MACHINE LEARNING BASED SPECTRUM SENSING FOR INTERFERENCE REDUCTION IN 5G COGNITIVE RADIO NETWORKS

Abstract

The rapid increase in global population has driven a surge in users of radio technology, leading to a shortage of available frequency spectrum for wireless systems. To optimize the use of limited spectrum, secondary (unlicensed) users can access the spectrum of primary (licensed) users when it is temporarily unused. These unused portions of spectrum are called spectrum holes or white spaces. Cognitive radios play a key role by performing spectrum sensing to detect when the spectrum is available for secondary users. Real-time spectrum detection is essential for allowing secondary users to access the spectrum without interfering with primary users. However, existing spectrum sensing methods often suffer from poor detection accuracy due to channel fading and noise. The work addressed the creation of machine learning-related algorithms that perform effective user classifications and spectrum sensing in 5G wireless cognitive radio networks to maximize spectrum usage and minimize interference. To classify users according to the patterns of activities and spectral behaviour, a hybrid sequential algorithm that merges Particle Swarm Optimization (PSO) and K-Means clustering were created to classify users. To perform spectrum sensing, a PSO-K Means-based algorithm has also been used to identify the spectrum holes through clustering sensed data in occupied and unoccupied frequency bands. This method employed PSO a population based optimization method to compute the initial centroids and give an optimal starting point in the clustering. K-means then grouped the sensed spectrum into two occupied and unoccupied. The hybrid PSO-K algorithm increased the primary user detection and allowed cognitive radios the opportunity to access the spectrum without resulting in interruptions. Extensive simulations in Python were conducted to evaluate the performance of the PSO-K algorithm in various 5G network scenarios. Results showed that PSO-K significantly outperformed traditional energy detection methods in terms of both performance and detection accuracy. The algorithm notably enhanced the probability of detection while reducing the probability of false alarms and missed detections. Analysis of detection accuracy demonstrated a 9.3% improvement compared to traditional energy detection techniques.