Published Nov 10, 2026 · Last updated Nov 10, 2026 · 7 min · IndieRF

Stability Circles: Which Side Is Stable, Then LNA Gain vs NF

The stable side of a load stability circle is the side holding the Smith chart centre when |S11| < 1, and the far side when |S11| > 1. On a synthetic FET at 10 GHz, a simultaneous conjugate match gives 12.26 dB at 2.03 dB NF, and moving Γ_S to Γ_opt gives 0.88 dB NF at 10.78 dB.

Find the centre of the Smith chart before you look at the stability circle. A matched load is ΓL=0\Gamma_L = 0, and there the input reflection is just S11S_{11}, so the centre is stable whenever ∣S11∣<1|S_{11}| < 1. The stable loads are then on the centre's side of the load stability circle. When ∣S11∣>1|S_{11}| > 1 the centre itself is unstable, and the stable side flips to the side without it. The source circle works the same way with ΓS\Gamma_S and ∣S22∣|S_{22}|.

The chart centre decides which side is stable

The load stability circle is every ΓL\Gamma_L where the input reflection reaches 1:

Γin=S11+S12S21 ΓL1−S22 ΓL,∣Γin∣=1\Gamma_\mathrm{in} = S_{11} + \frac{S_{12}S_{21}\,\Gamma_L}{1 - S_{22}\,\Gamma_L}, \qquad |\Gamma_\mathrm{in}| = 1

On one side of that circle ∣Γin∣<1|\Gamma_\mathrm{in}| < 1, the input returns less power than it receives, and the stage is stable. On the other side the input has negative resistance. The circle marks the boundary, and the shading needs one more fact. The centre gives it with one number: at ΓL=0\Gamma_L = 0, ∣Γin∣=∣S11∣|\Gamma_\mathrm{in}| = |S_{11}|.

The same answer comes out as a sign. With Δ=S11S22−S12S21\Delta = S_{11}S_{22} - S_{12}S_{21}, the inside of the load circle is stable when ∣S22∣<∣Δ∣|S_{22}| < |\Delta| and the outside when ∣S22∣>∣Δ∣|S_{22}| > |\Delta|. The centre rule is the one you can do in your head.

Four typed points on the Stability Circles page show both rules agree. Each is typed to be checked by hand.

Point|S11|CentreCircle holds the centre|S22||Δ|Stable side
2 GHz0.50stableno0.9000.602outside
3 GHz0.75stableyes0.8001.020inside
4 GHz1.20unstableno0.9001.098inside
5 GHz1.20unstableyes0.8060.573outside

At 3 GHz and 5 GHz the circle sits around the chart centre both times. At 3 GHz the inside is stable, and at 5 GHz the outside is, because ∣S11∣|S_{11}| crossed 1. A port with ∣S11∣>1|S_{11}| > 1 into 50 Ω is an oscillator that hasn't met its resonator yet, and oscillator designers build one on purpose.

In the clip, one step from 3 GHz to 5 GHz takes |S11| from 0.750 to 1.200, and the stable loads move from inside the load circle to outside it, with the chart centre inside the circle both times.

The typed cases on the Stability Circles page, stepped from 3 GHz to 5 GHz on the frequency slider. Open the 5 GHz point

μ says whether you need the circles at all

The single-number test is Edwards and Sinsky's μ\mu:

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

μ>1\mu > 1 means every passive load is stable, the same verdict as K>1K > 1 with ∣Δ∣<1|\Delta| < 1. With ∣S11∣<1|S_{11}| < 1, μ\mu is also the distance from the chart centre to the nearest unstable load. On IndieRF's synthetic FET at 10 GHz, μ=1.024\mu = 1.024: the nearest bad load sits at ∣ΓL∣=1.024|\Gamma_L| = 1.024, just off the edge of the chart, so any passive load is safe there with 2.4% to spare.

The same FET has μ=0.109\mu = 0.109 at 500 MHz, and it clears 1 only from 8.68 to 15.1 GHz. ∣S11∣|S_{11}| is 0.996 at 500 MHz, so a 50 Ω load is still on the stable side, by 0.4%. The 10 GHz matching networks present some load at 500 MHz too, and that out-of-band load is where this part can oscillate. Check the bias tee and the input network's low-frequency impedance against the 500 MHz circles before you trust a 10 GHz design.

Conjugate match against lowest noise at 10 GHz

At 10 GHz the FET is unconditionally stable, so both ports can be conjugate-matched at once. That gives the most gain the device has, GmaxG_\mathrm{max} = 12.26 dB, at 2.03 dB NF. Put the source at the noise optimum instead, ΓS=Γopt\Gamma_S = \Gamma_\mathrm{opt}, and conjugate-match the output to what that leaves: NF drops to NFminNF_\mathrm{min} = 0.88 dB and the transducer gain to 10.78 dB.

StateΓS\Gamma_SΓL\Gamma_LGTG_T (dB)NF (dB)Input VSWR
50 Ω on both ports007.501.223.03
Simultaneous conjugate match0.839∠122.1°0.855∠51.1°12.262.031.00
Lowest noise0.440∠99.9°0.614∠50.2°10.780.882.46

The trade is 1.48 dB of gain for 1.15 dB of NF. The bill for it arrives at the input: 2.46:1 VSWR, |Γ| 0.42 or 7.5 dB return loss, looking back into whatever filter sits in front. A filter designed for 50 Ω sees that mismatch and its passband ripple moves.

The second stage decides whether 1.15 dB of NF is real

Gain in front hides the second stage: with 20 dB in front, a 10 dB NF second stage adds about 0.3 dB to a 1 dB NF LNA. The rule breaks near 10 dB, where the same stage adds about 2.3 dB, and this LNA has only 10.8 to 12.3 dB. Friis divides the second stage's excess noise by the first stage's available gain, so the gain the lowest-noise point gives up comes back as noise.

Second stage NFCascade NF, conjugate matchCascade NF, lowest noiseLowest noise wins by
3 dB2.19 dB1.16 dB1.02 dB
10 dB3.28 dB2.96 dB0.32 dB

Behind a 3 dB NF second stage, the lowest-noise point keeps 1.02 of its 1.15 dB. Behind a 10 dB NF mixer it keeps 0.32 dB, and you paid 2.46:1 at the input for it. I put the source on Γopt\Gamma_\mathrm{opt} when the next stage is quiet, and when it is a mixer I add a second gain stage before I chase the last tenth of a dB at the first one.

The chart between the two points is the design. On the source chart, pick the noise circle you can live with (the page draws NFminNF_\mathrm{min} + 0.25, 0.5 and 1 dB by default) and slide ΓS\Gamma_S along it toward the conjugate point until the input VSWR suits the filter in front.

Where the numbers stop

Every number here is one bias point of one device. S-parameters and noise parameters both move with drain current, so a datasheet file at 10 mA describes a different device from the same part at 20 mA. The gate sets that current, and a depletion-mode FET needs its gate below 0 V: from a 0 to 2.5 V DAC, the gate-bias rescaler in the Op-Amp Lab gives −3.0 to −1.5 V with four E24 resistors and 0 V error at both ends. It is small signal: no compression, no IP3. The matching networks are ideal and lossless. Loss in front of the transistor adds to the NF decibel for decibel at 290 K, so a 0.3 dB input network turns 0.88 dB into about 1.18 dB, a quarter of the 1.15 dB the lowest-noise point bought.

The FET is IndieRF's synthetic part, simulated by the IndieRF Schematic Simulator. For a real transistor, download the .s2p with noise rows from the maker at your bias and drop it on either page.

References

  • D. M. Pozar, Microwave Engineering, 4th ed., ch. 12: stability circles, the conjugate match and noise circles.
  • G. Gonzalez, Microwave Transistor Amplifiers, 2nd ed., ch. 2 to 4.
  • M. L. Edwards and J. H. Sinsky, "A new criterion for linear 2-port stability using geometrically derived parameters," IEEE Trans. MTT, vol. 40, no. 12, 1992.
  • H. T. Friis, "Noise figures of radio receivers," Proc. IRE, vol. 32, no. 7, 1944.

Questions? Contact

Discussion

No comments yet. Start the thread with a measurement, a correction, or a worked example.