152 



than 1/10 a typical wavelength), the predictions of the propagation direction can 

 be very sensitive to subtle fluctuations in the measure record. In order to find a 

 stable solution, the window width must be increased to about 1/3 of the zero crossing 

 frequency. This includes more of the record in the window, allowing the propagation 

 direction to be more clearly defined. 



The examples in this Chapter were all computed using Matlab running under 

 the Linux operating system on an Intel 90MHz Pentium processor based PC. The 

 shallow water steady wave (Fig 6.2) took the longest to compute, about 64min., and 

 471 X 10^ floating point operations (flops). The steady deep water wave (Fig. 6.3) took 

 the least time, 12min. and 22x10^ flops, the standing wave (Fig. 6.4) took 16min. 

 and 131x10^ flops, the short crested wave (Fig. 6.6) took 13min. and 133x10^ flops, 

 and the field record (Fig. 6.8) took 17min. and 46x10^ flops. 



