正交分頻多工(Orthogonal frequency-division multiplexing, OFDM)是一種多載波調變的技術,其具備高速資料傳輸能力與良好的頻寬使用效率,對於多路徑衰減與延遲有較佳的穩定性,且能對抗頻率選擇通道,所以逐漸獲得重視與採用。在無線通訊系統,訊號在傳遞的過程中因環境影響所產生的多路徑效應會使訊號失真與衰減,而因傳送端與接收端有相對運動所造成的都卜勒效應則會使訊號載波偏移。因此,對於通道特性的了解顯得很重要。而為了要減少接收訊號受到的干擾,我們透過通道估測來知道通道的脈衝響應(Channel impulse response, CIR),然後再進行補償。本篇論文使用倒傳遞類神經網路(Back Propagation Neural Network, BPNN)來進行通道的估測與補償,之所以使用類神經網路來做通道估測和補償是因為類神經網路在處理複雜的工作時,不需要針對問題定義複雜的數學模式,而是藉由學習來面對複雜的問題與不確定的環境。傳統的倒傳遞類神經網路是採用梯度坡降法(Gradient Descent Method)來找尋最佳解,進而更新權重值。梯度坡降法通常收斂至區域最佳解(Local Optimum),因而可能搜尋不到全域最佳解(Global Optimum)。因此本篇論文將利用基因演算法(Genetic Algorithm, GA)來取代梯度坡降法來搜尋最佳解。最後本篇論文將傳統倒傳遞類神經網路、基於基因演算法的倒傳遞類神經網路、最小平方法(Least-Square, LS)以及最小均方誤差法(Minimum Mean-Square Error, MMSE)這四種方法在目前現有的OFDM通道環境下比較之間的位元錯誤率(Bit Error Rate, BER)和均方錯誤率(Mean Square Error, MSE)。
Orthogonal frequency division multiplexing (OFDM) is a multicarrier modulation technique; it is competent to high-speed data transmission capability and good bandwidth efficiency, better stability for multipath fading and delay, and resistance frequency selective channel. That gradually gains the attention and adoption. In Wireless communication system, due to the environmental impact generated the multipath effect caucused signals distortion and attenuation in transmitted process, and due to relative motion between transmitter and receiver caused the Doppler effect that make the signal carrier offset, Therefore, knowledge of the channel characteristics are very important. To remove the effect from received signal, the receiver needs to have knowledge of channel impulse response (CIR) by channel estimation, and then compensates signals. In this paper, we use back propagation neural network (BPNN) to estimate channel and compensate signals. The reason of use neural network to do channel estimation and compensation is that the neural network does not require definition of the problem of complex mathematical models that deal with complex problem and uncertain environment by learning. The traditional back propagation neural network used gradient descent method to find the optimal solution, and then updated the weight values. Gradient descent method usually converges to a local optimum, which may not find the global optimum. Therefore, we used the genetic algorithm (GA) to replace the gradient descent method to search for the optimal solution in this paper. Finally, we compare bit error rate (BER) and mean square error (MSE) of our proposed GA-based back propagation neural network with that of traditional back propagation neural network, least square (LS) algorithm and minimum mean square error (MMSE) algorithm in existing OFDM channel environments.