One goal of cosmic microwave background (CMB) data analysis is to combine d
ata at different frequencies, angular resolutions, and noise levels in orde
r best to extract the component with a Planckian spectral behaviour. A mult
ifrequency Wiener filtering method has been proposed in this context by Bou
chet, Gispert & Puget and in parallel by Tegmark Br Efstathiou. As shown by
Bouchet & Gispert, this Linear method is also convenient with which to est
imate a priori, given a sky model and an experimental description, the resi
dual errors on the CMB power spectrum assuming the foregrounds have been re
moved with this method. In this paper, we extend the method to the case in
which additional polarization data are available. In particular, we derive
the errors on the power spectra involving polarization and show numerical r
esults for the specifications of the future CMB space missions MAP and Plan
ck(1) when it is assumed that the Galactic synchrotron and dust emission ar
e respectively about 40 and 10 per cent polarized. We consider two underlyi
ng models for our study: we take a standard CDM model with tau = 0.1 for th
e extraction of E-mode polarization and ET cross-correlation; for B-mode po
larization we consider a tilted CDM model with n(s) = 0.9, n(t) = -0.1 and
T/S = 0.7. We find the following results. (1) The resulting fractional erro
rs on E-mode polarization and TE crosscorrelation power spectra are less th
an or similar to 10-30 per cent for 50 less than or similar to l less than
or similar to 1000 for Planck. The fractional errors are between 50 and 150
per cent for l less than or equal to 50. (2) The corresponding fractional
errors for MAP are greater than or equal to 300 per cent for most of the l
range. (3) The Wiener filtering gives extraction errors of less than or equ
al to twice the expected performance for the combined sensitivity of all th
e channels of Planck. For MAP, the corresponding degradation is similar or
equal to 4. (4) If, instead of individual modes, one considers band-power e
stimates with a logarithmic interval Delta l/l = 0.2 then the fractional er
ror for MAP drops to less than or similar to 100 per cent at the Doppler pe
ak around l similar or equal to 300 for the ET signal. (5) The fractional e
rror for B-mode polarization detection is less than or similar to 100 per c
ent with Planck for l less than or equal to 100. A band-power estimate with
Delta l/l = 0.2 reduces the fractional errors to less than or similar to 2
5 per cent for 20 less than or equal to l less than or equal to 100.