Echo cancellation methods and apparatus for voice processing systems

• US 5,164,989 A
• Filed: 12/11/1990
• Issued: 11/17/1992
• Est. Priority Date: 12/11/1990
• Status: Expired due to Term

• Current Assignee: Avaya Incorporated
• Original Assignee: Octel Communications Corp. (Nokia Corporation)
• Inventors: Brandman, Yigal, Puri, Manoj
• Primary Examiner(s): NOT, DEFINED
• Assistant Examiner(s): Shehata, M.
• Application Number:
  US07/625,368
• Time in Patent Office: 707 Days
• Field of Search: 379/3, 379/406, 379/408, 379/410, 340/425
• US Class Current: 379/406.11
• CPC Class Codes: H04B 3/238 using initial training sequence

Abstract

In a voice processing system employing the telephone communications network, any echo of an outgoing signal is substantially eliminated from the incoming signal by appropriately combining an appropriate version of the outgoing signal with the incoming signal so that there is destructive interference with any echo of the outgoing signal. To determine the appropriate version of the outgoing signal to be combined with the incoming signal, a brief training signal is applied to the telephone transmission line. The echo of this signal returned by the transmission line is stored and convolved with the time-inverse of the training signal to produce a function which approximates the impulse response of the transmission line. This function is thereafter used to produce the version of the outgoing signal which is combined with the incoming signal for echo cancellation. Prior to use, the above-mentioned function may be tested for suitability, and if it is found to be unsuitable, a new function may be generated by repeating the process of the invention.

Claims

1. A voice processing system which applies an outgoing signal to a telephone connection and which receives and responds to an incoming signal from said telephone connection comprising:

echo modelling circuitry for applying a predetermined test signal to said telephone connection, said predetermined test signal, having the property that when said test signal is convolved with a time-inverse of said test signal a function that is approximately equal to an impulse is produced, said echo modelling circuitry analyzing the incoming signal subsequent to application of said test signal in order to determine how said test signal must be modified in accordance with an approximate impulse response that is calculated by said echo modelling circuitry so that when said test signal is combined with said incoming signal any echoes of said test signal present in said incoming signal are substantially eliminated, said echo modelling circuitry modifying said outgoing signal in accordance with said approximate impulse response to produce a modified version of said outgoing signal and subsequently combining said modified version of said outgoing signal with said incoming signal to substantially eliminate any echo of said test signal in said incoming signal.

2. The apparatus defined in claim 1 wherein said echo modelling circuitry is operative each time a new telephone connection is established in order to adapt said voice processing system for each said new telephone connection.

3. The apparatus defined in claim 2 wherein said echo modelling circuitry is operative shortly after each new telephone connection is made.

4. The apparatus defined in claim 1 wherein said test signal is a pulse containing a plurality of random, discrete signal frequencies.

5. The apparatus defined in claim 1 wherein said test signal is a finite length random sequence.

6. The apparatus defined in claim 5 wherein said finite length random sequence is selected such that its autocorrelation function is approximately an impulse.

7. A voice processing system which applies an outgoing signal to a telephone connection and which receives and responds to an incoming signal from said telephone connection comprising:

echo modelling circuitry for applying a training signal to said telephone connection, said training signal having the property that when said training signal is convolved with a time-inverse of said training signal, a result that is approximately equal to an impulse is produced, said echo modelling circuitry receiving and storing any echo of said training signal returned to said voice processing system from said telephone connection and subsequently convolving said echo with the time-inverse of said training signal to produce a function which approximates the impulse response of said telephone connection, said echo modelling circuitry thereafter using said function to process said outgoing signal to produce a version of said outgoing signal which approximates the echo of said outgoing signal that will be returned to said voice processing system from said telephone connection, and said echo modelling circuitry combining said version of said outgoing signal with said incoming signal so as to cause said version of said outgoing signal to substantially cancel said echo of said outgoing signal in said incoming signal.

8. The apparatus defined in claim 7 further comprising:

a portion of said echo modelling circuitry configured for testing the suitability of said function for use in processing said outgoing signal before allowing said function to be used by said echo modelling circuitry, wherein said echo modelling circuitry is further configured for computing a quantity indicative of the energy in the difference between (1) the convolution of said training signal and said function and (2) said echo of said training signal, comparing said quantity to a predetermined threshold value, allowing said function to be used only if said quantity is less than said predetermined threshold value, and causing repeated applying, receiving, storing, and convolving if said quantity is not less than said predetermined threshold value.

9. A method of operating a voice processing system which applies an outgoing signal to a telephone connection and which receives and responds to an incoming signal from said telephone connection comprising the steps of:

applying a training signal to said telephone connection, said training signal having the property that when said training signal is convolved with a time-inverse of said training signal a result that is approximately equal to an impulse is produced;

receiving and storing any echo of said training signal returned to said voice processing system from said telephone connection;

convolving said echo with the time-inverse of said training signal to produce a function which approximates the impulse response to said telephone connection;

thereafter using said function to process said outgoing signal to produce a version of said outgoing signal which approximates the echo of said outgoing signal that will be returned to said voice processing system from said telephone connection; and

combining said version of said outgoing signal with said incoming signal so as to cause said version of said outgoing signal to substantially cancel said echo of said outgoing signal in said incoming signal.

10. The method defined in claim 9 further comprising the step of:

testing the suitability of said function for use in processing said outgoing signal before allowing said function to be used.

11. The method defined in claim 10 wherein said testing step further comprises the steps of:

computing a quantity indicative of the energy in the difference between (1) the convolution of said training signal and said function and (2) said echo of said training signal;

comparing said quantity to a predetermined threshold value; and

allowing said function to be used in said using step only if said quantity is less than said predetermined threshold value.

12. The method defined in claim 11 further comprising the step of:

repeating said applying step, said receiving and storing step, and said convolving step if said quantity is not less than said predetermined threshold value.