Browsing by Subject "Analog computing"

Now showing 1 - 2 of 2

Open Access
Learning and inference for wireless communications applications using in-memory analog computing
(2024-07) Ali, Muhammad Atif
The exponential growth of wireless communication technologies has created a crucial need for more efficient and intelligent signal processing in decentralized devices and systems. Traditional digital computing architectures increasingly struggle to meet these rising computational demands, leading to performance bottlenecks and energy inefficiencies. The problem becomes more significant on edge devices with limited computing capabilities and severe energy limitations. Integrating machine learning algorithms with in-memory analog computing, specifically memristor-based architectures, provides a non-traditional computing paradigm and can potentially enhance the energy efficiency of edge devices. By leveraging the properties of memristors, which can perform both storage and computation, this research investigates ways to potentially reduce latency and power consumption in signal-processing tasks for wireless communications. This study examines memristor-based analog computing for deep learning and inference in three areas of (wireless) communications: cellular network traffic prediction, multi-sensor over-the-air inference for internet-of-things devices, and neural successive cancellation decoding for polar codes. The research includes the development of robust training techniques for memristive neural networks to cater for degraded performance due to noise in analog computations and offer acceptable prediction accuracy with reduced computational overhead for network traffic management. It explores in-memory computing for an Lp-norm inspired sensor fusion method with analog sensors and enables more efficient multi-sensor data fusion. Also, it investigates the incorporation of analog memristive computing in neural successive cancellation decoders for polar codes, which could lead to more energy-efficient decoding algorithms. The findings of the thesis suggest potential improvements in energy efficiency and provide insights into the benefits and limitations of using in-memory computing for wireless communication applications.
Embargo
Over-the-air multi-sensor inference with neural networks using memristor-based analog computing
(2024-12-05) Tegin, Büşra; Ali, Muhammad Atif; Duman, Tolga Mete
Deep neural networks provide reliable solutions for many classification and regression tasks; however, their application in real-time wireless systems with simple sensor networks is limited due to high energy consumption and significant bandwidth needs. This study proposes a multi-sensor wireless inference system with memristor-based analog computing. Given the sensors’ limited computational capabilities, the features from the network’s front end are transmitted to a central device where an 𝐿𝑝 -norm inspired approximation of the maximum operation is employed to achieve transformation-invariant features, enabling efficient overthe-air transmission. We also introduce a trainable over-the-air sensor fusion method based on 𝐿𝑝 -norm inspired combining function that customizes sensor fusion to match the network and sensor distribution characteristics, enhancing adaptability. To address the energy constraints of sensors, we utilize memristors, known for their energy-efficient in-memory computing, enabling analog-domain computations that reduce energy use and computational overhead in edge computing. This dual approach of memristors and 𝐿𝑝 -norm inspired sensor fusion fosters energy-efficient computational and transmission paradigms and serves as a practical energy-efficient solution with minimal performance loss.