Interferometers are the basis for precision optical measurements, but their performance is often limited by their sensitivity to disturbances. Kamyar Behrouzi, Tanveer Ahmed Siddique, and Megan Teng, along with colleagues at the University of California, Berkeley and Lawrence Berkeley National Laboratory, have been tackling this challenge with a new approach to broadband interferometry. Their research introduces a technique called generalized polarization common-path interferometry (GPCPI) powered by artificial intelligence to separate polarizations and improve phase stability. This innovation improves phase stability by orders of magnitude, paving the way for more accurate measurements in areas such as molecular diagnostics and drug discovery. By combining advanced optical techniques and deep learning models, the research team has demonstrated that normal and cancerous skin cells can be distinguished at the single-cell level, providing a powerful new tool for disease diagnosis.
AI powers robust common-path interferometer measurements
Interferometric techniques are essential for extracting phase information from optical systems, allowing accurate measurements of dispersion and sensitive detection of perturbations. Although phase sensing provides increased sensitivity compared to traditional spectroscopy, this sensitivity often makes the system more vulnerable to external factors such as vibrations, resulting in instability and noise. In this study, researchers demonstrate a broadband, AI-enhanced interferometry method called generalized polarization common-path interferometry (GPCPI) that relaxes polarization constraints commonly found in traditional common-path interferometry. This approach utilizes a combination of polarization diversity methods and machine learning algorithms to reduce the effects of environmental disturbances and improve measurement stability.
By adopting a common path configuration, GPCPI is resistant to external vibration and noise, achieves high sensitivity, and minimizes the influence of airflow and mechanical vibration on interference signals. Additionally, the method incorporates a polarization diversity method to capture multiple polarization states of the interfering beam to improve signal quality and reduce sensitivity to polarization drift. A key element of this approach is the implementation of machine learning algorithms trained to identify and suppress residual noise, further improving the signal-to-noise ratio. This study contributes to a new interferometer setup that combines the advantages of common-path interferometry with advanced signal processing techniques. The GPCPI method demonstrates a wideband operating range from 650nm to 1650nm and achieves a phase noise of 2.34 picoradians/√Hz at 1kHz. This research establishes the foundation for developing more robust and sensitive interferometric sensors for a variety of applications such as environmental monitoring, biomedical diagnostics, and high-precision metrology.
Design and setup of a polarization-independent common-path interferometer
This research paper details a new approach to quantitative phase and dispersion measurements using polarization-independent common-path interferometry. This system overcomes limitations of traditional methods, such as sensitivity to polarization, complex setup, or difficulty achieving high accuracy and stability. Existing interferometric techniques can be sensitive to environmental noise and require precise alignment. The researchers developed a common-path interferometer that features polarization independence, simplifying the setup and improving robustness. The design also incorporates a broadband light source, allowing simultaneous phase and dispersion measurements.
The ConvNeXt V2 deep learning model is employed for robust fringe analysis, allowing accurate phase extraction and unwrapping, even for noisy or incomplete fringe patterns. Accurately extract complex refractive index and dispersion properties from measured phase data using vector fitting. The performance of the system was verified by measuring the refractive index and dispersion of known materials (ethylene glycol and water solutions) and comparing the results with established values. The researchers also demonstrated the system's ability to measure the dispersion of metamaterials. This technique can be applied to a wide range of materials, including transparent media, metamaterials, and biological samples, and has potential applications in materials characterization, biosensing, metamaterials research, testing of optical components, and real-time monitoring of physical properties.
GPCPI demonstrates 10x improvement in phase stability
Scientists achieved a 10-fold improvement in phase stability using a new broadband interferometry method called generalized polarization common-path interferometry (GPCPI). This breakthrough relaxes the traditional polarization constraints of common-path interferometry, allowing simultaneous measurements of amplitude and phase. The researchers measured phase patterns subjected to external shocks and found that GPCPI significantly reduced fluctuations compared to state-of-the-art interferometry techniques. Experiments demonstrated that the conventional method reduced contrast by 87% immediately after a vertical impact, whereas GPCPI only reduced contrast by 50% with a shock decay time of 1.6 seconds.
Further analysis included recording the normalized contrast of the phase pattern for 14 seconds and showed a standard deviation of approximately 31% for the conventional method and 14% for the GPCPI during the non-shock period, confirming the improved stability of the new approach. Changes in the vertical phase pattern at successive timestamps revealed minimal changes for the GPCPI method and maintained a consistent pattern even when traditional methods showed large changes. Research extends to plasmonic metasurface-based refractive index sensing. Scientists created a plasmonic metasurface made of nanorods and encapsulated it inside a custom flow cell.
A bulk refractive index change was induced by flowing a mixture of water and ethylene glycol, and measurements using GPCPI were successful in detecting a minimum concentration of 20%, corresponding to a refractive index change of 0.02, based on transmittance measurements only. The designed plasmonic sensor achieved a sensitivity of approximately 1400 nm per refractive index unit. GPCPI integration with the ConvNeXt V2 deep learning model, which consists of 28 million parameters pre-trained on the ImageNet dataset, enables single-shot, real-time tracking of phase variations with minimal noise. This combination enables reliable cell classification and disease diagnosis at the single-cell level, allowing the differentiation of normal and cancerous skin cells through interference pattern analysis.
GPCPI enables stable single-cell dispersion imaging
Researchers have developed a new interferometry technique, generalized polarization common path interferometry (GPCPI), that significantly improves the stability and accuracy of broadband phase measurements. By relaxing the polarization constraints of traditional methods and incorporating deep learning algorithms, specifically a customized ConvNeXt V2 model, GPCPI achieves orders of magnitude improvement in phase stability while simultaneously measuring both amplitude and phase. This advancement enables real-time, noise-suppressed phase sensing even when the sample exhibits arbitrary polarization. The capabilities of this technique were demonstrated through characterization of plasmonic metasurfaces and, importantly, through hyperspectral single-cell dispersion imaging.
Analysis of interference fringes allows reliable classification of skin cells to distinguish between normal and cancerous types, providing a platform for biological research and medical diagnosis. The authors acknowledge that a limitation is that the performance of deep learning models depends on the quality and diversity of the training data. Future research will focus on expanding the application of GPCPI to areas such as molecular diagnostics, drug discovery, and quantum sensing, based on its proven versatility and wide applicability to optical metrology.
