Apostolos Fanakis
6 years ago
5 changed files with 349 additions and 397 deletions
Binary file not shown.
Binary file not shown.
Binary file not shown.
@ -0,0 +1,349 @@ |
|||
% AUTHOR : Apostolos Fanakis |
|||
% $DATE : 02-May-2018 17:21:18 $ |
|||
% $Revision : 1.00 $ |
|||
% DEVELOPED : 9.0.0.341360 (R2016a) |
|||
% FILENAME : low_pass_design_single.m |
|||
% AEM : 8261 |
|||
|
|||
%% ========== DESIGN SPECIFICATIONS START ========== |
|||
% Figures out design specifications according to my AEM number |
|||
|
|||
AEM = [8 2 6 1]; |
|||
|
|||
if AEM(3)<3 |
|||
design_param_m = 1; |
|||
elseif AEM(3)<7 |
|||
design_param_m = 2; |
|||
else |
|||
design_param_m = 3; |
|||
end |
|||
|
|||
specification_pass_frequency = (1.1*(3+design_param_m))*1000; % Hz |
|||
specification_pass_radial_frequency = ... |
|||
specification_pass_frequency*(2*pi); % rad/s |
|||
|
|||
specification_stop_frequency = 2.1*specification_pass_frequency; % Hz |
|||
specification_stop_radial_frequency = ... |
|||
specification_stop_frequency*(2*pi); % rad/s |
|||
|
|||
specification_min_stop_attenuation = 23+(max(1,AEM(3))-5)*(3/4); % dB |
|||
specification_max_pass_attenuation = 0.6+((max(1,AEM(4))-5)/16); % dB |
|||
|
|||
clear design_param_m |
|||
|
|||
% ========== DESIGN SPECIFICATIONS END ========== |
|||
|
|||
%% ========== NORMALIZED DESIGN START ========== |
|||
% Calculates normalized design specifications and designs a normalized |
|||
% filter using them |
|||
|
|||
normalized_pass_radial_frequency = specification_pass_radial_frequency/ ... |
|||
specification_stop_radial_frequency; % rad/s |
|||
% normalized_stop_radial_frequency = 1; % Hz (stop_frequency/stop_frequency) |
|||
|
|||
% Calculates the filter's order using the eq. 9-137 |
|||
design_filter_order = ceil( ... |
|||
acosh(((10^(specification_min_stop_attenuation/10)-1)/ ... |
|||
(10^(specification_max_pass_attenuation/10)-1))^(1/2)) ... |
|||
/acosh(1/normalized_pass_radial_frequency)); |
|||
% Calculates epsilon parameter using the eq. 9-123 |
|||
epsilon_parameter = 1/(10^(specification_min_stop_attenuation/10)-1)^(1/2); |
|||
|
|||
% Calculates alpha using the eq. ?? wtf is this? =========^^^^^^^^^^^^^^^^^^&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&========== |
|||
alpha_parameter = asinh(1/epsilon_parameter)/design_filter_order; |
|||
|
|||
% Calculates the frequency at which half power occurs using the eq. 9-139 |
|||
design_half_power_radial_frequency = specification_stop_radial_frequency/ ... |
|||
(cosh(acosh(1/epsilon_parameter)/design_filter_order)); % rad/s |
|||
|
|||
% ----- |
|||
% Calculates stable poles, zeros, angles and other characteristic sizes |
|||
% using the Guillemin algorithm |
|||
% ----- |
|||
|
|||
% Initializes necessary variables |
|||
design_number_of_poles = idivide(design_filter_order,int32(2),'ceil'); |
|||
% Creates five vector arrays of dimensions [1 * number_of_poles] filled |
|||
% with zeros to store: |
|||
% - the Butterworth angles with reference to the negative horizontal axes, |
|||
% - the real parts of the poles, |
|||
% - the imaginary parts of the poles, |
|||
% - the radial frequencies (Omega0) of the poles and |
|||
% - the Q's of the poles (both the Chebyshev and Inverse Chebyshev poles |
|||
% have the same Q's) |
|||
design_butterworth_angles = zeros([1 design_number_of_poles]); |
|||
chebyshev_poles_real_parts = zeros([1 design_number_of_poles]); |
|||
chebyshev_poles_imaginary_parts = zeros([1 design_number_of_poles]); |
|||
chebyshev_poles_radial_frequencies = zeros([1 design_number_of_poles]); |
|||
inverse_chebyshev_poles_Q = zeros([1 design_number_of_poles]); |
|||
|
|||
% Calculates the Butterworth angles using the method suggested in chapter |
|||
% 9 (page 10) of the course notes and then uses them to calculate the |
|||
% inverse Chebyshev poles |
|||
if mod(design_filter_order,2)~=0 % Odd number of poles |
|||
% First pole has a zero angle |
|||
design_butterworth_angles(1,1)=0; |
|||
% The rest of the poles are scattered in the left half pane with |
|||
% equal angles |
|||
% Theta is a helper parameter |
|||
theta=180/design_filter_order; |
|||
|
|||
% Calculates the first pole's real part using the eq. 9-102 |
|||
chebyshev_poles_real_parts(1,1) = -sinh(alpha_parameter)* ... |
|||
cosd(design_butterworth_angles(1,1)); |
|||
% Calculates the first pole's imaginary part using the eq. 9-103 |
|||
chebyshev_poles_imaginary_parts(1,1) = cosh(alpha_parameter)* ... |
|||
sind(design_butterworth_angles(1,1)); |
|||
% Calculates the first pole's radial frequency using the eq. 9-150 |
|||
chebyshev_poles_radial_frequencies(1,1) = ... |
|||
(chebyshev_poles_real_parts(1,1)^2+ ... |
|||
chebyshev_poles_imaginary_parts(1,1)^2)^(1/2); |
|||
% Calculates the first pole's Q using the eq. 9-151 |
|||
inverse_chebyshev_poles_Q(1,1) = 1/ ... |
|||
(2*cos(atan(chebyshev_poles_imaginary_parts(1,1)/ ... |
|||
chebyshev_poles_real_parts(1,1)))); |
|||
|
|||
% Calculates the rest of the poles in the same way |
|||
for i=2:design_number_of_poles |
|||
design_butterworth_angles(1,i)=double((i-1)*theta); |
|||
% Pole's real part, eq. 9-102 |
|||
chebyshev_poles_real_parts(1,i) = -sinh(alpha_parameter)* ... |
|||
cosd(design_butterworth_angles(1,i)); |
|||
% Pole's imaginary part, eq. 9-103 |
|||
chebyshev_poles_imaginary_parts(1,i) = cosh(alpha_parameter)* ... |
|||
sind(design_butterworth_angles(1,i)); |
|||
% Pole's radial frequency, eq. 9-150 |
|||
chebyshev_poles_radial_frequencies(1,i) = ... |
|||
(chebyshev_poles_real_parts(1,i)^2+ ... |
|||
chebyshev_poles_imaginary_parts(1,i)^2)^(1/2); |
|||
% Pole's Q, eq. 9-151 |
|||
inverse_chebyshev_poles_Q(1,i) = 1/ ... |
|||
(2*cos(atan(chebyshev_poles_imaginary_parts(1,i)/ ... |
|||
chebyshev_poles_real_parts(1,i)))); |
|||
end |
|||
else % Even number of poles |
|||
% Theta is a helper parameter |
|||
theta=90/design_number_of_poles; |
|||
|
|||
for i=1:design_number_of_poles |
|||
design_butterworth_angles(1,i)=double(90)/ ... |
|||
double(design_filter_order)+double((i-1)*theta); |
|||
% Pole's real part, eq. 9-102 |
|||
chebyshev_poles_real_parts(1,i) = -sinh(alpha_parameter)* ... |
|||
cosd(design_butterworth_angles(1,i)); |
|||
% Pole's imaginary part, eq. 9-103 |
|||
chebyshev_poles_imaginary_parts(1,i) = cosh(alpha_parameter)* ... |
|||
sind(design_butterworth_angles(1,i)); |
|||
% Pole's radial frequency, eq. 9-150 |
|||
chebyshev_poles_radial_frequencies(1,i) = ... |
|||
(chebyshev_poles_real_parts(1,i)^2+ ... |
|||
chebyshev_poles_imaginary_parts(1,i)^2)^(1/2); |
|||
% Pole's Q, eq. 9-151 |
|||
inverse_chebyshev_poles_Q(1,i) = 1/ ... |
|||
(2*cos(atan(chebyshev_poles_imaginary_parts(1,i)/ ... |
|||
chebyshev_poles_real_parts(1,i)))); |
|||
end |
|||
end |
|||
|
|||
% Initializes array to hold the inversed poles |
|||
inverse_chebyshev_poles_radial_frequencies = ... |
|||
zeros([1 design_number_of_poles]); |
|||
% Calculates inverse Chebyshev poles by inversing the Chebyshev poles |
|||
% using the eq. 9-146 |
|||
for i=1:design_number_of_poles |
|||
inverse_chebyshev_poles_radial_frequencies(1,i) = 1/ ... |
|||
chebyshev_poles_radial_frequencies(1,i); |
|||
end |
|||
|
|||
% Initializes array to hold the transfer function zeros and a temporary |
|||
% helper variable |
|||
inverse_chebyshev_transfer_function_zeros = ... |
|||
zeros([1 length(design_butterworth_angles)]); |
|||
temp_index = 1; |
|||
% Calculates the transfer function's zeros using the eq. 9-143 |
|||
for i=1:2:design_filter_order |
|||
inverse_chebyshev_transfer_function_zeros(1,temp_index) = ... |
|||
sec((i*pi)/(2*design_filter_order)); |
|||
temp_index = temp_index + 1; |
|||
end |
|||
|
|||
% Clears unneeded variable from workspace |
|||
clearVars = {'theta', 'i', 'temp_index', 'alpha_parameter' ... |
|||
'epsilon_parameter', 'normalized_pass_radial_frequency'}; |
|||
clear(clearVars{:}) |
|||
clear clearVars |
|||
clear -regexp ^chebyshev_ |
|||
|
|||
% ========== NORMALIZED DESIGN END ========== |
|||
|
|||
%% ========== ZEROS-POLES GROUPING START ========== |
|||
|
|||
% ------------------------------------------------------------------------- |
|||
% Unit 1 has a pole pair with Q equal to 0.5822, radial frequency equal to |
|||
% 0.9631 rad/s and is grouped with a zero at 1.0824, resulting in a low |
|||
% pass notch. |
|||
% ------------------------------------------------------------------------- |
|||
% Unit 1 has a pole pair with Q equal to 1.8086, radial frequency equal to |
|||
% 0.7484 rad/s and is grouped with a zero at 2.6131, resulting in a low |
|||
% pass notch. |
|||
% ------------------------------------------------------------------------- |
|||
|
|||
% Two low pass notch units are required to implement the desired low pass |
|||
% filter. |
|||
|
|||
% ========== ZEROS-POLES GROUPING END ========== |
|||
|
|||
%% ========== UNITS IMPLEMENTATION START ========== |
|||
% AEM(3) = 6, so the circuit shown in 7.23 is going to be used for the low |
|||
% pass notch units. |
|||
|
|||
% Initializes necessary arrays, each array is 1X2, the first element (1,1) |
|||
% corresponds to the first unit and the second element (1,2) to second unit. |
|||
unit_low_pass_notch_resistors_1 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_2 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_3 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_4 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_5 = zeros([1 2]); |
|||
unit_low_pass_notch_capacitors = zeros([1 2]); |
|||
unit_low_pass_notch_gains_high = zeros([1 2]); |
|||
unit_low_pass_notch_gains_low = zeros([1 2]); |
|||
unit_frequency_scale_factors = zeros([1 2]); |
|||
unit_amplitude_scale_factors = zeros([1 2]); |
|||
unit_transfer_function = [tf(1) tf(1)]; |
|||
|
|||
for i=1:design_number_of_poles |
|||
% normalized_inverse_pole_radial_frequency = 1; |
|||
normalized_transfer_function_zero = ... |
|||
inverse_chebyshev_transfer_function_zeros(1,i)/ ... |
|||
inverse_chebyshev_poles_radial_frequencies(1,i); |
|||
|
|||
% According to the design method outlined in 7.6-B, at page 35 |
|||
unit_low_pass_notch_resistors_1(1,i) = 1; % Ohm |
|||
% Calculates the capacity of the normalized circuit capacitors using |
|||
% the eq. 7-150 |
|||
unit_low_pass_notch_capacitors(1,i) = 1/ ... |
|||
(2*inverse_chebyshev_poles_Q(1,i)); % Farad |
|||
% Calculates the resistance of R2 using the same equations (7-150) |
|||
unit_low_pass_notch_resistors_2(1,i) = 4* ... |
|||
inverse_chebyshev_poles_Q(1,i)^2; % Ohm |
|||
% Calculates the resistance of R5 using the eq. 7-152 |
|||
unit_low_pass_notch_resistors_5(1,i) = ... |
|||
(4*inverse_chebyshev_poles_Q(1,i)^2)/ ... |
|||
(normalized_transfer_function_zero^2-1); % Ohm |
|||
unit_low_pass_notch_resistors_4(1,i) = 1; % Ohm |
|||
% Calculates the resistance of R3 using the eq. 7-155 |
|||
unit_low_pass_notch_resistors_3(1,i) = ... |
|||
(normalized_transfer_function_zero^2)/ ... |
|||
(2*inverse_chebyshev_poles_Q(1,i)^2); % Ohm |
|||
% Calculates the gain of this unit in high frequencies using the eq. 7-143 |
|||
unit_low_pass_notch_gains_high(1,i) = ... |
|||
(unit_low_pass_notch_resistors_4(1,i))/ ... |
|||
(unit_low_pass_notch_resistors_3(1,i)+ ... |
|||
unit_low_pass_notch_resistors_4(1,i)); |
|||
% Calculates the gain of this unit in low frequencies using the |
|||
% eq. 7-146, 7-147, 7-148, setting s = 0 |
|||
unit_low_pass_notch_gains_low(1,i) = ... |
|||
unit_low_pass_notch_gains_high(1,i)* ... |
|||
normalized_transfer_function_zero^2; |
|||
|
|||
% Performs scaling |
|||
unit_frequency_scale_factors(1,i) = ... |
|||
specification_stop_radial_frequency* ... |
|||
inverse_chebyshev_poles_radial_frequencies(1,i); |
|||
% AEM(4) = 1, so the scaling will be performed to achieve a capacitor |
|||
% value of 0.1uF using the eq. 6-33 |
|||
unit_amplitude_scale_factors(1,i) = ... |
|||
unit_low_pass_notch_capacitors(1,i)/ ... |
|||
(unit_frequency_scale_factors(1,i)*0.1*10^(-6)); |
|||
|
|||
unit_low_pass_notch_resistors_1(1,i) = ... |
|||
unit_low_pass_notch_resistors_1(1,i)* ... |
|||
unit_amplitude_scale_factors(1,i); % Ohm |
|||
unit_low_pass_notch_resistors_2(1,i) = ... |
|||
unit_low_pass_notch_resistors_2(1,i)* ... |
|||
unit_amplitude_scale_factors(1,i); % Ohm |
|||
unit_low_pass_notch_resistors_3(1,i) = ... |
|||
unit_low_pass_notch_resistors_3(1,i)* ... |
|||
unit_amplitude_scale_factors(1,i); % Ohm |
|||
unit_low_pass_notch_resistors_4(1,i) = ... |
|||
unit_low_pass_notch_resistors_4(1,i)* ... |
|||
unit_amplitude_scale_factors(1,i); % Ohm |
|||
unit_low_pass_notch_resistors_5(1,i) = ... |
|||
unit_low_pass_notch_resistors_5(1,i)* ... |
|||
unit_amplitude_scale_factors(1,i); % Ohm |
|||
unit_low_pass_notch_capacitors(1,i) = 0.1*10^(-6); % Farad |
|||
|
|||
% Builds unit's transfer function |
|||
% Builds numerator and denominator of the transfer function using the |
|||
% eq. 7-146, 7-147 & 7-148 |
|||
unit_numerator = [1 ... |
|||
(((unit_low_pass_notch_gains_high(1,i)-1)/ ... |
|||
(unit_low_pass_notch_gains_high(1,i)* ... |
|||
unit_low_pass_notch_resistors_1(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)))+ ... |
|||
(2/(unit_low_pass_notch_resistors_2(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)))+ ... |
|||
(2/(unit_low_pass_notch_resistors_5(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)))) ... |
|||
(1/(unit_low_pass_notch_resistors_1(1,i)* ... |
|||
unit_low_pass_notch_resistors_5(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)^2)+ ... |
|||
1/(unit_low_pass_notch_resistors_1(1,i)* ... |
|||
unit_low_pass_notch_resistors_2(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)^2))]; |
|||
unit_denominator = [1 ... |
|||
2/(unit_low_pass_notch_resistors_2(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)) ... |
|||
1/(unit_low_pass_notch_resistors_1(1,i)* ... |
|||
unit_low_pass_notch_resistors_2(1,i)* ... |
|||
unit_low_pass_notch_capacitors(1,i)^2)]; |
|||
|
|||
unit_transfer_function(1,i) = tf(unit_numerator, unit_denominator); |
|||
unit_transfer_function(1,i) = unit_transfer_function(1,i)* ... |
|||
unit_low_pass_notch_gains_high(1,i); |
|||
end |
|||
|
|||
% Clears unneeded variable from workspace |
|||
clearVars = {'normalized_transfer_function_zero', 'i'}; |
|||
clear(clearVars{:}) |
|||
clear clearVars |
|||
|
|||
% ========== UNITS IMPLEMENTATION END ========== |
|||
|
|||
%% ========== GAIN ADJUSTMENT START ========== |
|||
|
|||
total_fried_units_attenuation = unit_low_pass_notch_gains_low(1,1)* ... |
|||
unit_low_pass_notch_gains_low(1,2); |
|||
unit_adjustment_gain = 1/total_fried_units_attenuation; |
|||
% We arbitrarily choose to use a 10KOhm series resistor in the adjustment |
|||
% unit |
|||
unit_adjustment_feedback_resistor = 10*10^3*unit_adjustment_gain; |
|||
|
|||
% ========== GAIN ADJUSTMENT END ========== |
|||
|
|||
%% ========== TRANSFER FUNCTIONS START ========== |
|||
|
|||
total_transfer_function = series(unit_transfer_function(1,1), ... |
|||
unit_transfer_function(1,2)); |
|||
total_transfer_function = total_transfer_function*unit_adjustment_gain; |
|||
|
|||
%{ |
|||
plot_transfer_function(total_transfer_function, ... |
|||
[10 ... |
|||
specification_pass_frequency ... |
|||
design_half_power_radial_frequency/(2*pi) ... |
|||
specification_stop_frequency]); |
|||
%} |
|||
|
|||
%ltiview(unit_transfer_function(1,1)); |
|||
%ltiview(unit_transfer_function(1,2)); |
|||
%ltiview(total_transfer_function); |
|||
ltiview(unit_transfer_function(1,1), unit_transfer_function(1,2), ... |
|||
total_transfer_function); |
|||
|
|||
% Clears unneeded variable from workspace |
|||
clear -regexp _numerator$ |
|||
clear -regexp _denominator$ |
|||
clear -regexp _transfer_function$ |
|||
|
|||
% ========== TRANSFER FUNCTIONS END ========== |
@ -1,397 +0,0 @@ |
|||
% AUTHOR : Apostolos Fanakis |
|||
% $DATE : 02-May-2018 17:21:18 $ |
|||
% $Revision : 1.00 $ |
|||
% DEVELOPED : 9.0.0.341360 (R2016a) |
|||
% FILENAME : low_pass_design_single.m |
|||
% AEM : 8261 |
|||
|
|||
%% ========== DESIGN SPECIFICATIONS START ========== |
|||
% Figures out design specifications according to my AEM number |
|||
|
|||
AEM = [8 2 6 1]; |
|||
|
|||
if AEM(3)<3 |
|||
design_param_m = 1; |
|||
elseif AEM(3)<7 |
|||
design_param_m = 2; |
|||
else |
|||
design_param_m = 3; |
|||
end |
|||
|
|||
specification_pass_frequency = (1.1*(3+design_param_m))*1000; % Hz |
|||
specification_pass_radial_frequency = specification_pass_frequency*(2*pi); % rad/s |
|||
|
|||
specification_stop_frequency = 2.1*specification_pass_frequency; % Hz |
|||
specification_stop_radial_frequency = specification_stop_frequency*(2*pi); % rad/s |
|||
|
|||
specification_min_stop_attenuation = 23+(max(1,AEM(3))-5)*(3/4); % dB |
|||
specification_max_pass_attenuation = 0.6+((max(1,AEM(4))-5)/16); % dB |
|||
|
|||
clear design_param_m |
|||
|
|||
specification_pass_radial_frequency = 1000; % rad/s |
|||
specification_stop_radial_frequency = 1400; % rad/s |
|||
specification_min_stop_attenuation = 18; % dB |
|||
specification_max_pass_attenuation = 0.25; % dB |
|||
% ========== DESIGN SPECIFICATIONS END ========== |
|||
|
|||
%% ========== NORMALIZED DESIGN START ========== |
|||
% Calculates normalized design specifications and designs a normalized |
|||
% filter using them |
|||
|
|||
normalized_pass_radial_frequency = specification_pass_radial_frequency/ ... |
|||
specification_stop_radial_frequency; % rad/s |
|||
% normalized_stop_radial_frequency = 1; % Hz (stop_frequency/stop_frequency) |
|||
|
|||
% Calculates the filter's order using the eq. 9-137 |
|||
design_filter_order = ceil( ... |
|||
acosh(((10^(specification_min_stop_attenuation/10)-1)/ ... |
|||
(10^(specification_max_pass_attenuation/10)-1))^(1/2)) ... |
|||
/acosh(1/normalized_pass_radial_frequency)); |
|||
% Calculates epsilon parameter using the eq. 9-123 |
|||
epsilon_parameter = 1/(10^(specification_min_stop_attenuation/10)-1)^(1/2); |
|||
|
|||
% Calculates alpha using the eq. ?? wtf is this? =========^^^^^^^^^^^^^^^^^^&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&========== |
|||
alpha_parameter = asinh(1/epsilon_parameter)/design_filter_order; |
|||
|
|||
% Calculates the frequency at which half power occurs using the eq. 9-139 |
|||
design_half_power_radial_frequency = specification_stop_radial_frequency/ ... |
|||
(cosh(acosh(1/epsilon_parameter)/design_filter_order)); % rad/s |
|||
|
|||
% ----- |
|||
% Calculates stable poles, zeros, angles and other characteristic sizes |
|||
% using the Guillemin algorithm |
|||
% ----- |
|||
|
|||
% Initializes necessary variables |
|||
design_number_of_poles = idivide(design_filter_order,int32(2),'ceil'); |
|||
% Creates five vector arrays of dimensions [1 * number_of_poles] filled |
|||
% with zeros to store: |
|||
% - the Butterworth angles with reference to the negative horizontal axes, |
|||
% - the real parts of the poles, |
|||
% - the imaginary parts of the poles, |
|||
% - the radial frequencies (Omega0) of the poles and |
|||
% - the Q's of the poles |
|||
design_butterworth_angles = zeros([1 design_number_of_poles]); |
|||
poles_real_parts = zeros([1 design_number_of_poles]); |
|||
poles_imaginary_parts = zeros([1 design_number_of_poles]); |
|||
poles_radial_frequencies = zeros([1 design_number_of_poles]); |
|||
inverse_poles_Q = zeros([1 design_number_of_poles]); |
|||
|
|||
% Calculates the Butterworth angles using the method suggested in chapter |
|||
% 9 (page 10) of the course notes and then uses them to calculate the |
|||
% inverse Chebyshev poles |
|||
if mod(design_filter_order,2)~=0 % Odd number of poles |
|||
% First pole has a zero angle |
|||
design_butterworth_angles(1,1)=0; |
|||
% The rest of the poles are scattered in the left half pane with |
|||
% equal angles |
|||
% Theta is a helper parameter |
|||
theta=180/design_filter_order; |
|||
|
|||
% Calculates the first pole's real part using the eq. 9-102 |
|||
poles_real_parts(1,1) = -sinh(alpha_parameter)* ... |
|||
cosd(design_butterworth_angles(1,1)); |
|||
% Calculates the first pole's imaginary part using the eq. 9-103 |
|||
poles_imaginary_parts(1,1) = cosh(alpha_parameter)* ... |
|||
sind(design_butterworth_angles(1,1)); |
|||
% Calculates the first pole's radial frequency using the eq. 9-150 |
|||
poles_radial_frequencies(1,1) = (poles_real_parts(1,1)^2+ ... |
|||
poles_imaginary_parts(1,1)^2)^(1/2); |
|||
% Calculates the first pole's Q using the eq. 9-151 |
|||
inverse_poles_Q(1,1) = 1/ ... |
|||
(2*cos(atan(poles_imaginary_parts(1,1)/poles_real_parts(1,1)))); |
|||
|
|||
% Calculates the rest of the poles in the same way |
|||
for i=2:design_number_of_poles |
|||
design_butterworth_angles(1,i)=double((i-1)*theta); |
|||
% Pole's real part, eq. 9-102 |
|||
poles_real_parts(1,i) = -sinh(alpha_parameter)* ... |
|||
cosd(design_butterworth_angles(1,i)); |
|||
% Pole's imaginary part, eq. 9-103 |
|||
poles_imaginary_parts(1,i) = cosh(alpha_parameter)* ... |
|||
sind(design_butterworth_angles(1,i)); |
|||
% Pole's radial frequency, eq. 9-150 |
|||
poles_radial_frequencies(1,i) = (poles_real_parts(1,i)^2+ ... |
|||
poles_imaginary_parts(1,i)^2)^(1/2); |
|||
% Pole's Q, eq. 9-151 |
|||
inverse_poles_Q(1,i) = 1/ ... |
|||
(2*cos(atan(poles_imaginary_parts(1,i)/poles_real_parts(1,i)))); |
|||
end |
|||
else % Even number of poles |
|||
% Theta is a helper parameter |
|||
theta=90/design_number_of_poles; |
|||
|
|||
for i=1:design_number_of_poles |
|||
design_butterworth_angles(1,i)=double(90)/ ... |
|||
double(design_filter_order)+double((i-1)*theta); |
|||
% Pole's real part, eq. 9-102 |
|||
poles_real_parts(1,i) = -sinh(alpha_parameter)* ... |
|||
cosd(design_butterworth_angles(1,i)); |
|||
% Pole's imaginary part, eq. 9-103 |
|||
poles_imaginary_parts(1,i) = cosh(alpha_parameter)* ... |
|||
sind(design_butterworth_angles(1,i)); |
|||
% Pole's radial frequency, eq. 9-150 |
|||
poles_radial_frequencies(1,i) = (poles_real_parts(1,i)^2+ ... |
|||
poles_imaginary_parts(1,i)^2)^(1/2); |
|||
% Pole's Q, eq. 9-151 |
|||
inverse_poles_Q(1,i) = 1/ ... |
|||
(2*cos(atan(poles_imaginary_parts(1,i)/poles_real_parts(1,i)))); |
|||
end |
|||
end |
|||
|
|||
% Initializes array to hold the inversed poles |
|||
inverse_poles_radial_frequencies = zeros([1 design_number_of_poles]); |
|||
% Calculates inverse Chebyshev poles by inversing the Chebyshev poles |
|||
% using the eq. 9-146 |
|||
for i=1:design_number_of_poles |
|||
inverse_poles_radial_frequencies(1,i) = 1/poles_radial_frequencies(1,i); |
|||
end |
|||
|
|||
% Initializes array to hold the transfer function zeros and a temporary |
|||
% helper variable |
|||
inverse_transfer_function_zeros = zeros([1 length(design_butterworth_angles)]); |
|||
temp_index = 1; |
|||
% Calculates the transfer function's zeros using the eq. 9-143 |
|||
for i=1:2:design_filter_order |
|||
inverse_transfer_function_zeros(1,temp_index) = sec((i*pi)/ ... |
|||
(2*design_filter_order)); |
|||
temp_index = temp_index + 1; |
|||
end |
|||
|
|||
% Clears unneeded variable from workspace |
|||
clearVars = {'theta', 'i', 'temp_index', 'alpha_parameter' ... |
|||
'epsilon_parameter', 'normalized_pass_radial_frequency', ... |
|||
'poles_real_parts', 'poles_imaginary_parts', 'poles_radial_frequencies'}; |
|||
clear(clearVars{:}) |
|||
clear clearVars |
|||
|
|||
% ========== NORMALIZED DESIGN END ========== |
|||
|
|||
%% ========== ZEROS-POLES GROUPING START ========== |
|||
|
|||
% Grouping is done "by hand". The first pole, that has a radial frequency |
|||
% equal to 0.9631, is grouped with the first zero (1.0824) and the second |
|||
% pole, that has a radial frequency equal to 0.7484, is grouped with the |
|||
% second zero (2.6131). |
|||
% Two low pass notch units are required to implement the desired low pass filter. |
|||
|
|||
% ========== ZEROS-POLES GROUPING END ========== |
|||
|
|||
%% ========== UNITS IMPLEMENTATION START ========== |
|||
% AEM(3) = 6, so the circuit shown in 7.23 is going to be used for the low |
|||
% pass notch units. |
|||
|
|||
% Initializes necessary arrays, each array is 1X2, the first element (1,1) |
|||
% corresponds to the first unit and the second element (1,2) to second unit. |
|||
unit_low_pass_notch_resistors_1 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_2 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_3 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_4 = zeros([1 2]); |
|||
unit_low_pass_notch_resistors_5 = zeros([1 2]); |
|||
unit_low_pass_notch_capacitors = zeros([1 2]); |
|||
unit_low_pass_notch_gains_high = zeros([1 2]); |
|||
unit_low_pass_notch_gains_low = zeros([1 2]); |
|||
|
|||
% The specifications for the first unit are: |
|||
% - inverse pole radial frequency = 0.9631 |
|||
% - zero at 1.0824 (zero > inverse pole radial frequency) |
|||
% - pole angle = 0.5822 |
|||
|
|||
% normalized_inverse_pole_radial_frequency = 1; |
|||
normalized_transfer_function_zero = inverse_transfer_function_zeros(1,1)/ ... |
|||
inverse_poles_radial_frequencies(1,3); |
|||
|
|||
% According to the design method outlined in 7.6-B, at page 35 |
|||
unit_low_pass_notch_resistors_1(1,1) = 1; % Ohm |
|||
% Calculates the capacity of the normalized circuit capacitors using the |
|||
% eq. 7-150 |
|||
unit_low_pass_notch_capacitors(1,1) = 1/(2*inverse_poles_Q(1,3)); % Farad |
|||
% Calculates the resistance of R2 using the same equations (7-150) |
|||
unit_low_pass_notch_resistors_2(1,1) = 4*inverse_poles_Q(1,3)^2; % Ohm |
|||
% Calculates the resistance of R5 using the eq. 7-152 |
|||
unit_low_pass_notch_resistors_5(1,1) = (4*inverse_poles_Q(1,3)^2)/ ... |
|||
(normalized_transfer_function_zero^2-1); % Ohm |
|||
unit_low_pass_notch_resistors_4(1,1) = 1; % Ohm |
|||
% Calculates the resistance of R3 using the eq. 7-155 |
|||
unit_low_pass_notch_resistors_3(1,1) = (normalized_transfer_function_zero^2)/ ... |
|||
(2*inverse_poles_Q(1,3)^2); % Ohm |
|||
% Calculates the gain of this unit in high frequencies using the eq. 7-143 |
|||
unit_low_pass_notch_gains_high(1,1) = (unit_low_pass_notch_resistors_4(1,1))/ ... |
|||
(unit_low_pass_notch_resistors_3(1,1)+unit_low_pass_notch_resistors_4(1,1)); |
|||
% Calculates the gain of this unit in low frequencies using the |
|||
% eq. 7-146, 7-147, 7-148, setting s = 0 |
|||
unit_low_pass_notch_gains_low(1,1) = unit_low_pass_notch_gains_high(1,1)* ... |
|||
normalized_transfer_function_zero^2; |
|||
|
|||
% The specifications for the first unit are: |
|||
% - inverse pole radial frequency = $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ |
|||
% - zero at $$$$$$$$$$$$$$$$$$$$ (zero > inverse pole radial frequency) |
|||
% - pole angle = $$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ |
|||
|
|||
normalized_transfer_function_zero = inverse_transfer_function_zeros(1,2)/ ... |
|||
inverse_poles_radial_frequencies(1,2); |
|||
|
|||
% According to the design method outlined in 7.6-B, at page 35 |
|||
unit_low_pass_notch_resistors_1(1,2) = 1; % Ohm |
|||
% Calculates the capacity of the normalized circuit capacitors using the |
|||
% eq. 7-150 |
|||
unit_low_pass_notch_capacitors(1,2) = 1/(2*inverse_poles_Q(1,2)); % Farad |
|||
% Calculates the resistance of R2 using the same equations (7-150) |
|||
unit_low_pass_notch_resistors_2(1,2) = 4*inverse_poles_Q(1,2)^2; % Ohm |
|||
% Calculates the resistance of R5 using the eq. 7-152 |
|||
unit_low_pass_notch_resistors_5(1,2) = (4*inverse_poles_Q(1,2)^2)/ ... |
|||
(normalized_transfer_function_zero^2-1); % Ohm |
|||
unit_low_pass_notch_resistors_4(1,2) = 1; % Ohm |
|||
% Calculates the resistance of R3 using the eq. 7-155 |
|||
unit_low_pass_notch_resistors_3(1,2) = (normalized_transfer_function_zero^2)/ ... |
|||
(2*inverse_poles_Q(1,2)^2); % Ohm |
|||
% Calculates the gain of this unit in high frequencies using the eq. 7-143 |
|||
unit_low_pass_notch_gains_high(1,2) = (unit_low_pass_notch_resistors_4(1,2))/ ... |
|||
(unit_low_pass_notch_resistors_3(1,2)+unit_low_pass_notch_resistors_4(1,2)); |
|||
% Calculates the gain of this unit in low frequencies using the |
|||
% eq. 7-146, 7-147, 7-148, setting s = 0 |
|||
unit_low_pass_notch_gains_low(1,2) = unit_low_pass_notch_gains_high(1,2)* ... |
|||
normalized_transfer_function_zero^2; |
|||
|
|||
% Clears unneeded variable from workspace |
|||
clear normalized_transfer_function_zero |
|||
|
|||
% ========== UNITS IMPLEMENTATION END ========== |
|||
|
|||
%% ========== DENORMALIZATION START ========== |
|||
|
|||
% Unit sizes rescale |
|||
unit_frequency_scale_factors = zeros([1 2]); |
|||
unit_amplitude_scale_factors = zeros([1 2]); |
|||
|
|||
% For unit 1 |
|||
unit_frequency_scale_factors(1,1) = specification_stop_radial_frequency* ... |
|||
inverse_poles_radial_frequencies(1,3); |
|||
% AEM(4) = 1, so the scaling will be performed to achieve a capacitor value |
|||
% of 0.1uF using the eq. 6-33 |
|||
unit_amplitude_scale_factors(1,1) = unit_low_pass_notch_capacitors(1,1)/ ... |
|||
(unit_frequency_scale_factors(1,1)*0.1*10^(-6)); |
|||
|
|||
unit_low_pass_notch_resistors_1(1,1) = unit_low_pass_notch_resistors_1(1,1)* ... |
|||
unit_amplitude_scale_factors(1,1); % Ohm |
|||
unit_low_pass_notch_resistors_2(1,1) = unit_low_pass_notch_resistors_2(1,1)* ... |
|||
unit_amplitude_scale_factors(1,1); % Ohm |
|||
unit_low_pass_notch_resistors_3(1,1) = unit_low_pass_notch_resistors_3(1,1)* ... |
|||
unit_amplitude_scale_factors(1,1); % Ohm |
|||
unit_low_pass_notch_resistors_4(1,1) = unit_low_pass_notch_resistors_4(1,1)* ... |
|||
unit_amplitude_scale_factors(1,1); % Ohm |
|||
unit_low_pass_notch_resistors_5(1,1) = unit_low_pass_notch_resistors_5(1,1)* ... |
|||
unit_amplitude_scale_factors(1,1); % Ohm |
|||
unit_low_pass_notch_capacitors(1,1) = 0.1*10^(-6); % Farad |
|||
|
|||
% For unit 2 |
|||
unit_frequency_scale_factors(1,2) = specification_stop_radial_frequency* ... |
|||
inverse_poles_radial_frequencies(1,2); |
|||
% AEM(4) = 1, so the scaling will be performed to achieve a capacitor value |
|||
% of 0.1uF using the eq. 6-33 |
|||
unit_amplitude_scale_factors(1,2) = unit_low_pass_notch_capacitors(1,2)/ ... |
|||
(unit_frequency_scale_factors(1,2)*0.1*10^(-6)); |
|||
|
|||
unit_low_pass_notch_resistors_1(1,2) = unit_low_pass_notch_resistors_1(1,2)* ... |
|||
unit_amplitude_scale_factors(1,2); % Ohm |
|||
unit_low_pass_notch_resistors_2(1,2) = unit_low_pass_notch_resistors_2(1,2)* ... |
|||
unit_amplitude_scale_factors(1,2); % Ohm |
|||
unit_low_pass_notch_resistors_3(1,2) = unit_low_pass_notch_resistors_3(1,2)* ... |
|||
unit_amplitude_scale_factors(1,2); % Ohm |
|||
unit_low_pass_notch_resistors_4(1,2) = unit_low_pass_notch_resistors_4(1,2)* ... |
|||
unit_amplitude_scale_factors(1,2); % Ohm |
|||
unit_low_pass_notch_resistors_5(1,2) = unit_low_pass_notch_resistors_5(1,2)* ... |
|||
unit_amplitude_scale_factors(1,2); % Ohm |
|||
unit_low_pass_notch_capacitors(1,2) = 0.1*10^(-6); % Farad |
|||
|
|||
% ========== DENORMALIZATION END ========== |
|||
|
|||
%% ========== GAIN ADJUSTMENT START ========== |
|||
|
|||
total_fried_units_attenuation = unit_low_pass_notch_gains_low(1,1)*unit_low_pass_notch_gains_low(1,2); |
|||
unit_adjustment_gain = 1/total_fried_units_attenuation; |
|||
% We arbitrarily choose to use a 10KOhm series resistor in the adjustment |
|||
% unit |
|||
unit_adjustment_feedback_resistor = 10*10^3*unit_adjustment_gain; |
|||
|
|||
% ========== GAIN ADJUSTMENT END ========== |
|||
|
|||
%% ========== TRANSFER FUNCTIONS START ========== |
|||
|
|||
% Builds numerator and denominator of the transfer function of each unit |
|||
% using the eq. 7-146, 7-147 & 7-148 |
|||
unit_1_numerator = [1 ... |
|||
(((unit_low_pass_notch_gains_high(1,1)-1)/ ... |
|||
(unit_low_pass_notch_gains_high(1,1)*unit_low_pass_notch_resistors_1(1,1)* ... |
|||
unit_low_pass_notch_capacitors(1,1)))+ ... |
|||
(2/(unit_low_pass_notch_resistors_2(1,1)* ... |
|||
unit_low_pass_notch_capacitors(1,1)))+ ... |
|||
(2/(unit_low_pass_notch_resistors_5(1,1)* ... |
|||
unit_low_pass_notch_capacitors(1,1)))) ... |
|||
(1/(unit_low_pass_notch_resistors_1(1,1)* ... |
|||
unit_low_pass_notch_resistors_5(1,1)*unit_low_pass_notch_capacitors(1,1)^2)+ ... |
|||
1/(unit_low_pass_notch_resistors_1(1,1)* ... |
|||
unit_low_pass_notch_resistors_2(1,1)*unit_low_pass_notch_capacitors(1,1)^2))]; |
|||
unit_1_denominator = [1 ... |
|||
2/(unit_low_pass_notch_resistors_2(1,1)* ... |
|||
unit_low_pass_notch_capacitors(1,1)) ... |
|||
1/(unit_low_pass_notch_resistors_1(1,1)* ... |
|||
unit_low_pass_notch_resistors_2(1,1)*unit_low_pass_notch_capacitors(1,1)^2)]; |
|||
|
|||
|
|||
unit_1_transfer_function = tf(unit_1_numerator, unit_1_denominator); |
|||
unit_1_transfer_function = unit_1_transfer_function*unit_low_pass_notch_gains_high(1,1); |
|||
|
|||
unit_2_numerator = [1 ... |
|||
((unit_low_pass_notch_gains_high(1,2)-1)/ ... |
|||
(unit_low_pass_notch_gains_high(1,2)*unit_low_pass_notch_resistors_1(1,2)* ... |
|||
unit_low_pass_notch_capacitors(1,2))+ ... |
|||
2/(unit_low_pass_notch_resistors_2(1,2)* ... |
|||
unit_low_pass_notch_capacitors(1,2))+ ... |
|||
2/(unit_low_pass_notch_resistors_5(1,2)* ... |
|||
unit_low_pass_notch_capacitors(1,2))) ... |
|||
(1/(unit_low_pass_notch_resistors_1(1,2)* ... |
|||
unit_low_pass_notch_resistors_5(1,2)*unit_low_pass_notch_capacitors(1,2)^2)+ ... |
|||
1/(unit_low_pass_notch_resistors_1(1,2)* ... |
|||
unit_low_pass_notch_resistors_2(1,2)*unit_low_pass_notch_capacitors(1,2)^2))]; |
|||
unit_2_denominator = [1 ... |
|||
2/(unit_low_pass_notch_resistors_2(1,2)* ... |
|||
unit_low_pass_notch_capacitors(1,2)) ... |
|||
1/(unit_low_pass_notch_resistors_1(1,2)* ... |
|||
unit_low_pass_notch_resistors_2(1,2)*unit_low_pass_notch_capacitors(1,2)^2)]; |
|||
|
|||
unit_2_transfer_function = tf(unit_2_numerator, unit_2_denominator); |
|||
unit_2_transfer_function = unit_2_transfer_function*unit_low_pass_notch_gains_high(1,2); |
|||
|
|||
tmpHpRC = double(1)/double(0.1*10^(-6)*4.14*10^3); |
|||
unit_3_transfer_function = tf([0 tmpHpRC], [1 tmpHpRC]); |
|||
|
|||
total_transfer_function = series(series(unit_1_transfer_function,unit_2_transfer_function), unit_3_transfer_function); |
|||
total_transfer_function = total_transfer_function*unit_adjustment_gain; |
|||
%plot_transfer_function(unit_1_transfer_function, [1 10]); |
|||
%plot_transfer_function(unit_2_transfer_function, [1 192.82 222.817 234 238.73]); |
|||
%plot_transfer_function(0.656*series(unit_1_transfer_function,unit_2_transfer_function), ... |
|||
% [1 159.155 192.82 222.817 234 238.73]); |
|||
%{ |
|||
plot_transfer_function(total_transfer_function, ... |
|||
[1 ... |
|||
specification_pass_radial_frequency/(2*pi) ... |
|||
design_half_power_radial_frequency/(2*pi) ... |
|||
specification_stop_radial_frequency/(2*pi)]); |
|||
%} |
|||
|
|||
%ltiview(unit_1_transfer_function); |
|||
%ltiview(unit_2_transfer_function); |
|||
%ltiview(unit_3_transfer_function); |
|||
%ltiview(total_transfer_function); |
|||
%ltiview(unit_1_transfer_function, unit_2_transfer_function, ... |
|||
% unit_3_transfer_function, total_transfer_function); |
|||
|
|||
% Clears unneeded variable from workspace |
|||
clear tmpHpRC |
|||
clear -regexp _numerator$ |
|||
clear -regexp _denominator$ |
|||
clear -regexp _transfer_function$ |
|||
|
|||
% ========== TRANSFER FUNCTIONS END ========== |
Loading…
Reference in new issue