Neural network filter for adaptive noise substraction in optical heterodyne signals

• US 20040047639A1
• Filed: 09/05/2002
• Published: 03/11/2004
• Est. Priority Date: 09/05/2002
• Status: Abandoned Application

• Current Assignee: Agilent Technologies, Inc.
• Original Assignee: Agilent Technologies, Inc.
• Inventors: McAlexander, William Ian
• Application Number:
  US10/235,701
• Publication Number:
  US 20040047639A1
• Time in Patent Office: Days
• US Class Current: 398/202
• CPC Class Codes: H04B 10/60 Receivers
• CPC Class Codes: H04B 10/64 Heterodyne , i.e. coherent receivers where, after the opto-electronic conversion, an electrical signal at an intermediate frequency [IF] is obtained

Abstract

An artificial neural network is utilized as a filter in an optical heterodyne balanced receiver system. Such an artificial neural network can be adapted or trained to correct for non-ideal behavior and imperfections in the optical heterodyne system.

Claims

1. A data processing unit for use in an optical heterodyne balanced receiver system to correct for non-ideal behavior and imperfections in said system, the data processing unit comprising:

a central processing unit; and

a data storage unit connected to said central processing unit, said data storage unit having computer readable instructions stored thereon for causing said central processing unit to;

implement an artificial neural network in said data processing unit;

receive data to be corrected into one or more input nodes of said artificial neural network; and

correct said data for said non-ideal behavior and imperfections using said artificial neural network.

2. The data processing unit according to claim 1, wherein said computer readable instructions further cause said central processing unit to apply an input node weight to said data for at least one input node.

3. The data processing unit according to claim 2, wherein said computer readable instructions further cause said central processing unit to sum said weighted data from said at least one input node at one or more hidden nodes of said artificial neural network, and apply a hidden node weight to said sum to produce a weighted sum for at least one hidden node.

4. The data processing unit according to claim 3, wherein said computer readable instructions further cause said central processing unit to sum said weighted sum from said at least one hidden node at one or more output nodes of said artificial neural network to produce corrected data.

5. The data processing unit according to claim 3, wherein said computer readable instructions further cause said central processing unit to apply a transfer function to said sum of weighted data from said at least one input node and to said sum of weighted sum from said at least one hidden node.

6. The data processing unit according to claim 5, wherein said transfer function is a non-linear transfer function.

7. The data processing unit according to claim 3, wherein said computer readable instructions further cause said central processing unit to adjust said input node weight and said hidden node weight for said at least one input node and said at least one hidden node.

8. The data processing unit according to claim 7, wherein said computer readable instructions further cause said central processing unit to add a correction factor to the current value of said input node weight and said hidden node weight.

9. The data processing unit according to claim 1, wherein said computer readable instructions further cause said central processing unit to add at least one input node for tracking at least one parameter of interest associated with said optical heterodyne system, said at least one parameter of interest including wavelength, polarization, and temperature.

10. The data processing unit according to claim 4, wherein said computer readable instructions further cause said central processing unit to subtract an output of said one or more output nodes from data that has not been corrected to produce an optical heterodyne signal.

11. A method of correcting for non-ideal behavior and imperfections in an optical heterodyne balanced receiver system, comprising:

inputting data to be corrected to one or more input nodes of an artificial neural network; and

correcting said data for said non-ideal behavior and imperfections using said artificial neural network.

12. The method according to claim 11, wherein said step of inputting data to be corrected includes normalizing said data to be corrected by a predefined normalization factor.

13. The method according to claim 11, further comprising inputting data to be corrected to one or more input nodes of at least one additional artificial neural network.

14. The method according to claim 13, further comprising subtracting an output of one or more output nodes of said at least one additional neural network from data that has not been corrected to produce an optical heterodyne signal.

15. An optical heterodyne system, comprising:

an optical coupler;

detectors adapted to detect output signals from said optical coupler;

an artificial neural network configured to receive data representing one of said detected output signals; and

a data processing unit configured to subtract an output of said artificial neural network from data representing another one of said detected output signals to isolate an optical heterodyne signal.

16. The system according to claim 15, wherein said optical heterodyne system is a balanced receiver system.

17. The system according to claim 15, wherein said artificial neural network includes at least one input node, at least one hidden node, and at least one output node.

18. The system according to claim 17, wherein a value of said at least one input node is a normalized version of said data representing one of said detected output signals.

19. The system according to claim 17, wherein a weight associated with said at least one input node and said at least one hidden node is adjusted by adding a correction factor to a current value of each weight.

20. The system according to claim 19, wherein said correction factor includes a product of an error, a momentum factor, and a derivative of said output of said artificial neural network with respect to said current value of each weight.

21. The system according to claim 20, wherein said error is based on a difference between said output of said artificial neural network and said data representing another one of said detected output signals.

22. The system according to claim 17, wherein at least one of said input nodes represents a parameter of interest associated with said optical heterodyne system, including wavelength, polarization, and temperature.

23. The system according to claim 15, wherein said data processing unit is further configured to train said artificial neural network.

24. The system according to claim 23, wherein said training is performed using data derived from a single input to said optical coupler.

25. The system according to claim 23, wherein said training is performed using data representing portions of said detected output signals where said heterodyne signal is considered to be substantially zero.

26. The system according to claim 15, wherein said optical coupler includes a beam splitter.

27. The system according to claim 15, wherein said optical coupler has two or more inputs.

28. The system according to claim 15, further comprising at least one additional artificial neural network configured to receive said data representing another one of said detected output signals.

29. The system according to claim 15, further comprising at least one additional artificial neural network configured to receive data representing a further one of said detected output signals.

30. The system according to claim 15, wherein said detectors include two or more detectors.