الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 155 |  |  |
| | | from | 155 | | |
Abstract
Optical wireless communication is a promising technology due to the huge unregulated bandwidth in the optical band. In adaptive optical communication systems, modulation formats and data rates are changed continuously according to channel conditions and customers’ needs. Due to the spectral efficiency requirements, a reduction of peer-to-peer information between the transmitter and the receiver is very necessary. This can be achieved through blind modulation format recognition (MFR) schemes.
This thesis is mainly concerned with efficient blind MFR based on techniques such as Hough transform (HT), orthogonal triangular-matrix decomposition (OTD), deep learning (DL), transfer learning (TL), and chaotic encryption. The HT is used to project constellation diagrams onto another space for efficient feature extraction. For different modulation classification schemes, constellation diagrams are obtained at optical signal-to-noise ratios (OSNRs) ranging from 5 to 30 dB. Another scheme involves decomposing each of the acquired constellation images into an orthogonal matrix (Q) and an upper triangular matrix (R), and then the HT algorithm is applied. For constellation diagrams, these two proposed schemes provide unique signatures. Different deep learning (DL) classifiers based on convolutional neural networks (CNNs) are used for the MFR task. For each modulation format, a study of the effect of changing the number of samples, phase noise (PN), and decimation on the accuracy of the classifiers is presented. The obtained results reveal that the proposed scheme succeeds in identifying the wireless optical modulation format blindly with a classification accuracy up to 100%, with low number of samples, even at low OSNR values less than 10 dB.
Different deep-tuned CNN-based transfer learning (TL) models are constructed and utilized to effectively classify modulation constellation diagrams with minimal computations and achieve maximum identification accuracy. Accuracy, loss, precision, recall, F1-score, confusion matrix, precision and recall curve, and ROC curve were used to assess the model. All of the examined evaluation metrics show that the created multi-classification framework outperforms all other traditional techniques. The suggested framework classification accuracy increased to 98.9% with the deep-tunned CNN-based HT model, which is deemed superior to standard classifiers. This is attributed to the usage of a well-developed HT method and CNN-based TL models.
Moreover, a scheme for automatic modulation format classification through the utilization of the constellation images with chaotic Baker map (CBM), wavelet image fusion, and correlation score as a classification metric is introduced. The CBM is introduced for image pixel permutation. Image fusion is used for fusing several permuted constellation diagrams of the same modulation format into one reference image for comparison using a correlation strategy. The proposed scheme was evaluated using equal error rate (EER) based on false acceptance rate (FAR) and false rejection rate (FRR), area under the receiver operating characteristic (ROC) curve (AROC), and decidability. The modulation classification simulation scenario resembles that utilized in cancelable biometric systems. Simulation and comparison results obtained for the proposed scheme ensure high AROC values, low EER values, and high decidability values. The results prove good classification performance for all studied modulation formats.