Title: The BINGO Project V: Further steps in Component Separation and Bispectrum Analysis

Abstract: Observing the neutral hydrogen distribution across the Universe via redshifted 21cm line intensity mapping constitutes a powerful probe for cosmology. However, the redshifted 21cm signal is obscured by the foreground emission from our Galaxy and other extragalactic foregrounds. This paper addresses the capabilities of the BINGO survey to separate such signals. Specifically, this paper looks in detail at the different residuals left over by foreground components, shows that a noise-corrected spectrum is unbiased, and shows that we understand the remaining systematic residuals by analyzing nonzero contributions to the three-point function. We use the generalized needlet internal linear combination, which we apply to sky simulations of the BINGO experiment for each redshift bin of the survey. We present our recovery of the redshifted 21cm signal from sky simulations of the BINGO experiment, including foreground components. We test the recovery of the 21cm signal through the angular power spectrum at different redshifts, as well as the recovery of its non-Gaussian distribution through a bispectrum analysis. We find that non-Gaussianities from the original foreground maps can be removed down to, at least, the noise limit of the BINGO survey with such techniques. Our component separation methodology allows us to subtract the foreground contamination in the BINGO channels down to levels below the cosmological signal and the noise, and to reconstruct the 21cm power spectrum for different redshift bins without significant loss at multipoles $20 \lesssim \ell \lesssim 500$. Our bispectrum analysis yields strong tests of the level of the residual foreground contamination in the recovered 21cm signal, thereby allowing us to both optimize and validate our component separation analysis. (Abridged)