A Ka-Band SATCOM Transceiver in 65-nm CMOS With High-Linearity TX and Dual-Channel Wide-Dynamic-Range RX for Terrestrial Terminal

In this article, a $Ka$ -band satellite communication (SATCOM) transceiver is first presented using a standard CMOS technology. The proposed $Ka$ -band SATCOM transceiver consists of a high-linearity transmitter (TX) and dual-channel receiver (RX); both TX and RX are based on direct-conversion architecture. By implementing the dual-channel RX, multiple multiplexing modes, including polarization multiplexing and frequency multiplexing, can be enabled depending on the application. The RX variable gain is distributed in both RF blocks and baseband blocks to achieve a wide input dynamic range. The LNA employs a dual-coupling transformer for a low-noise figure (NF) and wideband input matching; moreover, it enhances the amplifier unilateralization characteristic to mitigate the loading effect from switched attenuators. An adjacent channel interference (ACI) cancellation scheme is proposed to further enhance the RX linearity in the frequency multiplexing mode. In the TX, single-turn high-quality-factor transformer is employed to realize matching network and four-way power combining. A prototype of the SATCOM transceiver is fabricated in a standard 65-nm CMOS process. Under 1.05-V supply voltage, the TX achieves a $P_{\mathrm {SAT}}$ of 20.5 dBm and an average output power of 12 dBm with 2% error vector magnitude (EVM) and a 37.6-dB ACPR. The dual-channel RX achieves an NF of 5.4 dB, an IIP3 of −30 dBm on the high-gain mode, the ACI cancellation is measured of 7.9 dB with a 100-MHz signal bandwidth.

SECTION I. Introduction

The artificial satellites have been invented for more than half a century to obtain a wide spherical view on the earth and realize long-distance instant communication around the world. The satellites can be mainly categorized by dedicated functions of earth imagery mapping, meteorological monitoring, space observation, global positioning, broadcasting, and communication [1]–​[3]. Recent advancements in satellite communication (SATCOM) and spaceflight technology provide ever-promising high-speed low-latency global Internet access by using massive satellite constellations. Compared to the conventional long-distance optical fiber network or microwave relay, the SATCOM provides realistic, highly portable, and low-cost user-end terminal deployment in the mountain or insular area [4]–​[9].

Among the various communication methods, SATCOM is the most unique one with its repeat station orbiting the earth. According to the orbit altitudes, the satellites can mainly be divided into three types: high earth orbit, medium earth orbit (MEO), and low earth orbit (LEO). The high earth orbit mostly indicates the geosynchronous orbit (GSO) or the geostationary orbit (GEO) for a special case. GEO satellites are popular for communication purposes due to their fixed position, wide earth surface coverage, and visibility. The LEO satellites are becoming more popular and attractive owing to the high demand on low-latency wideband Internet using SATCOM. The LEO satellites are much closer to the earth than a GEO one; both the smaller signal propagation time delay and lower free-space path loss (FSPL) featured in LEO are essential to a reliable high-speed SATCOM link establishment. Another promotion on the development of LEO satellites is that the advancement of economy space launch to the LEO makes the large-scale LEO satellite constellation possible for private companies. A SATCOM system usually consists of two segments: the space segment and the ground segment. The space segments are mostly the communication payloads of satellites used as transponders; the ground segment can be service provider hub or user terminal (UT) known as earth station or terrestrial terminal. This work presents a $Ka$ -band SATCOM transceiver for terrestrial terminals on a standard CMOS technology, which can eventually lead to the next-generation low-cost highly integrated compact solution for user-end SATCOM products.

The silicon-based millimeter-wave wireless transceivers have been significantly researched for improving integration level, system performance, and frequency of operation, especially for applications in 60-GHz WiGig [10]–​[20], 5G new radio [21]–​[37], and sensing radar [38]–​[42]. These systems are optimized for in-door short-range usage, phased-array enhanced beamforming, and signal detection, respectively. In contrast, the integrated SATCOM transceivers are seldom reported due to the particular requirements on linearity, dynamic range, and noise performance. Designed for SATCOM Ka frequency bands with downlink at 17.7–21.2 GHz and uplink at 27.5–31 GHz, as shown in Fig. 1, this article introduces a $Ka$ -band SATCOM transceiver [43] with expanded discussion on the SATCOM link budget, system requirements, detailed circuit design, and more measurement results. A four-path power-combining power amplifier (PA) enables a highly linear transmitter (TX), which guarantees the high average power ($P_{\mathrm {avg}}$ ) low error vector magnitude (EVM) output signal in the uplink. For the downlink, the adaptive receiver (RX) gain enables a broadened dynamic range based on the received signal level. The RX input port is designed to maintain a wideband low-noise figure (NF) matching with enhanced immunity on the switched load impedance. An adjacent channel interference (ACI) cancellation block is implemented for dual-channel operation. This article is organized as follows. Section II discusses the SATCOM link budget and system architecture. Section III describes the in-depth design of the SATCOM transceiver. Section IV introduces the detailed measurement results. Section V concludes this article.

Fig. 1.

SATCOM scenario and the terrestrial/non-terrestrial network.

SECTION II. SATCOM Terrestrial System Design

Substantial research results and developments have been reported in the silicon-based millimeter-wave transceiver operating in 28-/39-/60-/77-GHz frequency bands driven by exponentially increased data traffic capacity. There are few reports on the CMOS SATCOM transceiver since the specific requirements on SATCOM link usually leads to III–V compound individual monolithic microwave integrated circuit (MMIC) components, which eventually making the SATCOM terminal cumbersome and expensive. In order to overcome the large FSPL between satellite and UT, the high-power highly linear uplink and low-noise downlink are required. The SATCOM link budget and hybrid-integrated transceiver architecture are discussed in this section.

A. SATCOM Link Budget Design

The key difference of an established SATCOM data link compared to conventional wireless link is that the satellite payload behaves as a transponder between the service provider terminal and the UT. The SATCOM total data link quality will be determined by both uplink and downlink. The uplink is analyzed at first as an example. As shown in Fig. 2(a), by applying the Friis transmission formula, the carrier-to-noise power density ratio can be expressed as follows:\begin{align*} \left ({\dfrac {C}{N_{0}} }\right)_{\mathrm{ up}}=&P_{t,{\mathrm{ ES}}} \cdot G_{\mathrm{ TXA,ES}} \cdot \dfrac {G_{\mathrm{ RXA,Sat}}}{T_{\mathrm {Sat\_{}total}}} \cdot \left ({\dfrac {\lambda _{\mathrm{ up}}}{4\pi d} }\right)^{2} \cdot \dfrac {1}{L_{\mathrm{ oth}}} \cdot \dfrac {1}{k} \\ {}\tag{1}\\ d=&\sqrt {\left ({r+h }\right)^{2} - r^{2}\cdot \mathrm {cos}^{2}\,\theta _{\mathrm {EL}}}- r\cdot \mathrm {sin}\,\theta _{\mathrm {EL}}\tag{2}\end{align*} where $({C}/{N_{0}})_{\mathrm{ up}}$ denotes the uplink carrier power-to-noise power density ratio, and $P_{t,{\mathrm{ ES}}}$ , $G_{\mathrm{ TXA,ES}}$ , $G_{\mathrm{ RXA,Sat}}$ , $T_{\mathrm {Sat\_{}total}}$ , $\lambda _{\mathrm{ up}}$ , $d$ , and $k$ denote the earth station transmitted power, earth station TX antenna gain, satellite RX antenna gain, satellite RX noise temperature, uplink wavelength, satellite-to-earth station slant distance, and Boltzmann’s constant, respectively. The slant distance $d$ is dependent on the satellite elevation angle $\theta _{\mathrm {\,EL}}$ and the orbit altitude $h$ . The earth radius $r$ is a constant of 6371 km. $L_{\mathrm{ oth}}$ is the sum of the additional losses, for example, the antenna polarization error loss $L_{\mathrm{ Pol.}}$ and the antenna pointing error loss $L_{\mathrm{ Point}}$ . The essential parameters for a high-quality uplink decided by earth station are mainly $P_{t,{\mathrm{ ES}}}$ and $G_{\mathrm{ TXA,ES}}$ , while $G_{\mathrm{ TXA,ES}}$ is proportional to the effective aperture area of an antenna. Since the antenna physical aperture size cannot be designed excessively huge for a portable and reasonable installation, the most significant parameter is $P_{t,{\mathrm{ ES}}}$ , which should be carefully driven as high as possible.

Fig. 2.

(a) SATCOM link budget for both uplink and downlink. (b) Calculated system ${C}/{N_{0}}$ as a function of the elevation angle.

A similar calculation for downlink can be shown as follows:\begin{align*} \left ({\dfrac {C}{N_{0}} }\right)_{\mathrm{ dn}}= P_{t,{\mathrm{ Sat}}} \cdot G_{\mathrm{ TXA,Sat}} \cdot \dfrac {G_{\mathrm{ RXA,ES}}}{T_{\mathrm {ES\_{}total}}} \cdot \left ({\dfrac {\lambda _{\mathrm{ dn}}}{4\pi d} }\right)^{2} \cdot \dfrac {1}{L_{\mathrm{ oth}}} \cdot \dfrac {1}{k} \\ {}\tag{3}\end{align*} where $({C}/{N_{0}})_{\mathrm{ dn}}$ , $P_{t,{\mathrm{ Sat}}}$ , $G_{\mathrm{ TXA,Sat}}$ , $G_{\mathrm{ RXA,ES}}$ , $T_{\mathrm {ES\_{}total}}$ , and $\lambda _{\mathrm{ dn}}$ denote the downlink carrier-to-noise power density ratio, the satellite transmitted power, the satellite TX antenna gain, the satellite RX antenna gain, the earth station RX noise temperature as a function of elevation angle at 19 GHz [44], and the downlink wavelength, respectively. For earth station, the key parameter to decide the link quality is RX total noise temperature $T_{\mathrm {ES\_{}total}}$ .

The SATCOM total link quality of uplink and downlink is then \begin{equation*} \dfrac {1}{\left ({{C}/{N_{0}} }\right)_{\mathrm{ tot}}}=\dfrac {1}{\left ({{C}/{N_{0}} }\right)_{\mathrm{ up}}} + \dfrac {1}{\left ({{C}/{N_{0}} }\right)_{\mathrm{ dn}}}.\tag{4}\end{equation*}

It can be found that, for SATCOM earth station transceiver design, the dominant two parameters are TX output power $P_{t,{\mathrm{ ES}}}$ and RX noise temperature $T_{\mathrm {ES\_{}total}}$ .

Assuming that an LEO SATCOM constellation is used with distance $d = 500$ km, a typical 74-$cm$ -size aperture antenna is used with a 45-dBi gain at 29 GHz and a 42-dBi gain at 19 GHz in the UT, and the antenna effective aperture area and gain in LEO satellite are assumed 6 dB higher than in the UT. The RX overall NFs are 3 dB in the satellite and 5 dB in the UT. A 10-dB additional loss in each link is considered, including the antenna pointing error, the polarization error, and the insertion loss. The TX total gain and NF are assumed 60 and 14 dB in both uplink and downlink. For a linear operation, the UT TX OIP3 and RX IIP3 are 20 dBW and −30 dBm, respectively. Based on system consideration in Section II-B, the uplink $({C}/{N_{0}})_{\mathrm{ up}}$ is calculated 113 dB$\cdot $ Hz with the output power $P_{t,{\mathrm{ ES}}}$ of 3 dBW. The downlink $({C}/{N_{0}})_{\mathrm{ dn}}$ is calculated 114 dB$\cdot $ Hz with the output power $P_{t,{\mathrm{ Sat}}}$ of 6 dBW. The up/down link total CNR supports 300-MS/s 256-APSK modulation [45]. The calculated system ${C}/{N_{0}}$ as a function of the elevation angle is shown in Fig. 2(b). The total link ${C}/{N_{0}}$ is degraded by increased link distance and antenna noise temperature; 60° elevation angle tuning will result in a 6-dB CNR degradation. Further increase in the output transmitted power in an earth station of the service provider can enhance the link quality with a wider system dynamic range and higher data rate.

B. SATCOM Terrestrial Solution

The two dominant parameters for a terrestrial terminal or an earth station are concluded as the TX output power $P_{t,{\mathrm{ ES}}}$ and the RX noise temperature $T_{\mathrm {ES\_{}total}}$ . From the analyzed link budget, the $P_{t,{\mathrm{ ES}}}$ is assumed to be 3 dBW, which is feasible by using only one commercially available external GaN solid-state amplifier at 6-dB power backoff [46]. An external GaAs or InP LNA can be used to further enhance the RX overall noise performance [47]. The digital-intensive high-integrated CMOS System-on-Chip (SoC) core chip with the external PA LNA blocks results in a low-cost compact-area high-performance solution for the SATCOM system, including the block upconverter (BUC) and the low-noise block converter (LNB). Fig. 3(a) shows the block diagram of the possible SATCOM terrestrial solution with hybrid system integration. It consists of an external multiple-input multiple-output (MIMO) antenna system in light purple, a GaN/GaAs high-performance amplification block in light blue, and a digital intensive CMOS in light orange. The SATCOM CMOS SoC is focused in this work, and the CMOS GaN/GaAs hybrid integration system link budgets are given in Fig. 3(b) and (c) [46], [47]. The CMOS TX is required to be highly linear to drive the external PA. The CMOS RX gain is subject to the received signal power level; the dynamically controlled gain is required to maintain the best signal-to-noise ratio (SNR) for demodulation. Considering the atmospherical and cosmic effects, such as the rain attenuation and sun outage, the controllable gain range can be further extended.

Fig. 3.

(a) SATCOM transceiver block diagram with a hybrid system integration. (b) CMOS GaN hybrid TX link budget. (c) GaAs CMOS hybrid RX link budget.

From the perspective of satellite application and CMOS implementation, there are a few considerations.

1. The millimeter-wave CMOS PA output power is constrained by limited supply voltage in standard CMOS technology, and multiple-path power combining is required to be capable of driving a GaN amplifier.

2. Due to the strong end-user traffic asymmetry in the downlink and the uplink, the downlink dominates around 80% in the data traffic, and a multiple-channel RX is then desired for boosting the downlink speed. With multiple receiving channels, the antenna polarization can also be employed for enriching the resource allocation method.

3. The satellite direct-to-home (DTH) TV service can be improved by receiving from multiple satellites, and the received signal in one channel will be jammed by another channel if the power difference is too high. As a result, an ACI tolerant RX is required in the multisatellite application.

4. The dynamic range of the CMOS SATCOM RX chip is contingent on the following factors in practical use; hence, the gain range is required to be wide.

   1. The preceding antenna and external LNA in different applications may be different and result in a wide input power range.

   2. The atmospherical and cosmic effects on SATCOM link, such as the rain attenuation and the sun outage, require the established link to be high power for the robustness.

   3. The SATCOM link power and modulation can be adaptively adjusted based on the throughput for optimized resource allocation and battery saving.

   4. The dual-channel RX in FDMA mode will have a wide-range input power from satellites in different directions.

Table I shows a target specification of the CMOS $Ka$ -band SATCOM transceiver for terrestrial usage. The transceiver is designed for LEO satellites orbiting the earth at a 500–2000-km distance. The up/down total link supports 256 APSK modulation with a 300-MHz bandwidth. The SATCOM LNB G/T is assumed of 15 dB/K, and the NF of CMOS RX targets 5 dB. The RX design target also includes a 30-dB dynamic range and a 10-dB ACI cancellation. The TX targets a linear output at 12 dBm.

TABLE I SATCOM Transceiver Target Specification

SECTION III. SATCOM Transceiver Circuit Design

Fig. 4 shows the entire block diagram of the $Ka$ -band SATCOM transceiver, including the RF front-end and the analog baseband. The transceiver is composed of one 29-GHz TX for uplink and two 19-GHz RXs for downlink. Both the TX and the RX are in direct-conversion architecture with differential analog baseband. The direct-conversion architecture is chosen for the system integration simplicity and higher dynamic range over the digital IF architecture. The TX includes a pair of $RC$ low-pass filters (LPFs), a double-balanced mixer, an RF variable-gain amplifier (RF-VGA), a drive amplifier (DA), and a four-way two-stage power-combining PA. The TX LO is generated by using a poly-phase filter (PPF) from an external LO input. On the RX side, the received 19-GHz RF signal is amplified with a three-stage low-noise amplifier (LNA) followed by an RF-amplifier and an RF-VGA. Switched gain controls are distributed in the LNA and the RF-VGA. The RF signal is down-converted to baseband through a double-balanced mixer. PPF and external LO are also used for the quadrature LO generation in RX. The RX analog baseband circuitry consists of a differential two-stage active LPF and a 10-bit digital-to-analog converter (DAC)-controlled VGA. The active LPF employs Akerberg Mossberg topology for wide bandwidth characteristics up to 300 MHz. The RX includes an ACI cancellation negative feedback network by up-converting the RX1 high-power-level adjacent channel filtered baseband signal to the RF domain and summing with the RX2 input. The mixers in RX1 and ACI cancellation block are driven with anti-phase LOs for generating desired ACI-canceling RF signal.

Fig. 4.

CMOS $Ka$ -band SATCOM transceiver block diagram.

A. Receiver

The SATCOM RX gain, noise, and linearity link budget are planned first, as shown in Fig. 5. The RF signal is amplified through a single-ended LNA and divided into I-path and Q-path with differential circuit topology. The total gain is 18 dB in the RF path and 20 dB in the baseband path. The variable gain stages are also distributed in both RF and baseband paths. The main purpose of variable gain in the RF stage is to prevent the baseband circuit from being saturated when the gain provided by external LNA is too high. In this case, the LNA design is critical from the perspective of system overall NF and linearity. Fig. 6 shows the proposed LNA with switched attenuation and unilateral input network. The LNA is based on a common-source configuration, and ground-shorted switches are inserted between the LNA stages. In order to support a wide input power range, the first gain stage and the last gain stage are enhanced by inductive source degeneration for higher linearity, and the transistor sizes are also chosen larger for the capability of high-power input signal. Since the variable attenuation is enabled by the ground-shorted switch, the dramatic load impedance change will cause severe input matching mismatch if the signal has significant reverse path. The weakly coupled transformer feedback provides possible solution of mitigating the reverse signal path without degrading NF [48], [49]. The criterion of the amplifier unilateralization is \begin{equation*} h_{12(\mathrm {Core})} = -h_{12(\mathrm {XFMR})}\tag{5}\end{equation*} where $h_{(\mathrm {Core})}$ and $h_{(\mathrm {XFMR})}$ represent the hybrid $h$ -parameters of LNA first-stage core network and the transformer network, as shown in Fig. 6. Considering the gate–drain capacitance $C_{\mathrm {\mathrm{ gd}}}$ , the calculated LNA core network $h_{12(\mathrm {Core})}$ is \begin{equation*} h_{12(\mathrm {Core})} = - \dfrac {C_{\mathrm{ gd}}}{C_{\mathrm{ gd}}+\dfrac {C_{\mathrm{ gs}}}{1+sL_{s1}\left ({sC_{\mathrm{ gs}}+g_{m} }\right)}}\tag{6}\end{equation*} and the transformer network $h_{12(\mathrm {XFMR})}$ is \begin{equation*} h_{12(\mathrm {XFMR})} = k_{1} \cdot n_{1}.\tag{7}\end{equation*} The calculated $h_{12(\mathrm {Core})}$ has a real part of 0.2 approximately, which is dominant over the imaginary part. For a transformer turn ratio $n_{\mathrm {1}}=1$ , the magnetic coupling coefficient $k_{1}$ is then decided 0.2 in order to satisfy the unilateralization criterion. The LNA second-stage input also employs a customized inductor for low-noise impedance matching since the transmission line (TL) is optimized for its characteristic impedance, and therefore, the quality factor is relatively low [50], [51]. The inductors $L_{\mathrm {D1}}$ and $L_{\mathrm {D2}}$ are magnetically coupled for enhancing the inductance and saving the area. The input transformer is modeled using EM simulation; the transformer is simulated with certain diameter and width deviations to ensure reliability under possible process variation. The remaining stages use TLs in the matching blocks. An identical standalone LNA circuit is fabricated for characterization. Fig. 7 shows that the three-stage LNA has a measured peak gain of 17.7 dB with a controllable gain range of 9.0 dB and a 3-dB BW of 7.6 GHz. The minimum NF is measured 2.1 dB in the high-gain mode and 7.0 dB in the low-gain mode. The LNA input reflection coefficient S11 is below −10 dB with the first-stage attenuation switch turned off over 17–21 GHz. By using the unilateralization coupling network, the LNA maximum S11 is degraded by 2 dB when the switch is turned on. By contrast, the simulated S11 at SW = 111 without unilateralization feedback is also shown in Fig. 7(a), and S11 is degraded to worse than −10 dB. The simulated stability k-factor [see Fig. 7(b)] shows that the LNA is in stable condition. Since the transformer unilateralization feedback is passive and implemented in the top thick metal layers with metal width wider than 4 $\mu \text{m}$ , the transformer feedback is relatively insensitive to process variation.

Fig. 5.

SATCOM RX gain, NF, and IIP3 level diagram.

Fig. 6.

Circuit schematic of RX LNA and RF-VGA.

Fig. 7.

RX three-stage LNA measured performance, simulated S11 at SW = 111 without unilateralization feedback, and simulated k-factor.

The single-ended RF signal after LNA is converted to differential through a Balun, as shown in Fig. 8. The differential RF signal is then divided into I-path and Q-path for down-converting to baseband frequency. A differential RF-VGA with the switched gain mode is employed for the additional RX variable gain range. By enabling cross-connection with anti-phase signal in the differential RF-VGA output, the RF-VGA can provide 12-dB variable gain in simulation. The down-conversion mixer employs passive double-balanced architecture for suppressing potential LO and RF leakage to the baseband.

Fig. 8.

Circuit schematic of the RX RF-VGA and double-balanced mixer.

Fig. 9 shows the detailed circuit schematics of the LPF in RX analog baseband. The active LPF is composed of two-stage second-order LPFs with their total roll-off rate of −24 dB/octave. Due to the limited voltage headroom in the deep sub-micrometer CMOS process, the LPF has a current input and behaves as a transimpedance filter. Since the active LPF has a common-mode voltage on half VDD, and the mixer terminal needs to keep a low potential for conducting the transistor, a level shifter (LS) is, therefore, used to shift the low voltage at the mixer to half VDD at the LPF. The two-stage transimpedance LPF has a simulated gain of 60 dB$\Omega $ . The LPF employs Åkerberg–Mossberg topology in order to control quality factor (${Q}$ ) and pole frequency independently. The quality factor and pole frequency can be expressed as \begin{align*} Q=&\sqrt {\frac {C_{1}R_{Q}^{2}}{C_{2}R_{f}R_{m}}} \tag{8}\\ f_{p}=&\dfrac {1}{\sqrt {C_{1}C_{2}R_{f}R_{m}}}\tag{9}\end{align*} respectively. The variable gain in LPF is realized through changing its second-stage input resistance $R_{G}$ . The variable quality-factor, bandwidth, and gain are digitally controlled with a 5-bit resistor bank and capacitor bank.

Fig. 9.

(a) Circuit schematic of RX baseband LPF. (b) RX LPF simulated gain and frequency characteristics.

The LPF is followed by a baseband VGA, as shown in Fig. 10. The VGA source degeneration resistance $R_{\mathrm {deg}}$ can be tuned to realize variable transconductance and, therefore, the variable gain. It realizes a simulated gain tuning range from −2 to 12 dB. Active inductances are used to cancel the output capacitance and enhance the bandwidth of VGA. The active inductances employ Gyrator realization with transistor parasitic capacitance. The dc offset cancellation (DCOC) in the VGA is added to eliminate the dc offset amplification. The dc feedback loop (DCFL) contains an integrator and two gain stages. The VGA with DCFL closed loop shows a high-pass characteristic, which can prevent the dc offset from saturating subsequent circuits.

Fig. 10.

(a) Circuit schematic of RX baseband VGA. (b) RX baseband VGA simulated gain and frequency characteristics.

The ACI cancellation block shown in Fig. 11 is used in the RX frequency multiplexing mode. An anti-phase cancellation signal is generated in case the high-level channel-1 signal becomes an adjacent interference of the low-level channel-2 signal. The channel-1 received baseband signal is filtered and up-converted with anti-phase LO. The ACI RF signal is amplified by a two-stage RF-amplifier, as shown in Fig. 12(a). The ACI cancellation block provides a 22-dB gain with a 17-dB variable gain range and a 7-GHz RF bandwidth, as shown in Fig. 12(b). Since the receiving antenna polarizations are the same in the frequency multiplexing mode, the RX two channels can be switched by sensing output power, and the ACI is, therefore, implemented in one channel. The ACI signal is tapped on RX1 by T-junction TL. By adding the ACI path, the RX NF and gain are degraded. From the simulation, when the ACI is enabled, the RX2 NF is 0.53 dB increased, and the gain is 1.2 dB decreased.

Fig. 11.

ACI cancellation scheme in the FDMA operation mode.

Fig. 12.

(a) Circuit schematic of the RX ACI cancellation block. (b) Simulated conversion gain of the ACI cancellation block.

B. Transmitter

As shown in Fig. 13, the SATCOM direct-conversion TX baseband input is connected to mixers through an $RC$ passive LPF with variable passband controlled by selecting capacitor bank. The LPF cutoff frequency can be tuned from 50 to 400 MHz with 6-bit binary control. The mixer is passive and with double-balanced topology; the mixer RF port dc offset is tunable in order to cancel the LO feedthrough (LOFT) caused by potential transistor mismatch [52]–​[54]. The up-converted RF IQ signal is added to two differential paths RFP and RFN through an RF adder with a 7-dB RF gain, as shown in Fig. 13.

Fig. 13.

Circuit schematic of the LPF, mixer, and RF adder in SATCOM TX.

The RF adder differential output RFP and RFN is connected to a differential PA through an RF VGA, as shown in Fig. 14. The RF VGA is based on current steering cascode topology [55], the current steering transistor is controlled by a 10-bit DAC, and the VGA realized a variable gain of 10 dB. The differential PA has three stages. The first stage contains two parallel single-ended DA. The SATCOM TX is designed for achieving high linearity due to the requirement of output power and drive ability. In the CMOS process, over 10-dBm OP1dB output power PA has been demonstrated [56]–​[61]; however, it is not high enough to drive the 3-dBW external amplifier. Further increasing the PA output power, larger devices and transformer-based power combining techniques can be used. Therefore, in this TX, the differential path unit PA final two stages are divided to differential again with four paths power combining in total, as shown in Fig. 14. Power matching is used in the PA load network to achieve high output power and linearity. Fig. 15(a) shows the load–pull result of the PA transistor pair, and the higher output power leads to a low output impedance. In the current-type power-combining load network, the load impedance seen by the two identical paths will be twice of $R_{L}$ . Since a transformer-based balun is required for converting differential to single-ended, the matching network can be easily realized by the combined network of a balun, a series TL, and a shunt capacitor. The realized matching network is shown in Fig. 15(b), and a high-quality-factor transformer is carefully designed for low-loss power matching. The transformer has a 1:1 turn ratio and vertically stacked structure for achieving high-quality factor and high coupling coefficient at millimeter-wave frequency [56], [61]. To improve the simulation and EM modeling accuracy, the standalone transformer is fabricated and measured to establish a semi-empirical model by fitting the EM-based model with the measured performance. The simulated PA transformer-based output network has a 2.5% impedance mismatch and a 0.7° phase mismatch seen by the transistor pairs, and the mismatch will not degrade the in-phase current-type power combining performance. For a reliable design, the PA is located at the corner of the chip with a strong power supply and ground, and it is also close to the decoupling capacitors to enhance the power integrity and PA stability. The PA transistor pair core utilizes a cross-coupled capacitive neutralization technique to enhance reverse isolation and to improve the power gain. Since the PA large power capability can be degraded by the hot-carrier-injection (HCI) effect, further consideration of the HCI healing technique can be useful for enhancing the PA reliability [62]. A standalone PA is fabricated, including the final two power stages and one DA stage. The standalone PA is measured with a peak gain of 23.7 dB with a 3-dB BW of 12 GHz. The standalone PA small-signal measurement result is shown in Fig. 15(c). The large-signal measurement result of the standalone PA is shown in Fig. 15(d); at class-A bias condition, the PA has a saturated output ($P_{\mathrm {SAT}}$ ) of 20.7 dBm and a 1-dB compression point ($P_{\mathrm {1\,dB}}$ ) of 15.9 dBm at 29 GHz.

Fig. 14.

Circuit schematic of the PA in SATCOM TX.

Fig. 15.

(a) PA matching network illustrated Smith chart. (b) PA matching network with transformer 3-D view. (c) PA measured small-signal performance. (d) PA measured large-signal performance.

SECTION IV. Measurement Results

The $Ka$ -band SATCOM transceiver is fabricated in a standard 65-nm CMOS process with nine metal layer stacks, including one thick metal layer and one ultra-thick metal layer. The chip micrograph is shown in Fig. 16. The transceiver occupies a chip area of 3 mm $\times 3$ mm. The VDD routing and the RF signal are mainly in the ultra-thick metal layer, the analog and digital circuits are designed in the thin metal layers for a compact area, and the GND routing uses all the metal layers. The TX is located at the top side, the dual-channel RX is located at the lower side, and the RF input/output ports and the LO ports are located at the left- and right-hand sides, respectively. The transceiver power breakdown is also shown in Fig. 16, the TX consumes 0.62 W, and the dual-channel RX consumes 0.93 W in total, including the ACI cancellation block. The RF blocks use a 1.05-V supply, and the analog baseband uses a 1.2-V supply. An evaluation PCB board is implemented with Megtron 6 material, and the chip is mounted using wire bonding. In the future CMOS III–V hybrid system integration, the ICs can be placed in the same cavity for a low wire bonding loss, and the thermally conductive adhesive glue can be used to attach the ICs with good thermal performance.

Fig. 16.

Die micrograph and power breakdown table.

The SATCOM transceiver’s small-signal performance is first evaluated on a probe station using a vector network analyzer (Keysight N5247A). Fig. 17 illustrates the general setup for evaluating both TX and RX using N5247A, two of the ports are connected to the input/output of the TX or RX on-chip through RF probes, and the external LO signal is also generated and controlled by N5247A with a typical phase noise of −97 dBc/Hz at a 1-MHz offset at the $Ka$ -band. The external LO power is 3 dBm for both TX and RX. The RF and LO frequency offsets are set to 100 MHz. The digital control and power supply are connected using eye-pass probes on the other two sides of the chip. An external LNA is used at the baseband output to keep high enough gain in RX NF measurement. For setup simplicity, the double-sideband (DSB) noise is measured only. The measurement setup is calibrated to the probe tips of signal input and output. Fig. 18(a) shows that the measured peak conversion gains of the TX are 26.1 and 35.1 dB on low- and high-gain modes, respectively. The TX 3-dB bandwidth on the low-gain mode is from 26.9 to 32.6 GHz and 27.0 to 32.4 GHz on the high-gain mode. The TX large-signal characteristic is evaluated using a power meter (Keysight E4417A). Fig. 18(b) shows the measured TX $P_{\mathrm {SAT}}$ and $P_{\mathrm {1\,dB}}$ over the operation frequency from 27 to 31 GHz. $P_{\mathrm {SAT}}$ and $P_{\mathrm {1\,dB}}$ are measured higher than 20 and 15 dBm, respectively. Fig. 19(a) shows that the RX measured peak conversion gain on the high-gain mode is 34.3 dB, and the 3-dB bandwidth is from 16.9 to 22.5 GHz. The measured total variable gain range is 50 dB. Fig. 19(b) and (c) shows the RX measured input third-order intercept point (IIP3) and NF on different gain conditions. On the high-gain mode, the RX achieves an IIP3 of −30 dBm and an NF of 6 dB at 19 GHz. In the 17–21-GHz operation band, the IIP3 varies from −28.3 to −31.1 dBm, and the NF is between 5.4 and 7.0 dB. The signal-to-noise and distortion ratio (SNDR) of the RX can be calculated from the measured RX gain, intermodulation distortion (IM3), and NF. Fig. 20 shows the RX achieves a peak SNDR of 35.3 dB at a −48.5-dBm input power on the high-gain mode and 40.1 dB at a −20.0-dBm input on the low-gain mode with a 300-MHz signal bandwidth. Fig. 21 shows the measured RX analog filter gain and frequency characteristics normalized to the lowest gain mode; it has a variable gain of 20 dB and is capable of variable bandwidth up to 300 MHz. For the digital-to-RF system integration, a 10-dBm full-scale-input differential ADC with a sampling rate higher than 600 MS/s can be used in the RX, considers a 30-dB signal SNR for demodulation, a 20-dB power range, a 6-dB quantization noise margin, and a 6-dB signal PAPR margin, and the required ADC ENOB is estimated of 10 bit. The ADC ENOB requirement can be relaxed by a variable gain driver preceding the ADC, and the required driver OIP3 is estimated at 30 dBm.

Fig. 17.

Measurement setup using a vector network analyzer.

Fig. 18.

Measured and simulated (a) TX conversion gain, (b) TX $P_{\mathrm {SAT}}$ and $P_{\mathrm {1\,dB}}$ , (c) TX conversion gain compression, and (d) power compression.

Fig. 19.

RX measured and simulated (a) conversion gain, (b) IIP3, and (c) NF.

Fig. 20.

RX measured SNDR in (a) high-gain mode and (b) low-gain mode.

Fig. 21.

RX measured baseband frequency characteristics and gain control.

The SATCOM transceiver is also evaluated using modulated signals of QPSK, 32APSK, 64APSK, and 256APSK. Modulation schemes regulated in SATCOM standard DVB-S2X are included in the measurement. In the TX measurement, the baseband IQ differential signals are generated by an arbitrary waveform generator (AWG) (Keysight M8195A) with a symbol rate of 150 MBaud and a roll-off factor of 0.1, the 29-GHz RF output signal is down-converted to a 3-GHz IF frequency using an external mixer and evaluated by an oscilloscope (Keysight DSO90804A), and the output average power is measured by a power meter directly at the RF port. The external LO in the EVM measurement uses Keysight E8257D with a typical phase noise of −130 dBc/Hz at a 1-MHz offset at the $Ka$ -band. The external LO power is 3 dBm for both TX and RX. Fig. 22 shows the TX measured EVM at different average output power in both QPSK and 64APSK modulation schemes, and the power level is tuned by adjusting AWG signal level and external attenuators at the baseband input. The measured average output power is 17.5 dBm with EVM of −22.1 dB in QPSK and 15.3 dBm with EVM of −20.7 dB in 64APSK. Fig. 23 shows the summary of the TX measured the best constellation, EVM, average output power, and ACPR in modulation schemes of QPSK, 16APSK, 64APSK, and 256APSK. The measured TX EVMs achieve lower than −42 dB with ACPR around 45 dB. To achieve the target EVM in the digital-to-RF system integration, consider a 40-dB signal SNR, a 15-dB power range, and a 6-dB signal PAPR margin, and the corresponding DAC bit for TX is 10 bit. In the RX measurement, the 19-GHz modulated RF signals are generated by up-converting AWG (Keysight M8195A) baseband signal to 19 GHz using a vector-signal generator (VSG) (Keysight M8267D), the symbol rate is 150 MBaud, and the roll-off factor is 0.1. The baseband output signal is evaluated by an oscilloscope (Keysight DSO90804A). The RF power is attenuated using external attenuators, and the input average power is measured by a power meter directly at the RF port. Fig. 24 shows the RX measured EVM at different average input powers, the power level is tuned by adjusting VSG signal level and external attenuators at the RF input, and the RX gain is also tuned to maintain the best EVM performance with different input power levels. The 64APSK modulation is used in the high- and medium-gain modes, and the QPSK modulation is used in the low-gain mode. The measured average input power is −53 dBm with EVM of −25.1 dB in 64APSK; the dynamic range is more than 35 dB with EVM under −25 dBm. The RX is also capable of high-power input; it achieves EVM of −12.6 dB at a −1-dBm average input power in QPSK modulation. Fig. 25 shows the summary of the RX measured the best constellation, EVM, and average input power in different modes. The measured RX EVMs achieve lower than −25 dB with the variable gain setting.

Fig. 22.

TX measured EVM as a function of the average output power.

Fig. 23.

TX measured constellation summary.

Fig. 24.

RX measured EVM as a function of the average input power.

Fig. 25.

RX measured constellation summary.

The dual-channel RX ACI cancellation is measured with two signal sources, as shown in Fig. 26: one is a 19.05-GHz continuous-wave (CW) generated by the signal generator, and another is the modulated RF signal at 19.3 GHz. The dual-channel RX operates in the FDMA mode, the modulated RF signal will be demodulated in RX1, and the CW signal is meant to be down-converted by RX2. The two signal sources are combined and connected to both RX1 and RX2 input ports. In this setup, the 19.05-GHz CW down-converted baseband is the desired output, and the 19.3-GHz modulated RF signal is the interference signal located in the adjacent channel. The RX2 baseband output is observed using a Spectrum Analyzer without external LO phase control. To keep the same true time delay between the ACI path and the RX2 path, the RF cable from the divider feed to RX2 is adjusted slightly longer than to RX1. Fig. 27(a) shows the measured spectrum in RX2 baseband output; when the ACI cancellation is enabled, the adjacent channel signal down-converted spectrum at 300 MHz is canceled to a lower power level. The cancellation reduces the total input power including, the undesired ACI signal; the linearity for signal in the desired channel is, therefore, improved. Fig. 27(b) shows the effective cancellation results against ACI signal RF bandwidth. For ACI signal bandwidth of 100 MHz, the effective cancellation is 7.9 dB. The effective cancellation is more than 10 dB for ACI signal bandwidth of 60 MHz. The channel power is measured by the Spectrum Analyzer. Fig. 27(c) shows the ACI cancellation scheme improves the RX input 1-dB gain compression point by 4 dB.

Fig. 26.

ACI cancellation measurement setup.

Fig. 27.

ACI cancellation measurement results. (a) Measured baseband spectrum. (b) ACI cancellation with different RF bandwidth. (c) RX linearity improvement with ACI cancellation.

The performance comparison of $Ka$ -band millimeter-wave transceivers for SATCOM and 5G is listed in Table II. The SATCOM RX demonstrates low NF and high linearity characteristics with a minimum NF of 5.4 dB and an IIP3 of −30 dBm on the high-gain mode over 17–21 GHz. The RX achieves a 50-dB variable gain range and wide dynamic range. The dual-channel RX realizes 7.9-dB ACI cancellation with a signal bandwidth of 100 MHz; the cancellation scheme further enhances the RX linearity in the FDMA multiplexing mode. The SATCOM TX shows highly linear characteristics by achieving 12-dBm average output power using QPSK and 9.5-dBm output power using 64APSK with EVM below 2%.

TABLE II Performance Comparison of $Ka$ -Band Millimeter-Wave Transceivers for SATCOM, 5G, and Beyond

SECTION V. Conclusion

This article first presents a $Ka$ -band CMOS transceiver for SATCOM. The transceiver consists of a highly linear TX and dual-channel wide-dynamic-range RX. To increase RX dynamic range, the RF stages and baseband stages can offer a variable gain of 50 dB in total. A dual-coupling transformer is used in LNA for a low-noise and wideband design; it also mitigates the loading effect from switched attenuators. An additional ACI cancellation scheme is implemented to further enhance the RX linearity in the frequency multiplexing mode. The SATCOM TX employs four-way power combining and a high-quality-factor transformer to maintain highly linear RF output. The transceiver achieves a $P_{\mathrm {SAT}}$ of 20.5 dBm and an average output power of 12 dBm with an EVM of below 2%. The dual-channel RX achieves an NF of 5.4 dB and an IIP3 of −30 dBm on the high-gain mode; the ACI cancellation is measured of 7.9 dB with a 100-MHz signal bandwidth.