Overcoming the Challenges of mmWave, On-Wafer Load-Pull Measurements for 5G
Hybrid-active load-pull overcomes the challenges in mmWave power amplifier design by removing the uncertainty of unclosed contours to enable designing for peak performance.
Fifth-generation mobile represents the next evolution in wireless communications. With an emphasis on connectivity, 5G is expected to bring together data, voice, video, IoT, connected cars, smart homes, smart cities, augmented reality and industrial automation. 5G will achieve this aggressive goal by deploying technologies over multiple frequency bands, from low MHz to high GHz. Research in the 450 MHz to 6 GHz bands is targeting long-range communication and in the 28 to 30 and 37 to 39 GHz bands for high data rates. While posing unique challenges, the mmWave bands promise to bring many advantages, including larger bandwidth, greater capacity, increased security and longer battery life.
A critical enabler in the 5G infrastructure is the power amplifier (PA), which must be properly designed for optimum performance, i.e., maximizing power and efficiency while maintaining appropriate linearity. A useful design tool for maximizing performance is load-pull.
Load-pull is the process of changing the load impedance presented to a device under test (DUT), commonly a transistor, to measure its performance characteristics under varying large-signal conditions. The impedance is systematically changed while parameters such as output power, gain and efficiency are measured or calculated. Contours representing fixed performance values (e.g., x dBm output power or y percent efficiency) are then plotted to visualize the point of maximum performance, the rate at which the performance changes and trade-offs between various parameters (see Figure 1).
But how does load-pull work? First, consider a DUT as a two-port network (see Figure 2). A signal a1 is injected into port 1 of the DUT. A portion of the signal is delivered to the DUT while another portion is reflected as b1, due to the mismatch between the input impedance of the DUT and the source impedance of the input network. A modified signal b2 exits port 2 of the DUT and is delivered to the load, while a portion of it is reflected back as a2, due to the mismatch between the output impedance of the DUT and the load impedance of the output network. The magnitude and phase of that reflection, represented as ΓL, is
Load-pull changes the magnitude and phase of ΓL by changing the reflected signal a2. Any load impedance, which can be calculated as
can be presented to the DUT as long as the signal a2 can be achieved. There are two common methodologies to vary the impedance presented to the DUT: passive load-pull and active load-pull.
Passive load-pull uses mechanical impedance tuners to change the magnitude and phase of the reflected signal a2 and vary the impedance presented to the DUT (see Figure 3a). The magnitude and phase of the load impedance are adjusted by varying the position of a probe (or slug) in both x and y axes along a 50 Ω airline (see Figure 3b). The magnitude of the reflection is controlled by moving the probe vertically within the airline, while phase is controlled by moving the probe horizontally along the airline. By moving the probe up and down, left and right, it is possible to present nearly any impedance to the DUT, as long as the magnitude of a2 remains sufficiently large so that the desired
can be achieved. It is important to note that ΓL is less than 1, since a2 is always smaller than b2 due to losses between the output of the DUT and the tuner.
Open-loop active load-pull (see Figure 4) does not rely on a mechanical tuner to reflect part of b2 back as a2; rather, it uses a signal generator with magnitude and phase control to create a new signal a2. When amplified by an external amplifier, any a2 and, hence, any ΓL can be achieved. At first glance, active load-pull may seem superior to passive load-pull since it has no theoretical ΓL limitation; however, a practical limitation is the power required to achieve the signal a2 actually delivered to the output of the DUT. Active tuning has several advantages over passive tuning, including speed, as there are no mechanical moving parts, and increased Smith chart coverage, as a2 is directly generated, enabling
to be greater than 1.
The limitation is the maximum output power of the amplifier. Referring to Figure 4, the mismatch between the 50 Ω amplifier and the non-50 Ω DUT causes a portion of the signal to be reflected back toward the amplifier; the larger the mismatch, the larger the signal that is reflected. Under extremely mismatched conditions, it is possible that only 10 percent of the signal available will actually be delivered to the output of the DUT, requiring a large amplifier.
Hybrid-active load-pull overcomes this limitation by pre-matching the DUT impedance from highly mismatched to moderately mismatched, lowering the power required to deliver the same signal a2 to the output of the DUT.
When performing load-pull, it is preferable to be able to close the measurement contours to ensure the DUT’s maximum performance has been achieved. Without closed contours, it is possible for the optimum performance condition to be missed and the wrong conclusion formed.
With a passive load-pull system, the net magnitude of reflection achievable at the DUT reference plane can be calculated as
Assuming a typical tuner VSWR and coupler, cable and probe losses at 30 GHz, VSWRtuner = 20:1, ILcoupler+cable+probe = 2.5 dB, the maximum achievable magnitude of reflection is reduced from Γ = 0.9 at the tuner reference plane to Γ = 0.5 at the DUT reference plane. Modern GaN transistors have output impedances of 1 to 2 Ω, which can be represented by Γ values from 0.96 and 0.92, respectively. Figure 5 shows actual passive load-pull measurement data for a GaN transistor on-wafer at 30 GHz with a maximum output power of 30.66 dBm. Notice how the contours do not close, so it is uncertain how the transistor would perform if further tuning could be performed.
Hybrid-active load-pull overcomes this limitation in the passive load-pull’s measurement range by adding an active injection signal to increase a2 and, therefore, increase Γ. A commercial hybrid-active load-pull system is shown in Figure 6a and a typical test setup in Figure 6b. The relationship between the transistor, the system impedance, the injection power and the tuning range is
where ZL is the impedance presented to the DUT, ZSys is the system impedance and ZDUT is the DUT’s output impedance. K is defined as
where Pa2 is the active tuning power injected into the output of the DUT at the DUT reference plane, Pb2 is the DUT’s output power and Z0 = 50 Ω. The net reflection achievable at the DUT reference plane is
With a driver amplifier output of 40 dBm and using the same passive impedance tuner to transform the system impedance from 50 Ω to 23.17 + j28.12 Ω, it is possible to achieve Γ = 0.85 and successfully close the output power contours. The contours shown in Figure 7 demonstrate that a maximum output power of 31.12 dBm can be achieved by the same GaN transistor, which is 0.46 dB or approximately 11 percent more power than initially determined through passive load-pull with incomplete contours.
As companies accelerate development of 5G technologies and compete for best-in-class solutions, the optimization of power, efficiency and linearity will become more essential. Small advantages of a few dB in power or a few percentage points in efficiency may mean the difference between best-in-class and “never was.” Hybrid-active load-pull helps overcome the challenges in mmWave PA design by removing the uncertainty of unclosed contours. This enables ideal matching and gives those that adopt the methodology an edge in the marketplace.