Advanced Signal Processing Algorithms for IoT Devices: Real-Time Audio Classification with Resource-Constrained VLSI Architecture

Abstract: This study focuses on the investigation of enhancement of VLSI architectures for supporting real-time audio classification in IoT systems. Audio classification is considered an important element of IoT because there are a lot of cases when the detection of sound is absolutely crucial for the efficient functioning of the IoT system. Since IoT devices are restricted in terms of power supply, memory size, and computational power, this report preliminarily reviews dedicated VLSI architectures sufficient to meet these requirements. It also constrains the study to VLSI choices like ASICs, FPGAs and DSPs where each of the choices has its merits. For instance, ASICs provide power and specific application traditional solutions, FPGAs offer flexible prototyping and DSPs stand out as optimal for repetitive signal related operations. MFCC’s and STFT are pointed out to be critical feature extraction roles for efficient, accurate audio representation and low complexity with the help of power optimization techniques as clock gating and dynamic voltage scaling. Nevertheless, some existing issues have not been solved, such as making better tradeoff between the latency and accuracy and coping with the scaling requirements of the new applications. Future trends such as neuromorphic computing and TinyML have potential in improving efficiency and biologically inspired signal processing with advanced hardware software codesign are expected to lower power consumption and expand the scope of VLSI. In summary, improvements on VLSI for IoT application such as audio classification make real time processing feasible meaning that IoT devices may infiltrate more fields. Innovative architectures, as well as emerging technologies are quickly enhancing the horizons of these capabilities, as this report shows.