Communications – electrical: acoustic wave systems and devices – Receiver circuitry
Patent
1988-06-24
1990-08-21
Tarcza, Thomas H.
Communications, electrical: acoustic wave systems and devices
Receiver circuitry
367901, 367903, 379406, G01S 1500
Patent
active
049512692
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an echo canceller with a short processing delay and a decreased number of multiplications.
It is well known that the frequency domain adaptive filter (FDAF) is superior to the time domain adaptive, filter (TDAF) concerning the of computations required. This is more realizable when the filter length is increased, as in the case of the acoustic noise or acoustic echo cancelling in speech communication systems having a long impulse response. The reduction in the computation number is obtained by replacing the convolution in the time domain by multiplication in the frequency domain, using a fast Fourier transform (FFT).
However, to realize a linear convolution of the time domain in the frequency domain, circular convolution must be considered which requires that the FFT use overlapped-save method and a block LMS algorithm. If the length of the block data introduced each time to the system is long, then the input-output, system propagation delay will also be long. If the length of the shifted data is short, then the propagation delay will be short but the multiplication number will be increased. Note, in the FDAF, the FFT length will be too long when the acoustic impulse response is too long. The impulse response of an acoustic echo path is about several hundred millisconds.
Accordingly, for the above reasons, a new echo canceller having a decreased computation number and a short propagation delay is required.
2. Description of the Related Art
The prior art will be explained with reference to FIGS. 9-14.
An echo canceller in the time domain is, as is well known and as shown in FIG. 9, constructed by a digital filter having a finite impulse response (FIR) type having coefficients h.sub.k which are the estimated values of an impulse response h.sub.k of an echo path to be estimated, where k=0, 1, . . . N-1, and N is the length of the FIR digital filter.
Each coefficient h.sub.k is obtained by an adaptive control in which the difference signal e(n) between an echo signal y(n) passed through the echo path to be estimated and the estimated output y(n) of the FIR filter is adaptively controlled to be zero. Known adaptive controls are, a successive adaptive control method and a block adaptive control method. In the successive adaptive control method, the coefficients h.sub.k are adaptively updated upon each input of one sample of the input data x(n). In the block adaptive control method, the coefficients are not updated until L samples of the input data have been input, so that the coefficients are updated as a lump at one time when L samples of input data have been received.
In the successive adaptive control method, the necessary multiplication number for each input sample is 2N. That is, N multiplications are necessary for calculating the output signal y(n), and N multiplications are necessary for updating the coefficients, resulting in the need for 2N multiplications.
In the block adaptive control method, a fast Fourier transform (FFT) is introduced to effect the processing shown in FIG. 9 in the frequency domain, so that the necessary number of multiplications can be reduced. The FFT by the block adaptive control is well known and is described in, for example, "Fast Implementation of LMS Adaptive Filters" IEEE TRANSACTION ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, VOL. ASSP-28, No. 4 AUGUST 1980.
The basic concept thereof is as follows. The convolution expressed by: ##EQU1## can be considered, as shown in FIGS. 10B-10E, as a sum of products of the input signal x(n) (FIG. 10A) and the respective coefficients, where the coefficient series is shifted by one sample for each input signal x(n). In the figure, N represents the length of the impulse response. Until L samples after one updating , the coefficients are not updated by the adaptive control.
Now, consider a series of the input samples x(n) with a periodic period (N+L-1), each period including (N+L-1) samples x(n-N+1) through x(n+L-1) of the input data x(n), as shown in FIG
REFERENCES:
patent: 4355368 (1982-10-01), Zeidler et al.
patent: 4593161 (1986-06-01), Desblache et al.
patent: 4807173 (1989-02-01), Sommen et al.
"A Unified Approach to Time- and Frequency-Domain . . . "; Gregory Clark et al., IEEE Transactions on Acoustics, Speech & Signal Processing, vol. 31, No. 5; Oct. 1983.
"Unconstrained Frequency-Adaptive Filter"; D. Mansour et al.; 726-734, IEEE Transactions on Acoustics, Speech & Signal Processing; vol. 30, No. 5; Oct. 1982.
"The Discrete Fourier Transform Applied to Time Domain Signal Processing"; pp. 13-22, F. J. Harris; IEEE Communications, May 1982.
Amano Fumio
Asharif Mohammad R.
Sakai Yoshihiro
Unagami Shigeyuki
Fujitsu Limited
Swann Tod
Tarcza Thomas H.
LandOfFree
Echo canceller with short processing delay and decreased multipl does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Echo canceller with short processing delay and decreased multipl, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Echo canceller with short processing delay and decreased multipl will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1684234