الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 136 |  |  |
| | | from | 136 | | |
Abstract
This era is considered as the kingdom of Artificial Intelligence (AI) and Machine Learning (ML), specially into emerging technologies and signal processing.
In this thesis, the utilization of Machine Learning for wireless channel estimation has been investigated. For better accuracy, Deep Learning (DL) has been proposed to estimate the channel. The channel estimation method can be varied depending on the system. In this study, a multi-carrier system called Universal Filtered Multi-Carrier (UFMC) is applied. The usage of DL for channel estimation was proposed for UFMC systems. Furthermore, the odd-indexed frequency domain samples are proposed to be exploited to enhance the channel estimation in UFMC systems.
The DL model has been trained offline using dataset generated from MATLAB simulations for certain channel model and different Signal-to-Noise Ratio (SNR) regions. The metric of the validation stage is Normalized Mean-Squared Error (NMSE) between the perfect channel (actual channel coefficients) and the output from the DL model which is the enhanced channel estimation. The model then is deployed into UFMC system to measure its performance from Bit-Error-Rate (BER) point of view.
The proposed channel estimation technique is evaluated versus conventional channel estimation techniques, such as: Least-Squared (LS). The simulations showed that the proposed DL-aided channel estimation overperforms the conventional LS channel estimation. Moreover, the usage of the odd-indexed frequency domain samples provide enhancement to the channel estimation from the DL-based approach. The overall SNR gain for the proposed technique
is 5-6 dBs and 2-3 dBs on average for the NMSE and BER, respectively.
As a conclusion, the utilization of DL model and incorporation of odd-indexed samples are improving the channel estimation in UFMC system.