基於符號補償的新型數位預失真架構

Abstract

在現代通訊系統中，功率放大器（ Power Amplifier, PA）作為一個核心元件，扮演著放大輸入信號以達成必要傳輸功率的重要角色。然而，由於物理限制及設計挑戰，功率放大器常常表現出非線性特性，進而導致訊號失真，並損害系統性能。為了克服這個問題，數位預失真（ Digital Predistortion， DPD）技術在功率放大器的線性化策略中被廣泛運用。數位預失真技術主要可以劃分為記憶性數位預失真與非記憶性數位預失真兩類。記憶性數位預失真在模型準確度上優於非記憶性數位預失真，但是其需要大量的計算資源來實現。因此，對於那些資源有限或注重計算效率的應用場景，非記憶性數位預失真提供了一個有效的解決方案。基於查找表（ Look-Up Table， LUT）的非記憶性數位預失真，是當中一種主流的方法，能克服功率放大器的非線性特性，且相比於記憶性方法，可以明顯地減少計算資源的需求。

從過去的研究中發現在查表後再添加一個濾波器能更接近具有記憶性的結果。然而，由於信號是透過過取樣（ Oversampling）基頻符號 (symbol) 生成的，因此若在查表後才加入濾波器將需要大量乘法器，換言之，會消耗大量運算資源。因此，在本篇論文中，為了降低資源使用量，我們提出在基頻符號生成階段對信號進行振幅和相位的補償，藉此達成資源節省的目標。

In modern communication systems, the Power Amplifier (PA) is a core component, playing a critical role in amplifying input signals to achieve the necessary transmission power. However, due to physical limitations and design challenges, power amplifiers often exhibit nonlinear characteristics, leading to signal distortion and degrading system performance. To overcome this issue, Digital Predistortion (DPD) techniques have been widely employed in linearization strategies for power amplifiers. DPD techniques can be mainly divided into memory-based and memoryless types. While memory-based DPD offers superior model accuracy compared to memoryless DPD, it requires significant computational resources to implement. Therefore, for scenarios with limited resources or a focus on computational efficiency, memoryless DPD provides an effective solution. Memoryless DPD based on Look-Up Tables (LUT) is a mainstream approach among these, capable of overcoming the nonlinear characteristics of power amplifiers and significantly reducing the computational resource demand compared to memory-based methods.

In previous research, we found that adding a filter after the lookup table brings results closer to those with memory. However, since the signal is generated by oversampling the baseband symbols, adding a filter after the lookup table would require a large number of multipliers, in other words, consume a considerable amount of computational resources. Therefore, in this paper, to reduce resource utilization, we propose compensating for amplitude and phase during the baseband symbol generation stage, thereby achieving our goal of resource conservation.