90 likes | 115 Views
Learn about MMSE approximation and least squares method in signal processing, including windows, sampling, and MATLAB implementation details.
E N D
MMSE Optimal Design:The Least Squares method Prof. Siripong Potisuk
MMSE Approximation • Measurable feature of the approximation error • An averaging integral of the squared error over the time period of interest • Can be used to prove convergence of the continuous-time Fourier series • The mean-square error, denoted by EN, is the average power in the error signal and defined as
Windowing Method Revisited • Uses a truncated Fourier series expansion of the desired amplitude response, Hd(f). • Produces a filter that is optimal in the sense that it minimizes the objective function below: where Hr(f) is the actual amplitude response, which is real and has even or odd symmetry.
Least Squares Method • Uses frequency samples of desired frequency response • The filter coefficients obtained by minimizing a discrete version of the weighted objective function defined as • {F0,F1,. . . ,FP} is a set of P+1 uniformly-spaced distinct frequencies with F0= 0, FP= Fs, and F0 < F1 <. . . < FP such that F1 = iFs/P • w(i) > 0 is a weighting function
Special case: N = M+1 = P • Known as the frequency sampling method • The causal impulse response is the IDFT of equally-spaced samples of desired magnitude sequence. • Theactual magnitude response will approximate or interpolate (i.e., passes through the N frequency samples of) the desired magnitude sequence. • It can be shown that
MATLAB Implementation • Function B = firfs(N, Hk) • MATLAB user-defined function for FIR filter • design using the frequency sampling method • (text, pp.290) • Input Arguments: N = number of filter taps (must be an odd number) = M+1 where M is a filter order (even number for Type 1) Hk = sampled desired frequency response for k = 0,1,…,M/2 • Output: • B = FIR filter coefficients
Example 7.12 Design a type 1 linear-phase FIR filter with coefficients satisfying lowpass response characteristics with 7 taps and a cutoff frequency of c = 0.4 radians using the frequency sampling method.
Example 7.13 Design a type 1 linear-phase FIR filter with coefficients satisfying bandpass response characteristics with 25 taps and lower and upper cutoff frequencies of 1000 Hz and 3000 Hz, respectively, using the frequency sampling method. Plot both the impulse response and the frequency response.