Home | People | CWT Research | Publications | Presentations | Resources

Intelligent Machines

Intelligent Machines can be thought of as an optimization process to produce the best outcome given a set of inputs (via monitoring of ongoing actions). Science Magazine describes it as,

"Adaptive goal-directed behavior involves monitoring of ongoing actions and performance outcomes, and subsequent adjustments of behavior and learning."

   -Science

Biologically-inspired algorithm are highly adept at multi-objective optimization. They use genetic algorithms to evolve the radio. These algoritms treat the radios like a biological system. The traits (knobs) of the radio are genes in a chromosome, and the GA manipulates the chromosome to create a better set of traits adapted for the current environment.

History

While there were discussions and debate going on beforehand, Turing's paper "Computing Machinery and Intelligence" was a groundbreaking paper in the development of modern artificial intelligence thought. Before this, we can look to the work of Max Black who defined the concept of vagueness in logic ("Vagueness: an Exercise in Logical Analysis"), published in 1937, which is the first published work in the direction of Fuzzy Logic (see Zadah's paper, too). McCulluch and Pitts in 1943 then defined a way of looking at brain neurons as mathematical functions (this is also often credited with the first view of a computer as a finite state machine).

AI has had a roller-coaster of a ride. These historical facts are just some of the major ones that began the field. A lot of work and ideas has developed over the past 50 - 60 years. We are now getting to the point when the computational power and our understanding of intelligence are advanced enough to really see these algorithms in useful action.

There are a number of successful systems based on AI techniques, although they are usually narrowly-defined problems in which they succeed.

The following list describes the most popular and well-known AI algorithms and techniques:

Applicaiton to Radios

All of the intelligent algorithms are trying to optimize a performance metric.In the cognitive radio case, we are given a QoS need by the applications in use, and we have environmental context through our sensors and awareness capability. We want to adapt the radio parameters to optimize the performance with respect to the inputs. Radio optimization depends on many different objectives that define the quality of service and user satisfaction that is user, application, and context sensitve. Figure 1 shows a basic system diagram of a transmitter physical and MAC (medium access control) layers to show how many parameters we have to adjust to optimize the quality of service.

Figure 1. Basic block diagram of a radio transmitter.

The parameters on these layers we can monitor and measure to determine the QoS include: BER, spectrum occupancy, SINR, symbol rate, FER, data rate, delay, computational complexity, and power consumption. No one of these metrics will be enough to measure the QoS of an application by itself.

Application of AI Techniques

The external data provides environmental context to the solution, and Neural networks are great at pattern recognition are greate pattern recognizers. Here, we can use temporal statistics of a radio envionmnet to classify a signal's modulation type. Figure 2 shows part of the neural network, and Figure 3 shows some of the feature space results of different modulations, which show how these features can be used by the neural network to distinguish modulations.

Figure 2. One Class One Network Neural Net use to realize modulation classification.Figure 3. Feature space represntation of different modulations.

Biologically-Inspired Radio Optimization

We use genetic algorithms to evolve the radio given input from the environment, the user, the application, the radio hardware (capabilities and limitaitons), and the regulatory policy. The genetic algorithm treats the radio like a biological system that evolves over generations. The traits (knobs) of the radio are genes in a chromosome, and the GA manipulates the chromosome to create a better set of traits adapted for the current inputs.

References