HyperLynx Design Space Exploration

Initially, users modify the design and re-simulate changes one at a time. This works well for simple studies and is easy for new simulation users to understand. This method works best when there are only one or two design parameters (variables) to be studied and when the user can readily determine the parameter values to use for the next study based on the results of previous ones.

When the number of combinations of design parameters (permutations) to be studied becomes more than a handful, a more consistent and organized methodology is needed. Swept-parameter analysis lets the user define the parameters and associated values to be studied, and then uses an automated methodology to simulate the behavior of each case, building a table of the results so that the best can be identified. Key to this methodology is the selection of a quantitative metric, or figure of merit - a numeric value derived from the simulation results that serves as an indication of the design's behavior. Swept-parameter analysis comes in two categories - full analysis, where all combinations of all design variables are explored, and reduced set analysis, where only a subset of all combinations are explored.

Reduced set analysis is important because the number of permutations in a full analysis quickly grows as variables and values are added to the mix. Full analysis evaluates all combinations of all variables, whether they make sense or not. Reduced set analysis helps keep simulation times manageable under these conditions, by allowing users to leverage their design intuition and specifying ways to limit the number of cases explored.

For large studies, the number of cases to be explored can grow into the millions, and traditional exploration is no longer possible. In these cases, the subset of all permutations that will be simulated must be picked very efficiently using advanced mathematical sampling methods, because only a small percentage of the total permutation count can be practically simulated. This is what advanced optimization tools do - the user defines the range of design variables and values to be explored (the design space), the output metrics used to compare design behavior (the responses), the values for those responses that define design success or failure (the design targets) and, for each response, whether the intent is to maximize or minimize its value (the design goals).

The response surface is defined as the n-dimensional space created by the value of each metric over the range of inputs - simply put, a plot of the output values. When only one variable exists, the response surface is an 2D (x-y) plot, or a curve. When two input variables exist, the response surface is a 3D plot, with the output metric plotted on the Z axis. When three input variables exist, the response surface becomes 4 dimensional, and so on. These higher order response surfaces are hard for humans to internalize, but the additional complexity doesn't make much difference to computers and mathematical algorithms, other than the time, memory and calculations required. Thus, advanced optimization tools (HL-DSE is one of them) combine response surface-based analysis with sophisticated numerical algorithms that use simulation experiments to automatically investigate a large design space as efficiently and comprehensively as possible.