Abstract
Provision of multimedia contents over the Internet of things (IoT) presents significant challenges to wireless networks owing to nodes diversity. Therefore, efficient utilization of resources is required to meet the growing diversity in user's behavior and wireless services that mark the future of wireless communication. Automatic classification of wireless signals based on their modulation schemes is a crucial technology that enables wireless transceivers to utilize the resources efficiently. Traditional feature-based approaches lack generalization, and versatility by relying on single classifier predictions. In this paper, two adaptive boosting (Adaboost) based wireless signal classifiers called SigmaBoost and KNNAdaboost are proposed. Adaboost generates an optimal prediction rule by combining the prediction of many weak component classifiers (CCs). However, sometimes it overfits on real-world scenarios with noisy data. That makes the choice of CC vital for its success in classifying wireless signals. Therefore, two well-known classifiers support vector machine (SVM) and k-nearest neighbor (KNN) are used as CCs in proposed schemes respectively. In SigmaBoost an adaptive decrementing mechanism is introduced, which decreases the value of Gaussian radial function (RBF) of SVM kernel as boosting progresses. It not only ensures the generation of weak RBFSVM component classifiers but also improves diversity across the ensemble. Signal spectral features and higher-order cumulants are used as input features. The experimental results demonstrate significant gain in the performance of SigmaBoost, as compared to KNNAdaboost and other single classifier-based approaches. Hence, SigmaBoost can be used in multimedia-enabled IoT for quick discrimination of wireless signals to ensure better radio spectrum management.
Introduction
The Multimedia Internet of Things (mIoT) represents a significant evolution in connectivity, supporting data-heavy applications such as smart cities, autonomous vehicles, and real-time health monitoring. These diverse applications require efficient spectrum management to handle the massive influx of multimedia content. A key technology for this is Automatic Modulation Classification (AMC), which allows transceivers to identify the modulation scheme of incoming signals blindly, without prior coordination.
Traditional AMC approaches often rely on single classifiers that struggle with generalization in noisy environments. Likelihood-based methods are accurate but computationally expensive, while feature-based methods using a single classifier often fail to handle the variability of real-world channels. This research proposes bridging this gap using Adaptive Boosting (AdaBoost) to combine multiple "weak" classifiers into a robust classification system.
Methodology
The proposed AMC framework follows a structured pipeline: pre-processing, feature extraction, and classification.
The core innovation is SigmaBoost, which utilizes Support Vector Machines (SVMs) with an adaptive decrementing mechanism. Typically, SVMs are "strong" learners, which can lead to overfitting in a boosting framework. To address this, the authors introduced a mechanism that starts with a large Gaussian radial basis function (RBF) kernel parameter ($\sigma$), intentionally "weakening" the initial SVMs. As boosting iterations proceed, $\sigma$ is systematically reduced, allowing the ensemble to capture increasingly complex patterns while maintaining diversity and avoiding overfitting.
The system uses a hybrid feature set to distinguish between various modulation schemes:
- Spectral Features: Five statistical measures of amplitude, phase, and frequency variance.
- Higher-Order Cumulants (HOC): 2nd and 4th-order statistical moments used to distinguish between complex modulation schemes like QAM16 and QAM64, which often appear similar under simpler metrics.
Experimental Results
The models were evaluated using a dataset mimicking harsh conditions, including Additive White Gaussian Noise (AWGN), multi-path fading, and Carrier Frequency Offset (CFO). Eight modulation schemes were tested: 2ASK, 4ASK, 2FSK, 4FSK, BPSK, QPSK, QAM16, and QAM64.
- Superior Accuracy: SigmaBoost achieved a remarkable 99% classification accuracy at an SNR of 3dB (using 4096 samples). In comparison, KNNAdaboost reached approximately 90% accuracy only at a much higher SNR of 15dB.
- Sample Efficiency: Even with short signal bursts (128 samples), SigmaBoost maintained 70% accuracy at 0dB SNR, demonstrating its suitability for rapid classification in dynamic environments.
- Robustness: SigmaBoost consistently outperformed standard single-classifier SVMs and Naive Bayes models across the entire SNR spectrum, approaching the performance limits of Maximum Likelihood methods with significantly lower computational overhead.
Conclusion
SigmaBoost provides a powerful solution for signal classification in the mIoT. By leveraging the strengths of ensemble learning and adaptive kernel tuning, it achieves high accuracy and robustness against channel impairments. While high-order QAM schemes remain challenging, the proposed approach offers a significant step forward in efficient radio spectrum management. Future work will explore real-world implementation on Software Defined Radios and further optimization for power-constrained IoT devices.
Summary updated