Unique word detection in the presence of frequency offset

This paper proposes a new approach to unique word detection in the presence of frequency offsets for TDMA communication systems. It is based on an information theoretic criterion for model selection, known as the minimum description length criterion. Different from the conventional scheme of filter bank plus multiple detection, the new approach employs a signal subspace decomposition technique to form the test statistic. It offers a probability of detection independent of the frequency offset, and a constant false alarm rate regardless of the noise characteristics. It is also computationally efficient and suitable for DSP implementation.