Seeking help with gradient descent optimization for Nonlinear Transmission Line

Question

Syed Abid Sahdman 2024-10-22

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/2161300-seeking-help-with-gradient-descent-optimization-for-nonlinear-transmission-line

编辑： Walter Roberson 2024-10-22

I am trying to optimize the parameters L and Cjo of this NLTL circuit using Gradient Descent Optimization. I am trying to get a peak voltage of 20V. But I am getting unrealistic values and iteration is not running for more than once. For Example, this is an output I am getting:

Iteration 1: Error = 124.4584

Optimized Parameters:

1.0e-05 *

0.2700 0.0000

I am not sure if I am defining the desired result or the compute gradient correctly. Could someone help me fix the issues? I am providing the circuit and the type of waveform I want to generate with peak of 20V. I am providing the code that I have:

% Define initial parameters
params = [2.7e-6, 61.3e-12];  % [L_value, Cjo_value]
learning_rate = [1e-3, 1e-6];  % Separate learning rates for L and Cjo
tolerance = 1e-9;
max_iterations = 10;
desired_error_limit = 1e-5;
% Objective function to minimize
objective_function = @(params) run_simulation_and_get_error(params);
% Gradient descent loop
for iter = 1:max_iterations
    % Calculate the current error
    current_error = objective_function(params);
    
    % Compute the gradient for each parameter
    gradient = compute_gradient(params, objective_function);
    
    % Update parameters using gradient descent with different learning rates
    params(1) = params(1) - learning_rate(1) * gradient(1);  % L_value update
    params(2) = params(2) - learning_rate(2) * gradient(2);  % Cjo_value update
    
    % Check for convergence
    if norm(gradient) < tolerance || current_error <= desired_error_limit
        if current_error <= desired_error_limit
            disp('Converged based on desired error limit');
        else
            disp('Converged based on gradient norm');
        end
        break;
    end
    
    fprintf('Iteration %d: Error = %.4f\n', iter, current_error);
end
Error using fread
Invalid file identifier. Use fopen to generate a valid file identifier.

Error in solution>update_netlist_file (line 63)
fileContent = fread(fid, '*char')';

Error in solution>run_simulation_and_get_error (line 49)
update_netlist_file(netlistFile, L_value, Cjo_value);

Error in solution>@(params)run_simulation_and_get_error(params) (line 8)
objective_function = @(params) run_simulation_and_get_error(params);
% Final optimized parameters
fprintf('Iteration %d: Error = %.4f\n', iter, current_error);
disp('Optimized Parameters:');
disp(params);
% Function to run LTspice simulation and get error
function error = run_simulation_and_get_error(params)
% Parameters
L_value = params(1); % Extract L from params
Cjo_value = params(2); % Extract Cjo from params
% File paths
netlistFile = ['C:\Users\s007s364\Downloads\'...
    'LTspice_2_Matlab\oscillatory.net'];
rawFile = ['C:\Users\s007s364\Downloads\'...
    'LTspice_2_Matlab\oscillatory.raw'];
ltspiceExe = 'C:\Program Files\ADI\LTspice\LTspice.exe';
% Update the netlist file with the new L and Cjo values
update_netlist_file(netlistFile, L_value, Cjo_value);
% Run LTspice simulation
run_ltspice_simulation(ltspiceExe, netlistFile);
% Extract results from LTspice output
simulationResults = extract_simulation_results(rawFile);
% Get the desired target waveform
targetWaveform = get_desired_results(simulationResults);
% Compute the error between simulation results and the target waveform
error = compute_error(simulationResults, targetWaveform);
end
% Function to update the netlist with new L and Cjo values
function update_netlist_file(netlistFile, L_value, Cjo_value)
% Read the netlist file
fid = fopen(netlistFile, 'r');
fileContent = fread(fid, '*char')';
fclose(fid);
% Update the L parameter in the netlist
newContent = regexprep(fileContent, 'L\s*=\s*\d+\.?\d*e?-?\d*', sprintf('L=%.6f', L_value));
% Update the Cjo parameter in the .model statement
newContent = regexprep(newContent, 'Cjo\s*=\s*\d+\.?\d*e?-?\d*', sprintf('Cjo=%.6fe-12', Cjo_value));  % Convert Cjo to picofarads
% Write the updated content back to the netlist file
fid = fopen(netlistFile, 'w');
fwrite(fid, newContent);
fclose(fid);
end
% Function to run LTspice simulation
function run_ltspice_simulation(ltspiceExe, netlistFile)
% Execute LTspice simulation
command = sprintf('"%s" -Run -b "%s"', ltspiceExe, netlistFile);
disp(['Command: ', command]);
[status, cmdout] = system(command);
if status ~= 0
    error('LTspice simulation failed: %s', cmdout);
else
    disp('LTspice simulation completed successfully.');
end
% Pause to ensure the simulation has time to complete
pause(5);
end
% Function to extract simulation results
function results = extract_simulation_results(rawFile)
% Use LTspice2Matlab or another method to read the .raw file
raw_data = LTspice2Matlab(rawFile);
variable_to_plot = 22;  % Adjust based on your netlist output variable
voltage = raw_data.variable_mat(variable_to_plot, :);  % Extract voltage data
results = voltage;  % Return the full waveform (not just max voltage)
end
% Function to compute the error between simulation and desired results
function error = compute_error(simulationResults, targetWaveform)
% Compute Mean Squared Error (MSE) between simulated and target waveform
error = mean((simulationResults - targetWaveform).^2);
end
% Function to create a target waveform that peaks at 20V
function targetWaveform = get_desired_results(simulationResults)
% Scale the waveform to peak at 20V based on the structure of simulation output
max_simulation_voltage = max(simulationResults);
scaling_factor = 20 / max_simulation_voltage;  % Scale to 20V peak
targetWaveform = simulationResults * scaling_factor;
end
% Function to compute gradient (finite differences)
function grad = compute_gradient(params, objective_function)
delta = [1e-6, 1e-9];  % Set appropriate delta values for L and Cjo
grad = zeros(size(params));
for i = 1:length(params)
    params_plus = params;
    params_plus(i) = params_plus(i) + delta(i);  % Use appropriate delta for each parameter
    grad(i) = (objective_function(params_plus) - objective_function(params)) / delta(i);
end
end