Disclosed in the present disclosure is an FPGA implementation device and method for an FBLMS algorithm based on block floating point. The method includes: blocking, caching, and reassembling a reference signal, by an input caching and converting module, converting into a block floating point system and performing FFT; filtering, by a filtering module, in a frequency domain and performing dynamic truncation; caching, by an error calculating and output caching module, a target signal on a block basis, converting into a block floating point system, subtracting an output result output from the filtering module from the converted target signal to obtain an error signal, converting the error signal into a fixed point system to obtain a final cancellation result; obtaining, by a weight adjustment amount calculating module and a weight updating and storing module, an adjustment amount of a frequency domain block weight and updating the frequency domain block weight.