Lab 4

The 2.4 GHz LNA

Run the course LNA of Chapter 9 as a testbench of your own: S-parameters and noise from ngspice, read and checked with scikit-rf, the noise circles that the four noise parameters draw, a two-tone test in Xyce, and the same again at the fast corner and 85 °C.

  • Build the course LNA’s netlist from Chapter 9’s script and run its sp analysis with noise and its two-tone harmonic balance yourself.
  • Read the results into a scikit-rf Network with its noise parameters, and refer them to the 290 K of the noise-figure definition.
  • Measure the gain, the input match and its bandwidth, NF and NFmin, the stability factors and the IIP3, and compare each with Chapter 9.
  • Draw the noise circles and say why the 50 Ω source is not the noise optimum.
  • Repeat the checks in a process corner and explain what moves.
Contents
Lab 4 at a glance  
Goal The course LNA’s gain, match, noise, stability and linearity from your own runs, at the typical corner and at ff, 85 °C.
Chapters Chapter 9 (Sections 9.4, 9.5 and 9.7, Tables 9.2–9.4, Worked example 9.1); Chapter 4 for NF and IIP3
Time 3 hours
Files tools/labs/lab4/: lna_lab.py; it imports tools/figs/ch09_lna.py (the netlist) and tools/figs/ch04_noise_linearity.py (harmonic-balance post-processing)
Run time ngspice sp under a second; Xyce two-tone about 4 s; lna_lab.py all about 10 s

Hand calculation

The course LNA (Figure 9.5) has \(L_g\) = 13.98 nH off chip with Q = 40, \(L_s\) = 0.823 nH, M1 of 128 µm at 4.70 mA, and a pad with ESD diodes. Before you simulate:

  1. Input Q and match bandwidth. Eq. (9.10) gives the loaded Q of the input network, \(Q_L = \omega_0(L_g + L_s)/R_S\). Compute it at 2.44 GHz. For a series resonant input matched to \(R_S\), the reflection is \(\Gamma = jX/(2R_S + jX)\) with \(X \approx 2Q_LR_S\,\Delta f/f_0\), so \(\vert S_{11}\vert\) = −10 dB where \(X = 2R_S/3\). Show that the −10 dB bandwidth is \(2f_0/(3Q_L)\) and evaluate it.
  2. Noise figure. Add up Eq. (9.9) with the four terms of Worked example 9.1: \(R_{Lg}/R_S\) = 0.107, \(R_g/R_S\) = 0.0672, \(R_{Ls}/R_S\) = 0.0268 and the channel term 0.176. Give F and NF.
  3. The hot corner. Every term of \(F - 1\) is thermal noise, so to first order it scales with the absolute temperature (the reference source stays at 290 K). Starting from the simulated \(F - 1\) = 0.309 at 27 °C, predict the NF at 85 °C.
  4. Linearity at the corner. Section 9.5 found that OIP3 = IIP3 + G stays near +12.7 dBm because the drain swing sets it. If OIP3 did not move, what IIP3 would a gain of 17.36 dB give?

Steps

Work in tools/labs/lab4 of the course files. Inside the container, run the commands as shown; from your own computer, put ../common/iic.sh in front of each one (Lab 0 shows how). Every step writes its netlist into out/, runs the simulator there, and prints what it measured.

  1. S-parameters and noise.

    cd /foss/designs/rfic-design/tools/labs/lab4
    python3 lna_lab.py sp                 # from your computer: ../common/iic.sh python3 lna_lab.py sp
    

    The script builds the netlist with lna_lines() of Chapter 9’s script, from the design values at its top, writes it as out/lna_tt27_sp.cir (tag = corner and temperature) and runs ngspice -b on it. Open the netlist. You will find every part of Figure 9.5: Lg with its series resistance Rlg for Q = 40, the pad capacitance and the two ESD diodes, the DC block XCin, the 1:16 bias mirror, M1 and M2 as 16 RF cells each (rfmode=1), the EM-fitted π models of \(L_s\) and \(L_d\), and XC1, XC2 of the output match. Its .control block runs an operating point (out/lna_tt27_op.txt), an sp analysis from 1 to 4 GHz with noise, written with wrs2p as out/lna_tt27.s2p and with wrdata as the noise-parameter table out/lna_tt27_noise.txt, and a second sp from 0.1 to 30 GHz for the stability check (out/lna_tt27_wide.s2p).

  2. Read the results with scikit-rf [30, 215]. lna_lab.py does this in load_sp() and analyse_sp(); do it once by hand to see each piece. Start python3 in the lab folder:

    import skrf as rf, numpy as np
    lna = rf.Network("out/lna_tt27.s2p")
    lna["2.44GHz"].s_db                   # |S11| |S12| / |S21| |S22| in dB
    from lna_lab import load_sp
    net, wide, nf_ngspice, op = load_sp("tt27", 27)
    k = np.argmin(abs(net.f - 2.44e9))
    net.nfmin_db[k], net.g_opt[k], net.rn[k]       # the four noise parameters, at 290 K
    10 * np.log10(net.nf(50.0))[k]                  # NF for a 50 ohm source, from the noise parameters
    wide.stability.min()                            # Rollett's K, 0.1-30 GHz
    

    load_sp() attaches the noise parameters with Network.set_noise_a(). ngspice refers its noise figures to the circuit temperature, 27 °C here, so the script first converts them to the 290 K of the definition, as Chapter 4 does: \(F_{290} = 1 + (F - 1)\,T/T_0\) for \(F\) and \(F_{min}\), and \(R_n\,T/T_0\). Compare net.nf(50.0) with ngspice’s own nf column after the same conversion (nf_ngspice above): scikit-rf recomputes the noise figure from \(F_{min}\), \(R_n\) and \(\Gamma_{opt}\) alone, so agreement shows that the four parameters describe the two-port’s noise completely.

  3. The noise circles. For a source reflection \(\Gamma_s\) every noise figure above \(NF_{min}\) is reached on a circle around \(\Gamma_{opt}\) (Chapter 6, Eq. (6.11)). net.nf_circle(nf) returns the circle for a noise figure nf in dB at every frequency; take the column of 2.44 GHz. The script draws four of them, including the one through the 50 Ω point, into out/lab4_sp_tt27.png (Figure L4.1b).

  4. Two-tone test in Xyce.

    python3 lna_lab.py iip3
    

    It writes out/lna_tt27_2tone.cir, Chapter 9’s harmonic-balance netlist: the same LNA (lna_lines() without the ports) between a 50 Ω source with two tones at 2.440 and 2.441 GHz and a 50 Ω load, .hb with five harmonics per tone, and .step over nine powers from −40 to −20 dBm per tone. Xyce [195] runs with the PSP plugin, and rfmode=1 works in Xyce (see the PDK note in Section 9.7). The script reads out/lna_tt27_2tone.cir.HB.FD.prn, refers the outputs to available powers and fits lines of slope 1 and 3 through the three lowest points, as the chapter does. It also prints the slopes it actually finds there and between −27.5 and −20 dBm.

  5. One corner. Rerun both steps with fast MOSFETs at 85 °C, the corner where Table 9.3 found the worst match and the lowest IIP3:

    python3 lna_lab.py sp iip3 --corner ff --temp 85
    

    Only the mos_tt word of the .lib line and the temperature change; the passives stay typical and the reference current stays 312.5 µA, as in Chapter 9.

  6. Everything at once. python3 lna_lab.py all runs steps 1, 4 and 5, stores the checkpoint numbers in lab_numbers.json and draws Figure L4.1 and Figure L4.2 into out/.

Two panels. (a) S21, S11 and S22 in decibels from 1 to 4 gigahertz with the 2.4 gigahertz band shaded: S21 peaks at 17.6 decibels at 2.44 gigahertz; S11 has a deep notch just below the band; S22 a notch near 2.47 gigahertz. On a second axis the noise figure falls from about 2.7 decibels at 1 gigahertz to 1.17 decibels in the band and rises to about 6 at 4 gigahertz; NFmin rises slowly from 0.6 to 1.3 decibels. (b) Noise circles on the Smith chart at 2.44 gigahertz: four nested circles around Gamma opt, a dot at about 0.46 on the positive real axis, for 0.93, 1.17, 1.33 and 1.83 decibels; the 1.17 decibel circle passes through the centre of the chart, marked 50 ohms.
Figure L4.1. The LNA is power-matched at 50 Ω, which costs 0.34 dB against its noise optimum. Lab 4, ngspice sp read with scikit-rf (tt, 27 °C; NF at 290 K). (a) S-parameters, NF for a 50 Ω source and NFmin; shaded: 2.400–2.4835 GHz. (b) Noise circles at 2.44 GHz from Network.nf_circle(): NFmin + 0.1, + 0.5 and + 1 dB, and the circle through the 50 Ω source.
Output power per tone against input power per tone from minus 40 to minus 20 dBm, for tt at 27 degrees and ff at 85 degrees: the fundamentals rise with slope 1 from about minus 22 dBm, the IM3 products with slope 3 from about minus 92 dBm, steepening above minus 30 dBm. Dotted extrapolations meet at IIP3 of minus 4.9 dBm for tt and minus 5.7 dBm for ff at 85 degrees.
Figure L4.2. The two-tone test gives IIP3 = −4.9 dBm at tt and −5.7 dBm at ff, 85 °C. Lab 4, Xyce harmonic balance, tones at 2.440 and 2.441 GHz; fundamental and IM3 (mean of the two products) per tone, with lines of slope 1 and 3 through the three lowest points.

Checkpoint

Table L4.1. Lab 4 results (SG13G2, tt, 27 °C unless stated; band 2.40–2.49 GHz on the 10 MHz grid; NF at 290 K). Tolerance: ±1 in the last digit shown, except where stated.

Quantity Lab 4 Chapter 9
Hand calculation: \(Q_L\), −10 dB bandwidth \(2f_0/3Q_L\) (±5 %) 4.54, 358 MHz 4.54
Hand calculation: NF from Eq. (9.9); NF at 85 °C from \(F - 1\) (±0.02 dB) 1.39 dB; 1.36 dB 1.39 dB
Hand calculation: IIP3 at ff, 85 °C if OIP3 stayed at +12.7 dBm −4.7 dBm  
Supply current, power; VG, VX 4.70 mA, 5.64 mW; 0.664, 0.498 V 4.70 mA, 5.64 mW; 0.664, 0.498 V
S21 at 2.44 GHz; band minimum 17.61 dB; 17.43 dB 17.61 dB; 17.43 dB (Tables 9.2, 9.4)
S11 at 2.44 GHz; band maximum −20.97 dB; −15.39 dB −20.97 dB; −15.39 dB
−10 dB return-loss band 2.230–2.609 GHz, 379 MHz 2.23–2.61 GHz, 379 MHz
S22 at 2.44 GHz, band maximum; S12 −19.48, −12.71 dB; −38.12 dB −19.48, −12.71 dB; −38.12 dB
NF at 2.44 GHz (scikit-rf, 50 Ω); band maximum 1.17 dB; 1.18 dB 1.17 dB; 1.18 dB
NF from ngspice’s nf column, converted to 290 K 1.17 dB  
NFmin, \(R_n\), \(\Gamma_{opt}\) at 2.44 GHz 0.83 dB, 12.5 Ω, 0.459∠−2.4° NFmin 0.83 dB
K minimum (0.1–30 GHz), at; μ minimum above 1 GHz 5.23 at 2.51 GHz; 1.02 5.23 at 2.51 GHz; 1.02
Two-tone: IIP3, OIP3 (±0.1 dB) −4.9 dBm, +12.7 dBm −4.9 dBm, +12.7 dBm
Two-tone: gain of the slope-1 fit; slopes at −40 to −35 dBm 17.61 dB; 1.00 and 3.04 17.62 dB (one tone at −40 dBm); 1.00 and 3.04
IM3 slope between −27.5 and −20 dBm (±0.05) 3.55 3.57 (Section 9.7)
ff, 85 °C: current; S21 band 5.01 mA; 17.12–17.37 dB 5.01 mA; 17.12–17.37 dB (Table 9.3)
ff, 85 °C: NF band maximum; S11 band maximum 1.38 dB; −13.7 dB 1.38 dB; −13.7 dB
ff, 85 °C: IIP3, OIP3 (±0.1 dB); NFmin; K minimum −5.7 dBm, +11.7 dBm; 1.00 dB; 4.49 −5.7 dBm

Every row that Chapter 9 also reports agrees with it, the IIP3 and the slopes to the digits shown. Only the steep IM3 slope between −27.5 and −20 dBm, where the fifth-order products take over (Section 9.7), comes out 0.02 below the chapter’s; the chapter’s sweep continued to −5 dBm, and that last digit of a difference quotient is within the harmonic balance’s convergence tolerance. The match bandwidth of the hand calculation is 6 % narrower than the simulation: \(Q_L\) counts only \(L_g + L_s\), while the pad capacitance and \(C_{gd}\) change how fast the input reactance moves with frequency. At ff, 85 °C the hand calculation’s −4.7 dBm misses by 1 dB: the output intercept did not stay constant (Questions).

Questions

  • \(\Gamma_{opt}\) is 0.459 at almost zero angle, so the optimum source is about 135 Ω, real. Why does the inductively degenerated stage not put its noise optimum at 50 Ω, and which element of the design would you change to move it there? What would that do to the input match?
  • Hand calculation 3 predicted 1.36 dB at 85 °C from the temperature alone, and the simulation gives 1.37 dB at 2.44 GHz at ff. Which effects of the fast corner push the noise figure up and which push it down?
  • At ff, 85 °C the gain fell by 0.25 dB but the IIP3 by 0.8 dB, so OIP3 dropped by about 1 dB. Look at \(V_X\) in the printout (0.498 V at tt, 0.564 V at ff, 85 °C) and explain with the drain-swing argument of Section 9.5.
  • Rollett’s K is above 4 everywhere, yet μ approaches 1.02 above 1 GHz. Which S-parameter makes μ so close to one, and why does that not mean the amplifier is close to oscillating?

Extension

Replace the Q = 40 chip inductor by an on-chip spiral: set qlg=11.3 in LNA at the top of lna_lab.py (the Q of Chapter 7’s largest spiral) and rerun python3 lna_lab.py sp. Compare NF and NFmin with Figure 9.7: which of the two moves, and why does the other stay put? Then try the design trade of Table 9.2: change ld to 3.0 (nH) without rematching and run sp iip3. How much of the table’s 0.9 dB gain increase and 1.4 dB IIP3 loss do you get, and what does the mismatch at the output cost you?


Back to top

SkillSurf · RFIC Design with Open-Source Tools, draft v0.1 (November 2026). Every source is listed on the References page.

This course was written with the help of AI tools and may contain mistakes. If you find one, please tell us through SkillSurf or email skillsurfedu@gmail.com, and we will correct it.

This site uses Just the Docs, a documentation theme for Jekyll.