Summary The main objective of this study is to develop real‐time, energy‐efficient embedded computing for machine learning based target detection. Convolutional neural network (CNN) model based detection, a machine learning technique, can provide higher detection accuracy than constant false alarm rate (CFAR) detection techniques even if it results in higher processing costs. In this study, we achieve three significant improvements for real‐time radar target detection by considering computational cost. The first improvement is to reduce the computational cost of the CNN model. The second achievement is the design of heterogeneous computing optimizations. The third of them is to support energy‐efficient computing solutions for mobile sensors. Compared to the initial CNN model, layer improvements decreased the number of operations by 8.5x. Real‐time operations are satisfied by hardware‐specific improvements like vectorization and parallelization. The embedded NVIDIA Jetson GPU and Intel MYRIAD VPU, which have power consumption of 15 Watts and 1 Watt, respectively, have been used to execute the energy‐efficient target detection. The most energy‐efficient solution is achieved by using Jetson AGX Xavier GPU with 32‐bit single precision and 15 Watts of power consumption.

This study purposes a real-time computing of deep learning-based multi-target detection in defense-purpose edge sensors. Our study suggests two fundamental optimizations to accelerate target detection inference model: algebraic enhancements and post-training quantization. Comprehensive benchmark results show that our computing design achieves real-time multi-target detection on energy-efficient edge devices.

Summary In the direction of arrival (DoA) estimation, typically sensor arrays are used where the number of required sensors can be large depending on the application. With the help of compressed sensing (CS), hardware complexity of the sensor array system can be reduced since reliable estimations are possible by using the compressed measurements where the compression is done by measurement matrices. After the compression, DoAs are reconstructed by using sparsity promoting algorithms such as alternating direction method of multipliers (ADMM). For the given procedure, both the measurement matrix design and the reconstruction algorithm may include computationally intensive operations, which are addressed in this study. The presented simulation results imply the feasibility of the system in real‐time processing with energy efficient implementations. We propose employing parallel programming to satisfy the real‐time processing requirements. While the measurement matrix design has been accelerated 16 with CPU based parallel version with respect to the fastest serial implementation, ADMM based DoA estimation has been improved 1.1 with GPU based parallel version compared to the fastest CPU parallel implementation. In addition, we achieved, to the best of our knowledge, the first energy‐efficient real‐time DoA estimation on embedded Jetson GPGPUs in 15 W power consumption without affecting the DoA accuracy performance.

Summary Wideband software‐defined radio (SDR) applications include data and time intensive operations such as wideband spectrum, signal detection, digital down conversion (DDC), and analog demodulation. Each of these processes need to be performed in order to produce the audio signal from the wideband signals. However, serial implementation of SDR applications do not provide necessary rate speed to obtain the sound and speech data in real time. In this work, we propose a real time SDR implementation using a heterogeneous architecture with CPUs and GPUs. To obtain the sound and speech data from the signals received and processed in SDR algorithms in real time, we also provide necessary optimizations. We test the proposed design using both single CPU core, multiple CPU cores, and GPUs. High performance is observed with our proposed algorithm in experimental tests. We also provide test results in a mobile setup where resources such as power and size are limited. For this purpose, we provide test results on NVIDIA Jetson GPUs and portable laptop CPUs. This work shows a proof of concept that the sound of a signal can be detected in the wideband spectrum and can be played back continuously in a real time with the help of the parallel programming suitable for low power consumption.

The purpose of this study is to calculate the inter-carrier interference in visible light communications due to clipping in Asymmetrically Clipping Optical - Orthogonal Frequency Division Multiplexing (ACO-OFDM) signal and also to examine the distribution of this interference. Busgang's model, a technique widely used in the literature, can only provide statistical information about the interference caused by clipping. The method proposed in this paper can numerically determine the instantaneous value of the interference. Taylor series expansion is used to approximate the actual interference caused by clipping and satisfactory performance is achieved.