Radio Frequency Interference (RFI) is a growing concern in the radio astronomy community. Single-dish telescopes are particularly susceptible to RFI. Several methods have been developed to cope with RF-polluted environments, based on flagging, excision, and real-time blanking, among others. All these methods produce some degree of data loss or require assumptions to be made on the astronomical signal. We report the development of a real-time, digital adaptive filter implemented on a Field Programmable Gate Array (FPGA) capable of processing 4096 spectral channels in a 1 GHz of instantaneous bandwidth. The filter is able to cancel a broad range of interference signals and quickly adapt to changes on the RFI source, minimizing the data loss without any assumption on the astronomical or interfering signal properties. The speed of convergence (for a decrease to a 1%) was measured to be 208.1 us for a broadband noise-like RFI signal and 125.5 us for a multiple-carrier RFI signal recorded at the FAST radio telescope.