Diabetes is a widespread and serious disease and noninvasive measurement has been in high demand. To address this problem, a power spectral density-based method was offered for determining glucose sensitive sub-bands in the nearinfrared (NIR) spectrum. The experiments were conducted using phantoms of different optical properties in-vitro conditions. The optical bands 1200-1300 nm and 2100-2200 nm were found feasible for measuring blood glucose. After that, a photoplethysmography (PPG)-based low cost and portable optical system was designed. It has six different NIR wavelength LEDs for illumination and an InGaAs photodiode for detection. Optical density values were calculated through the system and used as independent variables for multiple linear regression analysis. The results of blood glucose levels for 24 known healthy subjects showed that the optical system prediction was nearly 80% in the A zone and 20% in the B zone according to the Clarke Error Grid analysis. It was shown that a promising easy-use, continuous, and compact optical system had been designed.