In this letter, we develop a novel scheme for joint channel estimation, training design, transmit (Tx) power allocation, and receive (Rx) power splitting in multiple-input multiple-output (MIMO) simultaneous wireless information and power transfer (SWIPT) systems, which maximizes the average signal-to-error-plus-noise ratio (SENR) with average harvested energy constraints at the receiver. To solve this challenging design problem, we first convert it into a more tractable form, which enables to derive the optimal solution efficiently. To further lower the required complexity, we also propose a suboptimal solution in a closed form. The performance of the proposed scheme is demonstrated through numerical simulations.