Data Directed Estimation Techniques for Single-Tone HF Modems

This paper is concerned with data directed estimation (DDE) techniques for use in single-tone HF modems. The proposed scheme uses least-squares adaptive techniques for dealing with fading and multipath radio HF channels. The DDE is based on direct estimation of channel parameters and data blocks. Training data is imbedded into the transmission in the form of blocks of training bits alternated with blocks of source data at a fixed (though selectable) fraction of the transmission. Two DDE techniques, i.e., linear and nonlinear, are discussed in this paper. In the linear DDE, the source data block is estimated by a Levinson recursive algorithm. In the nonlinear DDE, only the end symbols of the data block are kept, quantized, and subtracted from the simultaneous equations. The symbols left in the block are estimated by a Levinson algorithm again, and so forth. The procedure described here is used recursively until all the known symbols are obtained. In both linear and nonlinear DDE techniques, the channel is estimated by a steepest-descent algorithm. As with a linear equalizer, the linear DDE lacks the ability to cope with the fading dispersive HF radio channels. However, the nonlinear DDE works well under the same channel conditions. The analysis shows that the nonlinear DDE can achieve performance similar to the decision-feedback equalization (DFE) technique with known channel conditions. The channel estimator used in the DDE techniques will converge faster than the DFE (see Section VII) if a similar adaptive algorithm is utilized.