By modulating the optical power of the light-emitting diode (LED) in accordance with the electrical source and using a photodetector to convert the corresponding optical variation back into electrical signals, visible light communication (VLC) has been developed to achieve lighting and communications simultaneously, and is now considered one of the promising technologies for handling the continuing increases in data demands, especially indoors, for next generation wireless broadband systems. During the process of electrical-to-optical conversion using LEDs in VLC, however, signal distortion occurs due to LED nonlinearity, resulting in VLC system performance degradation. Artificial neural networks (ANNs) are thought to be capable of achieving universal function approximation, which was the motivation for introducing ANN predistortion to compensate for LED nonlinearity in this paper. Without using additional training sequences, the related parameters in the proposed ANN predistorter can be adaptively updated, using a feedback replica of the original electrical source, to track the LED time-variant characteristics due to temperature variation and aging. Computer simulations and experimental implementation were carried out to evaluate and validate the performance of the proposed ANN predistorter against existing adaptive predistorter schemes, such as the normalized least mean square predistorter and the Chebyshev polynomial predistorter.

We demonstrate a triple-pass scheme for coherent transfer of optical frequency and the delay effect on the fiber phase noise compensation. It is theoretically proved that the delay effect consists of both fiber delay and servo delay. The delay effect confines the servo bandwidth within 1/8𝜏 and induces a residual fiber phase noise after noise compensation. For a 25-km-long fiber, the servo bandwidth is found to be around 1 kHz, and the fiber phase noise is suppressed approaching to the theoretical limitation. The triple-pass scheme enables the simultaneous transfer of optical frequency to multiple remote users. The performance of noise compensator in the triple-pass scheme can achieve a similar level result compared with that in the double-pass scheme.

No abstract

No abstract

Image segmentation can be viewed as an essential step for extracting information from the images un der investigation. Among many developed segmenta tion methods, the technique of clustering has been ex tensively studied. However, determining the number of clusters of an image is inherently a difficult problem, es pecially when a priori information on the image is un available. This study proposes a support vector ma chine approach for clustering images. To help deter mine the number of clusters, a greedy strategy is de signed which extends or condenses the number of clus ters by evaluating the clustering results from support vector machines. Comparisons on the effectiveness of the proposed method on various parameters settings are conducted. Experimental results are provided to illus trate the feasibility of the proposed approach.