Method and system for real-time signal classification

Abstract

A method to achieve an accurate, extremely low power state classification implementation is disclosed. Embodiments include a sequence that matches the data flow from the sensor transducer, through analog filtering, to digital sampling, feature computation, and classification.

We claim: 
     
         1 . A method for real-time signal classification, the method to be executable on a low-power processor, the method comprising:
 providing a managed bandwidth signal;   establishing a time bound;   windowing, within the time bound, the managed bandwidth signal to produce a vector;   performing, within the time bound, feature computation using fixed point arithmetic on the vector to produce a model-compatible vector;   projecting, within the time bound, the model-compatible vector having a first multi-dimensional factor space onto a second multi-dimensional factor space where the second multi-dimensional factor space is smaller than the first multi-dimensional factor space to produce a lower dimensional factor vector; and   evaluating, within the time bound and using log representations, the lower dimensional factor vector against a model to produce a classification result.   
     
     
         2 . The method of  claim 1  wherein providing a managed bandwidth signal further includes:
 receiving a signal from an accelerometer sensor; and 
 filtering the signal t place a limit on the frequency of the signal. 
 
     
     
         3 . The method of  claim 2  wherein filtering further includes shifting a center point of the signal. 
     
     
         4 . The method of  claim 2  wherein filtering further includes scaling the amplitude of the signal. 
     
     
         5 . The method of  claim 1  wherein providing a managed bandwidth signal further includes:
 receiving a data from a digital sensor; and 
 filtering the data to produce a managed bandwidth signal. 
 
     
     
         6 . The method of  claim 1  wherein windowing further includes grouping a sequence of samples to form a vector that represents dynamic behavior of the signal over time. 
     
     
         7 . The method of  claim 1  wherein performing feature computation further includes transforming the vector from a time domain to a frequency domain. 
     
     
         8 . The method of  claim 1  wherein performing feature computation further includes transforming the vector from a time domain to a frequency domain. 
     
     
         9 . A system for real-time signal classification, the system having a lower power processor, the system comprising:
 means for providing a managed bandwidth signal;   means for establishing a time bound;   means for windowing, within the time bound, the managed bandwidth signal to produce a vector;   means for performing, within the time bound, feature computation using fixed point arithmetic on the vector to produce a model-compatible vector;   means for projecting, within the time bound, the model-compatible vector having a first multi-dimensional factor space onto a second multi-dimensional factor space where the second multi-dimensional factor space is smaller than the first multi-dimensional factor space to produce a lower dimensional fact vector; and   means for evaluating, within the time bound and using log representations, the lower dimensional factor vector against a model to produce a classification result.   
     
     
         10 . The system of  claim 9  wherein the means for providing a managed bandwidth signal further includes:
 means for receiving a signal from an accelerometer sensor; and 
 means for filtering the signal to place a limit on the frequency of the signal. 
 
     
     
         11 . The system of  claim 10  wherein the means for fdtering further includes means for shifting a center point of the signal. 
     
     
         12 . The system of  claim 10  wherein the means for fdtering further includes means for scaling the amplitude of the signal. 
     
     
         13 . The system of  claim 9  wherein the means for providing a managed bandwidth signal further includes:
 means for receiving data from a digital sensor; and 
 
     
     
         14 . The system of  claim 9  wherein the means for windowing further includes means for grouping sequence of samples to form a vector that represents dynamic behavior of the signal over time. 
     
     
         15 . The system of  claim 9  wherein the means for performing feature computation further includes means for transforming the vector from a time domain to a frequency domain. 
     
     
         16 . The system of  claim 9  wherein the means for performing feature computation further includes means for digitally filtering the vector. 
     
     
         17 . The system of  claim 9  wherein the means for performing feature computation further includes means for computing time derivatives of the vector. 
     
     
         18 . The system of  claim 9  where in the means for performing feature computation further includes means for performing vector additions, vector products, and vector scaling operations, and means for computing vector magnitudes.