The best military minds in history have all had a knack for predicting what the enemy might do and then coming up with ways to counteract it. Simply put, brilliant military minds are very capable of predicting what’s coming based on observation. In a modern military environment that relies so much on technology, is it possible to translate that predictive capability to machines? Absolutely.
Bayesian analysis can be applied to signal processing technologies for more accurate defense analysis. Though the theory behind it, known as Bayes’ theorem, has been around since the mid-18th century, Bayesian analysis is comparatively new in military signal processing. Yet it has proven to be remarkably helpful.
The Bayes Theorem
The Reverend Thomas Bayes was not only a Presbyterian minister, he was also an accomplished statistician and philosopher. His theorem stated that it was possible to use statistics and mathematics to predict the probability of a future event based on past conditions and circumstances.
The theorem begins with a hypothesis. For example, if X and Y conditions exist, Z is likely to occur as a result. The accuracy of any such predictions relies in the accuracy of the data describing the X and Y conditions. Understanding this is critical when applying Bayes’ theorem to signal processing for military defense.
Signal Processing Basics
Rock West Solutions, a California company that provides data and signal processing services for defense contractors, explains that signal processing is a science through which usable data is extracted from a variety of different signals – e.g., acoustic, images, etc. It is accomplished with a combination of hardware and software.
The holy grail of signal processing is always to remove as much noise as possible so that usable data is pure. The more noise you can remove, the more reliable the data is. This is where Bayesian analysis can be quite helpful. Properly devised and applied Bayesian filters can go a long way toward eliminating noise.
Think of the Bayesian filter in terms of the spam folder in your e-mail program. A Bayesian e-mail filter analyzes e-mails based on known data points. If enough of those data points match up, the filter predicts that the e-mail in question is spam and marks it as such.
The thing about Bayesian e-mail filters is that they get better with use. If you regularly train a Bayesian filter, its accuracy increases. The more training, the more accurate it becomes. Why? Because it has more accurate data on which to base its prediction.
Bayesian Analysis and Defense
We can apply the same principles of the Bayesian e-mail filter to signal processing in the defense arena. You start with a hypothesis intended to predict what the enemy might do. Then you use past data points to create rules for filtering out noise from intelligence signals.
Bayesian rules can be established that account for past observations and signal noise. For example, if a software system analyzing a particular signal recognizes data points X, Y, and Z within that signal, a rule can be established that says those data points are noise. They can be removed from the signal before predictive capabilities are applied.
What has been described here is by all means rudimentary. There is a lot more to it than this basic illustration affords. Yet it is enough to get the point across. Bayesian analysis improves military predictive capabilities through signal processing based on observable data from past circumstances. The more accurate the data applied to Bayesian analysis is, the more accurate signal processing becomes. And accurate signal processing generally means improved predictive results over time.