REINFORCEMENT LEARNING AND Q LEARNING FOR RESOURCE ALLOCATION IN MIMO NETWORK WITH INTELLIGENT REFLECTIVE SURFACES

In the last decade, modern wireless communication system is widely developed to enhance the channel performance and overcome the issue of fading. However, with new infrastructures of modern cites, the issue of multipath reflections from buildings and obstacles has become a significant problem that reduces the quality of the signal and the bit error rate. In such circumstances, many paths to the signal between the user and the node could be generated that limit the accuracy of modulation/ de-modulation process. Therefore, it is a subject of this research to invoke a hybrid technique based on combing two intelligent algorithms of R-learning and Q-learning processes. Such technique could be applicable to modern wireless communication systems based on IRS and MIMO networks. In this matter the use of R-learning starts the action to determine wither is desired according to the criteria of signal quality or not. The Q-learning comes to localize the user with best action through generating a criterion based on minimum BER response. It is observed from such combination; a significant reduction is achieved in the BER of the received signal that almost realizes an enclosed resonance to the ideal case without noise or fading effects. Later, an analytical study is introduced by increasing the number of IRS elements as hardware solution to be compared with the achieved results from the previous solution. It is found that a significant enhancement is accomplished in the signal quality with increasing the number of IRS elements, however, the system complexity is increased rapidly. This gives an indication on how much such algorithms is an effective solution to maintain minimum hardware complexity with low cost.