Custom instrumentation helps build models for more advanced RF amplifiers

High peak-to-average ratios inherent in the modulation schemes for 4G and those being put forward for future 5G systems are driving the circuitry controlling RF power amplifiers to become more modeling-oriented. These techniques help avoid the need to use expensive devices and deal with the problems caused by power supplies and heat.

During a dual-keynote session at NI Week in Austin, Texas, researchers Zoya Popovic of the University of Colorado at Boulder and Christian Fager of Chalmers University in Sweden described some of the techniques that may be needed to improve the efficient of RF power amplifiers and how these changes will drive the test equipment needed to characterize devices and circuits.

Some of the techniques, such as Doherty amplification and envelope tracking or supply modulation, have already made it into the market although there are still improvements to be made, said Popovic. She pointed to the sensitivity of envelope tracking to power-supply quality. “If you want to change output voltage by 1 per cent what does that mean in terms of how precise the supply needs to be? It needs to be very precise at such a high bandwidth that they just can’t make one, although there are tricks that people play,” she explained.

Worth the complexity

“Is all the complexity worth it?” she asked. An analysis of one design showed that components could run 86 per cent cooler, greatly increasing efficiency. “So, yes, once you have done the work, the complexity is worth it.”

Popovic described another emerging technique: outphasing. Working on the basis that amplifiers operate at maximum efficiency when saturated, this technique uses phase changes to adjust output amplitude according to the sum of a vector in the complex domain using two fully saturated matched amplifiers. This Class F amplifiers produces a lot of harmonics. “So you need to understand what happens to the harmonics and you have to characterize in a very non-linear way. Unfortunately, non-linear models are often not good enough when you are talking about modulating carriers,” she claimed.

In order to perform network analysis, Popovic’s team built a custom large-signal network analyzer (LSNA) – in fact, a combination of two LSNAs. “LSNAs are difficult to buy,” she said.

To deal with the high signal rates involved, the LSNA needs to use sub-sampling of the signal, which naturally produces a number of aliases in the spectrum. “So we need post-processing to deal with that, which NI hardware is good at,” she said, pointing to the team’s used of modular instruments produced by the instrumentation company. The instrument has allowed the team to look at complex interactions between PAs in an outphasing configuration among others. “There are things that go on with a non-isolated combiner, so we measured that,” Popovic said. “We would also like to combine this with the AWR [RF EDA] tools so we can do hardware-in-the-loop type experiments.”

New semiconductor materials

New semiconductor materials put forward for use in cellular such as gallium nitride bring additional issues to consider, said Fager. “There are heating and trapping and detrapping issues with some of these new semiconductors. Because linearity must always be met and at higher speeds, we need to work on faster techniques to adapt the system. The issue is that tranditional offline analysis techniques do not help much with analysis. In general, research on realtime adaptation is lacking in my opinion.”

Fager said traditional bench-top instruments can be used to look at the problems but they tend to be slow to use, demanding that the user set up a condition, capture it and then program in a new state – emulating the behavior of a supply-modulated device at speeds that are orders of magnitude slower than real time. “This is doable but also very cumbersome. And it’s not really the best way to do it,” said Fager.

Fager described a system used by Chalmers based on the vector-signal transceiver (VST) CompactPXI instrument which uses a combination of processor and user-programmable FPGA to control the experiment and process measurements. “This allows measurement in real-time. We have two loops: one operating at 120MHz that deals with the transmission and reception of data; and another loop that runs at slightly lower speed that takes care of the adaptation. With this we can investigate interesting effects.”

The analysis has suggested new, model-based supply modulation algorithms.

Model-based control

“Today adaptation is very much a wait-and-see approach. We see what comes out of the PA and say ‘oops, we’d better do something about that’. A delay between error and recovery is inevitable with this approach. I call it reactive control. But if you have a way of estimating how a change will happen you can pre-empt the error.

“In an amplifier, if you increase input power it should come as no surprise that the temperature will rise. So, we used a model of the temperature change: a model that predicts from input data what will happen to the amplifier,” said Fager. “With GaN power amplifiers we have problems when the temperature changes.

“We have observed long-term memory effects due to trapping in GaN devices. LDMOS stabilises quickly. But GaN, when the power is reduced, it takes a long time to be restored to its previous performance. That’s not a surprise to people in the semiconductor industry but it plays a big role in the linearity of RF power amplifiers,” Fager explained. “That is something that an adaption loop can deal with by using the time constant of the change in performance under temperature changes as part of the model. With an improved model we can improve the speed of linearization.”