We consider a collaborative communication system consisting of several nodes jointly communicating to a common destination node. Once each node determines the message to send, a space-time code across the sources is used to transmit the signal to the destination. As there may be frequency offsets and unknown frequency selective channels, we assume a training sequence is sent in the preamble to estimate these parameters. In this paper, we derive an algorithm that designs training sequences that minimize the Cramer-Rao Bound for frequency estimation in a frequency-selective environment for a multiple node collaborative communication system.