100 likes | 318 Views
Genetic Algorithm for Parameter Optimization. Objective: to implement a Genetic Algorithm into myspice to allow parameter optimization of multiple parameters and multiple goal specifications. Skyler Weaver. Introduction.
E N D
Genetic Algorithm for Parameter Optimization Objective: to implement a Genetic Algorithm into myspice to allow parameter optimization of multiple parameters and multiple goal specifications. Skyler Weaver
Introduction • Optimization allows you to set circuit outputs and solve for parameters • Circuit outputs are “goals” • Many optimization routines require a good initial guess • Many local minima • Slow gradient F(x1,x2,x3,…) Genetic Algorithm doesn’t have this problem!!!
The concept Create population of randomized parameters Choose variables Determine fitnessof individualsand rank them “mate” fittestindividuals tocreate new population Solution? Solutionfound Mutate newpopulation
The approach Bit array represents “DNA” : RRRR CCCC LLLL mother father 1011 0100 1011 1001 0111 1010 } FFMF MFMF FMMM crossover 1001 0110 1010 } 0000 0100 0001 mutation 1001 0010 1011 Get values for R, C, L and run simulation
Does it work? DC Goals: dc[3] = 0.500000 (0.500000) 100.0% dc[5] = -0.050000 (-0.050000) 100.0% AC Goals: ac[4] = 0.500000 (0.500037) @ 1000000.0 Hz ac[4] = 0.400000 (0.399643) @ 2000000.0 Hz R1 supply div 10.645582 R2 div 0 9.999998 R3 in out 164.898438 Vdd supply 0 1.000000 0.000000 Vsig in div 0.000000 1.000000 C1 out 0 625.375000pF L1 supply out 81.607788uH Vdd supply 0 1 0 Vsig in div 0 1 R1 supply div ~ 1 1000 R2 div 0 ~ 1 1000 R3 in out ~ 1 1000 L1 supply out ~ 1e-6 1e-3 C1 out 0 ~ 1e-12 1e-9 ~ DC div 0.5 ~ DC Vdd -0.05 ~ AC out 0.5 1e6 ~ AC out 0.4 2e6 Positive mutation occurred Successful sexualcrossover