TL;DR
This paper presents an FPGA-based system for real-time pre-processing of self-mixing interferometry (SMI) signals. By combining median and wavelet filtering with a normalization step, the system effectively reduces noise and prepares SMI signals for accurate displacement and parameter extraction, crucial for high-quality sensing applications.
What Problem Does This Paper Address?
Self-mixing interferometry (SMI) is a powerful technique for measuring displacement and other parameters using a laser diode with optical feedback. However, the raw SMI signals are often corrupted by noise, making precise information retrieval challenging. Existing pre-processing methods, while effective, can sometimes alter critical signal features like sharp transitions, impacting accuracy. For industrial applications requiring real-time, high-quality sensing, a compact and efficient processing solution is needed, moving beyond PC-based analysis.
Methods
The authors propose an FPGA-based pre-processing system that integrates two key stages: de-noising and normalization.
- De-noising: This stage employs a two-pronged approach:
- Median Filter: A 5-point median filter is used to address "sparkle-like" noise, which typically affects a small number of sampling points. The implementation involves a series of compare-swappers to sort the input data and select the median value.
- Wavelet Transform Filter: This method, previously developed by the authors, is implemented to remove other noise components. It involves discrete wavelet transform, single-level reconstruction, spatial filtering of high-frequency content (preserving sharp transitions), and recombination to produce a processed signal. The core filtering blocks utilize Distributed Arithmetic (DA) for efficient hardware implementation.
- Normalization: After filtering, the signal is normalized to a range of [-1, 1]. This is achieved by removing the direct current (DC) component, effectively centering the signal around zero. The process involves calculating the minimum and maximum values of the filtered signal and scaling it accordingly.
The entire design is implemented using System Generator for DSP, which allows for high-level modeling in Simulink and conversion to Verilog HDL for FPGA deployment.
Datasets
Not discussed in the paper. The paper focuses on the design and simulation of the FPGA system using simulated and experimental SMI signals.
Key Insights and Ideas
- Hybrid Filtering for Robustness: Combining a median filter for impulse noise with a wavelet filter for broader noise reduction offers a more robust de-noising strategy for SMI signals.
- Preserving Signal Integrity: The wavelet filter's ability to spatially filter high-frequency content while preserving sharp transitions is critical for maintaining the accuracy of displacement measurements derived from SMI signals.
- FPGA for Real-Time Industrial Sensing: Leveraging FPGAs provides a compact, high-speed, and integrated solution for signal processing, making SMI systems more viable for real-time industrial applications where PC-based processing is impractical.
- DA for Efficient FIR Filtering: Using Distributed Arithmetic for FIR filter implementation significantly reduces hardware resource usage and improves processing efficiency on FPGAs.
Key Results
The paper demonstrates the effectiveness of the proposed FPGA-based pre-processing system through hardware co-simulation. Figure 7 in the paper shows:
- (a) Original Experimental SMI Signal: A noisy SMI signal with clear fringe patterns.
- (b) Filtered SMI Signal: The signal after applying both the median and wavelet filters, showing a significant reduction in noise.
- (c) Normalized SMI Signal: The filtered signal scaled to the range of [-1, 1], ready for further analysis.
The results indicate that the FPGA-based pre-processing effectively filters SMI signals and normalizes them, confirming the system's performance for practical SMI applications.
Limitations and Future Work
- Limitations: The paper primarily relies on hardware co-simulation to verify the design. While this is a strong indicator of performance, actual hardware implementation and testing on a physical SMI system would provide further validation. The specific performance metrics (e.g., processing speed in Hz, power consumption) are not explicitly detailed.
- Future Work: The authors suggest that this FPGA-based pre-processing method can be integrated into a compact SMI system for real-time sensing applications. Further optimization of the FPGA design for even higher performance or lower power consumption could be explored.
Summary updated