A 2 kW S-band RF source for multipactor research utilizing GaN HEMTs Open Access

The multipactor phenomenon may occur when the mean free path of an electron within a structure is larger than the distance between two surfaces, in this case the top and bottom of a rectangular waveguide operated in its fundamental mode, and the movement of electrons is synchronized with the RF signal.^1,2 Conditions for the multipactor effect are fairly specific, but space based systems are particularly susceptible.³ In addition to physical damage to RF circuits,⁴ significant signal degradation is a possible result of multipactor. Pressure, electric field, and residual gas type affect how and when multipactor occurs.

This paper focuses on efforts in multipactor research conducted in a WR284 waveguide multipactor test cell with a narrow gap section on the order of a few mm⁵ with 2-surface multipactor of particular interest at 2.85 GHz necessitating a high power RF generator and traveling wave ring resonator to increase power to suitable levels. An impedance transformer is used to transition from the low impedance waveguide to the high impedance test section to generate sufficient voltage across a small enough gap. Through simulation and the literature, it was found to require a certain level of power, starting at ∼2 kW for a 2 mm gap,^6–8 to initiate the multipactor event through secondary electron emission in copper with more power required for larger gaps. This is primarily driven by the secondary electron yield (SEY) and where the first crossover point (SEY > 1) occurs. Beyond that, the balance of field and phase must be correct in order to achieve multipactor, generally of an odd order, N.

The purpose of multipactor research is not to explicitly design systems capable of generating a multipactor event, but rather finding ways to mitigate or prevent the detrimental phenomenon from presenting itself, for which systems capable of generating the phenomenon are required, particularly, in space based RF systems where in situ repairs are—as of now—impossible. The multipactor effect, at best, causes de-tuning and loss of Q in RF systems, and at worst, will cause significant damage to physical RF structures and accompanying circuits.^1,4

Traditionally, high power amplifiers—particularly at microwave frequencies—used vacuum electron tube topologies, which are physically large designs most suited to power outputs much higher than those considered here.^9,10 Topologies, such as Traveling-Wave-Tubes (TWTs), may reach power densities of 0.7 W/cm³ with GaN on SiC HEMTS¹¹ reaching 40 W/cm. Additionally, vacuum tube technologies commonly require high voltage sources for the cathodes (or anodes depending on design), whereas the HEMT used here simply requires 50 V DC for the drain bias. Unlike waveguide based amplifiers (i.e., TWTs), solid state amplifiers themselves can be a fraction of the wavelength, ∼10 mm at 2.85 GHz, with packaging mainly dictating physical size. Recently, with the advent of compact, efficient semiconductor solutions for high power microwave, the miniaturization of high power amplifiers has become possible. One such emerging commercial technology is the Gallium Nitride (GaN) (a III–V wide bandgap semiconductor) High Electron Mobility Transistor (HEMT), which boasts efficiencies higher than vacuum tubes while being a fraction of the size.¹² Silicon carbide transistors and Laterally Diffused Metal Oxide Semiconductor (LDMOS) amplifiers, while power dense, are not particularly suited to GHz operation mainly related to the relatively large parasitic capacitances present in such devices.¹³ With gate shorting technologies being developed, GHz capable LDMOS power amplifiers are becoming available; however, they are as of yet limited to mid-3 GHz operation.^14,15 GaN is mainly limited in its low thermal conductivity commonly being grown on a SiC substrate to aid in heat distribution.¹⁶ In general, the highest power GaN amplifiers are best suited to pulsed operation where the average power dissipation is lower than in continuous wave cases where LDMOSs will be more suitable at lower frequencies. Gallium Arsenide (GaAs), another well established III–V semiconductor, is well suited to lower power, low noise RF systems where linearity and noise performance are most critical.¹¹ GaN’s wide bandgap and voltage hold off enable compact, high power amplifiers in the GHz regime extending into the high power mm-wave band in the future^17,18 or beyond albeit at lower power.^19,20 One additional benefit to GaN is the inherent radiation hardness, making it a good fit for satellite communications.^21,22

Compared to a basic commercial solution, such as a unit produced by RFHIC²³ intended for airport radar use or similar, the cost including estimated labor for the presented case will be similar. Furthermore, full lab grade systems available at the time of the drafting of the manuscript are roughly five times more expensive than that. Physically, the RFHIC unit is similar in size to the presented source when power supplies and an exciter—VCO and pre-amplification—are added; however, rack-mounted lab grade systems tend to be much larger.

Starting with a low noise oscillator, the system consists of four amplifying stages with auxillary components in place in-between, see Fig. 1. In detail, the RF signal is generated by a Mini-Circuits ZX95-3063C-S + Voltage Controlled Oscillator (VCO), which is buffered with a ZX60-3011 + Low Noise Amplifier (LNA) before being fed to a ZFSWA2-63DR + Transistor–Transistor Logic (TTL) switch controlling the RF flow and setting the RF pulse duration via external TTL (5 V) pulse. While a VCO presents a relatively budget friendly option, other options, such as a Phased Locked Loop (PLL) (e.g., Maxim MAX2871) or low noise laboratory source, may be utilized as well. To achieve the stability needed in the traveling ring resonator, care was taken in assuring low noise on both VCC and tune voltage lines for low phase noise, referred to as frequency pushing. Noise on the power and control voltage lines from both the switch mode power supplies used and electromagnetic interference (EMI) from the environment had initially caused relatively high levels of phase noise on the VCO output causing the output frequency to be unstable. Thus, the VCO was installed in a shielded enclosure to reduce these EMI effects. It is worth noting that the VCO tuning sensitivity can directly affect phase noise with more sensitive VCOs being more susceptible to external noise sources. Hence, the VCO used here has a sensitivity of ∼30 MHz/V, which would shift the output frequency ∼300 kHz with a typical power supply ripple of 10 mV. After the pulse is generated, what follows are simply additional gain stages to reach the required 2 kW output power. A 30 dB continuously variable micrometer-adjusted attenuator is placed in line to control the output level. The majority of the gain in the system, 45 dB, is provided by a ZHL-16W-43-S+ power amplifier, capable of 16 W maximum output; however, only 6 W is needed to drive the final stages for the rated output. The Qorvo TGA2585-SM, a 10 W GaN Monolithic Microwave Integrated Circuit (MMIC) amplifier, may be substituted. Four parallel CGHV31500F GaN HEMTs from Cree/Wolfspeed are used as the final output stage with a fifth providing enough power to drive them. Additionally, a custom power sequencing and biasing control board was designed. Subsections II A–II C will provide information on each part of the system in more detail.

Gallium Nitride High Electron Mobility Transistors (GaN HEMTs) from Cree/Wolfspeed (model CGHV31500F) are used for the main power stages as they provide good gain, 12.5 dB, with high drain efficiency, ∼67%, at 500 W output per device.²⁴ The transistors are specified for 50 V drain, approximately −3 V gate threshold voltage (−2.7 V gate quiescent), and 500 mA drain bias current. While rated for 100 µs pulse width at 10% duty cycle, longer pulse widths of up to 500 µs are possible if the duty cycle is kept low. In a typical multipactor research scenario, single shots with long off times between shots have very low duty cycles, thus possibly enabling longer pulse lengths. Each GaN amplifier is mounted in a Cree/Wolfspeed test fixture, which provides mounting of the transistor, input and output SMA connectors, drain/gate connections with power filtering, and charge storage onboard. The bandwidth of the system is mainly limited by the GaN amplifiers with a bandwidth of 2.7 GHz–3.1 GHz, and while some gain may be available outside of this range, the output power will be reduced. An appropriate oscillator—VCO or otherwise—for the frequency range should be selected.

Due to the power levels being considered, off-the-shelf splitters/combiners were unavailable and were custom designed in microstrip technology (Fig. 2). As a microstrip substrate, a 1.524 mm thickness Rogers 4350B²⁵ was selected with a permittivity of ∼3.5 to keep trace widths wide, which ensured low ohmic losses along with 2 ounce copper pour. Alternative materials, such as FR4 or alumina, were rejected, based on their high dielectric loss or large permittivity (∼9), respectively. The splitter and combiner are of a corporate branchline design, which was designed in NI’s AWR Microwave Office, with all four ports having equal phases. A coupled line directional coupler was additionally designed (Fig. 3); however, directivity is limited with both coupled lines on the same side of the main through line. Thus, a circulator was additionally used on the output to both prevent damage to the source from reflections at the load and give another point to measure the reverse power. The splitters have a loss of ∼0.1 dB, which is omitted in the block diagram (Fig. 1), as this small loss is easily overcome by the gain of the system.

Due to the power sequencing and bias requirements of GaN HEMTs (Fig. 4), a custom control board was designed and implemented (Fig. 5). GaN HEMTs require that a negative gate bias be applied before the drain voltage may be applied.²⁶ In addition to biasing and sequencing the HEMTs, the board also provides voltages needed to run other components, namely, the VCO, LNA, switch, and bias/sequencing modules. The five identical modules contain the circuitry necessary for biasing, drain power control, and temperature compensation. The modules are based on a circuit found in Ampleon AN11130²⁶ with modifications to suit the application here. The Ampleon circuit originally included provisions for generating the voltage rails needed, namely, an inverter for the negative rail, which was removed and a Low Drop Out regulator for a 5 V rail, which was also removed. A Murata 28 V isolated DC/DC converter with a −8 V linear regulator is used to generate the negative voltage needed for the gate biasing. Additionally, the control board provides +5 V, +8 V, and +12 V. Each biasing module is provided 5 V, −8 V, and 50 V rails. The gate voltage minimum was adjusted from −4 V in the reference circuit to −6 V to match the operating conditions of the HEMTs. Additionally, the temperature compensation of the circuit was modified. The turn-on time of the external high-side N-MOS, which controls the drain bias, was increased to lower the current through it during turn-on preventing damage. Gate voltage feedback is critical as the GaN HEMTs may pull significant gate current (up to 80 mA), and as such a large voltage drop may develop across any resistor placed in the gate bias path, as is the case with the test apparatus.²⁴ Gate feedback is pulled from as close to the gate connection as possible with hook up wire soldered to a pad provided with an air core choke wound upon it to prevent RF feedback onto the DC line. Additionally, multiple ferrite cores are placed on the control and bias lines to provide further RF isolation.

RF detector diodes were used to measure the output power of the source, both directly at the output and from the directional coupler. During use, the coupler may be used to monitor the RF power being fed into a multipactor test cell. Initially, a maximum output of just under 2 kW was measured with cross-amplifier oscillation presenting itself. Having the four amplifiers at equal phases is prone to issues due to load-pull effects whereby the impedance seen by the amplifier output due to the reflected power can affect its characteristics including inducing oscillations.²⁷ Thus, the amplifiers were placed 90° out of phase using SMA adapters, which are close to a quarter-wavelength at 2.85 GHz to improve isolation between amplifier modules and mitigate load-pull effects. The exact phase is not too important as long as the signals are close to 90° out of phase within the amplifiers and are in phase at the combiner (Fig. 6). A true quadrature splitter/combiner architecture would improve isolation further enabling higher power output with less inter-module instabilities; however, this would require replacement of the corporate branchline-splitter/combiner with units based upon 90° hybrids or other architecture providing the necessary phase shifts and isolation.²⁸ 

A maximum saturated output of ∼3.5 kW was measured (Fig. 7) with a 100 µs pulse width. Long term operation at the saturation point may degrade performance and can cause damage to the GaN amplifiers. Again, at the low duty cycles being used here, this is likely not an issue but should be considered regardless. While the maximum power was measured at 3.5 kW, a more typical output power of 2 kW is readily achieved. The bandwidth of the system, mainly limited by the GaN HEMTs, was measured to be ∼2.45 GHz–2.95 GHz within the linear region of the transistors with output power dropping off at the band edges (Fig. 8). The trough at ∼2.7 GHz is speculated to be caused by an interaction between the microstrip based splitters and the phasing method used. A true quadrature phasing architecture, as explained previously, could fix this issue. The linearity of the amplifiers was also examined by increasing the input power until saturation was observed (Fig. 9) at ∼2.2 kW.²⁹ 

Direct detection of the multipactor effect was achieved feeding the source output into a ring resonator with a typical quality factor, Q, of ∼10. An Electron Multiplier Tube (EMT) was used to directly detect multipactor electrons within a stepped impedance transformer test section with a gap of 5.5 mm for the results shown here. The EMT is placed in the middle of the broadside of the WR-284 waveguide with a small hole (∼1 mm) through which any electrons may pass into the EMT. This measurement technique assumes the presence of electrons in the middle of the waveguide, where the electric field is highest; however, when fields exceed the second crossover point, detection of multipactor occurring beyond the center of the waveguide is necessarily impossible to measure directly. After the ring becomes detuned and the forward power decreases, multipactor may move back to the center of the waveguide where it may be then measured presenting as a time delay. In either case, the effects upon the tuning of the ring are still visible. More in-depth information and analysis is available elsewhere.⁵ With tuning, ∼6 kW was routinely achieved within the ring enabling threshold measurement of multipactor (Fig. 10). The microwave switch, cf. Fig. 1, is operated at 0 s with a pulse width of 100 µs. UV seeding with a 265 nm LED was used to initiate the multipactor event within the test section as unaided multipactor within the 5.5 mm gap used would require ∼10 kW. The UV LED is pulsed for ∼10 µs, and multipactor is typically seen within 5 µs–10 µs of the UV pulse being applied. Upon ignition of multipactor (Fig. 11), an increase in reverse power is observed, indicating a large reflection coefficient within the ring illustrating one possible detrimental result of the multipactor effect whereby the transmitted power is reflected back toward the source.

An RF source capable of at least 2 kW in the S-band was described with considerations necessary for reliability and protection of the relatively delicate GaN HEMTs. Offering high power density and efficiency compared to traditional vacuum tube systems and higher output power than LDMOS or GaAs, the pitfalls of depletion mode HEMTs—i.e., biasing/sequencing requirements and avalanche susceptibility—are evident. The source presented may be packaged into a compact form factor with all power supplies and associated hardware internal providing good power density compared to vacuum tube technologies. While efficiency was of little importance for this application, it should be noted that GaN HEMTs offer an efficient, linear solution to sourcing microwave power in the kW regime. Recommendations for VCO, LNA, switch, amplifiers, and power control system are provided with possible substitutions should the need arise. If a source of higher frequency is desired, one simply needs to source transistors at the required frequency—in addition to any other amplifiers in the chain—and adjust the splitter/combiner dimensions. It was demonstrated that multipactor can be stimulated using the source described here, although requiring a ring resonator—or other method of increasing effective power—for gaps larger than ∼2 mm.

This research was supported in part by a AFOSR - Air Force Office of Scientific Research MURI grant, No. FA9550-18-1-0062, “Multipactor and Breakdown Susceptibility and Mitigation in Space-Based RF Systems.”

Full schematics for the control board and biasing modules are available through the Texas Data Repository at https://doi.org/10.18738/T8/69H7EY.