在本論文中，將針對高頻無線通訊中的類比語音攪拌加密訊號以基因演算法從其聲紋所具有的特性，看成解方格拼圖的問題還原其原始訊號。實驗中所使用的語音訊號源取樣量化與攪拌模擬，乃至於各類數值與演算法運算，均在x86個人電腦上完成，使用的程式語言為C++，編譯器為Borland C++ Builder；在考量處理速度之下，我們採用8 bits的ADC並將二維的聲紋圖依攪拌使用的四個頻段接續轉為一維的問題處理，而被攪拌的聲紋單元邊界像素被用來做為解攪拌接圖的特徵元素；在一筆語音信號檔中，不含有能量的區間或含有較少能量的聲紋圖框無法完整以拼圖解攪拌還原是可以被理解的，我們對不同能量比例的聲紋圖框做了一系列的實驗統計，在能量佈滿整個聲紋圖框的解攪拌結果上，幾乎都呈現有九成以上的還原率。而且在基因演算法中疊代的世代次數並不用太高即可收斂完成，使用的時間在此NP-Complete的問題上相對並不會太久。從語音訊號的分析、攪拌的原理與模擬、基因演算法的模塑應用與實作結果統計，都在本論文中有詳細的論述。 In this paper, we will reconstruct an analog scramble-secured speech signal, which usually used in HF (High Frequency) radio transmission, using the feature of sonogram of it and treat it as solving a jigsaw puzzle by genetic algorithm (GA). In our experiment, including signal sampling, scrambling simulation, mathematical computing and algorithm applying were all accomplished at x86 PC platform, and with programming language of C++, compiler of Borland C++ Builder. An 8 bits ADC is used by reason of process speed. We also take 2-dimension sonogram as 1-dimension issue by connecting 4 frequency bands which used for scrambling. The boundary pixel of each scrambling element sonogram is used as main feature for descrambling, or reconstruction. In a speech signal source stream, it can be realized that if none of speech energy or small amount of energy located in a process frame is not possible to be fully reconstructed. We have done a series of experimental and statistical analysis according to different amount of energy in one frame. It shows that above 90% reconstruction rate if speech energy spread all over one processing frame. And it also shows that there are not many generations past after GA process to become a converge state. The speech signal analysis, analog scrambling method, scrambling/descrambling simulation and program implementation will all be illustrated in this paper in detail.
