University of Sheffield
Hardware-Efficient Automatic Modulation Classification and Blind SNR Estimation for Cognitive Radio Systems: A Novel DBSCAN-Based Approach with an Optimised Implementation
Abstract
dc:description.abstractArtificial Intelligence is a technology which has the potential to provide significant enhancements for digital communication systems, particularly through the concept of Cognitive Radio (CR). Automatic Modulation Classification (AMC) is a critical function of CR; it offers the ability to identify the modulation scheme of received signals to enable dynamic reconfiguration of physical layer hardware with the aim of maximising data rates and minimising error rates. Current research into AMC systems focuses primarily on achieving high classification accuracy, when the state-of-the-art algorithms are implemented in hardware, the resultant systems suffer from high utilisation and power consumption. To overcome this limitation, this thesis develops a novel, hardware-efficient AMC method based on the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm. Several novel optimisations to the DBSCAN algorithm are developed which improve hardware efficiency and overcome the inability of the algorithm to differentiate between same-order modulation schemes. Additionally, an automated heuristic method for hyperparameter selection is devised which results in up to a 9.8% increase in classification accuracy in comparison to traditional optimisation methods. Furthermore, a novel hardware implementation of insertion sort is proposed which enables real-time classification with low latency. The proposed optimisations result in a hardware implementation with approximately equivalent size to the state-of-the-art in terms of Flip-Flops and Look-Up Tables, as well as being 71.7% more power-efficient. This approach is also shown to achieve competitive, and in some cases superior, classification accuracy by achieving 100% accuracy at a Signal-to-Noise Ratio (SNR) as low as 10dB in certain cases. Finally, it is demonstrated that the same hardware architecture can be reused for the purpose of non-data-aided SNR estimation with competitive accuracy and is effective across a larger range of SNRs and modulation schemes than existing methods, further enhancing the efficiency of the proposed system.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Sheffield
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Gavin, Bill
- Advisors dc:contributor.advisor
-
- Ball, Edward
- Deng, Tiantai
- ORCID person:identifier.orcid
Rights
- Language dc:language.dcterms_iso639_2
- eng
Identifiers
dc:identifier.*- Dc Identifier Grantnumber
- MR/T043164/1