Deterministic Sensor Signal Detection Technique for Static and Short-Ranged Channels Using Real-Valued Independent Component Analysis

Abstract

Periodic noise in active sensors is not random and cannot easily be eliminated by successive integration of signal samples. These noise components, often, tend to severely degrade the signal-to-noise ratio of the sensor signals, thereby resulting in detection failures. To facilitate effective reduction of non-random noises, the proposed study presents a digital front-end procedure based on modified independent component analysis (ICA). In comparison with conventional demodulation-based techniques, the proposed method can suppress periodic noise whilst enabling fine detection of the signal phase and magnitude through use of only a few samples. A new cost-effective two-channel ICA-based technique and its initialization method were devised for individual sensor devices. The proposed preprocessing method employing partial in-phase and quadrature demodulation strengthens system immunity against Gaussian random noises, which inevitably exist in most sensing circuits. This paper also discusses results obtained via implementation of a customized logic design for the full scenario of the proposed denoising technique. With regard to the capacitive and visible light sensors, experimental results demonstrate that errors caused by noise components that occur close to the stimulating frequency are suppressed at rates 3.2-5.3 times higher compared to existing processing techniques.