Full-duplex (FD) wireless communications have the potential to double the ergodic capacity with respect to conventional half-duplex (HD) systems [1]. Since the capacity increase of FD communications is obtained by allowing simultaneous transmission and reception over the same frequency band, the high level of self-interference (SI) at the receiver side is drowning out the signal-of-interest (SOI). Consequently, SI cancellation has attracted a lot of attention in the recent literature [1]. Passive methods rely on isolation/shielding of the transmit and receive antennas, while active methods mainly rely on analog domain SI cancellation to avoid saturation of the analog-to-digital converter. After combined passive and active analog cancellation, the residual self-interference (RSI) must be cancelled down to the noise floor.The focus of this letter is on digital domain RSI cancellation for orthogonal frequency division multiplexing (OFDM) FD transceivers equipped with one transmit and one receive antenna. Previous studies have considered RSI channel estimation, followed by RSI reconstruction from the known SI data (in the time or frequency domain) and cancellation. First, least squares (LS) estimation of the RSI channel impulse response (CIR) has been used in [2] and [3] to mitigate linear and nonlinear effects, respectively. Since the presence of the SOI is ignored, several consecutive OFDM blocks are needed to obtain reliable LS estimates. Second, using orthogonal pilot symbols for the SI and SOI, channel frequency response (CFR) estimation is advocated in [4] and [5] to mitigate linear and nonlinear effects, respectively. Assuming one pilot per channel coherence time, the estimated parameters are then used during subsequent OFDM data blocks.Against this background, the main contribution of this letter is to introduce a RSI mitigation scheme for fast-moving devices, where the channel can vary from one OFDM block to the next. The proposed low complexity receiver performs joint RSI/SOI channel estimation, RSI cancellation, and decoding, based on a factor graph approach. The key ingredient is that that both the pilot and data OFDM blocks serve the purpose of RSI mitigation in the presence of nonlinear effects.