Why S-Parameters Instead of Z or Y at Microwave Frequencies
A 0.1 pF fringing capacitance turns a 10 GHz open into about 159 Ω, so the ideal opens and shorts behind Z and Y parameters cannot be built. S-parameters use matched Z0 loads, keep transistors stable, and stay bounded at |Sii| ≤ 1 for passive networks.
Engineer’s note: Z and Y parameters need ports that are perfectly open or shorted, and above a few GHz a perfect open or short doesn't exist; parasitics always show up. S-parameters measure reflection against a 50 ohm load, which is easy to make and stays well behaved across a wide band. That's why they became the common language of high-frequency work.
For practitioners
At low frequencies a two-port network is described with impedance (Z) or admittance (Y) parameters, and you can measure them with an open or a short on the far port. At microwave frequencies those terminations stop being what their names promise. A transistor under test may also oscillate with them. Scattering parameters avoid both problems by putting a matched load on the unused port, and that is why every network analyzer reports S-parameters.
What Z and Y parameters demand
Z and Y parameters relate port voltages and port currents. Each one is defined with a specific termination on the other ports:
Z-parameters need the other ports open-circuited, so their currents are zero. Y-parameters need them short-circuited, so their voltages are zero. The transfer terms work the same way: Z21 = V2/I1 with I2 = 0, and Y21 = I2/V1 with V2 = 0.
On paper these are clean definitions. On the bench they depend on an ideal open or an ideal short at the reference plane. That works at audio frequencies and still works at a few megahertz. At microwave frequencies it breaks down, because every real open has capacitance and every real short has inductance.
Worked example: a 0.1 pF open at 10 GHz
Take an open SMA-style end or an unconnected pad on a test fixture. The electric field fringes off the end of the center conductor, which acts as a small capacitance to ground. Assume 0.1 pF and compute its reactance at 10 GHz:
The "open" is about −j159 Ω, not infinity. It is only about three times the 50 Ω system impedance. You can express this as a reflection coefficient against Z0 = 50 Ω:
Γ = (Z − Z0) / (Z + Z0) = (−j159 − 50) / (−j159 + 50)
The magnitude is 1, since a pure reactance absorbs no power. The phase is about −35° when an ideal open would give 0°. So port 2 is far from the I2 = 0 condition the definition requires, and the Z21 you record is the response into a 159 Ω capacitor. The error also changes with frequency. The same 0.1 pF is about 318 Ω at 5 GHz and about 80 Ω at 20 GHz, so the termination drifts across the sweep.
The short has the mirror problem. Any physical short has some series inductance in the shorting path, and its reactance ωL grows with frequency. A short is not 0 Ω, so V2 is not zero and the Y-parameter definition is not met either. The figure shows both cases next to the third option, a resistive Z0 load. Unlike the open and the short, a resistive load can be built to stay close to its nominal value over a wide band.
Active devices and reactive terminations
The second problem matters most for transistors. An open or a short is a purely reactive termination with |Γ| = 1. An active device can oscillate when its ports see such a load. Once it oscillates, it is no longer a small-signal linear network. The measured "parameters" then describe a broken setup, and the oscillation can also damage the part.
A transistor whose ports are terminated in a resistive 50 Ω load is usually stable. The resistance absorbs energy that would otherwise reflect back and build up. That makes the matched load the safe measurement condition as well as the accurate one.
What S-parameters define instead
S-parameters describe waves, not voltages and currents. At each port the incident wave a and the outgoing wave b are linear combinations of V and I, normalized to the reference impedance Z0:
The parameters are ratios of these waves, with the unused port terminated in Z0:
The condition a2 = 0 means no wave comes back into port 2. A load equal to Z0 absorbs the wave leaving port 2 and reflects nothing. That condition is realistic at microwave frequencies. A good 50 Ω termination is a standard, well-characterized part, and the VNA calibration removes its remaining imperfection. Waves also fit the hardware better. A directional coupler separates incident from reflected waves directly. On a line that is long compared with a wavelength, voltage and current change with position, while the wave amplitudes keep their magnitude along a lossless matched line. The VNA measurement page covers how the instrument does this.
Bounded numbers and exact conversion
S-parameters have another practical advantage. For a passive network every reflection coefficient satisfies |Sii| ≤ 1. A passive network cannot return more power than it receives, so the numbers stay inside the unit circle of the Smith chart. They plot cleanly on a dB scale, and an instrument can represent them with a fixed range.
Z and Y values have no such limit. Near a parallel resonance an impedance can rise toward infinity. Near a series resonance an admittance does the same. Measuring and storing those numbers directly would need an enormous dynamic range.
Nothing is lost by measuring S. All three sets describe the same linear network, so you can convert between them exactly. For one port:
The two-port conversions follow the same pattern with matrices. The usual approach is to measure S with matched loads and calibrate at the reference plane. You then convert to Z or Y in software when a circuit model needs them. For what each S-parameter means physically, see S-Parameters Explained: S11, S21, S12 and S22.
In plain words · for students and beginners
S-parameters vs Z and Y in plain words
An everyday picture
Imagine you want to learn how a hallway handles sound. One method is to block the far end with a perfectly hard wall, or to leave it perfectly open to the outside, and then listen. The other method is to hang a thick curtain at the far end that soaks up all sound, so no echo comes back from there. Then you shout from your end and measure two things: what bounces back to you and what reaches the curtain.
Z and Y parameters use the first method: a perfect open or a perfect short on the other port. S-parameters use the curtain, which is a 50 Ω load that absorbs everything. The picture breaks down in one way. Radio waves also have phase (timing), and the "curtain" only absorbs perfectly when its resistance equals the reference impedance Z0, the impedance the measurement is based on.
What the analyzer actually sends and receives
A network analyzer (VNA) sends a wave a1 into port 1. Part of it comes back as b1, and part of it passes through and leaves port 2 as b2. That wave goes into a 50 Ω load and stops there, so nothing comes back into port 2 (a2 = 0).
S11 = b1/a1 is the fraction that bounces back. S21 = b2/a1 is the fraction that gets through. For a passive part, meaning one with no amplifier or power supply inside, |S11| can never be more than 1. You cannot get back more than you sent.
The same example, step by step
- A real "open" end of a connector still has a tiny capacitance (a part that stores charge between two conductors), about 0.1 pF in this example.
- At 10 GHz its reactance (how strongly it resists AC current) is 1 / (2π × 10,000,000,000 × 0.0000000000001), which is about 159 Ω.
- An ideal open is infinite ohms. 159 Ω is only about three times 50 Ω, so it is nothing like infinite.
- The Z-parameter you measure is therefore wrong, and the error changes with frequency.
- A 50 Ω resistor stays close to 50 Ω over a wide frequency range, so the S-parameter measurement condition is easy to meet.
Mistakes beginners make
- Thinking an unconnected connector is an open circuit at high frequency. It is really a small capacitor.
- Thinking S-parameters replace Z and Y. They describe the same network, and you can convert between them with exact formulas.
- Putting an open or a short on a transistor during a test. It can start to oscillate (produce its own unwanted signal). A 50 Ω load usually keeps it stable.
- Assuming a large Z value is a measurement error. Near a resonance (the frequency where a circuit naturally "rings") Z or Y can become huge, while S stays at or below 1 for passive parts.
Remember
- Z needs ideal opens and Y needs ideal shorts, and neither exists at microwave frequencies.
- S-parameters use a matched 50 Ω load, so a2 = 0 and S11 = b1/a1.
- A passive network always has |Sii| ≤ 1, so S values stay bounded.
Related pages on RF View
Frequently asked questions
Why are Z-parameters hard to measure at microwave frequencies?
Z-parameters need the other ports open-circuited, and a real open always has fringing capacitance. Just 0.1 pF has a reactance of about 159 Ω at 10 GHz, so the open is nowhere near an infinite impedance.
Why are S-parameters measured with 50 Ω loads?
S-parameters are defined with the unused ports terminated in the reference impedance Z0, usually 50 Ω. A resistive load can be built well across a wide band, and it absorbs the outgoing wave so nothing comes back into the port (a2 = 0).
Can I still get Z or Y parameters from a VNA?
Yes. S, Z and Y describe the same linear network, so you measure S-parameters and convert them with exact formulas such as Zin = Z0 (1 + S11) / (1 − S11). Near resonances the converted values can become very large and sensitive to small errors.
Why can a transistor oscillate during a Z or Y measurement?
An open or a short is a purely reactive termination, and an active device can oscillate when its ports see one. A resistive 50 Ω termination usually keeps the device stable, so the measured data describes a working amplifier.