Lossless Networks and the Unitary S-Matrix
A lossless network has a unitary S-matrix, S^H·S = I, so the squared magnitudes in each column sum to 1 and any two different columns are orthogonal. In a lossless reciprocal two-port |S21|² = 1 − |S11|², so |S11| = 0.6 forces |S21| = 0.8 (−1.94 dB).
Engineer’s note: When a vendor sends S-parameters for a filter or switch, check first that |S11|² + |S21|² never goes above 1. If it does, their VNA calibration was off, and if you feed that file to a simulator you'll watch a passive part amplify a signal with no amplifier anywhere in the circuit.
For practitioners
A lossless network can't absorb power. Every watt that goes in has to come back out of some port. Written in S-parameters, that one physical rule becomes a compact matrix condition: S has to be unitary. This page derives that condition, explains what it says about each column, and uses it to show how S11 sets S21 in a lossless reciprocal two-port.
Power balance in terms of a and b
Take an N-port with S-parameters referenced to real port impedances, using the usual power-wave normalization. The incident waves a and the outgoing waves b are related by b = S·a. With this normalization, |ai|² is the power incident on port i and |bi|² is the power leaving port i. If you work with peak phasors, both carry the same factor of 1/2, and it cancels out of the balance.
The net power delivered into the network is the total incident power minus the total outgoing power:
Here S^H is the conjugate transpose of S. A lossless network dissipates nothing for any excitation, so P_net = 0 for every possible vector a. The matrix I − S^H·S is Hermitian. A Hermitian matrix whose quadratic form is zero for every vector must be the zero matrix. So:
That is the unitary condition. A passive but lossy network only satisfies P_net ≥ 0, which means I − S^H·S is positive semidefinite and not zero.
What the unitary condition says about each column
Entry (i, i) of S^H·S is the sum of |Ski|² down column i. Setting it equal to 1 gives:
Column i describes what happens when you drive port i and terminate every other port in a matched load. Then ai is the only incident wave, and each bk = Ski·ai. The sum says that all of the incident power comes back out somewhere. Part of it reflects at port i (|Sii|²) and the rest goes to the other ports (|Ski|² for k ≠ i). None of it disappears inside the network.
This is a statement about columns, which means one excitation at a time. Rows of a unitary matrix also have unit norm, because S·S^H = I holds as well for a square matrix. Even so, the column reading is the one that maps directly to a bench setup.
Orthogonal columns and the phase constraint
The off-diagonal entries of S^H·S must be zero. For any two different columns i ≠ j:
This condition doesn't come from conserving power under single-port drive. It comes from driving two ports at once. When ports i and j are both excited, the outgoing waves from the two drives interfere. Without orthogonality, the cross terms would make the total output power depend on the relative phase of the drives. A lossless network can't let that happen, so the cross terms must cancel.
Orthogonality is where the phase information lives. The column-norm conditions only constrain magnitudes. Orthogonality ties the phases of different S-parameters together.
The lossless reciprocal two-port: S11 sets S21
For a two-port, the column conditions are |S11|² + |S21|² = 1 and |S12|² + |S22|² = 1. A reciprocal network also has S12 = S21. Subtracting the two conditions gives:
Orthogonality of the two columns gives S11·S21* + S21·S22* = 0. When S21 ≠ 0, this confirms |S11| = |S22| and also fixes the phases:
So in a lossless reciprocal two-port, one reflection magnitude fixes both |S22| and |S21|. All of the insertion loss is mismatch loss: −10·log10(1 − |S11|²).
The curve is very flat near a good match. At |S11| = −20 dB, |S21| is only −0.044 dB. At −10 dB it is −0.46 dB. The two magnitudes are equal at −3.01 dB, where the power splits evenly between reflection and transmission. As |S11| gets close to 0 dB, |S21| falls off quickly. It crosses −10 dB when |S11| reaches about −0.46 dB.
Worked example: |S11| = 0.6
Suppose a lossless reciprocal two-port has |S11| = 0.6, which is −4.44 dB.
- Reflected power fraction: |S11|² = 0.36.
- Transmitted power fraction: |S21|² = 1 − 0.36 = 0.64.
- |S21| = √0.64 = 0.8, and 20·log10(0.8) = −1.94 dB.
- Reciprocity and losslessness give |S22| = 0.6 as well.
The −1.94 dB of insertion loss is pure mismatch. If you match the input, it goes away. On a real part, any shortfall from |S11|² + |S21|² = 1 is the power the part actually dissipates. For more on how the two dB figures relate, see Return Loss vs Insertion Loss.
Why a lossless reciprocal three-port can't be matched everywhere
Suppose a three-port is lossless and reciprocal and matched at all three ports, so S11 = S22 = S33 = 0 and Sij = Sji. Apply orthogonality to each pair of columns:
- Columns 1 and 2 give S31·S32* = 0.
- Columns 1 and 3 give S21·S23* = 0.
- Columns 2 and 3 give S12·S13* = 0.
These three conditions force at least two of S12, S13 and S23 to be zero. Say S12 = S13 = 0. Then column 1 has |S21|² + |S31|² = 0, which breaks the unit-norm condition. The assumptions contradict each other. You need to give up one of them: allow loss, as with the resistor in a matched power divider, give up reciprocity, as in a circulator, or accept a mismatch at one port.
In plain words · for students and beginners
The unitary S-matrix in plain words
An everyday picture
Think of a sheet of perfectly clear glass. Light hits it. Some bounces back and the rest passes through. If the glass absorbs nothing, the light reflected plus the light transmitted equals the light that came in. A lossless network does the same thing with RF power. S11 tells you how much bounces back. S21 tells you how much gets through.
The picture breaks down in one place. Radio waves have phase, which is the timing of the wave's peaks. When you drive two ports at once, the waves can add or cancel. The "unitary" rule includes a second condition about phase, called orthogonality, that the glass picture doesn't show. Orthogonal here just means the phases are arranged so the cross effects cancel.
The worked example, step by step
- Send 1 W into port 1. Put a matched load on port 2, meaning a termination that reflects nothing.
- |S11| = 0.6. Power goes as the square of the wave size, so 0.6 × 0.6 = 0.36 W comes back out of port 1.
- Nothing is lost inside, so the rest, 1 − 0.36 = 0.64 W, leaves port 2.
- The wave size at port 2 is √0.64 = 0.8, so |S21| = 0.8.
- In decibels (a log scale for ratios), that is −4.44 dB for S11 and −1.94 dB for S21.
Check: 0.36 + 0.64 = 1. All the power is accounted for.
Mistakes beginners make
- Adding the magnitudes. 0.6 + 0.8 = 1.4, not 1. The rule adds the squares, because the squares are powers.
- Adding the dB values. −4.44 dB and −1.94 dB don't add to 0 dB. Convert to linear power fractions first, then add.
- Treating real parts as lossless. Real cables, filters and boards turn a little power into heat. On real hardware, |S11|² + |S21|² comes out a bit below 1. The gap is the power lost as heat.
- Expecting a perfect three-way splitter with no resistor. A lossless, reciprocal three-port can't be matched at all three ports. Reciprocal means the signal behaves the same in both directions.
Remember
- Lossless means S^H·S = I. In each column, the squared magnitudes add to 1.
- For a lossless reciprocal two-port, |S21|² = 1 − |S11|², and |S11| = |S22|.
- Do the power sum in linear units, never in dB.
Related pages on RF View
Frequently asked questions
What does it mean that the S-matrix of a lossless network is unitary?
It means S^H·S = I, where S^H is the conjugate transpose. Each column has squared magnitudes that add up to 1, so no power is lost, and any two different columns are orthogonal.
How do I get S21 from S11 for a lossless two-port?
If the two-port is lossless and reciprocal, |S21|² = 1 − |S11|². For example, |S11| = 0.6 gives |S21| = 0.8, which is −1.94 dB.
Is |S11| always equal to |S22| in a lossless two-port?
Yes, if the two-port is lossless and reciprocal. The two column power sums and S12 = S21 together force |S11| = |S22|.
Why can't a lossless reciprocal three-port be matched at all ports?
If you set all three diagonal terms to zero, the orthogonality conditions force at least two of the three transmission terms to zero. Then at least one column can't have squared magnitudes that sum to 1. A matched three-port therefore needs loss, like the resistor in a matched divider, or a non-reciprocal device.