The Mu Stability Factor: A Single-Number Stability Test

A two-port is unconditionally stable if and only if μ > 1, so one number replaces the K > 1 and |Δ| < 1 pair. Because μ is the distance from the Smith chart center to the nearest unstable load, a larger μ means more stability margin.

Engineer’s note: Mu is handy because it rolls K and delta into one number: above 1 means unconditionally stable. Just don't stop at the frequency you matched. PAs love to oscillate where nobody looked, tens of MHz down or several GHz up, and burn the chip. Compute mu across the whole range where the device still has gain, not just your band.

For practitioners

Why a single-number test matters

An amplifier is a two-port with feedback built in. That feedback is S12. Any S12 at all means the input reflection depends on the load, and the output reflection depends on the source. If some passive termination pushes |Γin| or |Γout| above 1, the device can oscillate.

The classic check is the Rollett pair. You compute K and the determinant Δ, then check two things:

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

with Δ = S11 S22 − S12 S21. Both conditions must hold. K > 1 alone is not enough. That is a common source of errors in sweep scripts and optimizer goals. Someone writes a goal on K, forgets |Δ|, and the optimizer finds a corner where K looks fine and the device is not actually stable.

Edwards and Sinsky gave a test that needs only one number. It is necessary and sufficient on its own. An optimizer can use it as one goal, and you can read it from one trace on a frequency plot.

The mu and mu-prime definitions

The load-side factor is:

μ=1−∣S11∣2∣S22−ΔS11∗∣+∣S12S21∣\mu = \frac{1-|S_{11}|^2}{|S_{22}-\Delta S_{11}^*|+|S_{12}S_{21}|}

The source-side factor is its mirror image, with ports 1 and 2 swapped:

μ′=1−∣S22∣2∣S11−ΔS22∗∣+∣S12S21∣\mu' = \frac{1-|S_{22}|^2}{|S_{11}-\Delta S_{22}^*|+|S_{12}S_{21}|}

Here S11* is the complex conjugate of S11, and Δ is the same determinant used in the K test.

The rule is simple. The two-port is unconditionally stable if and only if μ > 1. Testing μ' > 1 is equivalent: if one is above 1, so is the other. You only need to compute one of them for the verdict. The two values are generally not equal, though. Each one measures margin in its own plane, so check μ for load-side margin and μ' for source-side margin.

Two details in the formula are worth noting:

  • The numerator 1 − |S11|² must be positive for μ > 1. If |S11| ≥ 1 with matched terminations, the device is already unstable, and μ cannot exceed 1. This check is built into the formula.
  • The term |S12 S21| sits in the denominator. More feedback makes μ smaller. There is no division by |S12 S21|, so μ stays finite as S12 goes to zero. K does not: its denominator goes to zero.

The geometric meaning: distance to the nearest unstable load

The useful property of μ is that its value has a geometric meaning. μ is the distance from the center of the Smith chart to the nearest unstable point in the load plane. The Smith chart itself is the disk |Γ_L| ≤ 1, all the passive loads. If the nearest unstable point is farther than 1 from the center, no passive load can reach it. That is exactly why μ > 1 means unconditional stability.

This also explains why the number measures margin. Say μ = 1.05. The unstable region starts just outside the chart's edge. A small shift in S-parameters can move that boundary inside the chart: a process corner, a temperature change, a bias change, or a package parasitic the model left out. At μ = 2, the boundary sits twice the chart radius away from the center. The device can drift a lot before any passive load becomes dangerous.

K does not give this directly. K is a ratio of terms, not a distance, so comparing K = 1.5 with K = 3 on two devices does not compare their margins in any geometric sense. μ ranks margin by construction.

Worked example: a unilateral device

Take a device with S12 = 0, |S11| = 0.5 and |S22| = 0.5. With S12 = 0, the determinant is Δ = S11 S22, so |Δ| = 0.25.

Numerator:

1 − |S11|² = 1 − 0.25 = 0.75

Denominator. The feedback term |S12 S21| is zero. The other term simplifies:

S22 − Δ S11* = S22 − S11 S22 S11* = S22 (1 − |S11|²)

so its magnitude is |S22|(1 − |S11|²) = 0.5 × 0.75 = 0.375.

Result:

μ = 0.75 / 0.375 = 2

In general, for any unilateral device the factor (1 − |S11|²) cancels, and μ = 1/|S22|. In the same way, μ' = 1/|S11|, which is also 2 here. Both are well above 1, so the device is unconditionally stable. That matches intuition: with no reverse path, the load cannot affect the input, and a passive device with |S11| and |S22| below 1 cannot be made to oscillate.

Now the K test on the same device. Its denominator, 2|S12 S21|, is zero, so K is unbounded. The verdict is still correct, because K > 1 and |Δ| = 0.25 < 1. But "infinite" tells you nothing about margin. μ = 2 does: the nearest unstable load is at distance 2 from the chart center.

ΓL planeoriginstable|ΓL| = 1radius 1μ = distance to nearestunstable load = 2unstable loads(beyond radius 2)stability circleradius 1/|S22| = 2Whole Smith chart lies insidethe stable region, so μ > 1
Load-plane view of the mu stability factor for a unilateral device with |S11| = 0.5 and |S22| = 0.5, where the nearest unstable load sits at distance mu = 2 from the chart center. Concept drawing.

The figure shows this case in the Γ_L plane. The unit circle, radius 1, is the edge of the Smith chart. The boundary of the unstable region is drawn as a circle of radius 2 around the origin, and everything beyond it is unstable loads. The arrow from the origin to the nearest point on that circle has length 2, which is μ. The whole Smith chart lies inside the stable region, so μ > 1.

Using mu in practice

A few working rules follow from the definition:

  • Sweep across the whole band, not just the operating band. Stability is checked one frequency at a time. A device can have μ well above 1 in band and drop below 1 far out of band, where gain is still present. Check every frequency where the device has gain.
  • Set the goal with margin. μ > 1 is the boundary, not a design target. Since μ is a distance, a goal a bit above 1 gives real room for model error and spread between parts.
  • Check both factors when one side matters more. If the source termination is well controlled but the load is not, μ is the one to watch.
  • Compute μ from the same S-parameter set you trust for gain. A de-embedding error that shifts S12 phase changes μ directly.
μ > 1 proves stability only for the two-port as measured or modeled. Bias networks, ground inductance and coupling between stages are outside the two-port unless they are in the S-parameters. A stage that passes the μ test on its own can still oscillate in a cascade.

In plain words · for students and beginners

The mu stability factor in plain words

An everyday analogy

Picture a flat, round table in the middle of a large floor. Somewhere on the floor there is a hole. As long as the hole is off the table, nothing placed on the table can fall into it. μ tells you how far the nearest edge of the hole is from the center of the table, where the table's radius counts as 1.

  • μ greater than 1: the hole starts beyond the table's edge. Everything on the table is safe.
  • μ less than 1: part of the hole is on the table. Some spots are dangerous.

The table is the Smith chart, the chart of every passive load you could connect. A passive load is one that only absorbs or reflects power and never adds any. The hole is the set of loads that would make the amplifier oscillate, which means it creates its own signal with no input.

Where the analogy breaks down: a real table does not move. The amplifier's "hole" shifts with frequency, temperature and bias. That is why the same check has to be repeated at every frequency.

The worked example, step by step

Use the same device as above. It has no reverse path (S12 = 0, so nothing leaks from output back to input). Its input and output reflections are both 0.5 in size.

  1. Top of the fraction: 1 − 0.5² = 1 − 0.25 = 0.75.
  2. Bottom of the fraction: 0.5 × 0.75 = 0.375. The feedback part is zero, because S12 is zero.
  3. Divide: 0.75 / 0.375 = 2.
  4. Compare with 1: 2 is greater than 1, so the device is stable with any passive load.
0123boundary (μ = 1)conditionally stableor unstableunconditionally stablelarger μ = more stability marginexample: μ = 2S12 = 0, |S11| = |S22| = 0.5Mu stability factor μ (dimensionless)
Number line of the μ stability factor: the boundary is at μ = 1, values above 1 are unconditionally stable, and the worked example sits at μ = 2. Concept drawing.

On this number line, 1 is the boundary. Below 1, the device is only conditionally stable or unstable. "Conditionally" means some loads are safe and some are not. Above 1, it is unconditionally stable. Our example sits at 2, well to the right. The further right the value is, the more margin you have.

Mistakes beginners make

  • Treating 1.01 as safe. It passes, but just barely. Small changes in the real part can push it below 1.
  • Checking only one frequency. Stability has to hold at every frequency where the device has gain, not just the frequency you care about.
  • Mixing up the two tests. The older K test needs two conditions, K > 1 and |Δ| < 1. The μ test needs only μ > 1. Do not apply K's rules to μ.
  • Expecting μ and μ' to be equal. They give the same yes-or-no answer, but they measure distance on different sides of the device, so the numbers can differ.

Remember

  • μ > 1 means no passive load or source can make the two-port oscillate.
  • The value of μ is a distance, so a bigger number means more margin.
  • One number does the job of the two-part K and |Δ| test.

Frequently asked questions

What does mu greater than 1 mean for an amplifier?

It means the two-port is unconditionally stable at that frequency. No passive source or load can make it oscillate. The test is necessary and sufficient, so you do not need a second condition.

Is the mu test better than the Rollett K factor?

Both give the same stable or not-stable verdict when K is paired with |Δ| < 1. Mu does it with one number, and its value is a geometric distance, so it also ranks how much margin you have.

What is the difference between mu and mu prime?

Mu is the load-side factor and mu prime is the source-side factor. Either one being greater than 1 proves unconditional stability, but their values are generally different because they measure distance in different planes.

What is mu for a unilateral amplifier with S12 = 0?

The formula reduces to μ = 1/|S22|. With |S22| = 0.5 this gives μ = 2, and the source-side factor reduces to μ' = 1/|S11|.

References and further reading