# Information Based Sensor Management with Kenneth Hintz

Artech author Kenneth Hintz gave us insights on sensor systems and the use of information theory:

Search, track, or identify?  How do you put these sensing actions on an equal footing and decide among observing relevant data since a sensor system cannot perform all of these actions at the same time and cannot look in all directions simultaneously.  The solution is to use information theory and mission value as major components to a sensor management objective function.  Information theory can be used because it reduces these conflicting mission goals to a commensurate measure space with units of bits of information.  Mission value because the purpose of a system is to accomplish a mission, and all of the actions of a system, including sensing actions, can be measured in terms of how much that action contributes to the accomplishment of a mission…a dimensionless number.

While information theory is generally applied to the coding of signals transmitted through a bandwidth limited, noisy communications channel, information is a more general theory predicated on changes in our uncertainty about a random variable.  While there are several mathematical ways for computing information, changes in entropy is straightforward and applicable to the sensor management problem.  For example, search information can be based on changes in the spatial distribution of uncertainties about the presence of targets; track information can be based on changes in our uncertainty about the physical state vector of the target in track; and, identification information can be predicated on reduced uncertainty of a targets type from friendly/unfriendly, to class, to airframe, and finally to a target’s individual identification.  This example deals with sensor information, the decrease in uncertainty about a specific random variable in our model of the physical world and different sensors and sensing modes can provide different amounts of these informations.

Before we get too far, it is important to note that information is a dynamic quantity, not a thing.  That is, information is the change in uncertainty associated with something in our environment.  Information does not exist on its own.  The repository of the information we have is our knowledge of the situation we are interested in.  Information is not data and data may or may not contain information.  Proper selection of data from physical, social, cyber, or databases can provide information by decreasing our uncertainty about an item of interest, but improper choice of data can result in no gain in information or even a loss of information.  There is a temporal aspect to knowledge and it is that without sensing actions to keep reducing or maintaining a desired level of uncertainty, our uncertainty grows until we know little about the situation.  We need information to restore our knowledge to an acceptable level.  Furthermore, the same data may provide one person’s knowledge with information while contributing no information to another person’s knowledge because they are interested in different situations even though they are looking at the same world.  So that brings in context.  The same data may result in information in one context but no information in another context.  This must all be taken into account when deciding what data will provide the right situation information independent of which sensor is used to obtain that data.

But information is not enough.  The mission value of each decrease in uncertainty is different and calculable utilizing a method called a goal lattice.  The topmost, soft, difficult to measure, mutually exclusive mission values can be used to weight the available situation informations in order to determine which is the best next collection opportunity without deciding how to obtain that information.  Through a process of determining those subgoals and their values which are necessary for the accomplishment of the topmost goals, the mission values of the lowest real, measurable sensing actions can be computed.  These sensing action goal values can be combined with the admissible sensing actions which can obtain the sensor information from which to realize the situation information and chose the best next sensor collection opportunity.