Abstract:
Quantum Computing (QC) and Machine Learning (ML) have converged to build the
emerging field of Quantum Machine Learning (QML), a revolutionary area that combines
quantum computing and machine learning. The combination of quantum and classical
processing capability provides unmatched potential for tasks such as prediction, classi fication, and optimisation. Modulation classification is a critical task in various fields
including telecommunications and wireless technologies. This thesis explores the fu sion of Quantum Computing (QC) and Machine Learning through the lens of Quantum
Machine Learning (QML) to address the classification challenges created by synthetic,
channel-impaired waveforms. The modulation waveforms, containing multiple classes,
are generated in MATLAB. These waveforms are then converted to 2D images as quan tum techniques works well with images. We designed a quanvolutional neural network
(QNN) with strongly entangled layer in pennylane to decode complex patterns within the
data. To classify these extracted quantum feature maps, a custom Keras Convolutional
Neural Network (CNN) model is created within the Jupyter Notebook environment.
By using a novel approach — a hybrid quantum classical model, we are able to sig nificantly improve the classification accuracy of modulated signals in this work. Using
this methodology yields an impressive 93% validation accuracy. In comparison, data
without quantum pre-processing, when trained on a classical 2D CNN, achieves a vali dation accuracy of 73%. As a result, the quantum pre-processed model performs better
than its classical CNN equivalent, highlighting its great potential and performance in
modulation classification. One of the research’s distinctive characteristics is Quantum
Machine Learning’s capacity to handle large datasets with a small number of qubits.
This quantum advantage facilitates the processing of large-scale datasets, resulting in
higher performance even with a small number of training samples.
