(Nested Processes ... Each Process controlled by a Solver)
The transfer function H(s) is the Laplace transform of the output signal Yout(s) divided by the Laplace transform of the input signal Yin (s): that is H(s) = where each signal's transform is assumed to be a ratio of polynomials. Thus, H(s) can likewise be stated in the form:
Assuming the numerator and denominator can be factored, yields H(s) in the general form:
where
each Zi is known as a "zero" and the Pi as a "pole" of the transfer function.
Zi and Pi are complex points in the Laplace domain.
A realizable transfer function must have poles and zeros with their conjugate point. That is, poles and zeros come in pairs. If a pole or zero is located at the complex point si + jwi, then its conjugate is located at si - jwi. Thus, a generalized transfer function is stated as
Given n-data points from a Bode plot (see drawing below) that define the mainlobe of the desired transfer function, find the optimal Pole/Zero constellation such that H(s) has equal sidelobe peak amplitudes in a Bode plot and curve fits the following data in the mainlobe.
Increased ProductivityExample #3Source Code:
A Calculus-level program for this optimal matched filter transfer function is as follows:
Problem .Matched.Filter.Transfer.Function execute .Setup for i = 0 to 1 sidelobes = i [Include Zeros On Omega Axis ? ] Find gain, p.real, p.imag In .Laplace.Domain To Match error repeat End Model .Laplace.Domain if sidelobes gt 0 AND omega.zeros gt 0 then for ij = 1 to omega.zeros do side.limits( ij) = x.zeros( ij) * (1 + move( ij)) * (2**(ij-2)) old.zeros( ij) = x.zeros( ij) repeat Find x.zeros In .Stopband By HERA With Bounds side.limits To Minimize peak.diff for ij = omega.zeros, move( ij) = x.zeros( ij) - old.zeros( ij) close for ij = 1 to npoints do [ --- Calculate Transfer Function ---- ] x2 = freq( ij) ** 2 execute .Transfer.Function if den eq 0, den = 1e-8 h( ij) = gain * num / den error( ij) = y.out( ij) - h( ij) * y.in( ij) error( ij) = error( ij) / y.out( ij) [relative error ... line optional] repeat End Model .Stopband [locate sidelobe peaks] peak.diff = 0 for ijk = 1 to omega.zeros do step.limit = side.limits( ijk) Find x.peak In .Sidelobes By HERA With Bounds step.limit To Maximize y.peak peaks( ijk) = x.peak peak.ampl( ijk) = y.peak if ijk gt 1, peak.diff = peak.diff + (y.peak - peak.ampl( ijk-1))**2 repeat peak.diff = peak.diff + (y.peak - peak.ampl( omega.zeros))**2 End Model .Sidelobes [calculate sidelobe amplitude at frequency 'x.peak'] x2 = x.peak**2 execute .Transfer.Function y.peak = num / den End Model .Transfer.Function num = 1 den = 1 for ii = 1 to p.pairs, den = den * .Factor( x2, p.real( ii), p.imag( ii)) if omega.zeros gt 0 then for ii = 1 to omega.zeros, den = den * .Factor( x2, 0, x.zeros(ii)) close End Function .Factor( x.sq, sigma, omega) real.sq = sigma**2 Imag.sq = omega**2 sum = real.sq + imag.sq - x.sq if omega eq 0, exit with sum / real.sq temp.f = sum * sum - 4 * x.sq * imag.sq End with temp.f / (real.sq + imag.sq)**2 Procedure .Setup freq = .data( 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24) y.in = .data( 51.31, 37.79, 28.26, 21.1, 15.37, 11.32, 8.06, 5.83, 3.91, 2.69, 1.78, .96, .52, .31, .25, .21, .18, .15, .12, .10, .08, .07, .06, .06, .05) [y.out(w) = optimum response in the linear Van der Maas sense ] y.out = .data(1, .981, .926, .836, .73, .609, .484, .365, .26, .173, .105, .058, .027, .012, 7.2e-3, 4.4e-3, 2.7e-3, 1.7e-3, 1e-3, 6.2e-4, 3.8e-4, 2.3e-4, 1.4e-4, 8.8e-5, 5.4e-5) fmax = freq( npoints) npoints = 25 gain = 1 p.pairs = 5 omega.zeros = 3 allot h( npoints), error( npoints) allot p.real( p.pairs), p.imag( p.pairs) for i = 1 to p.pairs do [initial guess] p.imag( i) = ((i-1) / p.pairs + .11) * fmax p.real( i) = fmax close if omega.zeros gt 0 then allot x.zeros( omega.zeros), old.zeros( omega.zeros), move( omega.zeros), side.limits( omega.zeros), peaks( omega.zeros), peak.ampl( omega.zeros) for i = 1 to omega.zeros, x.zeros(i) = (1 + 2**(i-1) /10)*fmax [initial guess] close End
Note: This is a multi-level optimization or nesting of optimizers example problem. Each 'find' statement starts a solver and at times has up to three Find statements executing at once (ie. nested). This nesting power should allow companies to optimizes at many levels and combine all optimizations for a true optimize company profits when necessary. Nice!
Note:Match-n-Freq application is exactly this multi-level optimization problem. Download Match-n-Freq application file. Give it a test run and see the nested solvers at work.
Increased ProductivityExample
#3More Individual Process Examples:
where each signal's transform is assumed to be a ratio of polynomials. Thus, H(s) can likewise be stated in the form: 
