Pulse or digital communications – Equalizers – Automatic
Patent
1997-05-08
1999-11-30
Chin, Stephen
Pulse or digital communications
Equalizers
Automatic
333 18, 333 28R, H03H 730
Patent
active
059955449
DESCRIPTION:
BRIEF SUMMARY
The present invention is concerned with adaptive equalization, in particular with a method and an apparatus for self-training adaptive equalization in the receiver of a PRIV transmission system.
BACKGROUND
An adaptive digital equalizer is usually realized as a finite impulse response (FIR) filter, also known as a transversal filter. The filter output is obtained by linear combination of N signal samples stored in the equalizer delay line, i.e., coefficients and X.sub.n ={x.sub.n, . . . , x.sub.n-N+1 } is the vector of signal samples stored in the equalizer delay line at lime n. The filter coefficients are traditionally adjusted by the least-mean square (LMS) algorithm with the objective of minimizing noise and residual signal distortion at the filter output. In the LMS algorithm, an error signal is first computed. This error signal is then correlated with each signal sample stored in the delay line to generate an estimate of the gradient vector of the mean-square error (MSE). The vector of filter coefficients is finally updated by subtracting a term that is proportional to that estimate. In reference-directed mode of operation, the error signal is obtained as the difference between the equalizer output signal and a known reference signal. In decision-directed mode, the reference signal is replaced by the locally determined, most likely received signal among the discrete signals of the employed signal constellation.
In reference-directed mode, convergence of the equalizer coefficients to correct settings can always be achieved, provided the receiver has proper knowledge of a training sequence sent prior to random data signals. If after this training, the receiver makes decisions with sufficiently low probability of error, equalizer adaptation will operate reliably also in decision-directed mode. However, if the equalizer is not well trained at the beginning of decision-directed equalization, convergence will usually not be achieved except in the cases of binary modulation or pure phase modulation.
The sending of a training sequence and the means required to determine in a receiver the exact time when reception of a training sequence begins introduce complexities in the modem design and lead to communication overhead, which may not be desirable. The use of a training sequence may also be inappropriate for various situations. For example, in multipoint networks of modems it is usually not practical to send training sequences from a master modem to the receivers of slave modems, which may independently be activated or deactivated by their users.
When initial reference-directed equalizer adjustment by a training sequence is either undesirable or not possible, self-training equalization methods must be employed to train an equalizer from random data signals. These methods generally require much longer training periods than needed for reference-directed training. However, for many applications the length of these training times is quite tolerable.
Self-training methods usually rely on the definition of a pseudo-error, which on average will lead to correct coefficient adjustment although initially no reliable individual decisions can be obtained from the equalizer output signals.
A partial-response class-IV (PRIV) system is defined by its discrete-time channel symbol response, which in D-transform notation is given by h.sub.PRIV (D)=1-D.sup.2, where D denotes the operator for delay by one modulation interval T. Note that in a PRIV system, intersymbol interference is introduced in a controlled fashion. This distinguishes PRIV systems from full-response systems, where no intersymbol interference is present. If a(D)= . . . a.sub.n D.sup.n +a.sub.n+1 D.sup.n+1 + . . . is the sequence of data symbols transmitted at the modulation rate 1/T, a sequence of correlated signal samples b(D)= . . . b.sub.n D.sup.n +b.sub.n+1 D.sup.n+1 = . . . (a.sub.n -a.sub.n-2)D.sup.n +(a.sub.n+1 -a.sub.n-1)D.sup.n+1 + . . . is obtained at the output of an ideal PRIV system. For example, for a quaternary PRIV system, the input symbols a.s
REFERENCES:
patent: 5319674 (1994-06-01), Cherubini
patent: 5353310 (1994-10-01), Russell et al.
patent: 5467370 (1995-11-01), Yamasaki et al.
Cherubini Giovanni
Oel.cedilla.er Sedat
Ungerbock Gottfried
Burd Kevin M
Chin Stephen
International Business Machines - Corporation
Woods Gerald R.
LandOfFree
Adaptive equalization for PRIV transmission systems does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Adaptive equalization for PRIV transmission systems, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Adaptive equalization for PRIV transmission systems will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1682794