US 5,852,567 AGranted

Iterative time-frequency domain transform method for filtering time-varying, nonstationary wide band signals in noise

AssigneesHughes Electronics Corporation, National Instruments Corporation

InventorsXiang-Gen Xia, Shie Qian

Life of this patent

Estimating the term…

Loading patent…

Abstract

An iterative time frequency algorithm filters noisy wide band/nonstationary signals by projecting the noisy signal into the TF domain, masking the TF response, computing the inverse TF transform to extract a filtered signal, and repeating these steps until the projection lies within the mask. As a result, the TF domain properties of the extracted signal are substantially equal to the desired TF domain properties. Furthermore, the iterative approach is computationally simple because it avoids inverting matrices. The TF transform and its inverse must be selected such that the iterative algorithm is guaranteed to converge. Candidate transform pairs can be tested on known data, and if the TF transforms converge to the desired TF properties, the candidate pair can be selected. Alternately, the candidate pairs can be tested against a sufficient convergence condition, and if they satisfy the condition within an acceptable tolerance, they can be selected with confidence. Furthermore, the sufficient convergence condition can be solved directly to provide the TF transform and its inverse.

Claims17

Dependent claims

A method for filtering time-varying nonstationary wideband signals, comprising:

a) protecting a noisy time-varying nonstationary wideband signal s onto a time-frequency (TF) plane so that the noise is spread over the TF plane and the signal information is concentrated in a local area of the TF plane;
b) masking the TF plane to reject the noise that lies outside the local area;
c) generating an estimated signal s.sub.k+1 from the signal information and the noise in the local area;
d) repeating steps a, b, and c until the estimated signal S.sub.k+1 converges to a limit signal s;
and
e) extracting the limit signal s as an estimate of the signal information in the noisy signal s,
said method further comprising:
f) selecting a class of TF domain transforms for filtering the signals;
and
g) selecting a pair of transform and inverse transform matrices from the selected class, said signal s being multiplied by the transform matrix to project the signal onto the TF plane, and the masked signal information and noise in the local area being multiplied by the inverse transform matrix to generate the estimated signal s.sub.k+1,
wherein the pair of transform and inverse transform matrices are selected by:
h) selecting a candidate pair of transform and inverse transform matrices;
i) iterating steps a,b and c on a test signal t having a known information content and noise until an estimated test signal t.sub.k+1 converges to a test limit signal t;
j) when the signal t approximates the known information content, selecting the candidate pair of transform and inverse transform matrices for application to the noisy time-varying nonstationary wideband signals s, and otherwise repeating steps h and i.

11 more elements

A method for filtering time-varying nonstationary wideband signals, comprising:

a) projecting a noisy time-varying nonstationary wideband signal s onto a time-frequency (TF) plane so that the noise is spread over the TF plane and the signal information is concentrated in a local area of the TF plane;
b) masking the TF plane to reject the noise that lies outside the local area;
c) generating an estimated signal S.sub.k+1 from the signal information and the noise in the local area;
d) repeating steps a, b, and c until the estimated signal S.sub.k+1 converges to a limit signal s;
and
e) extracting the limit signal s as an estimate of the signal information in the noisy signal s,
said method further comprising:
f) selecting a class of TF domain transforms for filtering the signals;
and
g) selecting a pair of transform and inverse transform matrices from the selected class, said signal s being multiplied by the transform matrix to project the signal onto the TF plane, and the masked signal information and noise in the local area being multiplied by the inverse transform matrix to generate the estimated signal S.sub.k+1,
wherein the pair of transform and inverse transform matrices are selected by:
h) providing a condition on said transform and inverse transform matrices that when satisfied guarantees the convergence of the estimated signal s.sub.k+1 to the limit signal s;
i) selecting a candidate pair of transform and inverse transform matrices;
j) testing the candidate pair of transform and inverse transform matrices to determine whether they satisfy the condition;
and
k) when the condition is satisfied, selecting the candidate pair of transform and inverse transform matrices for application to the noisy time-varying nonstationary wideband signals s, and otherwise repeating steps i and j.

13 more elements

A method for filtering time-varying nonstationary wideband signals, comprising:

receiving a noisy time-varying nonstationary wideband signal sn! that includes signal information and noise;
windowing the noisy signal sn! to generate a signal vector s of length L;
selecting a class of time-frequency (TF) domain transforms;
computing a pair of transform G.sub.MNxL and inverse transform H.sub.LxMN matrices for the selected class of TF domain transforms, where M and N are the number of sampling points in the time and frequency domains, respectively;
generating a mask D.sub.MNxMN for the signal vector s;
initializing an estimate signal s.sub.0 =s;
iteratively computing;
a) C.sub.k+1 =D.sub.MNxMN G.sub.MNxL S.sub.k, and
b) S.sub.k+1 =H.sub.LxMN C.sub.k+1 where C.sub.k+1 is a vector of transform coefficients, until S.sub.k+1 converges to a limit signal s;
and
extracting the limit signal S as an estimate of the signal information in signal vector s.

8 more elements

A method for filtering time-varying nonstationary wideband signals, comprising:

receiving a noisy time-varying nonstationary wideband signal sn! that includes signal information and noise;
windowing the noisy signal sn! to generate a signal vector s of length L;
selecting a Discrete Gabor Transform (DGT) class of time-frequency (TF) domain transforms;
providing a convergence condition that is given by:
##EQU7## for k=0,1, . . . , N-1 and m=0,1, . . . , M-1 where M and N are the number of sampling points in the time and frequency domains, respectively, .DELTA.M and .DELTA.N are the time and frequency sampling lengths, respectively, where the signal is oversampled such that MN>L;
selecting a pair of discrete time-varying synthesis hn! and analysis .gamma.n! window functions for the DGT that satisfy the condition within a given tolerance;
computing a pair of transform G.sub.MNxL and inverse transform H.sub.LxMN matrices from the synthesis hn! and analysis .gamma.n! window functions;
generating a mask D.sub.MNxMN for the signal vector s;
initializing an estimate signal s.sub.0 =s;
iteratively computing;
a) C.sub.k+1 =D.sub.MNxMN G.sub.MNxL S.sub.k, and
b) S.sub.k+1 =H.sub.LxMN C.sub.k+1 where C.sub.k+1 is a vector of Gabor transform coefficients, until S.sub.k+1 converges to a limit signal .sub.s ;
and
extracting the limit signal s as an estimate of the signal information in signal vector s.

11 more elements

Dependencies are inferred from the claim text.