Feature extraction and data compression system and method for distributed sensor networks

A distributed sensor network for locating and classifying signal sources comprising:

• a base station, and
• at least one cluster having a plurality of sensor nodes with each sensor node including:
• a sensor for receiving a signal,
• an analog to digital converter for converting said signal into a digital signal,
• a field programmable gate array connected to said analog to digital converter, and programmed to divide a selected time increment of said digital signal into blocks of time increments of said signal that are smaller than said selected time increment, perform a first Discrete Cosine Transform on each said block, select a plurality of frequency peaks for each said block, select a plurality of subbands of said signal based on the frequency of occurrence of said frequency peaks in said blocks, perform a second Discrete Cosine Transform on said selected time increment of said digital signal, and
• a wireless transceiver for collaborating, in cooperation with said field programmable gate array, with other sensor nodes in said cluster to select a plurality of common subbands from subbands selected by said nodes, and for transmitting said common subbands of said digital signal to said base station,
• whereby said common subbands include features relevant to a source of said signal and said base station locates and classifies said source by processing said common subbands received from said sensor nodes.

A method of feature extraction and data reduction of a signal received by a sensor node in a cluster of sensor nodes in a sensor network with said sensor network including a base station, comprising the steps of:

• dividing a selected time increment of said signal into blocks of time increments of said signal that are smaller than said selected time increment,
• performing a first Fourier-based transform on each said block,
• selecting a plurality of frequency peaks for each said block,
• selecting a plurality of subbands of said signal based on the frequency of occurrence of said frequency peaks in said blocks,
• collaborating with other sensor nodes in said cluster to select a plurality of common subbands from subbands selected by said nodes in said cluster, after said step of selecting a plurality of subbands,
• performing a second Fourier-based transform on said selected time increment of said signal, and
• transmitting said common subbands of said signal for said time increment to said base station,
• whereby said common subbands include features relevant to a source of said signal and said common subbands reduce data relative to said signal for said selected time increment.

A method of feature extraction and data reduction of a signal received by a sensor node in a cluster of sensor nodes in a sensor network with said sensor network including a base station, comprising the steps of:

• dividing a selected time increment of said signal into blocks of time increments of said signal that are smaller than said selected time increment,
• performing a first Discrete Cosine Transform on each said block,
• selecting five frequency peaks for each said block,
• computing a histogram of said frequency peaks,
• selecting three subbands of said signal based on the frequency of occurrence of said frequency peaks in said blocks,
• collaborating with other sensor nodes in said cluster to select three of common subbands,
• performing a second Discrete Cosine Transform on said selected time increment of said signal,
• encoding said common subbands of said signal for said selected time increment, and
• transmitting said common subbands of said signal for said selected time increment to said base station,
• whereby said common subbands include features relevant to a source of said signal and said common subbands reduce data relative to said signal for said selected time increment.