LINEAR AND NONLINEAR PREDICTION TECHNIQUES FOR SHORT-TERM FORECASTINGOF HF FADING SIGNALS

There exist two major mechanisms which are responsible for the fading phenomenon at HF frequencies. They are the multiple-mode interference and distortions due to the ionospheric irregularities. Fading time ser ies produced bq the first of these mechanisms alone should typically r epresent a multiple-periodical process. This kind of signal may also b e produced by an autonomous dynamical system. The character of the tim e series produced by the second mechanism depends on the nature of the ionospheric irregularities. Recently, evidence has been accumulated t o show that sometimes the ionospheric turbulence on equatorial and mid dle latitudes represents a low-dimensional deterministic chaotic proce ss. These facts suggest that for both mechanisms the fading time serie s may have a deterministic nature and therefore is predictable. Accord ingly we apply the nonlinear predicting technique proposed by Farmer a nd Sidorovich [1987] to the fading time series obtained by the Univers ity of Illinois HF sounder. In its application the prediction techniqu e is modified to take into account specifics of the HF data. For compa rison, the conventional linear autoregression prediction technique is also tested. It is found that, in general, the nonlinear prediction an d the linear autoregressive forecasting allow prediction on a few corr elation times and work with roughly the same success.