Non linear adaptive filters

• US 4,792,915 A
• Filed: 05/09/1986
• Issued: 12/20/1988
• Est. Priority Date: 05/10/1985
• Status: Expired due to Term

• Current Assignee: British Telecommunications PLC (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Smith, Mark J., Carpenter, Robert B. P., Adams, Peter F.
• Primary Examiner(s): Malzahn, David H.
• Application Number:
  US06/861,412
• Time in Patent Office: 956 Days
• Field of Search: 364/724
• US Class Current: 708/322
• CPC Class Codes: H03H 21/0016 Non linear filters
• CPC Class Codes: H04B 3/237 using two adaptive filters, e.g. for near end and for end echo cancelling
• CPC Class Codes: H04B 3/238 using initial training sequence

Abstract

A non-linear adaptive filter is described having a linear filter connected in parallel with a non-linear filter. The linear filter provides fast adaption until it has modelled the linear contribution of each coefficient. Thereafter the non-linear filter continues to adapt until the error signal has been reduced to an acceptable level.

In a preferred embodiment a plurality of unit delay devices provide taps to sub-processing units each of which is arranged to be adapted according to a linear algorithm during an initial part of a training period, and to a non-linear algorithm after said initial part.

The filter has particular use in echo cancellers for data modems.

Claims

1. A non linear adaptive filter comprising:

means for a receiving samples of a noise source signal,

processing means for processing a predetermined number of said samples with values stored in said processing means to produce a cancellation signal,

means for combining the cancellation signal with an input signal to produce an error signal, and

adaptation means arranged to receive said error signal and adapt said stored values so that said error signal is substantially reduced over an initial training period wherein the adaptation means includes

a first adaptation means for adapting the processing means to cancel noise according to a linear algorithm

a second adaptation means for adapting the processing means to cancel noise according to a non linear algorithm, and

switching means arranged to select said first adaptation means for a first part of said initial training period, said switching means selecting said second adaptation means after said first part of said initial training period.

2. A non linear filter according to claim 1 wherein:

the processing means comprises a linear filter, a separate non linear filter and means for combining the output from each of said filters, and

the first adaptation means being selected to adapt said linear filter during the first part of the initial training period and the second adaptation means being selected to adapt said non linear filter thereafter.

3. A non linear filter according to claim 1 wherein the processing means comprises:

a random access memory device which stores all possible output values to provide a non linear filter and

addressing means which addresses said memory to cause output of value including coefficients of the linear filter after the first part of the initial training period.

4. A non linear filter according to claim 1 wherein the processing means comprises:

a random access memory device which stores all possible output values to provide a non linear filter,

the first adaptation means causing a plurality of values stored in said memory device to adapt according to a linear algorithm during the first part of the initial training period.

5. A non linear filter according to claim 1 wherein the processing means comprises:

a plurality of unit delay devices to provide a tap to each of a respective sub processing unit and each sub processing unit is adapted according to linear algorithm during the first part of the initial training period and each subprocessing unit is adapted according to a non-linear algorithm after the initial part of the initial training period.

6. A method of adapting a non linear filter during an initial training period in which a predetermined number of samples of a noise source signal are processed with values stored in a processor to produce a cancellation signal;

the processor adapting the cancellation signal value according to a linear algorithm during a first part of the initial training period and adapting the cancellation signal value according to a non linear algorithm after said first part of the initial training period.

7. An echo canceller in which received data includes noise due to transmitted data, the echo canceller comprising:

a digital filter arranged to receive samples of a noise source signal derived from the transmitted signal to produce a cancellation signal in which characteristics of the filter are adapted in response to means for generating an error signal derived from said noise source signal,

the digital filter including first means for adapting said characteristics according to a linear algorithm during a first part of an initial training period and second means for adapting said characteristics according to a non linear algorithm and switching means arranged to cause selection of said first means for said first part of said initial training period and to cause selection of said second means thereafter.

8. A non-linear adaptive signal filter having a reduced training period and comprising:

linear adaptive filter means for adaptively cancelling noise components from an input signal in accordance with a linear algorithm;

non-linear adaptive filter means for adaptively cancelling noise components from said input signal in accordance with a non-linear algorithm;

switch means coupled to both said filter means (a) for initially utilizing said linear adaptive filter means during a first time period T1 during which said linear adaptive filter means adaptively generates an error cancellation signal which has a value E at the end of period T1, and (b) for subsequently utilizing said non-linear adaptive filter means after period T1 while continuing also to use the final cancellation signal value E provided by the linear adaptive filter means in combination with the output of the non-linear adaptive filter means to cancel noise components from the input signal.

9. A method for decreasing the effective training period of a non-linear adaptive signal filter, said method comprising:

adaptively cancelling noise components from an input signal initially during said training period in accordance with a linear algorithm during a first time period T1 during which a linear adaptive filter means adaptively generates an error cancellation signal which has a value E at the end of period T1, and

thereafter during said training period adaptively cancelling noise components from said input signal in accordance with a non-linear algorithm utilizing a non-linear adaptive filter means after period T1 while continuing also to use the final cancellation signal value E provided by the linear adaptive filter means in combination with the output of the non-linear adaptive filter means to cancel noise components from the input signal.