1：Experimental Design and Method Selection:

The study is split into two parts: in-vitro studies to determine glucose representative sub-bands using power spectral density analysis, and in-vivo studies using a self-designed PPG-based optical system for blood glucose prediction. Methods include transmission spectrum measurement, pre-processing with filters, segmentation, power spectral density calculation, correlation analysis, and multiple linear regression.

2：Sample Selection and Data Sources:

In-vitro: Tissue phantoms with different scattering coefficients prepared using intralipid solution and black Indian ink, with glucose concentrations from 0–1000 mg/dl. In-vivo: 24 healthy volunteers (20 male, 4 female, average age 30 ± 9.72), with blood glucose levels measured by a conventional glucometer as reference.

3：72), with blood glucose levels measured by a conventional glucometer as reference. List of Experimental Equipment and Materials:

3. List of Experimental Equipment and Materials: NIR spectrophotometer (NIRQuest512, Ocean Optics, USA), LEDs (LED1070L, LED1200L, LED1300L, LED1450L, LED1550L, LED1650L, Thorlabs, USA), InGaAs photodiode (SD039-151-011, API, USA), step motor, filters (high pass, notch, low pass), operational amplifier, analog-to-digital converter (USB6210, Nat. Inst., USA), Labview 2016 software, intralipid solution (ClinOleic 20%), black Indian ink (Higgins), conventional glucometer (GlucoDr, AGM 2200, Allmedicus, South Korea).

4：Experimental Procedures and Operational Workflow:

In-vitro: Measure transmission spectrums of phantoms, pre-process with moving average and baseline removal filters, segment into 100 nm sub-bands with Hanning window, calculate power spectral density using Welch method, analyze correlations. In-vivo: Use self-designed optical system to illuminate fingertip with LEDs sequentially, record PPG signals for 10s per LED, pre-process signals with Saviztky Golay and baseline removal filters, detect peaks and valleys, calculate optical absorption difference, perform multiple linear regression analysis.

5：Data Analysis Methods:

Correlation analysis for sub-band power and glucose concentrations, multiple linear regression with stepwise method for predicting blood glucose levels from optical density values, Clarke Error Grid analysis for performance evaluation.