A K/Ka-Band Frequency Reconfigurable GaN LNA for Multi-Band Communication Applications

I.   Introduction

Advancements in satellite communication (SATCOM) technology have significantly improved latency and capacity for global internet access compared to conventional long-distance optical fiber networks or microwave relays [1]. The integration of 5G networks with non-terrestrial access presents an opportunity to improve communication quality and coverage in case of natural disasters and other emergencies [2]. A viable and cost-effective approach to incorporate satellite communications into 5G networks is reusing certain circuit components of existing 5G technologies [3].

The low noise amplifier (LNA) plays a crucial role in the receive chain of front-end circuits, serving as the first active component after the antenna. It significantly impacts the overall sensitivity and dynamic ranges of the system. Therefore, the utilization of multi-band LNAs in non-terrestrial networks (NTN) that integrate 5G technologies, as well as in systems requiring coverage across multiple frequency bands, offers potential cost-effectiveness.

Although broadband LNAs [4]–[7] have good frequency coverage, optimizing both gain and noise figure (NF) over the entire wideband range presents a significant challenge, and additional filters are required since unwanted signals might degrade the receiver sensitivity in some cases [8]. Alternatively, frequency reconfigurable and concurrent multi-band approaches are being extensively studied to optimizing the performance at each operating bands. Concurrent multi-band LNAs can provide stable and high stopband rejection [9]–[13], whereas multi-band passives are significantly intricate and challenging to implement. Frequency reconfigurable LNAs offer the ability to achieve performance optimization within each frequency band, albeit requiring additional tuning (e.g., diodes) or switchable (e.g., FET switches) devices for operation [14]–[21]. In [15] and [16], switches are employed in input matching networks to electronically reconfigure the frequency band. However, the parasitic resistance of the switch device has negative impact on the input matching loss and degrade NF performance, especially in high frequency operation. Wideband input matching network and magnetically tuned variable inductor are adopted in [21] for load switching, but the size of reconfigurable elements and matching network in this architecture is relatively large.

In this paper, a K/Ka-band frequency reconfigurable LNA (FR-LNA) is presented with a reconfigurable inter-stage and output matching network to be applied in a frequency reconfigurable hybrid phased array RX chain, as shown in Figure 1. The gain budget of the building blocks shows that the CMOS LNA needs to provide a gain of 25 dB. Two additional driver amplifiers, DA₁ and DA₂, are employed to compensate for the passive losses induced by the passive phase shifter and attenuator. The proposed FR-LNA can be reconfigured to operate at two frequency bands. Together with the wideband CMOS IC, the overall receiver can be applied for various scenarios, i.e., 5G NTN. A switch-based reconfigurable resonator (SBRR) is employed in both the inter-stage and output stage matching network of the FR-LNA, enabling frequency band switching. The utilization of reconfigurable components is straightforward to implement and features a simple structure, thus resulting in a compact chip size.

Figure  1.  RX block diagram of the hybrid phased array system.

DownLoad: Full-Size Img PowerPoint

II.   Analysis and Design of Proposed Reconfigurable Resonator

Figure 2 shows the design concept of the proposed FR-LNA, which consists of three amplification stages with reconfigurable inter-stage and output stages. The LNA is implemented in a commercial 0.15-$\mu$m GaN HEMT process with an $f_T$ of approximately 34.5 GHz.

Figure  2.  Design concept of the proposed frequency reconfigurable LNA.

DownLoad: Full-Size Img PowerPoint

1   Analysis of the Switch HEMT

As depicted in Figure 3, the equivalent circuit of a GaN HEMT switch at ON- and OFF-state can be modeled by lumped series RLC circuit [22]–[24]. In contrast to the conventional circuit model for switch devices in the low-frequency band, a series inductor is included to model the parasitic effects of interconnect metals at high frequencies, particularly in millimeter-wave bands. This addition is crucial as it can significantly impact the RF performance of the switch in terms of insertion loss (IL) and isolation (ISO), making it imperative not to overlook. Nevertheless, when a switch HEMT is employed to achieve switchable or reconfigurable reactance by integrating capacitors and inductors, the parasitic parameters of the switch can be harnessed rather than eliminated through additional circuitry. This strategy serves to simplify the overall circuit complexity.

Figure  3.  Layout and equivalent circuit of the switch HEMT.

DownLoad: Full-Size Img PowerPoint

Figure 4 illustrates the RLC circuit parameters of the HEMT switch extracted at 30 GHz for different values of the unit gate width (UGW) and the number of fingers. The gate bias voltage is set to 0 V and $-10$ V for turning on and turning off, respectively, using a 5-k$\Omega$ bias resistor. The OFF-state capacitance ($C_{OFF}$) of the HEMT switch increases linearly with the gate width, whereas the ON-state resistance ($R_{ON}$) exhibits an opposite trend as the gate width increases. The OFF-state resistance ($R_{OFF}$) can be considered negligible. The amount of parasitic inductance ($L_d$) remains relatively consistent at a given NOF. This consistency can be observed in Figure 2, where the primary contributor to parasitic inductance is the metal interconnection between the electrode’s PAD and the surrounding device.

Figure  4.  RLC circuit parameters of the switch HEMT versus various gate width (UGW) and number of finger (NOF).

DownLoad: Full-Size Img PowerPoint

The size of this interconnecting metal remains largely uniform when the NOF is unchanged. The figure of merit (FOM) can be derived from the product of $R_{ON}$ $\times$ $C_{OFF}$, with a typical value ranging from 400 to 700 fs, as shown in Figure 5. In addition, the optimal circuit parameter region of the switch is highlighted in gray.

Figure  5.  Figure of merit (FOM) of the switch HEMT.

DownLoad: Full-Size Img PowerPoint

2   Switch-based Reconfigurable Resonator

Switch HEMT can manipulate the reactance of capac-itors or inductors by leveraging its distinct impedance in the ON- and OFF-states. This capability allows for the modification of the resonant frequency of LC resonators. The schematic of proposed SBRR network is depicted in Figure 6, where the HEMT switch $M_s$ is connected in series with $C_s$ and $TL_{se}$ to create transmission zeros, en-suring rejection of unwanted signals in the lower out-of-band region while exhibiting inductance within the operating frequency band. The series resonant frequency in the ON-state ($f_{TZ1}$) and OFF-state ($f_{TZ2}$) are given by (1) and (2), respectively,

Figure  6.  Schematic and equivalent circuit of the proposed switch-based reconfigurable resonator (SBRR).

DownLoad: Full-Size Img PowerPoint

where $L_{eq}$ is the sum of the parasitic inductance $L_d$ of $M_s$ and the extracted inductance $L_{se}$ of transmission line $TL_{se}$. The input admittance $Y_{in}$ of the SBRR can be calculated as (3), where $R_{sw}$ is the parasitc resistance of switch (i.e., $R_{ON}$ and $R_{OFF}$), and $f_{TZ}$ is the transmission zero (i.e., $f_{TZ1}$ and $f_{TZ2}$).

Accordingly, the resonant frequency of the entire SBRR in the ON-state ($f_L$) and OFF-state (($f_H$)) could also be deduced as (4) and (5).

Figure 7 illustrates the simulated responses of the proposed SBRR at both states with a 5$\times$100 $\mu$m HEMT. Moreover, extra IL can be induced by the parasitic resistances, with $L_d$=163.8 pH, $C_{OFF}$=105 fF, the IL of SBRR under different sets of $R_{ON}$ and $R_{OFF}$ is shown in Figure 8(a) and Figure 8(b) with 50 $\Omega$ terminated. As can be seen, a low $R_{ON}$ is crucial to reduce the IL. As the compromise of the $R_{ON}$ and $C_{OFF}$ discussed in Section 2, a switch HEMT size of 5$\times$100 $\mu$m with a FOM of 414 fs is chosen in this design. The values of $R_{ON}$, $R_{OFF}$, $C_{OFF}$ and $L_d$ are extracted to be 3.94 $\Omega$, 0.116 $\Omega$, 105 fF and 163.8 pH, respectively. Subsequently, the remaining components of the resonator are tuned to achieve the desired resonant frequency according to equation (1) $-$ (5). By applying gate bias voltages of 0 V and $-10$ V, the resonant frequency reconfigured from 21 GHz to 30 GHz. The simulated out-of-band rejection in the higher frequency band is approximately $-28$ dBc around 25 GHz, which surpasses the rejection seen around 10 GHz in the lower frequency band. This is attributed to the low resistance when the switch is biased in the OFF-state, leading to a higher Q value of the resonator.

Figure  7.  Simulated responses of the proposed SBRR at both states with a 5$\times$100 $\mu$m HEMT.

DownLoad: Full-Size Img PowerPoint

Figure  8.  (a) Simulated IL in ON-state with different $R_{ON}$ and (b) simulated IL in OFF-state with different $R_{OFF}$.

DownLoad: Full-Size Img PowerPoint

III.   Circuit Design and Implementation

1   Input stage impedance matching network

To obtain optimal NF and maximum available gain (MAG), the size of each HEMT is chosen to be 2$\times$50 $\mu$m. In Figure 9, the MAG and noise figure minimum ($NF_{min}$) characteristics of a 2$\times$50 $\mu$m common-source HEMT are depicted. In consideration of the MAG and $NF_{min}$, the quiescent current has been set at 100 mA/mm with a 5 V drain supply. A source-degenerated inductor of 134 pH is connected in series with the HEMT. Subsequently, the noise contour for the input stage at 21 GHz and 30 GHz is de-picted in Figure 10(a). The impedance of $NF_{min}$ at 21 GHz and 30 GHz is 22.872 + j45.7 $\Omega$ and 15.9 + j16.8 $\Omega$, re-spectively, with $NF_{min}$ values of 0.57 dB and 1 dB. However, achieving optimal noise impedance matching simultaneously is challenging. Therefore, a simple high-pass matching network is selected to minimize IL. The impedance trajectory of the input matching network (IMN) is positioned at the plane of symmetry of the noise circle at 21 GHz and 30 GHz. By sacrificing a small amount of NF at the edges of the frequency band, the optimal NF is achieved at the mid-frequency, as shown in Figure 10(b). Nevertheless, the 1.07 dB and 1.38 dB NF at the two target frequencies of the input stage are still acceptable.

Figure  9.  Simulated results of the 2$\times$50 $\mu$m CPW HEMT. (a) Maximum available gain (MAG) under 5 V drain bias, (b) Maximum available gain (MAG) under 10 V drain bias and (c) Minimum noise figure ($NF_{min}$).

DownLoad: Full-Size Img PowerPoint

Figure  10.  Simulated (a) noise contour with IMN impedance trajectory and (b) NF and $NF_{min}$ of the input stage with a 50 $\Omega$ load.

DownLoad: Full-Size Img PowerPoint

2   Inter-stage impedance matching network

After determining the IMN, the output impedance of the first stage can be obtained. The value of the degenerated inductor in the second stage is 88.4 pH, which is smaller than that of the first stage to improve the overall gain of the LNA. The first inter-stage matching network (ISMN) is initially designed to transform the output impedance ($Z_{d1}$) of the pre-stage to the optimal source impedance of the post-stage ($Z_{s,opt2}$) at 21 and 30 GHz simultaneously.

To this end, a second-order bandpass network is considered with resistive load termination. Then, $Z_{d1}$ is modeled as a parallel RC network, with the capacitor $C_{d1}$ integrated into the bandpass network. The optimal source impedance for the post-stage is represented by a series RC model. Additionally, an extra inductor ($L_3$) is connected in parallel with the gate of the post-stage HEMT and in series with $C_s$. This configuration forms a high-pass network, transforming the impedance ($Z_{im}$) into the optimal impedance ($Z_{s,opt2}$). The ISMN1 is illustrated in Figure 11(a). To facilitate the implementation of component values on-chip, a Norton transformation is applied to the inductors in the band-pass network, resulting in a T-type network that simplifies the bias network. The simulated $|S_{21}|$ of the ISMN1 is depicted in Figure 11(b), along with the source impedance trajectory of the second stage. Figure 12(a) shows the proposed and the switch-based reconfigurable ISMN2, both terminated with the same impedance for performance comparison. The proposed ISMN2 comprises a second-order bandpass network and a high-order high-pass network for impedance transformation. This arrangement obviates the need for impedance scaling using Norton transformation, as the high-order high-pass network facilitates an effective transition from a high-impedance to a low-impedance region. The proposed SBRR is integrated into the bandpass network, as previously discussed. The transmission zero shifts with the gate control voltage of the switch HEMT, thereby reconfiguring the passband. It is evident that the operation frequency transitions to a high-band (HB) with the switch HEMT turned off, achieving significant rejection in the low-band (LB). The simulated $|S_{21}|$ of the proposed and switch-based ISMN2 is illustrated in Figure 12(b). The insertion loss of the switch-based solution outperforms that of the proposed SBRR method when the same switch device is used. However, regarding out-of-band rejection, the proposed SBRR exhibits superior rejection at lower frequencies in HB mode, attributed to the detrimental parasitic effects of the embedded switch device.

Figure  11.  (a) First inter-stage matching network (ISMN1) and (b) simulated insertion loss of ISMN1 with the source impedance trajectory of the second stage.

DownLoad: Full-Size Img PowerPoint

Figure  12.  (a) Schematic of the proposed and switch-based second inter-stage matching network (ISMN2). (b) Simulated $|S_{21}|$ of the proposed and switch-based ISMN2.

DownLoad: Full-Size Img PowerPoint

3   Output stage impedance matching network

A reconfigurable output matching network (OMN) is designed with the proposed SBRR and is compared with a switch-based solution, as depicted in Figure 13(a). Due to the slightly higher $Q$ value (8.7) at the lower band, a transformed second-order bandpass network, complemented by additional matching elements, has been adopted. The SBRR is integrated into the bandpass network, analogous to ISMN2. The parallel RC network, consisting of $R_{d3}$ and $C_{d3}$, can be derived from the output impedance of the third stage. The simulated $|S_{21}|$ of the proposed and switch-based OMN is illustrated in Figure 13(b). Similar to ISMN2, the proposed OMN exhibits superior rejection at lower frequencies in HB mode compared to the switch-based solution. The IL in LB mode is slightly greater than in HB mode, primarily attributed to the $R_{ON}$ of the switch, as discussed in Section II.

Figure  13.  (a) Schematic of the proposed and switch-based output matching network (OMN). (b) Simulated $|S_{21}|$ of the proposed and switch-based OMN.

DownLoad: Full-Size Img PowerPoint

4   Frequency Reconfigurable LNA Design

The complete schematic of the proposed FR-LNA is illustrated in Figure 14. All the inductors in the above impedance matching network are realized using transmission lines (TLs). The width of each TL is 20 $\mu$m. $TL_3$, $TL_8$ and $TL_{13}$ are utilized to realize inductive source degeneration, which facilitates impedance matching and ensure stable conditions.

Figure  14.  Complete schematic of the proposed FR-LNA.

DownLoad: Full-Size Img PowerPoint

The proposed SBRR is implemented in both the second inter-stage and output matching network, capacitors $C_{s1}$, $C_{s2}$, $C_{p1}$, and $C_{p2}$ are finely tuned to satisfy the matching condition. The gate and drain bias TLs are involved in impedance matching networks for a compact layout. The simulated gain of individual stages and the total LNA stages in the LB and HB modes are plotted in Figure 15(a) and Figure 15(b), respectively. Consequently, the operating frequency band of the proposed FR-LNA can be reconfigured by applying various control voltages. Moreover, transmission zeros are clearly observed in the lower out-of-band.

Figure  15.  Simulated gain of the proposed FR-LNA in (a) LB mode and (b) HB mode.

DownLoad: Full-Size Img PowerPoint

In [15], the switched input match is adopted to reconfigure the operating frequency band, but due to the limitations of switch FOM, it leads to increased IL, resulting in NF degradation. Moreover, the switched degenerated inductor further degrades NF. In [16], similarly to [15], a substrate-shielded inductor (SSI), utilizing a switch to reconfigure the reactance, is employed in the input matching network, which also contributes to NF degradation. The proposed SBRR in this design is integrated in the ISMN and OMN. Compared to the reconfigurable methods in [15] and [16], the proposed SBRR has minimal impact on the NF after the operating band is reconfigured, with only some reduction in gain resulting from the additional switch loss.

IV.   Experimental Results

The micrograph of fabricated chip is shown in Figure 16. On-wafer measurements are conducted using GSG probes for RF input/output and DC probes for bias and power supply. The RF characterization setup is calibrated using the short-open-load-through (SOLT) method on a Keysight PNA-X analyzer, while power calibration is performed using a power meter.

Figure  16.  Micrograph of the proposed FR-LNA.

DownLoad: Full-Size Img PowerPoint

The FR-LNA is biased at $V_D$ = 5 V, $V_g$ = $-1.5$ V. The measured and simulated S-parameters and NF of the FR-LNA are presented in Figure 17 and Figure 18. The LB mode at 21 GHz achieves a gain of 13.9 dB with a NF of 2.4 dB. In the HB mode at 30 GHz, the FR-LNA obtains a gain of 8.9 dB with a NF of 3.2 dB. The out-of-band rejection exceeds 63 dBc at 16 GHz and 18 dBc at 20 GHz for LB and HB modes, respectively. It features excellent rejection against unwanted low-frequency signals, particularly for X-band satellite communications.

Figure  17.  Measured (a) S-parameters, (b) NF, and (c) $OP_{1\mathrm{dB}}$ of the FR-LNA in LB mode.

DownLoad: Full-Size Img PowerPoint

Figure  18.  Measured (a) S-parameters, (b) NF, and (c) $OP_{1\mathrm{dB}}$ of the FR-LNA in HB mode.

DownLoad: Full-Size Img PowerPoint

Furthermore, the output 1-dB compression point ($OP_{1\mathrm{dB}}$) of 5.4 and 6.7 dBm are achieved for the LB and HB modes, as illustrated in Figure 17(c) and Figure 18(c). In addition, the input-referred intercept point (IIP3) of the LNA is measured using a two-tone signal with 100 MHz tone spacing, as shown in Figure 19. The measured IIP3 is 2.56 dBm and 14.77 dBm at 21 GHz and 30 GHz, respectively, implying a good dynamic range for the receivers. The total power consumption ($P_{DC}$) of the FR-LNA is approximately 210 mW.

Figure  19.  Measured fundamental output power and power of third-order intermodulation distortion products (IM3) under two-tone signal excitation with a 100 MHz tone spacing.

DownLoad: Full-Size Img PowerPoint

The measured and simulated results reveal a close agreement. However, in LB mode, there are some discrepancies between simulation and measured S-parameters. As discussed in Section II, the parasitic parameters of switch device affect the performance of the SBRR, thus the overall performance of the LNA. The variation in $f_L$ and $f_H$ is may attributed to deviations in $L_d$ and $C_{OFF}$ between the model and the actual device. After $C_s$ and $C_p$ are determined, only the parasitic parameters of switch device affect the tow resonant frequencies, $f_L$ and $f_H$. Specifically, in HB mode, the switch HEMT is turned off. If we assume that the value of $L_d$ becomes smaller and $C_{OFF}$ becomes larger than simulated values, these opposing variations may cancel each other out, causing the measured S-parameters to appear closer to the simulated results. Figure 20 shows the simulated $|S_{21}|$ of the FR-LNA with various $L_d$ and $C_{OFF}$ under HB and LB modes. As shown in Figure 20(a), the response of the LNA exhibits slight variation and remains close to the measured results. However, when it is reconfigured in LB mode (see Figure 20(b)), as the switch turned on, varying $L_d$ prominently shifts the frequency response. Therefore, the discrepancies between simulated and measured results in LB mode become more pronounced. Besides, the discrepancy in the magnitude of $S_{21}$ may be caused by a larger $R_{ON}$ than the PDK model (see Figure 20(c)).

Figure  20.  Simulated $|S_{21}|$ with different $L_d$ and $C_{OFF}$. (a) Both $L_d$ and $C_{OFF}$ are varied in HB mode. (b) Only $L_d$varied in LB mode, with $R_{ON}$ = 5.91 $\Omega$. (c) Only $L_d$ varied in LB mode, with $R_{ON}$ = 11 $\Omega$.

DownLoad: Full-Size Img PowerPoint

Table 1 summarizes the performance of the proposed FR-LNA and compares with prior works. Since few III-V reconfigurable LNAs demonstrated at K/Ka-band, the proposed FR-LNA is also compared with some wideband designs. Two figure-of-merits (FOMs) are calculated to evaluate the performance, including bandwidth, gain, NF, $IP_{1\mathrm{dB}}$, and $P_{DC}$, as described in [25]:

Table  1.  Performance Comparison with State-of-the-arts

Ref.	[4]	[5]	[8]	[16]	[20]	This work
Tech.	0.15-$\mu$m GaN	0.15-$\mu$m GaN	0.1-$\mu$m GaN	65-nm CMOS	0.1-$\mu$m GaAs	0.15-$\boldsymbol{\mu}$m GaN
Topology	wideband	wideband	wideband	reconfigurable	switchable	reconfigurable
Freq. (GHz)	28	36	37	28/39	28/39	21/30
BW (GHz)	6	2	38	9/16	4.6/3	4.15/3
Gain (dB)	25a	19.7	16–21.5	17.2/12.1	25.1/27.7	13.9/8.9
NF (dB)	2.4–2.9	3.7	2.2–4.4	2.8/4.4	2.4/2.8	2.4/3.2
IP1dB (dBm)	−0.9b	4.0b	−5 to -0.5	−13.6/−11	−18.3/−17	−7.4b/−1.2b
IIP3(dBm)	7.2a	N.A.	N.A.	−4/−2	−8.8/−8.5	2.56/14.77
OP1dB (dBm)	19.1	22.7	10–20	2.6b/0.1b	6.8b/10.7b	5.5/6.7
PDC (mW)	300	3000	1400	28.5	74	210
Area (mm2)	3	5.2	4.8	0.157	1.6	2.625
FOMD (dB)	−10.6	−19.67	−16.98	−15.22/−18.7	−14.06/−13.16	−16.06/−17.4
FOMDBW (dB)	−8	−18.67	−11.71	−12/−14.69	−11.85/−11.57	−14/−15.8
[a] Graphically estimated.　　[b] Calculated with OP1dB = IP1dB + Gain − 1.　　[c] Simulated results.

 | Show Table

DownLoad: CSV

where $L$ represents the gate length in $\mu$m, $f_0$ is the center frequency in GHz, $\Delta f$ is the bandwidth in GHz, $IP_{1\mathrm{dB}}$ is in mW, $G$ is the linear value of gain, and $F$ is the noise factor. The proposed GaN FR-LNA showcases reconfigurable frequency capability and demonstrates comparable FOMs to its CMOS and GaAs counterparts, demonstrating its potential for multi-band communication systems.