Recursive training with unitary modulation for correlated block-fading MIMO channels

We propose and analyze a communication scheme called recursive training with unitary modulation (RTUM) for correlated block fading channels with multiple inputs and multiple outputs (MIMO) and without channel state information (CSI). In RTUM, we employ unitary matrices as constellation symbols, and utilize a recursive procedure to perform channel estimation and demodulation/decoding in an alternating manner. Central benefits of RTUM include: low-complexity implementation; transformation of the original fading channel without CSI into a series of parallel subchannels, each with perfect receive CSI but additional Gaussian noise; and convenience of analysis. By inspecting the effective signal-to-noise ratio (SNR) of the RTUM, we are further able to identify three operating regimes that lead to distinct channel behaviors. For typical channels we establish a new rule of thumb that spectrally-efficient low-SNR "coherent" communication can be realized when the average channel SNR is larger than the mean square one-step prediction error of the block fading process.