This study is aimed at numerically predicting the noise of rotor ingesting a thick turbulent boundary layer at the tail end of an axisymmetric body of revolution (BOR) and elucidating the noise source mechanisms. The BOR consists of an ellipsoidal nose, a cylindrical centerbody and a tail cone with a half-apex angle of 20 degrees, and is at Reynolds number of 1.9 million based on its length. A large-eddy simulation (LES) methodology that involves wall-modeling in the nose and centerbody sections of the BOR but is wall-resolved in the tail-cone section is employed to compute the turbulent flow field. This approach is validated by comparisons of velocity statistics with the experimental data from Virginia Tech (Hickling et al., AIAA-2019-2571) and the results from a wall-resolved LES in a BOR flow without a rotor. The evolution of flow statistics and structures in the tail-cone boundary layer is analyzed in detail. The two-point correlations of velocity fluctuations and the pressure fluctuations on the tail-cone surface demonstrate significant growth of length scales in the downstream direction as the boundary layer thickens under strong adverse pressure gradient.A hybrid computational approach combining the partially wall-modeled LES with the Ffowcs Williams-Hawkings equation is employed to investigate the noise generation from a five-bladed rotor operating at the tail end of the BOR. Two rotor advance ratios, 1.44 and 1.13, are considered. The computed sound pressure spectra exhibit reasonable agreement with the experimental measurements of Hickling et al. (AIAA-2019-2571) at free-stream Mach number of 0.059 over a wide range of frequencies. The spectra show broadband noise with haystacking peaks. Consistent with experimental results, the frequencies of the haystacking peaks are blue-shifted by 8-12% from the blade passing frequency and its first three harmonics, and stronger sound is produced as the advance ratio decreases. A mixed Mach number scaling for the ingestion noise is identified, which successfully collapses the sound pressure spectra for the two advance ratios. An examination of acoustic dipole-source distributions on rotor blades reveals that the leading-edge region is acoustically more important at low frequencies, whereas the trailing-edge region become dominant at high frequencies. Large correlations are found between sectional dipole sources on neighboring blades, and the corresponding blade-to-blade coherence is significant at the frequencies of haystacking peaks and valleys between these peaks. The blade-to-blade correlations and coherence are related to successive blades cutting through the same coherent turbulence structure and are the source of haystacking acoustic peaks. The accuracy of the Sears theory for predicting the rotor noise is evaluated using the upwash velocity field from the LES. The Sears theory accurately predicts the haystacking peaks in the sound pressure spectra, but the accuracy deteriorates at higher frequencies.