The Convolutional Neural Network (CNN) has been used in many fields and has achieved remarkable results, such as image classification, face detection, and speech recognition. Compared to GPU (graphics processing unit) and ASIC, a FPGA (field programmable gate array)-based CNN accelerator has great advantages due to its low power consumption and reconfigurable property. However, FPGA’s extremely limited resources and CNN’s huge amount of parameters and computational complexity pose great challenges to the design. Based on the ZYNQ heterogeneous platform and the coordination of resource and bandwidth issues with the roofline model, the CNN accelerator we designed can accelerate both standard convolution and depthwise separable convolution with a high hardware resource rate. The accelerator can handle network layers of different scales through parameter configuration and maximizes bandwidth and achieves full pipelined by using a data stream interface and ping-pong on-chip cache. The experimental results show that the accelerator designed in this paper can achieve 17.11GOPS for 32bit floating point when it can also accelerate depthwise separable convolution, which has obvious advantages compared with other designs.

No abstract

This paper introduces an indoor positioning system based on visible light communication technology with three-dimensional positioning capability. Light-emitting diodes are employed as transmitters, with photodiodes as receivers to obtain the received signal strength (RSS) information. Based on the trilateration technique, the proposed algorithm is able to calculate horizontal coordinates of the receiver with RSS information, after which the height of the receiver is estimated. The system does not require other measurements such as time-of-arrival or angle-of-arrival, thus system design and costs are simplified and minimized. Basic framed slotted ALOHA is applied as the channel access method to enable asynchronous transmissions. In addition, Kalman and particle filters are used in order to realize target tracking. Results show that both filters help to increase the positioning accuracy and the particle filter exhibits a better performance than the Kalman filter, with a higher computational complexity.

In this letter, we propose an indoor visible light positioning technique that combines deep neural network based on the Bayesian Regularization (BR-DNN) with sparse diagonal training data set. Unlike other neural networks, which require a large number of training data points to locate accurately, we realize the high precision positioning with only 20 training points in a 1.8 m × 1.8 m × 2.1 m location area. Furthermore, we test a new optimization method of training data set, which is the diagonal set. To verify our ideas, we experimentally demonstrate three different training data acquisition methods that contain the common choice of training points (even set), arbitrary selection (arbitrary set), and diagonal selection (diagonal set). Experimental results show that the average localization accuracy optimized by the BR-DNN is 3.40 cm with the diagonal set, while the average localization accuracy is 4.35 cm for the arbitrary set and 4.58 cm for the even set. In addition, the training time and positioning time are only 11.25 and 8.66 ms due to a significant reduction of the sparse training data. All of the aforementioned experimental results show that the algorithm and training data optimization we proposed provide a new solution for real-time and high-accuracy positioning with the neural network.