No abstract

The widespread application of artificial neural networks has prompted researchers to experiment with FPGA and customized ASIC designs to speed up their computation. These implementation efforts have generally focused on weight multiplication and signal summation operations, and less on activation functions used in these applications. Yet, efficient hardware implementations of nonlinear activation functions like Exponential Linear Units (ELU), Scaled Exponential Linear Units (SELU), and Hyperbolic Tangent (tanh), are central to designing effective neural network accelerators, since these functions require lots of resources. In this paper, we explore efficient hardware implementations of activation functions using purely combinational circuits, with a focus on two widely used nonlinear activation functions, i.e., SELU and tanh. Our experiments demonstrate that neural networks are generally insensitive to the precision of the activation function. The results also prove that the proposed combinational circuit based approach is very efficient in terms of speed and area, with negligible accuracy loss on the MNIST, CIFAR-10 and IMAGENET benchmarks. Synopsys Design Compiler synthesis results show that circuit designs for tanh and SELU can save between 3.13 7.69 and 4.45 8.45 area compared to the LUT/memory based implementations, and can operate at 5.14GHz and 4.52GHz using the 28nm SVT library, respectively. The implementation is available at: https://github.com/ThomasMrY/ActivationFunctionDemo.

No abstract

A compact ultra-wideband (UWB) slot antenna based on a mesh-grid structure is designed. A genetic algorithm is used to optimize the mesh-grid structure as well as other parameters of the proposed antenna for good impedance matching in the UWB band. The optimized UWB antenna has a compact size of 24 mm  30 mm and is fabricated and measured. According to the measured results, the proposed antenna yields a wide bandwidth, defined by VSWR<2, ranging from 3.1 to 12.2 GHz and a nearly omnidirectional radiation pattern in the H-plane. The antenna gains within the matching band are measured and a gain variation from 3.1 to 5.9 dBi is obtained. Figure 5 Simulated and measured gain for the proposed antenna with and without an arrow patchABSTRACT: A procedure is described to enhance the accuracy of microwave measurements of the complex permittivity of a dissipative medium. Monopole probe measurements are used in conjunction with two real-valued neural networks, which are integrated together to reconstruct the complex permittivity from the measured reflection coefficients. The approach is tested over the frequency range from 2.5 to 5 GHz, for the real part of the permittivity in the range 3-10 and the imaginary part in the range 0-0.5. The performance of the network is also demonstrated for a reduced frequency range from 3.5 to 5 GHz. Less than 4% error was observed in the presence of white Gaussian noise with an SNR of 10dB.

No abstract