| function b=wtsgaus(p,N) |
| % wtsgaus: weights for gaussian filter with specified frequency response |
| % b=wtsgaus(p,N); |
| % Last revised 2003-3-14 |
| % |
| % Weights for gaussian filter with specified frequency response |
| % Specify te wavelength for the 0.50 respons, and the length of series, get |
| % the coefficients, or weights |
| % |
| %*** INPUT |
| % |
| % p (1 x 1)i period (years) at which filter is to have amp frequency response of 0.5 |
| % N (1 x 1)i length of the time series (number of observations) |
| % |
| %*** OUTPUT |
| % |
| % b (1 x n)r computed weights |
| % |
| % |
| %*** REFERENCES |
| % |
| % WMO 1966, p. 47 |
| % |
| %*** UW FUNCTIONS CALLED -- NONE |
| %*** TOOLBOXES NEEDED -- stats |
| % |
| %*** NOTES |
| % |
| % Amplitude of frequency response drops to 0.50 at a wavelength of |
| % about 6 standard deviations of the appropriate guassian curve |
| % |
| % N is used as an input to restict the possible filter size (number of weights) to no larger than the sample length |
| |
| if p>N; |
| error(['Desired 50% period ' num2str(p) ' is greater than the sample length ' int2str(N)]); |
| end; |
| |
| |
| % Check that period of 50% response at least 5 yr |
| if p<5; |
| error('Period of 50% response must be at least 5 yr'); |
| end; |
| |
| sigma=p/6; % Gaussian curve should have this standard deviation |
| |
| x=-N:N; |
| b=normpdf(x/sigma,0,1); |
| bmax=max(b); |
| bkeep = b>=0.05*bmax; % keep weights at least 5% as big as central weight |
| b=b(bkeep); |
| b=b/sum(b); % force weights to sum to one |
| |
