Methods and apparatus are provided for reducing power fluctuations during preamble training in a multiple antenna communication system using cyclic delays. A preamble having a legacy portion and a high throughput portion is transmitted (or received) on each of N antennas, wherein at least a portion of the preamble on a first of the N antennas is delayed relative to at least the portion of the preamble on a second of the N antennas, wherein the delay is non-orthogonal amount to introduce variation across the preambles transmitted on the N transmit antennas. The legacy portion may be, for example, an 802.11 a/g preamble.