IMPROVEMENT OF TOPOLOGY CONTROL PERFORMANCE WITH NODE RANK AND REINFORCEMENT LEARNING METHOD ON CDMA WIRELESS SENSOR NETWORK

• Main Author: Widodo Heru Kurniawan, Dwi
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/53233
• Institution: Institut Teknologi Bandung

A high-capacity and energy-efficient wireless sensor network will be needed to accommodate the exponential traffic increase in the 5th and 6th generation (5G / 6G) communication systems or the internet of things (IoT) generation. Currently, wireless sensor networks apply the low energy adaptive clustering hierarchy (LEACH) algorithm, where the cluster head (CH) selection is randomized. As a result, the distance between member nodes to CH can be far apart so that the sensor will transmit power with high power, which will further cause interference and waste energy. The LEACH algorithm also uses a static power control so that the received signal to interference and noise ratio (SINR) signals in the CH can be decreased if the interference increases. The LEACH algorithm needs to be improved to increase network capacity and save energy. This dissertation develops a new algorithm by re-selecting CH, using the node rank method and reinforcement learning power control. With the node rank method, the method selects CH from the sensor with the most significant rank contributing from the largest number of neighboring nodes and high residual energy to balance energy consumption. The sensor determines neighboring nodes from nodes that have received signal strength indicator (RSSI) values above the current CH's average RSSI value. With the proposed method, the distance of member nodes to CH will shorten. The decreased distance will reduce the sensor's transmit power. Interference is also further suppressed by transmission power control techniques using a dual policy multi-agent reinforcement learning algorithm using SINR as a control parameter. The use of the reinforcement learning (RL) method will enable the sensor to track SINR fluctuations due to interference changes so that the transmit power of the sensor will be adjusted according to the minimum requirement. Testing the proposed node rank and reinforcement learning is carried out by computer simulation using a randomly wireless sensor network model with uniform distribution. Wireless network propagation channels use the free space path loss model (for short-range representation) and the two-ray ground channel model (for longer distance representation). The simulation shows that the proposed node rank algorithm performance is better than the conventional LEACH. The indicator iv improvements are a network lifetime growth by 8%, first node death (FND) growth by 9%, half node death (HND) growth by 14%, and delivered packets by 11%. The performance improvement is due to the formation of a closer distance between the node and the CH. The simulation also shows that the proposed power control is better than static power control used by conventional LEACH. The improvements consist of an increase in the network lifetime by 129%, FND by 52%, HND by 35%, and delivered packets by 162%. It can be explained that with the SINR fluctuation tracking capability, the sensor transmit power will meet the minimum requirement. In addition to suppressing interference, low (minimum) transmit power will also save energy consumption simultaneously. Controlling the transmit power using SINR as a control parameter gives better results than the signal strength (RSSI) because SINR involves the level of interference in controlling the sensor's transmit power.