The High Frequency Instrument of Planck will map the entire sky in the millimeter and sub-millimeter domain from 100 to 857 GHz with unprecedented sensitivity to polarization ($\Delta P/T_{\tiny cmb} \sim 4\cdot 10^{-6}$) at 100, 143, 217 and 353 GHz. It will lead to major improvements in our understanding of the Cosmic Microwave Background anisotropies and polarized foreground signals. Planck will make high resolution measurements of the $E$-mode spectrum (up to $\ell \sim 1500$) and will also play a prominent role in the search for the faint imprint of primordial gravitational waves on the CMB polarization. This paper addresses the effects of calibration of both temperature (gain) and polarization (polarization efficiency and detector orientation) on polarization measurements. The specific requirements on the polarization parameters of the instrument are set and we report on their pre-flight measurement on HFI bolometers. We present a semi-analytical method that exactly accounts for the scanning strategy of the instrument as well as the combination of different detectors. We use this method to propagate errors through to the CMB angular power spectra in the particular case of Planck-HFI, and to derive constraints on polarization parameters. We show that in order to limit the systematic error to 10% of the cosmic variance of the $E$-mode power spectrum, uncertainties in gain, polarization efficiency and detector orientation must be below 0.15%, 0.3% and 1\deg\ respectively. Pre-launch ground measurements reported in this paper already fulfill these requirements.

In this article we present and study a scaling law of the $m\Gamma_m$ CMB Fourier spectrum on rings which allows us (i) to combine spectra corresponding to different colatitude angles (e.g. several detectors at the focal plane of a telescope), and (ii) to recover the $C_l$ power spectrum once the $\Gamma_m$ coefficients have been measured. This recovery is performed numerically below the 1% level for colatitudes $\Theta> 80^\circ$ degrees. In addition, taking advantage of the smoothness of the $C_l$ and of the $\Gamma_m$, we provide analytical expressions which allow to recover one of the spectrum at the 1% level, the other one being known.