We consider the effectiveness of foreground cleaning in the recovery of Cosmic Microwave Background ( CMB ) polarization sourced by gravitational waves for tensor-to-scalar ratios in the range 0 < r < 0.1 .
Using the planned survey area , frequency bands , and sensitivity of the Cosmology Large Angular Scale Surveyor ( CLASS ) , we simulate maps of Stokes Q and U parameters at 40 , 90 , 150 , and 220 GHz , including realistic models of the CMB , diffuse Galactic thermal dust and synchrotron foregrounds , and Gaussian white noise .
We use linear combinations ( LCs ) of the simulated multifrequency data to obtain maximum likelihood estimates of r , the relative scalar amplitude s , and LC coefficients .
We find that for 10,000 simulations of a CLASS-like experiment using only measurements of the reionization peak ( \ell \leqslant 23 ) , there is a 95 % C.L .
upper limit of r < 0.017 in the case of no primordial gravitational waves .
For simulations with r = 0.01 , we recover at 68 % C.L . r = 0.012 ^ { +0.011 } _ { -0.006 } .
The reionization peak corresponds to a fraction of the multipole moments probed by CLASS , and simulations including 30 \leqslant \ell \leqslant 100 further improve our upper limits to r < 0.008 at 95 % C.L .
( r = 0.010 ^ { +0.004 } _ { -0.004 } for primordial gravitational waves with r = 0.01 ) .
In addition to decreasing the current upper bound on r by an order of magnitude , these foreground-cleaned low multipole data will achieve a cosmic variance limited measurement of the E-mode polarization ’ s reionization peak .