Pattern Detectioin System For Massive Spectral Signals From Optical Biosensors

Porous silicon microcavity can be used as an important optical biosensors. Microcavity structure will limit the incident light with resonance wavelength, thereby forming a high quality factor resonant valley in the reflectance spectrum. We can get different reflectance spectrum by injecting the test solution into the porous silicon pores. Since the movement of resonant valley on the reflectance spectrum is proportional to the concentration of the solution, we can distinguish different test solution by the red shift.Optical biosensors based on porous silicon microcavity have been applied to environmental monitoring and medical testing. At present, researches about the application of optical biosensors based on porous silicon microcavity are limited to detect and analyze small amount of data from a single sensor locally. Practical application scenarios such as real-time environmental monitoring and remote medical testing will generate massive signal data from sensors without the ability to analyze data. There is no related research to give a solution about this condition nowadays.In this paper, a pattern detection system for massive spectral data based on cloud computing was proposed. This system separates the data acquisition from data processing. Remote clients are responsible for collecting data with sensors and the server will make uniform detection and analysis of data. Server uses FPGA as a signal processing unit to make quick pattern detections. FPGA can offer computational parallelism, high computational speed and low power dissipation. The system can achieve massive data processing capacity with PCI Express to provide high transmission bandwidth. The signal processing accelerator implemented in FPGA uses a FIFO-based streaming architecture integrating multiple pattern detection modules as data processing channels. This structure has a low hardware cost and good scalability. Pattern detection module which was designed and implemented based on the pattern detection algorithm of one-dimensional spectral data is responsible for signal processing with a dynamically configurable pattern. By modifying configuration registers, the pattern detection can be used for data with different data length and SNR. The whole system has good versatility, scalability and computational efficiency, providing convenience and support for further analysis.