Emerged as an upgrade of currently available Surface Plasmon Resonance (SPR) biosensing in terms of sensitivity, phase-sensitive SPR technology still requires the minimization of instrumental noises to profit from its projected ultra-low detection limit (10(-8) refractive index units and lower). We present a polarimetry-based methodology for the efficient reduction of main instrumental noises in phase-sensitive measurements. The proposed approach employs a sinusoidal phase modulation of pumping light and is based on selection of proper modulation amplitude and initial phase relation for the first two modulation harmonics (F1 and F2), which enables to subtract amplitude drifts in the difference (F1 - F2) signal while doubling the phase response. The resulting effect can be called self-noise-filtering, since it implies an inherent noise subtraction in every phase sensing measurement. This methodology allows one to tackle drifts related to instabilities of light sources and optical elements and thus drastically lower the detection limit of phase-sensitive SPR sensing even in relatively simple and noisy experimental implementations.