Estimating a signal based on samples derived from dot products and random projections
View Patent ↗A system and method for estimating a signal based on a stream of randomly generated samples. The method includes: (a) receiving a sample; (b) generating a sampling vector; (c) multiplying the sample and the sampling vector to obtain a current back projection; (d) computing a first intermediate vector that represents an average of the current back projection and previous back projections; (e) transforming the first intermediate vector to determine a second intermediate vector; (f) identifying locations where the second intermediate vector attains its k largest values; (g) computing an estimate for the transformation of the signal by solving a system of equations based on the identified locations, the received sample value, previously received sample values, the sampling vector and previously generated sampling vectors; (h) inverse transforming the transformation estimate to determine an estimate of the signal; and (i) storing the signal estimate.
1. A computer system for estimating a signal, the computer system comprising:
a processor; and memory storing program instructions that are executable by the processor to:
(a) receive a data value, wherein the data value provides information regarding a signal to be estimated;
(b) multiply the data value and a corresponding sampling vector to obtain a current back projection;
(c) compute a first intermediate vector that represents an average of the current back projection and previous back projections;
(d) apply a transformation to the first intermediate vector to determine a second intermediate vector;
(e) identify a set of index positions where the second intermediate vector attains its k largest values in terms of absolute value, wherein k is a positive integer;
(f) compute a fourth intermediate vector by applying an inverse of the transformation to the third intermediate vector;
(g) compute a dot product of the fourth intermediate vector with the random vector;
(h) repeat (a) through (h) until a separation between the dot product and the corresponding data value smaller than a threshold value;
(k) compute an estimate of the signal using the set of index positions, the received data value, previously received data values, the sampling vector and previous sampling vectors; and
(l) store the signal estimate in the memory.
2. The computer system of claim 1 , wherein (k) includes:
computing an estimate for the transformation of the signal by solving a system of equations based on the set of index positions, the received data value, the previously received data values, the sampling vector and the previous sampling vectors; and
applying the inverse of the transformation to said estimate for the transformation to determine the estimate of the signal.
3. The computer system of claim 1 , wherein the separation between the dot product and the corresponding data value is a percentage difference between the dot product and the corresponding data value.
4. The computer system of claim 2 , wherein the transformation of the signal is k-sparse.
5. The computer system of claim 2 , wherein the program instructions are executable to:
display a graphical representation of the signal estimate via a display device.