A semi-coherent technique for low-complexity frequency offset estimation is presented. The proposed estimation is based on discrete-time Fourier transform (DTFT). Since residual frequency offset can be compensated by channel estimation, frequency offset can be estimated only coarsely. We take advantage of the relationship between frequency resolution and accumulation period in DTFT in deriving the coarse estimator. Based on that, a novel method to balance the coherent and the non-coherent accumulation for frequency offset estimation is proposed. The proposed algorithm has low latency and complexity so that it is particularly suitable for packet traffic. The semi-coherent structure of the proposed algorithm is also scalable so that it can be used for both bursty and continuous traffic.