This paper proposes an iterative estimator for bit interleaved coded modulation (BICM) with an iterative soft-input soft-output (SISO) receiver over time-varying flat fading channels. More specifically, the system implements signal space diversity (SSD) where the QAM symbols are rotated. The use of an iterative Kalman filter shows good results using superimposed training sequence and Gauss-Markov channel model for fast tracking of the channel variation. The proposed soft iterative estimator uses the soft extrinsic information from the decoder as well as the superimposed training sequence to update the channel coefficients for data detection. To analyze the results, an analytical bound for the asymptotic bit error rate (BER) for BICM-ID-SSD over correlated fading channels is computed with perfect channel state information (CSI).