正交分頻多工(Orthogonal Frequency Division Multiplexing,OFDM)技術有著高效率頻寬效益以及多路徑通道的穩定傳輸數據，使這項技術成為現代無線通訊不可或缺的其一技術。然而，OFDM技術本身具有高峰值對均值功率比(Peak-to-Average Power Ratio,PAPR)的問題，造成功率放大器呈現非線性輸出失真現象，進而使調變訊號產生頻譜再生(Spectral Regeneration)而干擾鄰近通道的訊號品質。為了解決此問題，便提出在OFDM訊號經過功率放大器前，讓訊號先經過數位預失真器(Digital Pre-Distorter)處理，此技術讓OFDM訊號在經過功率放大器後仍然呈現線性輸出，而考慮到寬頻系統存在記憶性的問題，亦需要在預失真器裡加入記憶性參數。在本論文的系統架構中，將功率放大器表示成記憶性沙雷(Saleh)模型；預失真器則使用記憶性多項式做為預失真模型。本論文提出使用類神經網路技術於間接學習架構預失真器，透過實驗來比較過去文獻常用的自適應演算法和神經網路應用於數位預失真的效能表現。;Orthogonal Frequency Division Multiplexing (OFDM) technology has become indispensable in modern wireless communication systems because of its high- efficiency bandwidth and high transmission stability in multi-path channel environments. However, OFDM technology itself has the problem of high peak-to-average power ratio (PAPR), which causes the output of the power amplifier to exhibit nonlinear gain distortion, which in turn causes the modulation signal to produce spectral regeneration (Spectral Regeneration) and interfere with the signal transmitted by the adjacent channel. In order to solve this problem, before the OFDM signal passes through the power amplifier, the signal is processed by a Digital pre-distorter. This technology allows the OFDM signal to still show a linear gain output curve after passing through the power amplifier. Considering the memory problem of broadband systems, it is also necessary to add memory parameters to the Pre-distorter. In the system architecture of this paper, the power amplifier is represented as a memory Saleh model and the Pre-distorter uses a memory polynomial as the model. By comparing with the Pre-distortion technology in the past literature, this paper proposes to use the neural network technology to indirectly learn the architecture of Pre-distorter parameter adaptive iteration, which is more suitable for application scenarios and accelerates the convergence speed, thereby achieving a more ideal power Amplifier linearization compensation technology.
