Share
تطبيق الشبكة العصبية في تكوين الشعاع المتكيف لنظام الهوائي الذكي == Neural Network Application In Adaptive Beamforming For Smart Antenna System
Author name:
سليمان احمد غازي
Supervisor name:
عضيد حسن سلومي
General topic:
Electrical, Electronic and Communications Engineering
Specific topic:
Electrical Engineering
Degree:
Master
University:
Mustansiriyah University - College Of Engineering - Department Of Electrical Engineering
Language:
English
University location:
Baghdad
First pages:
34T458 - p.pdf
Abstract:
انظمة الهوائيات الذكية تعمل على تحسين الاداء وزيادة في السعة لانظمة الاتصالات المتنقلة. ويمكن لانظمة الهوائيات الذكية حل مشكلة نظام الاتصالات المتنقلة مثل التدخل في نفس القناة، والتاخير الحاصل نتيجة الانتشار والمسارات المتعددة التي تاخذها الاشارة عن طريق | The smart antenna systems are promise to provide performance improvement and capacity increasing for the base station of mobile communication systems. Smart antenna systems can solve the problem of mobile communication system such as co - channel interference , delay spread and multipath by an advance signal processing technology called beamforming. In this work artificial neural networks (Feed Forward Neural Network (FFNN) and feedback Elman Recurrent Neural Network (ERNN) ) are used for smart antenna adaptive beamforming for one and multiple targets.Neural network is used to calculate the optimum weights of antenna array to adapt the radiation pattern of the antenna array by directing multiple narrow beams toward the desired users and nulling interference or unwanted users. FFNN and ERNN are trained by supervised backpropagation learning algorithms, FFNN was trained by using Levenberg - Marquardt (lm),Resilient Back - Propagation (Rprop), Gradient descent with momentum and adaptive learning rate (gdx), Gradient descent and adaptive learning rate (gda) and biasing regulation(br), while lm and Rprop are used to train ERNN..The simulation results show that the best performance of smart antenna adaptive beamforming for one and multiple targets is obtained when the NN (FFNN and ERNN) trained by lm algorithm as compared with other algorithms, since it consider the fastest supervised training algorithm but with more memory requirement. Where the performance of FFNN training phase based (lm) algorithm for single target for five element uniform linear array antenna is [2.746641e - 14] at epoch 27 with best validation performance and best test performance approximately equal to the best training performance. While The performance of ERNN training phase based lm algorithm for single target is [1.121938e - 14] at epoch 38 with best validation performance equal to [1.682442e - 14] and the best test performance is [3.363946e - 14].The neural network model in adaptive beamforming are compared with smart antenna adaptive beamforming based on Least Mean Square (LMS) algorithm, and gave better performance than LMS.
