US 7,289,557 B2 · Adaptive equalization using a conditional update sign-sign least mean square algorithm

Abstract

An adaptive equalizer finite impulse response (FIR) filter for high-speed communication channels with modest complexity, where the filter is iteratively updated during a training sequence by a circuit performing the update: h(t+1)= h(t)+μ[sgn{d(t)}−sgn{z(t)−Kd(t)}]sgn{ x(t)}, where h(t) is the filter vector representing the filter taps of the FIR filter, x(t) is the data vector representing present and past samples of the received data x(t), d(t) is the desired data used for training, z(t) is the output of the FIR filter, μ determines the memory or window size of the adaptation, and K is a scale factor taking into account practical limitations of the communication channel, receiver, and equalizer. Furthermore, a procedure and circuit structure is provided for calibrating the scale factor K.

Claims10

Dependent claims

An apparatus comprising:

a circuit including a discrete-time FIR (Finite Impulse Response) filter comprising n multiplier units to implement a filter response [ h(t)]i, i=0, 1, . . . , n−1, where t is a time index, the FIR filter to filter a discrete-time sequence of input voltages x(t) to provide a sequence of filtered output voltages z(t) where
z
⁡
(
t
)
=
∑
i
=
0
n
-
1
⁢
[
h
_
⁡
(
t
)
]
i
⁢
⁢
x
⁡
(
t
-
i
)
;
and
a data generator coupled to the discrete-time FIR filter, the data generator to provide a discrete-time sequence of desired voltages d(t), t=1, 2, . . . , T;
wherein for t=1, 2, . . . , T, the filter response satisfies an update relationship [ h(t+1)]i=[ h(t)]i+μ[sgn{d(t)}−sgn{z(t)−Kd(t)}]sgn{x(t−i)}, i=0, 1, . . . , n−1, where μ and K are scalars and sgn{ } denotes sign.

34 more elements

An apparatus comprising:

a circuit including a discrete-time FIR (Finite Impulse Response) filter to filter a discrete-time input sequence of voltages x(t) where t is a discrete-time index, to provide, for t=1, 2, . . . , T, a voltage indicative of
z
⁡
(
t
)
=
∑
i
=
0
n
-
1
⁢
[
h
_
⁡
(
t
)
]
i
⁢
⁢
x
⁡
(
t
-
i
)
⁢
⁢
where
⁢
[
h
_
⁡
(
t
)
]
i
,
i=0, 1, . . . , n−1 are n weights indexed by t;
a data generator coupled to the filter to provide a discrete-time sequence of desired voltages d(t), t=1, 2, . . . , T;
a latch circuit coupled to the filter to provide, for t=1, 2, . . . , T, a voltage indicative of sgn{z(t)−Kd(t)} where K is a weight and sgn{ } denotes the sign function;
a digital summer coupled to the filter to provide, for t=1, 2, . . . , T, n voltages indicative of sgn{d(t)}−sgn{z(t)−Kd(t)}, i=0, 1, . . . , n−1;
a digital multiplier coupled to the filter to provide, for t=1, 2, . . . , T, n voltages indicative of μ[sgn{d(t)}−sgn{z(t)−Kd(t)}]sgn{x(t−i)}, i=0, 1, . . . , n−1 where μ is a weight;
a digital summer and a delay element coupled to the filter to provide to the FIR filter, for t=1, 2, . . . , T, n voltages indicative of [ h(t)]i+μ[sgn{d(t)}−sgn{z(t)−Kd(t)}]sgn{x(t−i)}, i=0, 1, . . . , n−1 so that for t=1, 2, . . . , T the weights [ h(t+1)]i, i=0, 1, . . . , n−1 are given by [ h(t+1)]i+[ h(t)]i+μ[sgn{d(t)}−sgn{z(t)−Kd(t)}]sgn{x(t−i)}, i=0, 1, . . . , n−1.

50 more elements

Dependencies are inferred from the claim text.

Adaptive equalization using a conditional update sign-sign least mean square algorithm

Life of this patent

Adaptive equalization using a conditional update sign-sign least mean square algorithm

Life of this patent

Abstract

Claims10