Low-Noise Balun Aids Phased-Array Radars

The phase shifter (Fig. 2) is an important part of the T/R module, with a low-noise active balun (LNAB) comprising the first block of the phase shifter following the input signal port. The balun provides impedance matching (at 50 Ω), low-noise amplification, and single-to-differential conversion to provide a quadrature signal to a quadrature-all-pass-filter (QAF) network at X- and Ku-band frequencies. The LNAB circuit is divided into two circuits. The first stage is an active balun which converts single-ended input signals to quadrature signals. Critical performance parameters include high, flat gain, and low noise figure, with good input-output impedance matching.

A number of structures and approaches have been proposed to achieve these goals for the active balun, including a two-stage cascode configuration with passive inductors.^7,8 In one of these designs,⁷ the LNAB provides S₂₁ (gain) of 10 dB, noise figure of 5.4 dB, and power dissipation of 66 mW at 14 to 15 GHz. In the other,⁸ the simulated S₂₁ performance is 10.2 dB, the noise figure is 4.6 dB, and the power consumption is 87.5 mW at 13 to 15 GHz. Unfortunately, passive inductors contribute to higher loss and larger circuit sizes, and two-stage cascode active baluns have been proposed with active inductors, although these suffered in linearity.^4,6 One design offered simulated input third-order intercept point (IIP3) of -30 dBm at 12 GHz. In another design, differential passive inductors were used in an active balun to decrease size and improve circuit performance.⁹

In pursuit of improved LNAB performance at X/Ku-band frequencies, a two-stage design was developed where the first stage is a single-stage low-noise amplifier (LNA) with cascode configuration. It employs resistive shunt-shunt feedback and peaking inductor for enhanced bandwidth.^10,11 The second stage is an active balun with output emitter-follower to convert single-ended input signals to quadrature output signals and provide wideband output impedance matching. The two-stage LNAB was simulated and optimized using the Spectre RF computer-aided-engineering (CAE) software from Cadence Design Systems based on the use of a 0.18-μm silicon-germanium (SiGe) BiCMOS semiconductor technology.

In the proposed LNAB design (Fig. 3), wideband input impedance matching was achieved with a shunt-shunt feedback resistor (R_f) in conjunction with a preceding inductive-capacitive (LC) passive network; flat, high-frequency gain was achieved with the aid of the peaking inductor (L_d1 technique. The second stage is a differential cascode pair with differential peaking inductor (L_d2) to provide the single-ended-to-differential signal conversion, additional common-mode rejection, and wideband output-impedance matching. Use of the differential inductors, L_d2 and L₃, save chip space and improve linearity.

Resistor R_FB is often chosen to be equal to R_s, which is the resistance of the voltage source and which is usually 50 Ω for RF systems. At low frequencies, input impedance z_in is dominated by R_FB. Traditionally, RF is chosen as close to 50 Ω as possible to ensure that the input impedance of the amplifier is well matched before the capacitive effect (jwc_{p, in}) becomes appreciable at higher frequencies. In general, for RF circuits, S₁₁ should be less than -10 dB over the frequency band of interest, as portrayed by Eq. 3:

20log|S₁₁| = 20log[(r – 1)² + (ωR_FBc_{p, in})²]/[(r + 1)² + (ωR_FBc_{p, in})²]^0.5 ≤ -10 dB   (3)

where r = R_FB/R_s and the corresponding input-matching bandwidth, f_10dB (which equals ω_10dB/2π) and the range of r, can be derived by Eq. 4:

f_10dB = (-9r² + 22r - 9r)^0.5/(6πR_FBc_{p, in})   (4)

From Eq. 4, it can be seen that the input matching bandwidth is directly affected by r, so a suitable value of r was chosen to determine the bandwidth of the LNA.

Figure 4(b) shows the resistive shunt-shunt feedback amplifier with wideband input-impedance matching network and emitter degeneration inductor. The input impedance is given by Eq. 5:

Z_in = {(1 + sc_{p, in}R_FB)[ sc_{p, in}(L_g + L_s) + g_mL_s] + R_FB c_{p, in}}/[(1 + sc_{p, in}R_FB) c_{p, in}]   (5)

Assuming that R_FB = Z₀, frequencies corresponding to Z_in = Z₀—i. e., a virtually perfect input impedance match—can be derived from Eq. 6:

s²c_{p, in}R_FB(L_g + L_s + sc_{p, in}(L_g + L_s + R_FBg_mL_s - c_{p, in}R_FB²) + g_mL_s = 0   (6)

There are two frequencies at which a perfect match can be obtained, and the two frequencies are directly affected by shunt-shunt resistor R_f, emitter degeneration inductor, L_s, and base input inductor, L_g. For wideband input impedance matching (S₁₁ < -10 dB) for X- and Ku-band frequencies, the active device (Q₁) emitter area is 0.15 × 20 μm, and values for R_f = 950 Ω, L_s = 0.091 nH, and L_g = 0.17 nH were chosen. These component choices set the two pole frequencies at 6.95 and 14.87 GHz, respectively.

The shunt peaking inductor architecture represents an effective methods for improving high frequency gain flatness. It can compensate for decreasing gain with increasing frequency, to achieve flat gain with frequency. The peaking inductor technique also helps reduce the noise figure as well. To analysis the effect of the shunt peaking inductor on gain, Figs. 5(a) and (b) depict a typical resistive shunt-shunt feedback amplifier with peaking inductor and its small-signal equivalent circuit, respectively. From Fig. 5(b), the small-signal gain can be expressed as Eq. 7:

A_V = s(V_out/V_in) = [(R_f – g_m)RR_f²]/[R_f(1 + R_ssc_{p, in} + R_s]

[R_f(1 + Rsc_cs) + R] - RR_sR_f(R_f – g_m   (7)

where

R = R_d||SL_d and as the frequency increases, the value of SL_d increases, making R increase. Increasing R improves the high-frequency gain to help achieve flat gain.¹¹ To achieve adequate gain, a +3.3-VDC supply provides 3.03 mA current, using an emitter area of 0.15 × 20 μm. Shunt peaking, which also helps flatten the gain response, relies on R_d and L_d. For gain flatness with less than 1-dB deviation, values of 650 Ω for R_d and 0.157 nH for L_d were chosen.

Figure 6 shows the small-signal model of a purely resistive-feedback LNA including all significant sources of noise. Noise contributions from r_b are included, but the effects of r_b are excluded from the impedance calculations to reduce complexity. The use of emitter degeneration inductor L_s to replace the degenerative resistor can reduce the noise figure effectively, and the shunt peaking inductor L_d can also suppress noise generation.

From Fig. 6, the equivalent input referred noise can be expressed as Eq. 8:

v²_{n, in}/Δf = 4kT(R_s + R_lg + r_b + [(4kT/R_FB) + 2qI_B • (R_s||R_FB + R_lg) + 2qI_BR_ls² + (4kT/(R_d||R_ld) + 2qI_C)/g_m′²| 1 + jω(R_s||R_FB + R_lg)c′_{p, in}²|   (8)

where R_lg, R_ls, and R_ld are the equivalent parasitic resistances of inductors L_g, L_s, and L_d, respectively. Also, g′_m= g_m/(1 + g_mR_ls) and c′_{p, in} = c_{p, in}/(1 + g_mR_ls. From Eq. 8, the value of R_ls is the quantity 2qI_BR_ls² is much less than the value of a standard resistor, so the noise figure is effectively reduced by emitter degeneration inductor L_s.

The second stage of the LNAB is the active balun architecture, which converts single-ended input signals to differential output signals. Those signals drive the QAF network, so that good (50 Ω) matching is essential at the output port. Figure 7 shows the schematic diagram of the active balun stage. It is comprised of a differential cascode configuration with output emitter-follower pairs. For good impedance matching of the differential cascode pairs and good current density, the same emitter area of devices Q₃ through Q₆ was chosen for this, with area of 0.15 × 20 μm. The active balun has total current consumption of 9.37 mA from a +3.3-VDC supply.

The tail current of the differential cascode pairs is generated by m₀ with a W/L ratio of 57.3/0.18 μm. Differential inductor L₃ is used as an emitter degeneration inductor of the differential cascode pair to improve LNAB linearity and reduce passive inductance and chip size. The value of L₃ is chosen at 0.11 nH in an effort to maintain gain. The output load of the active balun is a differential inductor, L_d2, which helped reduce chip area, and resistor, R_d2. For good height and gain flatness, values of 0.495 nH for _d2 and 1.08 kΩ for R_d2 were chosen. Good wideband output impedance matching to 50 Ω was achieved by using the output emitter-follower pairs with the tail current of 660 μA. An emitter region area equivalent to that of Q₇ – Q₈ was used, of 0.15 × 4.8 μm.

The LNAB was simulated and optimized using the Spectre RF computer-aided-engineering (CAE) software from Cadence Design Systems based on a 0.18-μm SiGe BiCMOS technology. Impedance matching in the first stage was achieved by shunt-shunt feedback resistor in conjunction with the LC passive matching network. Component values for the shunt-shunt resistor R_f and inductors L_g and L_s in the LC passive network were adjusted with the help of a Smith chart and parameter sweeping tools. A wideband range of input impedance-matching values with two poles (6.95 and 14.87 GHz) was realized for X- and Ku-band frequencies, with input reflection coefficient of better than -10 dB.

In the second stage of the LNAB, the same emitter region area of devices Q₇ and Q₈ was chosen to ensure good impedance matching of the differential configuration. The values of differential inductor L_d2 and load resistor R_d2 were adjusted for flat gain, low noise figure, and small chip area. The small chip size and good linearity were also achieved by adjusting the value of differential emitter degeneration inductor (L₃) and the length-to-width ratio of the MOS (M₀). Finally, the wideband output impedance-matching was achieved by the emitter-follower pair, and choosing a suitable emitting region area of Q₇ and Q₈ to make sure that the output reflection coefficient is less than -10 dB at X/Ku-band frequencies.

Simulated results of S-parameters, noise figure, output power at 1-dB compression (P1dB), and input third-order intercept point (IIP3) for the LNAB are shown in Figs. 9 through 12. The solid line and dotted line in Figs. 9 and 10 are for simulations without and with parasitic extraction, respectively. Figure 9(a) shows simulated S₁₁ without parasitic extraction, which is better than -10 dB across the full bandwidth. Post-simulation S₁₁ is less than -9 dB from 8.04 to 12.5 GHz. The good S₁₁ performance can be linked to the resistive shunt feedback in conjunction with the LC passive matching network.

The measured and simulated data are generally consistent for the LNAB. Pre-simulation of S₂₁ revealed 17.83 ± 0.39 dB over X/Ku-band frequencies. Post-simulation of S₂₁ ranges from 16.90 to 18.52 [Fig. 9(b)]. The LNAB’s S₂₂ characteristics are all better than -10 dB, while S₁₂, at better than -90 dB, shows the good isolation achieved in the design. As the table shows, the performance of the LNAB compares favorably with performance levels cited in recent work.^{4, 7-9}

Wang Wei, Professor

Zhong Wu, Engineer

Xu Wei, Engineer

Mu Mao, Engineer

Yuan Jun, Lecturer

College of Electronics Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, People's Republic of China.