We investigate hard decision based low- density parity-check (LDPC) codes over an impulse noise channel, which is modeled as the concatenation of a binary erasure channel (BEC) and a binary symmetric channel (BSC). We propose a decoding algorithm based on three-state quantization of the traditional sum-product algorithm, where the three states correspond to "1", "0", and "E" (Erasure). As such, the proposed algorithm dramatically reduces the implementation complexity, while still keeps satisfactory decoding performance as shown by simulation results.