant interconnection equations or inverting suitable cascade representations.

Measurement, calibration, and uncertainty

A vector network analyzer measures complex ratios of wave amplitudes. A source excites the device, directional couplers or bridges separate incident and reflected signals, and coherent receivers compare their magnitude and phase. A two-port instrument switches or routes the excitation so that both forward and reverse measurements can be obtained. The result is a frequency sweep of the scattering matrix at a specified power level and operating condition.

Raw receiver ratios include imperfections in the instrument and test arrangement. Finite coupler directivity allows incident energy to contaminate the reflected-wave measurement. Source and load mismatch create additional reflections, while tracking errors make the measured transmission and reflection responses frequency dependent. Calibration estimates these systematic errors using standards with known or sufficiently constrained electrical properties.

A common coaxial method is short-open-load-through, or SOLT, calibration. The short and open are not assumed to be ideal at all frequencies: practical calibration kits specify models that include delay and parasitic behavior. Through-reflect-line, or TRL, calibration instead uses transmission structures and a highly reflecting standard. It is particularly useful in planar and waveguide environments where manufacturing an accurately known broadband load is difficult.

Calibration has a finite domain of validity. In TRL, the phase difference between the line and through must avoid poorly conditioned values near integer multiples of 180∘180^\circ. Multiline methods use several line lengths to improve coverage and uncertainty. In all methods, connector repeatability, mechanical alignment, temperature variation, and cable movement can leave residual errors after calibration.

The uncertainty of a measurement depends strongly on what is being measured. Residual directivity limits the credibility of very small reflections. Receiver noise and leakage limit the depth of a measurable filter stopband. Source or receiver compression corrupts high-level measurements, while insufficient signal power worsens phase and magnitude precision. Averaging and narrower receiver bandwidth can reduce random noise, but they do not remove an incorrect calibration-standard model.

For active devices, the bias voltage, bias current, temperature, and stimulus level are part of the measurement conditions. S-parameters measured at low power around a specified operating point need not describe the same amplifier near compression. A reproducible data set therefore requires more than a matrix of numbers.

A measurement uncertainty statement may include magnitude-phase covariance and correlations across frequency. Such correlations become important when measured data are differentiated to obtain group delay, transformed into the time domain, or fitted to a broadband model. Verification standards that were not used to establish the calibration provide an independent check of the complete measurement process.

Multiport networks, differential signals, and applications

For an NN-port network, the scattering matrix contains N2N^2 complex entries at each frequency, although reciprocity and symmetry may reduce the number of independent quantities. Multiport characterization is essential when unwanted paths matter as much as intended transmission. A directional coupler, for example, is normally described through four ports representing input, through, coupled, and isolated paths. Coupling and directivity cannot be inferred from the input-to-output response alone.

High-speed digital interconnects often use two conductors as a differential pair. Their four single-ended terminals can be transformed into differential and common-mode variables. With equal real single-ended reference impedances, one common power-normalized convention is

ad=a1−a22,ac=a1+a22,a_d=\frac{a_1-a_2}{\sqrt{2}}, \qquad a_c=\frac{a_1+a_2}{\sqrt{2}},

with corresponding expressions for outgoing waves. The transformation is applied at both ends of the pair to obtain mixed-mode S-parameters.

In conventional mixed-mode notation, Sdd21S_{dd21} describes differential transmission and Sdd11S_{dd11} differential reflection. Terms such as Scd21S_{cd21} describe conversion from differential excitation into common-mode output. This conversion can result from unequal trace lengths, asymmetric vias, connector geometry, or unequal coupling to nearby conductors. A pair with acceptable differential insertion loss can still produce substantial common-mode radiation.

Reference impedances require care in mixed-mode data. For two uncoupled 50 Ω50\,\Omega single-ended ports, conventional physical differential and common-mode impedances are 100 Ω100\,\Omega and 25 Ω25\,\Omega, respectively. These values follow from the definitions of differential voltage and current, not from a universal requirement that every differential structure have 100 Ω100\,\Omega impedance. Port pairing and polarity must also be recorded, since reversing one conductor changes the signs of relevant mixed-mode terms.

For an antenna, one-port S11S_{11} quantifies feed mismatch but does not determine radiation efficiency or radiation pattern. Power accepted at the feed may be radiated or dissipated. In antenna arrays, multiport S-parameters describe mutual coupling; the active input reflection of an element depends on the amplitudes and phases applied to the other elements. This is especially important in phased arrays, where those excitations change with beam direction.

The same framework supports characterization of microwave filters, packages, switches, resonators, and on-chip interconnects. It also extends naturally to modes in an optical waveguide, where appropriately normalized amplitudes represent optical power. The mathematical resemblance does not remove the need to specify polarization, propagation mode, and reference planes: these define the channels represented by the matrix.

Broadband models, time-domain interpretation, and limitations

A conventional S-parameter data set describes a linear, time-invariant relationship at each measured frequency. For an active device, this is usually a small-signal linearization around a particular bias point. It does not, by itself, describe harmonic generation, intermodulation, gain compression, or changes caused by a strong modulated signal. Those effects require nonlinear models or measurement frameworks that relate multiple frequencies and operating conditions.

Nor does an ordinary scattering matrix fully describe noise. A noisy network can be represented as

b=Sa+c,\mathbf b=\mathbf S\mathbf a+\mathbf c,

where c\mathbf c is a vector of internally generated noise waves. Its correlation matrix contains information absent from S\mathbf S. Two amplifiers can therefore have similar gain and reflection data but different noise performance. Noise parameters or noise-wave correlations supplement the deterministic network description.

Broadband S-parameters can be transformed into a time-domain response through an inverse Fourier transform. This is useful for locating discontinuities in cables and interconnects: a reflection arriving after a particular round-trip delay indicates a corresponding propagation distance. The technique connects frequency-domain measurements to time-domain reflectometry, but the inferred response is limited by the measured bandwidth, frequency spacing, and treatment of unmeasured frequencies.

For uniformly spaced frequency samples, the transform record repeats over an interval related to 1/Δf1/\Delta f, where Δf\Delta f is the frequency spacing. The ability to distinguish closely spaced events is governed roughly by the reciprocal of the measured bandwidth. Windowing suppresses sidelobes at the cost of broadening features. Missing low-frequency or direct-current information is especially significant when reconstructing step responses.

Time-domain gating can suppress a response associated with a distant connector or fixture, but it is not equivalent to complete network de-embedding. Overlapping responses may not be separable, and removing part of an impulse response modifies the frequency-domain data. Interpretation also depends on propagation velocity and dispersion, so converting delay directly into distance requires a suitable line model.

Circuit simulators frequently need a compact model rather than a table of samples. Vector fitting and related rational-approximation methods represent frequency responses through poles and residues, sometimes with explicit delays. Such models must preserve relevant physical properties. A passive measured component can acquire artificial gain through an unconstrained fit, while a noncausal interpolation can produce a response before the excitation arrives.

Data are commonly exchanged in Touchstone files, which record frequency-dependent network parameters and associated conventions. Useful metadata include port order, reference impedances, units, magnitude-phase or real-imaginary format, bias conditions, and calibration planes. A file extension such as .s2p identifies a two-port scattering data set, but does not establish its accuracy, linear operating range, or validity outside the measured frequency band.