In　orthogonal　frequency-division　multiplexing　(OFDM)　systems,　channel　estimation　is　required　for　guaranteeing　the　transmission　quality　over　rapidly　changing　wireless　channels.　At　the　same　time,　it　is　crucial　to　recover　the　clipping　signal　when　resolving　the　high　peak　average　power　ratio　(PAPR)　of　the　transmitted　OFDM　signal.　In　this　paper,　we　apply　the　deep　learning　(DL)　technique　to　boost　the　performance　of　channel　estimation　in　the　protograph　low-density　parity-check　(PLDPC)-coded　bit-interleaved　coded　modulation　(BICM)-aided　OFDM　(BICM-OFDM)　systems.　Specifically,　we　propose　a　channel　estimation　residual　network　(CERNet)　to　estimate　the　channel　state　information　(CSI).　Furthermore,　a　signal　recovery　residual　network　(SRRNet)　is　designed　to　recover　the　clipped　signal.　Our　proposed　residual　networks　can　capture　the　frequency-specific　features　of　the　channel　matrices　and　the　correlation　of　the　clipping　noise　over　the　received　data　signal　adequately　without　relying　on　explicit　priori　channel　information　to　further　improve　the　channel　estimation　performance.　Both　analytical　and　simulation　results　show　that　the　CERNet　achieves　superior　performance　compared　with　the　conventional　estimation　scheme,　and　the　joint　CERNet　and　SRRNet　can　further　improve　the　system　performance.　IEEE