PAPR Control in Digital Predistortion Systems Using PC-CFR and RLS Forgetting Factor Adjustment

Modern wireless communication relies on a technique called orthogonal frequency-division multiplexing, or OFDM, to send vast amounts of data through the air. This method splits a signal into many smaller sub-signals that travel simultaneously, allowing for high-speed internet and clear video calls. However, a fundamental quirk of this technology is that these sub-signals occasionally line up perfectly, creating sudden, massive spikes in power. Imagine a calm river that suddenly surges with a towering wave every few seconds; this is the nature of the signal. These spikes, known as a high peak-to-average power ratio, create a significant problem for the hardware that amplifies the signal for transmission. If the amplifier is set to handle these rare, extreme peaks, it wastes a tremendous amount of energy running inefficiently most of the time. If it is set to run efficiently, the peaks get crushed, distorting the message and causing interference with neighboring radio channels.

To solve this, engineers use a two-step process to tame the signal before it leaves the transmitter. First, they trim the highest peaks just enough to keep the amplifier safe, a process called crest factor reduction. Then, they apply a digital correction, known as digital predistortion, which intentionally warps the signal in the opposite direction of the amplifier's natural distortion, so that the final output is clean and linear. The challenge lies in balancing these two steps. If the initial trimming is too aggressive, it damages the signal's shape. If the digital correction is not tuned correctly, it can cause the peaks to grow back, undoing the work of the first step. Finding the perfect setting for both processes simultaneously is difficult because changing one often negatively impacts the other, and the best setting depends on the specific conditions of the radio equipment at that moment.

Researchers at the Communication University of China investigated a method to automatically find this balance by coordinating the two stages. They focused on a specific type of peak trimming that uses carefully shaped pulses to cancel out the spikes without introducing too much noise, and a mathematical learning algorithm that adjusts the digital correction in real time. The team discovered that the settings for the learning algorithm play a critical role in whether the signal peaks grow back after correction. Specifically, they found that the algorithm's "memory"—how much weight it gives to past data versus new data—directly influences the height of the remaining peaks. A setting that remembers too little of the past causes the signal to fluctuate wildly, creating new peaks, while a setting that remembers too much might not adapt quickly enough to changing conditions.

The researchers developed a step-by-step procedure to test different combinations of these settings. They began by gradually increasing the amount of peak trimming until the signal quality, measured by how closely the output matched the original message, began to degrade. They stopped at the point just before the quality got worse, identifying the maximum amount of trimming that the system could handle without damaging the data. With that limit set, they then adjusted the memory of the learning algorithm to find the specific setting that kept the signal peaks as low as possible while maintaining the best possible signal quality. This approach treats the two controls not as independent knobs, but as a linked system where one is adjusted to create a safe environment for the other to work.

In their experiments, the team used a real-world radio setup with a powerful amplifier and a field-programmable gate array, a type of reconfigurable computer chip, to process the signals. Before applying their coordinated method, the system produced a signal with a peak-to-average power ratio of 12.985 dB, which exceeded their safety limit of 12.5 dB. By applying their new procedure, they successfully reduced the peak ratio to 11.593 dB, bringing it safely within the required range. More importantly, this reduction did not come at the cost of signal clarity. The quality of the corrected signal improved significantly, with the error between the intended and actual signal dropping by 3.660 dB. Additionally, the interference leaking into adjacent radio channels decreased, with the upper and lower channels improving by 2.554 dB and 2.416 dB respectively.

The study demonstrates that simply turning down the volume on the peaks is not enough; the digital correction must be tuned to match the level of trimming. The researchers found that pushing the trimming too far, even if it lowers the peaks further, actually harms the overall signal quality and does not necessarily improve the interference levels. Their method identifies the sweet spot where the signal is clean, the peaks are controlled, and the interference is minimized. By updating the system's safety threshold based on the new, lower peak levels, the procedure ensures that the transmitter remains efficient and stable for future operations. This work provides a practical way to manage the complex trade-offs in modern wireless transmitters, ensuring that high-speed data can be sent efficiently without wasting energy or causing interference.