The paper discusses the optimal amount of training required to achieve high data rates in multiple-antenna wireless communication links. The authors derive a lower bound on the capacity of a fading channel that is learned through training and maximize this bound as a function of the received signal-to-noise ratio (SNR), fading coherence time, and number of transmitter antennas. They find that the optimal number of training symbols is equal to the number of transmit antennas when the training and data powers can vary, but may be larger than the number of antennas when the powers are equal. The paper also shows that training-based schemes can be optimal at high SNR but suboptimal at low SNR. The results are illustrated through figures that show the capacity lower bound and optimal training intervals as functions of the number of transmit and receive antennas, fading coherence time, and SNR.The paper discusses the optimal amount of training required to achieve high data rates in multiple-antenna wireless communication links. The authors derive a lower bound on the capacity of a fading channel that is learned through training and maximize this bound as a function of the received signal-to-noise ratio (SNR), fading coherence time, and number of transmitter antennas. They find that the optimal number of training symbols is equal to the number of transmit antennas when the training and data powers can vary, but may be larger than the number of antennas when the powers are equal. The paper also shows that training-based schemes can be optimal at high SNR but suboptimal at low SNR. The results are illustrated through figures that show the capacity lower bound and optimal training intervals as functions of the number of transmit and receive antennas, fading coherence time, and SNR.