IndieRF Amplifier Design

Amplifier from an .s2p: Γ_S, Γ_L, gain and noise

At 10 GHz the synthetic FET below reaches G_max = 12.26 dB with Γ_S = 0.839∠122.1° and Γ_L = 0.855∠51.1°, at 2.03 dB NF. Move Γ_S to Γ_opt and NF drops to 0.88 dB for 10.78 dB of gain. Drop your own .s2p and drag both points: the gain and noise circles, G_T, G_P, G_A and NF follow, and Match builds the networks.

or drop one here. The file stays in this browser.
Example:

IndieRF synthetic FET (common source, gm 60 mS), simulated by the IndieRF Schematic Simulator. It is not a real part. For a real transistor, download the .s2p from the maker and drop it here.

Open in RF Schematic SimulatorOpens the 10 GHz LNA example, built on this FET with a bias tee.

Where to put Γ_S and Γ_L

For the most gain, press Simultaneous conjugate match. At 10 GHz the FET gets G_max = 12.26 dB with Γ_S = 0.839∠122.1° and Γ_L = 0.855∠51.1°, at 2.03 dB NF. For the least noise, press Lowest noise: Γ_S moves to Γ_opt = 0.440∠99.9°, NF drops to 0.88 dB, and G_T is 10.78 dB with the output still conjugate-matched. That trade costs 1.48 dB of gain for 1.15 dB of NF, and an input VSWR of 2.46.

The gain matters one level up, because gain in front hides the next stage. A 10 dB NF second stage adds about 2.3 dB to a 1 dB NF first stage behind 10 dB of gain and about 0.3 dB behind 20 dB, so I check the cascade before I give up gain for the last tenth of a dB of NF. The chart between those two points is the design: pick a noise circle you can live with, and a gain circle that crosses it.

The note Stability circles: which side is stable, then LNA gain vs NF works this trade through a 3 dB and a 10 dB NF second stage, and shows which side of each stability circle is shaded and why.

G_T counts everything from the source's available power to the load. G_P depends on Γ_L alone and assumes the input takes all the power offered to it. G_A depends on Γ_S alone and assumes the load takes all the power available. So the load chart carries operating-gain circles and the source chart carries available-gain circles, and G_T equals G_P when the input is conjugate-matched and G_A when the output is.

A conjugate match needs K > 1 and |Δ| < 1. At 3 GHz the FET has K = 0.440, so the button refuses and says why. There the bar compares G_T with MSG = |S21/S12|, and only the shaded side of each chart gives a number. The Stability Circles page shows K and μ across the file.

Definitions

Every gain is a linear power ratio from the file's S-parameters, with both ports kept: no unilateral approximation. With D = (1 − S11·Γ_S)(1 − S22·Γ_L) − S12·S21·Γ_S·Γ_L:

  • Γ_in = S11 + S12·S21·Γ_L / (1 − S22·Γ_L), Γ_out = S22 + S12·S21·Γ_S / (1 − S11·Γ_S)
  • G_T = (1 − |Γ_S|²)·|S21|²·(1 − |Γ_L|²) / |D|²
  • G_P = |S21|²·(1 − |Γ_L|²) / ((1 − |Γ_in|²)·|1 − S22·Γ_L|²)
  • G_A = |S21|²·(1 − |Γ_S|²) / ((1 − |Γ_out|²)·|1 − S11·Γ_S|²)
  • Operating-gain circle, Γ_L plane: g = G_P/|S21|², C2 = S22 − Δ·S11*, centre = g·C2* / (1 + g·(|S22|² − |Δ|²)), radius = √(1 − 2K·|S12·S21|·g + |S12·S21|²·g²) / |1 + g·(|S22|² − |Δ|²)|. The available-gain circle is the same with 11 and 22 swapped, in the Γ_S plane, with g = G_A/|S21|².
  • Noise: F = Fmin + 4·r_n·|Γ_S − Γ_opt|² / ((1 − |Γ_S|²)·|1 + Γ_opt|²), r_n = Rn/Z0, with F the linear noise factor. Circle: N = (F − Fmin)·|1 + Γ_opt|² / (4·r_n), centre = Γ_opt/(N + 1), radius = √(N·(N + 1 − |Γ_opt|²)) / (N + 1).
  • Simultaneous match: B1 = 1 + |S11|² − |S22|² − |Δ|², C1 = S11 − Δ·S22*, Γ_S = (B1 − √(B1² − 4|C1|²)) / (2C1), the root inside the chart. Γ_L is the same with 11 and 22 swapped.
  • Input VSWR comes from |Γ_a| = |(Γ_in − Γ_S*) / (1 − Γ_S·Γ_in)|, the mismatch a lossless match leaves at the source. Output VSWR is the same with Γ_out and Γ_L.

Noise rows

Touchstone 1 files write Rn divided by the option-line R, and Touchstone 2 files write it in ohms. The page reads both. Γ_opt is referred to the option-line R, so a file with a different [Reference] gets it re-referred to port 1. When the noise rows sit on another grid from the S rows, the page takes the nearest row, prints its frequency, and does not interpolate. Outside the span of the rows it turns noise off. A file with no noise rows still gets every gain.

IndieRF serves no vendor files. For a real transistor, download the .s2p with noise parameters from the maker at your bias point and drop it on the page. The file stays in your browser.

Build it

Build this match opens Smith Chart & Matching for each side with the frequency filled in. To present Γ_S to the device, the network matches the conjugate of Z_S, so the link passes Z_S* and Z_L*. Match's load end is the device on both sides, so the output network goes in with Match's source end toward your load.

A depletion-mode GaN or GaAs 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 20 k, 7.5 k, 12 k and 10 k on E24, and 0 V error at both ends.

Limits

One bias point, one frequency, small signal. No compression, IP3, efficiency or bias network. The result is narrowband: the file says nothing about the networks away from this point, and a stage that is stable here can still oscillate at a frequency the file does not cover. The matching networks here are ideal and lossless: loss in a real input network adds to NF.

Checked against scikit-rf 2.2: an ideal lossless network on each side of the 2-port, with |S21|² of the cascade read as G_T, in linear watts, to 1e-9. Every gain and noise circle is sampled and each point evaluated from the S-parameters and the noise parameters directly. References: D. M. Pozar, Microwave Engineering, 4th ed., ch. 12; G. Gonzalez, Microwave Transistor Amplifiers, 2nd ed., ch. 2 to 4.

Checked against scikit-rf: the cases, the tolerance and what the check does not cover.

Question or a correction?