THE AUTOMATIC CLASSIFICATION OF MODULATION TYPES BY PATTERN RECOGNITION.
STANFORD UNIV CALIF STANFORD ELECTRONICS LABS
Pagination or Media Count:
This report presents the preliminary results of an investigation into the use of pattern-recognition techniques to rapidly and automatically identify the type of modulation on a high-frequency radio signal. Classes of modulation initially considered include double-sideband AM, upper and lower single-sideband suppressed carrier, CW, high- and low-speed teletype single-channel FSK, multichannel FSK, and on-off keying Morse code. The spectrum of the signal is measured by a digital analyzer whose outputs are classified by a pattern recognizer. The spectrum analyzer and classifier are realized on a PDP-8 digital computer. The new nearest neighbor type of pattern recognizer has been developed that significantly increases classification accuracy. The decision surfaces of this classifier asymptotically approach the Bayes decision surfaces with simple set size. Mis-classification rates of 5 to 10 percent have been obtained with signals recorded in a typical HF environment. Important characteristics of the system are the ability to recognize the presence of a signal when the modulation format is unknown and the ability to recognize the presence of a new signal that has not been previously encountered. Author
- Radio Communications