Previous article: ADS Tutorial (V) Real Time Tuning

In this section, we will take a look at how to perform optimization in ADS.

In the previous step, we created a low-pass filter and performed preliminary optimization using Real Time Tuning. To get better performance, we can perform Optimization in ADS. First, select “Optim/Stat/DOE” in the Library Palettes on the left side of the main design window:

Then click and put in the Optimization Controller, then click and put in the two optimization objectives.

Then we need to define the specific contents of the two GOALs:

In the first GOAL, we set the constraint on dB(S(2,1)) (the corresponding name can be viewed directly on the left side of the plot obtained by simulation, without having to guess it yourself). First, we define Expression as dB(S(2,1)), which represents the dB value of the S parameter S(2,1). Then, select the type of analysis (Analysis) to be optimized. Here, select SP1, which represents S parameter analysis. Then Weight represents the weight of the goal, which determines the importance of the goal in the overall optimization. The default value is 1.0. Next, select Sweep Variables as freq. Then set Limit lines. The specific content is shown in the figure. It is hoped that the designed low-pass filter has a low insertion loss (< -1 dB) in the passband (0.01 GHz to 0.2 GHz) and a high attenuation (< -25 dB) in the stopband (0.35 GHz to 0.5 GHz). Then click “OK”.

Similarly, we set the second GOAL. The constraint here requires that S(1,1) must be less than -15 dB in the frequency range of 0.01 GHz to 0.2 GHz. This means that in this frequency band, the reflection of the input port needs to be very small and the impedance matching must be good.

Next, we need to set up the Optimization Controller:

In “Setup”, there are multiple optimization types. We select the “Random” type this time and set the Number of iterations to 5000. The explanations of each optimization type and other contents are as follows:

• Random: Uses a random search algorithm for optimization. It is suitable for global optimization problems and can explore the entire design space.
• Gradient: Use the gradient descent algorithm for optimization. It is suitable for continuous and differentiable objective functions and converges quickly to the local optimal solution.
• Random Minimax: Combines random search and minimum-maximum optimization methods. It is suitable for optimization problems that need to take into account multiple objectives.
• Gradient Minimax: Combines gradient descent and minimax optimization methods. Suitable for multi-objective optimization, using gradient information to speed up convergence.
• Quasi-Newton: Use the quasi-Newton method for optimization. Suitable for optimization problems that require fast convergence, using second-order derivative information.
• Least Pth: Use the least Pth method for optimization. Applicable to optimization of specific types of objective functions.
• Minimax (minimum maximum): uses the minimum maximum optimization method to balance the optimization problem of multiple objective functions.
• Random Max: Uses random search to find the maximum value. Useful for global maximization problems.
• Hybrid: Combining multiple optimization algorithms for optimization, suitable for complex optimization problems.
• Discrete: An optimization method for discrete variables. Applicable to optimization problems where the variables take discrete values.
• Genetic: Use genetic algorithms for optimization. Applicable to complex global optimization problems and simulate biological evolution processes.
• Simulated Annealing: Use the simulated annealing algorithm for optimization. It is suitable for global optimization problems and simulates the physical annealing process.
• Simulated Annealing Corana: Corana is an improved simulated annealing algorithm. It improves the efficiency and convergence of simulated annealing.

In addition, Number of iterations sets the maximum number of iterations for the optimization process, and Desired error sets the desired error at which the optimization process stops. When the optimization result reaches this error level, the optimization process stops (0.0 means reducing the error as much as possible).

On the “Display” page, we can choose to optimize the content presented by the controller. In our experiment, first click “Clear All” to clear all options, and then select “OptimType” and “MaxIters”:

The resulting interface is as follows:

Next, we need to define the variables inside the component. We select “Simulate – Simulation Variables Setup”:

Then select the “Optimization” interface, select the parameters in the design, and then set their maximum and minimum values. There are multiple formats here, which can be selected as needed:

Click “OK” to complete the settings. After completing the Optimization settings, we click Execute Optimization in the toolbar, and then ADS will automatically start the optimization. This window displays a lot of information, such as optimization type, status, number of iterations , etc. On the right side of the window, the historical changes in the error during the optimization process and the contribution of each optimization target to the total error are displayed. It also shows the error and contribution of the dB value optimization targets of S (2,1) and S(1,1).

Wait for execution, and the optimized result will pop up after execution:

Since the termination reason at this time is that the number of iterations has reached the set number, it may not meet the goal we set. We can select and set different ranges for the parameters to let it try more different combinations, and then click to optimize again:

At this time, the termination reason becomes Goals are satisfied, which means that the optimization is successful and has met our current design requirements.

At this point, we click and observe the sub-design, and find that the parameters of the components in the sub-design have changed to a combination that meets the requirements:

Finally, click to perform a simulation to verify whether the previous operation is successful.

Next article: ADS Tutorial (VII) Monte Carlo Simulations