Traditional microwave transmitter systems can be simplified by replacing up-converting mixers and associated filters with a vector modulator that directly modulates data onto a carrier signal using orthogonal I and Q control inputs for simultaneous amplitude and phase control. The versatility of the vector modulator allows unlimited modulation control and methods, from simple BPSK or QPSK phase modulation schemes, to dual independent data rate modulations on the non-interfering I and Q channels, to Gaussian minimum shift keying schemes that fully utilize simultaneous amplitude and phase control. This article describes the architecture, design, control, layout and measured performance (24 to 44 GHz) of a versatile, broadband, high performance and high reliability Ka-band vector modulator MMIC targeted for space applications.
Depending on the frequency of operation, performance criteria and technology of choice, there are various ways to implement the vector modulator and its components. Before describing the design tradeoffs and optimization of the Ka-band MMIC, a review of the basic function of a vector modulator is appropriate. Figure 1 shows a typical vector modulator architecture with a 90° splitter, two attenuators and an in-phase combiner. A hybrid splitter divides the input signal into in-phase (I, 0°) and quadrature (Q, 90°) components. The attenuators control the amplitude and sign (±) of each I and Q path; ideally, the amplitude is between –1 and +1. The orthogonal vectors I and Q are combined by an in-phase summer to create a vector of any phase and amplitude. Direct vector control of the carrier allows simultaneous phase, amplitude and frequency modulations.
Since a quarter-wavelength line at Ka-band is relatively short in a GaAs MMIC (approximately 0.8 mm), simple distributed elements, such as Lange or Wilkinson couplers, can be used as components in the design. The input 90° splitter is easily implemented as a single Lange coupler. Likewise, the output summer can be implemented with a single Wilkinson combiner. The critical part of the design involves the two identical (I/Q) attenuators.
A simple reflective attenuator consists of two identical variable resistors and a Lange coupler (see Figure 2). With two identical PHEMTs used as variable resistors and connected to the through and coupled ports of the Lange coupler, the input and output matches will be good since the reflections from the resistors will cancel. The normal isolated port of the coupler becomes the through path receiving the summed reflections from the two equally biased PHEMT devices. Using an ideal resistance element, the reflective attenuator output is dependent on the range of resistance values around 50 ?, with 50 ? providing the minimum reflection. For example, a 10 ? resistance presents an ideal 3.5 dB insertion loss. This mirrors the 3.5 dB loss with a 180° phase shift of a 250 ? resistance. Unfortunately, the parasitic capacitances of the PHEMT devices, especially at high Ka-band frequencies, make it difficult to achieve an ideal resistance for the reflective attenuator. Compensating the parasitic capacitor with an inductor is generally too narrowband a solution for typical MMIC wafer processing variations.
A balanced attenuator is a broadband design consisting of four Lange couplers, two of which form complementary reflective attenuators for cancellation of the PHEMT capacitance (see Figure 3). Canceling the parasitic capacitance of the variable resistances makes the balanced attenuator broadband and less sensitive to MMIC wafer processing variations, such as varying threshold voltages or device characteristics. Figure 4 illustrates the balanced sum of the two complementary reflective attenuators, each consisting of a Lange coupler and two matched PHEMT devices. One reflective attenuator is biased for a particular voltage, for instance ON, and the other attenuator is biased at a complementary voltage, OFF. Two more Lange couplers are connected to sum one reflective attenuator and its mirror image reflective attenuator. The resulting vector is a controlled attenuation along a line with a 180° phase reversal on either side of the minimum attenuation.
The magnitudes of the in-phase and quadrature output signals of the input Lange splitter, adjusted with the balanced attenuators, are summed with a simple output Wilkinson combiner, which allows magnitude and phase modulation of the RF carrier. The balanced vector modulator architecture is very broadband and robust against MMIC processing variations.1,2
Optimizing the Design
While small flaws or anomalies in the vector modulator can be compensated for in the I and Q control signals, it is best to optimize the linearity and performance of the initial design. Maximizing bandwidth, minimizing insertion loss, maintaining good input and output match, and minimizing phase and amplitude imbalances, are some important criteria for the design tradeoffs.
Since overall insertion loss is less problematic than unbalanced losses, it is best to minimize phase and amplitude imbalances in the various components. Using symmetry in the physical design maintains phase balance and eliminates amplitude imbalances due to undesired parasitic coupling. Imbalances in the couplers, poor match (VSWR) (especially across various control states), poor isolation and imbalances in the matched PHEMTs of the attenuator can be sources of error. An equal power split in the Lange coupler is important both for the initial input phase splitter and for linearity in the two balanced attenuators. The Sonnet EM simulator was used to verify that undesired coupling between layout components was minimal, as well as to verify the layouts of the Lange and Wilkinson couplers. Fortunately, the PHEMT devices within an MMIC tend to be extremely well matched such that they add or cancel appropriately in the balanced attenuator design.
The penalty for having an extremely broadband and versatile vector modulator architecture is a moderate inherent insertion loss. When a Wilkinson coupler is used as a power combiner, its insertion loss is minimal when adding two coherent signals (that is, in-phase). In the vector modulator, the Wilkinson combiner sums the outputs of the balanced I and Q attenuators, facilitates good impedance match, and provides isolation for the I and Q ports. Since the two vectors are orthogonal (90°), there will be a 3 dB insertion loss for an ideal Wilkinson combiner. For the general-purpose vector modulator, one loses another 3 dB of insertion loss due to the RF input Lange splitter, as illustrated in Figure 5. The vector control region resembles a square where the corners are used for a BPSK or QPSK modulation with the I and Q vectors at a maximum adding together for minimal insertion loss. In the general case, represented by a circle, another 3 dB is lost when I or Q is maximum while its orthogonal vector is a minimum. In addition to the inherent 3 dB minimum — or 6 dB general — insertion loss for an ideal vector modulator, there are additional microstrip losses, losses due to the PHEMT devices and loss due to imperfections in the design.
The largest source of controllable loss in the vector modulator is the choice of the PHEMT devices. A device that is too small will have a high ON resistance that limits its reflection (that is, attenuation). Likewise, a device that is too large will have too much capacitance, also limiting the attenuation range. There is an ideal device size for a particular MMIC process and design frequency range. Additional losses such as microstrip losses or variations of the PHEMT characteristics across multiple wafer fabrications are expected to be small and can be compensated for at the system level in the amplifier chain and the I and Q vector modulator control inputs.
Optimizing the Control
While many applications may not need a rigorous idealization of the vector modulator, many small nonlinearities or variations across frequency, wafer processing, or design imbalances can be compensated for by optimizing the I and Q input controls based on measured performance. The following is a simple illustration of optimizing the complementary balanced attenuator inputs (I/IN or Q/QN) based on ideal Lange couplers combined with measured 0.15 mm × 150 mm PHEMT data. Realizing that there are multiple combinations of I/I- and Q/Q- to achieve a given amplitude and phase, a linear attenuation along a straight line on the polar (Smith) chart will be considered the optimal goal.
The Smith chart in Figure 6 illustrates the ideal paired complementary control signals (I/IN) for a balanced attenuator assuming four ideal equal split (that is, 3 dB) Lange couplers and four matched PHEMT devices. The “A” curve shows the reflection of a grounded PHEMT device measured from –0.5 to +0.5 V. For this GaAs process, the 150 mm PHEMT device is a good compromise between ON resistance and OFF capacitance. The canceling of the parasitic capacitance of the devices to achieve the ideal straight line (AA) results from combining the switch reflection (A curve) with a mirror image of itself (A* curve). For minimum insertion loss, a good choice is to use the voltage that brings the closest approach to the 50 ? center of the Smith chart. For this example, a bias of –0.27 V for I (A) and IN (A*) would be ideal. While any combination of equal I and IN voltages will produce a minimum, the ideal minimum bias point (that is, the closest approach to 50 ?) will depend on the threshold voltage and the particular device characteristics, both its resistance and capacitance. For the minimum insertion loss (–2.5 dB in this example), applying +0.5 V to I (A) mirrors the magnitude and phase of approximately –0.4 V applied to IN (A*). Likewise, applying –0.5 V to I (A) and approximately –0.4 V to IN (A*) results in the same minimal insertion loss but with a 180° phase shift. Ideal complementary control voltages for I and IN to achieve the straight-line vector attenuation (AA) are shown in the data plot. As one device and its mirror image approach the extreme ON or OFF position, the magnitude change is least sensitive to control voltage changes, while for minimum magnitudes it is most sensitive to control voltage changes. If a little phase skew from the ideal line is allowed, a slightly lower insertion loss (–1.7 dB) can be achieved if –0.5 V is applied in combination with +0.5 V for the complementary control inputs (I/IN). While this example illustrates a way of conceptualizing the ideal complementary control voltages, optimal control for critical applications should be determined based on measured performance.
Making the vector modulator design as linear and symmetric as possible will simplify the complementary control signal requirements. Applying complementary control signals similar to the curves illustrated may minimize small imbalances of the vector modulator over frequency and wafer fabrications. However, there are multiple combinations of complementary control signals to achieve a given vector, making the balanced design and its control broadband versatile and robust to device processing variations.
The MMIC was laid out to fit a standard 2 × 2 mm GaAs die for the TriQuint 0.15 mm PHEMT 3MI process prototype chip option (PCO). In the PCO, the full wafer cost is shared by a multi-project (that is, multi-customer) wafer fabrication, so die size choices are limited. While more complicated compaction of the design could be accomplished,3,4 the current design is simple and fits in the available die size (see Figure 7). Coplanar-to-microstrip launches, at the input (left) and output (right), allow for wafer probe testing as well as wire bond connections to the die. An important part of the layout is to maintain symmetry and phase balance as well as maintaining 50 ? impedance matches. The input Lange splitter is followed by symmetric 50 ? microstrip transitions tapering down to the two balanced I and Q attenuators. Each balanced attenuator consists of four Lange couplers, two in a splitter/combiner pair and two in the reflective attenuators. Complementary inputs (I/I- or Q/Q-) drive two pairs of matched PHEMT devices for each balanced attenuator. Finally, a compact folded Wilkinson combiner (right) sums the I and Q balanced attenuators to generate the RF output. The layout was performed with Agilent’s ADS tool using the TriQuint 0.15 mm 3MI PHEMT library. TriQuint provided final design rule checks (DRC).
The Ka-band vector modulator MMIC shows good performance over a broad frequency range (24 to 44 GHz). It was tested on a Cascade Microtech probe station with an Agilent 8510 network analyzer to measure S-parameters over a range of complementary I/I- and Q/Q-DC voltages. A typical array, measured at 32 GHz, shows a 10.4 dB insertion loss (S21) for a general-purpose vector modulator (see the center circle in the plot of Figure 8). The insertion loss from 24 to 44 GHz is less than 10.9 dB. As a BPSK or QPSK modulator, the insertion loss (S21) is typically 6 to 7 dB. Figure 9 shows the broadband input and output match from 24 to 44 GHz.
Using a commercially available TriQuint 0.15 mm PHEMT GaAs process, a balanced Ka-band MMIC vector modulator was produced with good performance, excellent bandwidth and with an architecture that makes it robust to MMIC processing variations. The TriQuint foundry was chosen for its high reliability for space-qualified MMICs, its low cost prototyping option (PCO), and the availability of test and inspection services. Results for the 2 ¥ 2 mm GaAs MMIC were very good with 10.4 to 10.9 dB insertion loss from 24 to 44 GHz and good input/output match. Two quadrature complementary control signals are required to modulate the amplitude and phase of this direct modulator device. The power consumption is minimal (mW) as the only active devices are eight 0.15 × 150 mm PHEMT switches.
The author would like to acknowledge the help and support of Narayan Mysoor, Brian Cook, Leon Alkalai, Bud Lovich, Samad Hayti and Elizabeth Kolawa of NASA’s Jet Propulsion Laboratories (JPL). Co-workers at the Johns Hopkins University Applied Physics Laboratory who have helped with this project include Paul Ostdiek, Barry Tossman, Mark Bernacik, Jon Bruzzi, Bob Bokulic, Bob Wallis and others. Everyone at TriQuint is always helpful, particularly Lisa Howard, the foundry manager. This work was conducted by the JHU/APL under NASA/JPL Contract #1243213 Mars Technology Program.
1. D.S. McPherson and S. Lucyszyn, “Vector Modulator for W-band Software Radar Techniques,” IEEE Transactions on Microwave Theory and Techniques, Vol. 49, No. 8, August 2001, pp. 1451–1461.
2. C.Y. Ng, M. Chongcheawchamnan and I.D. Robertson, “A Balanced Vector Modulator for LMDS Applications,” 6th IEEE High Frequency Postgraduate Student Colloquium Digest, 9–10 September 2001,
3. A.E. Ashtiani, T. Gokdemir, A. Vilches, Z. Hu, I.D. Robertson and S.P. Marsh, “Monolithic GaAs-InGaP HBT Balanced Vector Modulators for Millimeter-wave Wireless Systems,” IEEE Radio Frequency Integrated Circuits (RFIC) Symposium Digest, 11–13 June 2000, pp. 187–190.
4. C.Y. Ng, M. Chongcheawchamnan and I.D. Robertson, “Miniature Ka-band I/Q Vector Modulator Using 3D-MMIC Technology,” 33rd European Microwave Conference Digest, Vol. 2, 7–9 October 2003,
John E. Penn works in the space department of the Applied Physics Laboratory and teaches a MMIC design course at the Johns Hopkins University.