Partial equalization for digital communication systems

• US 6,512,789 B1
• Filed: 04/30/1999
• Issued: 01/28/2003
• Est. Priority Date: 04/30/1999
• Status: Expired due to Fees

• Current Assignee: Conexant Systems Incorporated (Synaptics Incorporated)
• Original Assignee: PCTEL, Inc.
• Inventors: Mirfakhraei, Khashayar
• Primary Examiner(s): CORRIELUS, JEAN B
• Application Number:
  US09/304,015
• Time in Patent Office: 1,369 Days
• Field of Search: 375/229, 375/230, 375/232, 375/219, 375/224, 375/285, 375/346, 375/350, 375/231, 708/323, 708/300, 333/18, 333/28.R
• US Class Current: 375/229
• CPC Class Codes: H04L 2025/03414 Multicarrier
• CPC Class Codes: H04L 2025/03445 Time domain
• CPC Class Codes: H04L 2025/03522 Frequency domain
• CPC Class Codes: H04L 25/03114 non-adaptive, i.e. not adjustable, manually adjustable, or adjustable only during the reception of special signals

Abstract

A training process for a filter such as included in a time domain equalizer for an xDSL transceiver includes a novel spectral estimation process for a channel. The spectral estimation determines the taps of the filter using an overdetermined set of equations based on the auto-correlation estimates of the received signal. A weighting function such as a sigmoidal function is applied to the AC coefficients to change the relative weighting of the AC coefficients. Upon solving for the taps of the filter using a fitting criterion such as the least square error criterion, the filter significantly reduces the impulse response of the channel.

Claims

1. A communication method comprising:

measuring a frequency response of a channel;

determining an auto-correlation sequence from the frequency response;

applying a weighting function to coefficients from the auto-correlation sequence to thereby determine weighted coefficients; and

using the weighted coefficients in an overdetermined system of equations to identify taps of a digital filter that reduces an impulse response of the channel.

2. The method of claim 1, wherein the weighted coefficients corresponding to higher index values in the auto-correlation sequence have magnitudes that are relatively lower than magnitudes of the corresponding unweighted auto-correlation coefficients.

3. The method of claim 2, wherein the weighting function is a sigmoidal function.

4. The method of claim 1, further comprising using the digital filter in a time-domain equalizer for an xDSL transceiver.

5. The method of claim 1 wherein using the weighted coefficients in the overdetermined system of equations, comprises solving the overdetermined system of equations using a fitting criterion to the identify taps of the digital filter.

6. The method of claim 5, wherein the fitting criterion identifies taps that cause least square error the overdetermined systems of equations.

7. The method of claim 1, wherein measuring the frequency response comprises:

receiving a received signal that corresponds to repeated copies of a transmitted signal that was transmitted over a channel;

averaging samples of the received signal that correspond to repetitions of a selected value in the transmitted signal, the averaging providing averaged sample values;

repeating the averaging step for a plurality of values in the transmitted signal;

performing a Fourier transform on the averaged sample values; and

comparing Fourier components from the transform to a known frequency content of the transmitted signal.

8. The method of claim 7, wherein the transmitted signal includes a pseudo-random sequence of values that are repeated.

9. The method of claim 1 wherein using the weighted coefficients in the overdetermined system of equations, comprises solving the overdetermined system of equations to identify denominator taps of an IIR filter that approximates the response of the channel, wherein the digital filter is a FIR filter that has taps equal to the identified denominator taps.

10. The method of claim 9, wherein solving the overdetermined system of equations uses a fitting criterion to the identify taps of the IIR filter.

11. The method of claim 1, wherein measuring the frequency response comprises:

receiving a reverb signal from the channel, wherein the reverb signal as transmitted had a known frequency distribution over at least a portion of a band width of the channel;

selecting replacement data for a first potion of the band width, wherein the reverb signal as transmitted did not include frequency components in the first portion of the band width; and

determining the frequency response using the replacement data and data from the reverb signal as received.

12. The method of claim 11, wherein selecting the replacement data comprises:

selecting replacement data that indicates no attenuation or distortion in the channel if the first portion is a low frequency band; and

selecting replacement data that indicates complete attenuation in the channel if the first portion is a high frequency band.

13. The method of claim 11, where the reverb signal as transmitted has a uniform frequency distribution over a second portion of the band width, the second portion corresponding to transmission frequencies for a remote transmitter.

14. A communication system comprising:

a time domain equalizer including a digital filter that reduces an impulse response of a channel;

a converter that samples a received signal via the channel during training of the time domain equalizer; and

a handshake block that trains the time domain equalizer, using a training process comprising;

measuring a frequency response of the channel;

determining an auto-correlation sequence from the frequency response;

applying a weighting function to coefficients from the auto-correlation sequence to thereby determine weighted coefficients; and

using the weighted coefficients in an overdetermined system of equations to identify taps of the digital filter.

15. The system of claim 14, wherein the weighted coefficients corresponding to higher index values in the auto-correlation sequence have magnitudes that are relatively lower than magnitudes of the corresponding unweighted auto-correlation coefficients.

16. The system of claim 15, wherein the weighting function is a sigmoidal function.

17. The system of claim 14, wherein the time-domain equalizer is for an xDSL transceiver.

18. The system of claim 14, wherein the handshake block solves the overdetermined system of equations using a fitting criterion.

19. The system of claim 18, wherein the fitting criterion identifies taps that cause least square error the overdetermined systems of equations.