A new model of HF atmospheric radio noise (ARN) that leads to observed probability distributions of noise is briefly described. The new ARN model is then applied to the detection of known signals in the presence of noise to determine optimal receiver structure for ARN. Probabilities of errors for a linear matched filter and nonlinear (-1/log) and inverted(log)^{2}correlator receivers and a phase coherent binary detector are determined and compared. It is shown that the inverted(log)^{2}correlator combined with the linear matched filter is an optimal receiver in the presence of impulsive noise in tropics. The time statistics of burst noise is applied to design of signals for improved reliability.