Multistage median cascaded canceller

• US 20030198340A1
• Filed: 04/22/2002
• Published: 10/23/2003
• Est. Priority Date: 04/22/2002
• Status: Active Grant

• Current Assignee: Leidos, Inc. (Leidos Holdings, Inc.), United States Secretary of the Navy
• Original Assignee: Science Applications International Corporation
• Inventors: Goldstein, Jay S., Gerlach, Karl R., Picciolo, Michael
• Granted Patent: US 7,167,884 B2
• Time in Patent Office: Days
• US Class Current: 379/406.01
• CPC Class Codes: G01S 7/292 Extracting wanted echo-signals

Abstract

An adaptive signal processing system utilizes a a Multistage Weiner Filter having an analysis section and a synthesis section, which includes a main input channel for receiving a main input signal, an auxiliary input channel for receiving an auxiliary input signal, and a processor. The processor includes an algorithm for generating a data adaptive linear transformation, followed by computing an adaptive weighting w_med of the data, and then applying the computed adaptive weighting w_med to a function of the main input signal and the auxiliary input signal to generate an output signal. The system includes a plurality of building blocks arranged in a Gram-Schmidt cascaded canceller-type configuration for sequentially decorrelating input signals from each other to thereby yield a single filtered output signal. Each building block includes a local main input channel which receives a local main input signal, a local auxiliary input channel which receives a local auxiliary input signal, and a local output channel which sends a local filtered output signal. Each building block generates an adaptive weight w_med that preferably uses just the sample median value of the real part of a complex weight w_med, with the imaginary part set to zero, of the ratio of local main input weight training data to local auxiliary input weight training data. Each building block generates a local output signal utilizing the adaptive weight w_med. The effect of non-Gaussian noise contamination on the convergence MOE of the system is negligible. In addition, when desired signal components are included in weight training data they cause little loss of noise cancellation.

Claims

1. In a Multistage Weiner Filter having an analysis section and a synthesis section, the filter including:

a main input channel for receiving a main input signal;

an auxiliary input channel for receiving an auxiliary input signal; and

a processor;

the improvement comprising the processor includes an algorithm for generating a data adaptive linear transformation followed by computing an adaptive weighting w_med of the data and for applying the computed adaptive weighting w_med to a function of the main input signal and the auxiliary input signal to generate an output signal.

2. A Multistage Weiner Filter as in claim 1, wherein the adaptive weighting w_med comprises a sample median value of the real part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal, and a sample median value of the imaginary part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal.

3. A Multistage Weiner Filter as in claim 1, wherein the adaptive weighting w_med comprises a sample median value of the real part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal.

4. A Multistage Weiner Filter as in claim 1,

wherein the adaptive weighting w_med is obtained from the equation:

$w_{\mathrm{med}}=\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{real}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]+j<br><br>\left\{\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{imag}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]\right\}$

where K is the number of weight training data samples, z is the main input signal, j is a unit imaginary number, and x is the auxiliary input signal.

5. A Multistage Weiner Filter as in claim 4, wherein the output signal, r, is obtained from the equation:

6. A Multistage Weiner Filter as in claim 5, wherein the processor includes a reduced rank truncation to decrease training data requirements and reduce processing time.

7. In a Multistage Weiner Filter for receiving a plurality of input signals corresponding to a common target signal and for sequentially decorrelating the input signals to cancel the correlated noise components therefrom, the filter having an analysis section and a synthesis section, and the filter further including:

a data adaptive linear transformation to apply to an input data vector;

a plurality of building blocks arranged in a cascaded configuration for sequentially decorrelating each of the input signals from each other of the input signals to thereby yield a single filtered output signal;

wherein each building block includes;

a local main input channel which receives a local main input signal,

a local auxiliary input channel which receives a local auxiliary input signal, and

a processor;

the improvement comprising the processor includes an algorithm for calculating an adaptive weighting w_med, and

generating a local output signal, utilizing the adaptive weighting w_med.

8. A Multistage Weiner Filter as in claim 7, wherein the adaptive weighting w_med comprises:

a sample median value of the real part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal, and a sample median value of the imaginary part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal..

9. A Multistage Weiner Filter as in claim 7, wherein each building block supplies the local output signal to a local output channel.

10. A Multistage Weiner Filter as in claim 7,

wherein the adaptive weighting w_med is obtained from the equation:

$w_{\mathrm{med}}=\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{real}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]+j<br><br>\left\{\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{imag}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]\right\}$

where K is the number of weight training data samples, z is the local main input signal, j is a unit imaginary number, and x is the local auxiliary input signal; and

the local output signal, r, from each building block is obtained from the equation;

r=z−

w*_medx.

11. A Multistage Weiner Filter as in claim 10, wherein the processor includes a reduced rank truncation to decrease training data requirements and reduce processing time.

12. An adaptive signal processing method, comprising the steps of:

receiving a plurality of input signals corresponding to a common target signal;

transforming the input signals to a data stream;

applying the data stream to a steering vector to commence processing with non-adaptive calculations by forming an initial estimate of the desired signal;

isolating the rest of the data by applying the transform x₀(k)=Bx(k), where B is a blocking matrix for finding a projection of the data orthogonal to the steering vector;

applying the initial data and the orthogonal projection of the data to at least one adaptive stage processor, said applying including;

determining the projection of the data in the direction most correlated with the desired signal;

determining the orthogonal projection to the direction most correlated with the desired signal;

applying said projections to the data;

inputting the input signals into a plurality of building blocks arranged in a cascade configuration for sequentially decorrelating each of the input signals from each other of the input signals;

in each building block, calculating an adaptive weighting w_med and generating a local output signal, utilizing the adaptive weighting w_med; and

generating a single filtered output signal.

13. An adaptive signal processing method as in claim 12, wherein the adaptive weighting w_med from each building block is obtained by calculating a sample median value of the real part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal and calculating a sample median value of the imaginary part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal.

14. An adaptive signal processing method as in claim 12, wherein in each building block the adaptive weighting w_med is obtained by calculating a sample median value of the real part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal.

15. An adaptive signal processing method as in claim 12, wherein the adaptive weighting w_med in each building block is obtained by calculating a sample median value of the imaginary part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal.

16. An adaptive signal processing method as claimed in claim 12,

wherein the adaptive weighting w_med in each building block is obtained from the equation:

$w_{\mathrm{med}}=\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{real}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]+j<br><br>\left\{\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{imag}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]\right\}$

where K is the number of weight training data samples, z is the local main input signal, j is the unit imaginary vector, and x is the local auxiliary input signal.

17. An adaptive signal processing method as claimed in claim 16, wherein the local output signal, r, in each building block is obtained from the equation:

18. An adaptive signal processing method as claimed in claim 17, wherein the processor includes a reduced rank truncation to decrease training data requirements and reduce processing time.

19. A multistage median cascaded canceller, comprising:

a means for receiving a plurality of input signals corresponding to the same target signal;

a means for transforming the input signals to a data stream;

a means for applying the data stream to a steering vector to commence processing with non-adaptive calculations by forming an initial estimate of the desired signal;

a means for isolating the rest of the data by applying the transform x₀(k)=Bx(k), where B is a blocking matrix for finding a projection of the data orthogonal to the steering vector;

a means of determining and applying a data adaptive linear transformation via a cascade of orthogonal subspace projections to create an efficient input data transformation;

a means for applying the initial data and the orthogonal projection of the data to at least one adaptive stage processor, said adaptive stage processor including a plurality of building blocks arranged in a cascade configuration for sequentially decorrelating each of the input signals from each other of the input signals; and

a means for generating a single filtered output signal; and

wherein each building block includes a means for receiving a local main input signal, a means for receiving a local auxiliary input signal, and a processing means for calculating an adaptive weighting w_med, and generating a local output signal, utilizing the adaptive weighting w_med.

20. A multistage median cascaded canceller as in claim 19, wherein the adaptive weighting w_med comprises:

a sample median value of the real part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal, and a sample median value of the imaginary part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal.

21. A multistage median cascaded canceller as in claim 19, wherein the adaptive weighting w_med comprises a sample median value of the real part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal.

22. A multistage median cascaded canceller as in claim 19,

wherein the adaptive weighting w_med in each building block is obtained from the equation:

$w_{\mathrm{med}}=\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{real}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]+j<br><br>\left\{\underset{k=1<br><br>\text{\hspace{1em}<br><br>}<br><br>\mathrm{to}<br><br>\text{\hspace{1em}<br><br>}<br><br>K}{\mathrm{MED}}<br><br>\left[\mathrm{imag}\left(\frac{{z<br><br>\left(k\right)}^{*}}{{x<br><br>\left(k\right)}^{*}}\right)\right]\right\}$

where K is the number of weight training data samples, z is the local main input signal, j is the unit imaginary vector, and x is the local auxiliary input signal; and

generates the local output signal, r, by solving the equation;

r=z−

w*_medx.

23. A multistage median cascaded canceller as in claim 22, wherein the processor includes a reduced rank truncation to decrease training data requirements and reduce processing time.

24. A multistage median cascaded canceller for receiving a plurality input signals corresponding to a common target signal and for sequentially decorrelating the input signals to cancel the correlated noise components therefrom, comprising:

an analysis section;

a synthesis section;

a processor configured for;

applying the data stream to a steering vector to commence processing with non-adaptive calculations by forming an initial estimate of the desired signal;

isolating the rest of the data by applying the transform x₀(k)=Bx(k), where B is a blocking matrix for finding a projection of the data orthogonal to the steering vector;

applying the initial data and the orthogonal projection of the data to at least one adaptive stage processor, said adaptive stage processor including a plurality of building blocks arranged in a cascade configuration having N input channels and N−

1 rows of building blocks, for sequentially decorrelating each of the input signals from each other of the input signals to thereby yield a single filtered output signal;

wherein each building block includes;

a local main input channel which receives a local main input signal,

a local auxiliary input channel which receives a local auxiliary input signal, and

wherein the processor is further configured to calculate an adaptive weighting w_med and generate a local output signal, utilizing the adaptive weighting w_med; and

wherein an end building block supplies the local output signal to a separate local output channel for follow on processing.

25. A multistage median cascaded canceller as in claim 24, wherein the Nth input channel is supplied for follow on processing.

26. A multistage median cascaded canceller as in claim 24, wherein said adaptive weighting w_med comprises:

a sample median value of the real part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal, and a sample median value of the imaginary part of the ratio of a main input weight training data signal to an auxiliary input weight training data signal.

27. A multistage median cascaded canceller as in claim 24, wherein said adaptive weighting w_med comprises a sample median value of the real part of a ratio of a main input weight training data signal to an auxiliary input weight training data signal.

28. A multistage median cascaded canceller as in claim 24, wherein said adaptive weighting w_med is obtained from the equation:

29. A multistage median cascaded canceller as in claim 28, wherein the processor includes a reduced rank truncation to decrease training data requirements and reduce processing time.