正交分頻多工(Orthogonal Frequency-Division Multiplexing, OFDM)是一種多載波(multi-carrier)調變(modulation)的技術,其技術具備高速資料傳輸能力與良好的頻寬使用效率(bandwidth efficiency),以及對於多路徑(multi-path)衰減(fading)與延遲(delay)有較佳的穩定性,並且擁有能夠對抗頻率選擇性(frequency selective)通道的能力,所以其技術逐漸獲得重視與採用。在無線通訊系統中,其訊號在傳遞的過程中,因為容易受到環境影響所產生的多路徑效應,其將會使訊號失真(distortion)與衰減,以及由於傳送端與接收端之間有相對運動關係影響所造成的都卜勒效應(Doppler effect),其將會使訊號發生載波偏移(carrier offset)的現象。由此可知,對於通道特性的了解顯得很重要。由於為了要減少接收訊號受到的干擾,本論文使用改良式倒傳遞類神經網路(Back Propagation Neural Network, BPNN)來進行通道的估測與補償訊號。一般研究者隨意設定倒傳遞類神經網路架構與其學習速率(learning rate)參數,因此其效能並不理想。由於使用者設計不佳將會嚴重影響系統效能,因此本論文將提出改善倒傳遞類神經網路效能的方法。本研究是在瑞利衰減通道(Rayleigh fading channel)通道環境下,對於倒傳遞類神經網路的位元錯誤率(Bit Error Rate, BER)進行分析,並且分別探討領航(pilot)資料配置的方式、學習速率的選擇和領航資料配置的數量之效能。最後將本論文所提出的改良式倒傳遞類神經網路相較於傳統的最小平方法(Least-Square, LS)和最小均方誤差法(Minimum Mean-Square Error, MMSE)進行通道估測與補償訊號之效能,其研究目標是將本論文所提出的改良式倒傳遞類神經網路所估測的效能接近最小均方誤差法所估測的效能。
Orthogonal frequency division multiplexing (OFDM) is a multi-carrier modulation technique; it is competent to high-speed data transmission capability and good bandwidth efficiency, better stability for multi-path fading and delay, and has the ability to be able to resist the frequency selective channel, so that widely gains the attention and adoption. In mobile communication systems, the multi-path channel causes signal distortion and attenuation, and the relative motion between transmitter and receiver causes the Doppler effect that produces signal carrier offset. Therefore, knowledge of the channel characteristics is very important. To remove these effects from received signal, in this thesis, we use back-propagation neural network (BPNN) to estimate channel and compensate carrier offset. Several researchers use the BPNN architecture and set its learning rate parameter arbitrarily, so its performance is not very good. Since not exactly using the structure of BPNN, it would seriously affect system performance. This thesis proposes a method to improve the performance of BPNN for channel estimation in OFDM systems. The study on OFDM systems is in Rayleigh fading channel environments and we obtain the bit error rate performance of the proposed BPNN. The configuration and the number of the pilots and the selection of the learning rate are discussed in the thesis. Finally, we compare the performance of the proposed improved BPNN with the least squares (LS) and minimum mean square error (MMSE) algorithms. The purpose of this thesis is to improve the performance of optimization BPNN to approach the performance MMSE algorithm.