Design optimization through an analytical approach using numerical algorithms helps engineers to develop optimal designs through simulation.
The recent trend in system design is increasing number of subsystems to achieve several functions with uncompromised quality. The industry calls for optimized designs, be it for efficiency, size, weight, or noise emissions. For example, in the design of a switching power supply, the filter component values are calculated based on the worst case values of voltages and currents. The other variables of the filters are also considered with significant margins. So, the product can be bigger in size, heavier in weight and costlier too. This is just a simple example used to understand the purpose of design optimization.
In the traditional approach, the system optimization requires several prototype iterations, many samples of various components, increased testing cost, extended project timelines and resources. This is because, in a manual approach, verifying the impact of variation for each component parameter based on an optimization objective, requires several samples of components. In the process, the prototype is tested with each sample added in the design and so, the prototype should be serviced, and the entire test process is executed. If the change shows a positive trend for the optimization objective, the parameters are further altered in the same direction and the hardware is tuned again. This process is continued till the optimization objective is achieved. It is clear from this procedure that it involves huge manual effort in addition to samples, testing, time, etc.
Filter Design for a Switching Power Supply:
Let us look at the design of EMI filters for a Switching Power Supply. The filters should be designed to attenuate noise that is being injected to the power supply source so that their limit meets the standards such as FCC and CISPR 22. There are several frequencies in a switching power supply. The switching frequency, any LC resonances in case of resonant converters, frequencies from device parasitic elements and the source frequency noise as shown in Figure 2. Calculating the noise magnitudes at different frequencies is not trivial. Even though engineers often use several iterations to select the most appropriate filter, they end up adding filters having substantial margins at times. Such filters might affect the overall system performance by making the response sluggish.
So, designing appropriate filters is essential. In a hardware design and testing approach, the prototype is first built and tested at the required test conditions. Expensive measurement instruments such as spectrum analyzers are used to examine the noise profile. With these measurements, the filters are designed and tuned. In this manual method, there are always constraints in experimenting all the test scenarios on a hardware prototype. Hence, this approach may not cover all the possible combinations of noise patterns. This eventually leads to more hardware iterations during complete system testing.
Availability of Simulation:
Various simulation tools are available in the market that allow engineers to execute performance studies on the designs at different abstraction levels. In the system design process, understanding the comprehensive performance through system simulation is the order of the day. The system simulation platforms are equipped with tools and applications that can accurately model the system behavior. Apart from modeling the system behavior, a wide range of simulations can be done to prove the system performance under different conditions of load, source, and other effects. Extending the system simulation capability to design optimization opens the gateway to optimizing designs during simulation phase.
Design Optimization using Simulation:
In a simulation environment, various component parameters can be altered and their impact on optimization objectives can be checked. Engineers that have the complete system modeled in a simulation tool, with the capability to vary different parameters, can analyze the effect of the parameter changes through simulation. Take the case of the filter design for Switching Power Supply. The engineer has the design fully modeled and simulated for measurement of noise magnitudes at different frequencies. With these noise measurements, the appropriate filters can be designed and tested in the simulation environment. Further tweaking can be done by varying the filter parameters and the noise levels on the input side can be checked. This process can be iterated to achieve an optimized design as shown in Figure 3.
In this process, general purpose engineering calculation tools can be used where the time domain simulations are done in some tool and the parameters from the simulation are used for optimization through equations. This results in solving large matrices and extra effort for engineers to author and validate those equations. Other tools can also provide this functionality through dedicated signal-flow type blocksets but require significant modeling efforts.
And by the way, simulation tools such as Saber offers an automated approach for finding the worst case scenarios and for performing design optimization. The designs that are used for nominal analysis are used for optimization without any additional manual exertion on the design schematics to match the tool’s flow. The optimization objective can be maximizing system efficiency, minimizing noise emissions, reducing switching transients, optimizing dead times between switches, etc. In Saber, the utility is called the Worst Case Analysis (WCA) tool.
Worst Case Analysis:
The Worst Case Analysis tool has built-in numerical algorithms to automatically search for the most optimal combination of the circuit parameters that satisfy the optimization objective, as shown in Figure 4. Engineers can select the components whose parameters influence the objective function and assign a range for the algorithm to search for the optimum combination that meets the goal.
Details of nomenclature used in the design optimization in Saber:
1) Performance Objective or Objective Function: A user defined function such as min(dead_time), max(efficiency), min(rise_time), etc. This is a function applied on the measurements from a simulation, to achieve the design goal for optimization. The numerical algorithms existing in the tool performs a search by varying all the selected parameters until the goal is reached.
2) Analyses and measurement: The design is simulated for the specified analysis to reveal the performance of the design for each parameter combination. For each circuit parameter combination, the selected analysis is run and measurements are applied on the signal waveforms. These measured values are substituted in the objective function. So, performing appropriate measurements is the key to creating a meaningful objective function and hence for arriving at optimal designs.
3) Parameters: These are the parametric values of the components in the design that are varied for optimizing the design. The optimization algorithm varies the component values to achieve the optimization goals. These parameters can be stochastic in nature in which case, the WCA tool can extract these parameters automatically for optimization, on the push of a button. You can add additional parameters and a range can be specified for the parameters. The numerical algorithm varies each parameter in the given range while making sure that the objective function is satisfied.
Saber also has the capability to perform Sensitivity Analysis, a powerful analytical approach for automatically finding the appropriate design parameters having the highest influence on desired output for optimization. This Sensitivity Analysis helps for deciding the parameters to be included in the optimization process and can be performed upfront. This is an analysis that determines the impact of parameters on measurements/signals for specified perturbation.
Design optimization through an analytical approach using numerical algorithms helps engineers to develop optimal designs through simulation. The manual process involves huge efforts and is often limited to only a few design parameters. To achieve an optimized design, which is only increasing in importance, the use of simulation is required.
-- Balaji Siva Prasad Emandi is a Corporate Application Engineer for Saber in the Verification Group at Synopsys. Balaji enthusiastically supports pre- and post-sales customers in the domain of simulation and development of Power Electronic subsystems.