RF PoL: A Special Variety of Point-of-Load Converters

兰博

RF PoL: A Special Variety of Point-of-Load Converters [Copy link]

The air around us is a sea of radio emissions from tens of kHz to tens of GHz, with transmitter power levels ranging from milliwatts to megawatts for the largest “long wave” broadcasters. We’re probably all familiar with Wi-Fi, Bluetooth, Zigbee, and the many RF-based TV remote controls like car keys, but the blast zone is 5G, which typically uses the 2.5 – 3.7 GHz band and extends up to 70 GHz at mmWave. This is driven not only by a surge in personal cell phone users around the world, but also by Internet of Things (IoT) connections in home, commercial, and industrial settings. Self-driving cars will also join the fun, with reports predicting that by 2025, total traffic from connected devices will be around 160 exabytes, or more than 1 million per square kilometer in some places.

The potential data rates of up to 10-50 Gb/s promised by 5G are enabled by higher operating frequencies compared to earlier standards, but this comes at the expense of range, which is about 1.5 km at 70 GHz. This means that all the associated infrastructure needs more "cells". 5G cells are divided into coverage categories, from highest power to lowest power, with correspondingly smaller ranges: "metro", "micro", "pico" and "femto". Metro cells using multiple-input multiple-output (MIMO) technology transmit at powers in excess of 100 W, while femto cells operate in the milliwatt range.

All units contain an RF power amplifier

A common feature of all cell types is the RF power amplifier, or “PA” in Figure 1. This provides the power amplification required to drive the antenna by receiving a small RF signal, thereby achieving the highest efficiency and radiated signal levels. At 4G and lower frequencies, LDMOS transistors have been commonly used and are capable of delivering kW-level outputs, but for the higher frequencies and lower power requirements of smaller, distributed 5G cells, Gallium Nitride (GaN) devices are now preferred due to their lower losses, large voltage handling capabilities, and excellent thermal performance.

Figure 1: Typical cell radio section outline

In Figure 1, the “power” of the PA stage is shown simplistically, but it is a critical factor for optimal efficiency and spectral purity of the transmission. LDMOS PAs typically require 26-32 V and GaN 28-65 V, and the supply must be able to respond to rapid load changes without significant overshoot and undershoot, while maintaining its nominal voltage accurately. The load varies from transmit to transmit, as the transmit power is modulated by the data and dynamically adjusted for interference, energy, and connection management. Even with GaN, RF PA efficiency is only around 60%, so for larger “metro” cells, the PA supply may need to be able to provide over 200 W or 4 A at 50 V. The power supply will always be switching-mode types, and therefore will generate its own noise, which must be at very low levels,

Given all these limitations, one approach that can be taken is to use a non-isolated point-of-load regulator, or RF PoL. PoLs may be more familiar than CPUs generating sub-1 V rails at tens or hundreds of amps, but RF PA applications have the same requirements for load voltage accuracy, with fast dynamic response and low noise. However, an RF PoL with high output voltage is designed differently than the CPU version. It may be a similar topology, typically a buck converter, but the higher output voltage directly affects the achievable efficiency, device ratings, and control loop design. In particular, the higher voltage swings and fast switching edges required for high efficiency can introduce noise issues.

Example of evaluating RF PoL

A suitable RF PoL is available, such as Qorvo’s ACT43850. This is a buck controller using wide bandgap (WBG) transistors with an input range of up to 150 V, an output programmable from 20 to 55 V, and a current rating of up to 20 A. A feature of the Qorvo device is its high surge current capability, exceeding 20 A, which is well suited for the peak power needs of the RF PoL. Evaluating the performance of such a part includes checking the control loop stability for a given load, which can be done by measuring the response time and any overshoot or undershoot during transient loads. 0 to 20 A+ is not a realistic load condition, so monitoring the output voltage transient with this large signal loop response can produce misleading results. A more realistic setup to show the true small signal response, verifying control loop stability, is to apply a higher current and superimpose smaller load steps, such as a 20 A vs. 2 A step, with a custom test arrangement. In practice, since 20 A is the peak rating of the part and cannot be maintained for thermal reasons, a step size of 0 A – 20 A – 22 A – 20 A -0 A can be applied at an appropriate duty cycle to limit the temperature rise. Figure 2 shows example results using these values on a Qorvo RF PoL.

Figure 2: Load transient test of RF PoL

The output voltage is in yellow, here a 0 V – 50 V – 0 V pulse, and the red curve is the output current. This does show a decreasing slope from the initial 20 A, but that is the effect of the AC coupled probe. In reality it is a constant 20 A, with a 2 A step in the middle of the graph, then a drop to 20 A, then to zero. The output voltage transient for the 2 A step in blue is about 20 s in duration, and the excursion is about 150 mV, or just +/-0.3% of Vout, which is a very trustworthy performance, with +/-5% being more typical for low voltage PoLs. The loop response time of this particular RFPoL is so fast that one could even consider implementing “envelope tracking” via its digital interface. A demonstration of the large signal response can be seen when the load is switched to 0 – 20 A and 20 A – 0 A, with the excursions being far from symmetrical. On a positive load step it is underdamped but fast, while on a negative step it responds more damped but slower to the unusual “kink” in the waveform. This indicates that the control loop is operating outside its linear range, which may temporarily saturate the error amplifier and then require time to re-establish the correct operating bias. In the case of the Qorvo part tested, this has synchronous rectification, so the asymmetry is not caused by a change in the buck operating mode or discontinuous and continuous conduction. It responds more damped but slower to the unusual "kink" in the waveform. This indicates that the control loop is operating outside its linear range, which may temporarily saturate the error amplifier and then require time to re-establish the correct operating bias. In the case of the Qorvo part tested, this has synchronous rectification, so the asymmetry is not caused by a change in the buck operating mode or discontinuous and continuous conduction. It responds more damped but slower to the unusual "kink" in the waveform. This indicates that the control loop is operating outside its linear range, which may temporarily saturate the error amplifier and then require time to re-establish the correct operating bias. In the case of the Qorvo part tested, this has synchronous rectification, so the asymmetry is not caused by a change in the buck operating mode or discontinuous and continuous conduction.

Output noise is critical in RF PoL applications

Accurately measuring high-frequency noise on the output of a switching regulator is always difficult. Common-mode and differential-mode components combine to pick up distortion results. For RF PA applications, what actually matters is the purity of the transmitted signal and the associated level of "spurs" - extraneous line emissions around the carrier frequency. Therefore, measuring the converter's spectral noise density under actual operating conditions using a spectrum analyzer is a better comparison metric. The RF PoL running at full peak power must be AC-coupled to the typical 50-ohm analyzer input. This is done, for example, by using a passive high-frequency, high-impedance probe with near-zero capacitance to avoid noise peaks caused by circuit loading and probe resonance.

The probe may typically have a x20 attenuation, so a preamplifier is needed to significantly boost the signal above the spectrum analyzer noise floor and match its 50 ohm impedance. The cables between the probe, preamplifier, and analyzer should be verified and included in any calibration procedures for the test setup. This can be facilitated by a known, accurate noise source to ensure that the results are trustworthy. A suitable calibration and test setup is depicted and shown in Figure 3.

Figure 3: Accurate noise measurement test setup for RF PoL

Typical results are shown in Figure 4, where the noise floor from the test setup is superimposed, showing good test margins.

Figure 4a/4b: Noise spectral power density plot of a typical RF PoL

Although unrelated to the curves of Figure 4, the typical impact of RFPoL noise on the RF transmit spectrum can be seen in the example of Figure 5, where a PoL switching frequency “spur” is introduced at 500 kHz, approximately 67 dBm below the multiple lower carrier.

Figure 5: Typical battery output spectrum showing RF PoL switching noise “spurs”.

The PoL noise output can be modified through clock frequency “dithering” techniques to spread and reduce peaks in the noise spectrum – a feature that can be remotely enabled in the Qorvo PoL via I2C and the GUI. Synchronizing the PoL clock to an external signal can also avoid undefined “bouncing” effects.

in conclusion

The DC power supply at the RF PA level is critical to achieve optimal performance. When verifying the load transient response and output noise level, a point-of-load converter designed for this application can be a good solution. There are other potential applications for RF PoL in areas such as medical, measurement, laser power supply, etc.