mohatta2015.pdf

94 LINEAR EQUALIZATION
4.6 THE LITERATURE
A survey of early work on linear equalization can be found in [Luc73, Bel79]. Early
work on LE for cochannel interference suppression can be found in [Geo65].
According to [Mon84], LE with multiple receive antennas is considered as early
as [Bra70j. In [Bal92], an MF front end is used to collapse the multiple streams
of data into one (symbol-level equalization). Cochannel interference can be suppressed
by modeling it as spatially colored noise [Win84, Cla94]. To avoid temporal
processing (multiple processing delays), multiple antennas can be used to suppress
ISI due to dispersion (as well as cochannel interference) [Won96]. A mix of temporal
and spatial processing to suppress ISI from dispersion is also possible [Fuj99].
Different forms of interference suppression result, depending on what aspects of the
interference are known or estimated (amplitude, channel response, symbol values)
[Aff02, Han04].
As for different criteria, development of minimum distortion linear equalization
is provided in [Pro89]. Work on minimum BER linear equalization can be found
in [WanOO]. Block linear equalization for TDM is considered in [Cro92] and is
extended to include cochannel interference suppression in [Gin99].
When the number of processing delays is limited, different strategies can be
used to select the delays. Such strategies apply to both LE and the forward filter
of DFE, so literature on both is discussed here. One strategy is to find a set of
delays that minimizes MSE [Rag93b, LeeOl] or maximizes SNR, (or an approximation
to it [Ari97] (DFE)). An order-recursive approach can be used, in which
processing delays are added one at a time to maximize SNR, or an approximation to
it [Kha05, Zhi05] (LE), [Sui06, Kut07] (LE, DFE). Another approach is to find the
locations that have the largest weight magnitudes [Bun89] (DFE) or are expected
to have the largest weight magnitudes [Lee04] (LE, DFE). Another is a mirroring
approach [Kut05, Ful09], which can be related to approximate inverse channel filtering
[Ful09]. This approach is similar to the idea of placing fingers where copies
of interfering symbols are present [Has02, Sou]. A matching pursuit-based strategy
is proposed in [Zhi05|. Strategies for addressing dispersive cochannel interference
are discussed in [Ari99].
Sometimes the cochannel interference can be better modeled as noncircular (improper)
noise. For example, BPSK interference occupies only one dimension in
the complex plane. When this occurs, there are two, equivalent approaches for
formulating the linear equalization problem. One is linear conjugate linear (LCL)
filtering [Bro69], also known as widely linear filtering [Pic95], in which the equalizer
processes both the received signal and its conjugate. The other is to break apart the
complex received signal into its real and imaginary components [Bro69]. Such filtering
has been applied to BPSK cochannel interference [Yoo97] and GMSK cochannel
interference in GSM [Ger03, Mey06]. In this latter context it is sometimes called
single antenna interference cancellation (SAIC) because it allows interference suppression
similar to that obtained with two receive antennas but without the need
for a second antenna. Such filtering can also be applied to CDMA systems employing
BPSK [BuzOl], including early versions of the US CDMA (IS-95) standard that
employ BPSK with QPSK spreading [Bot03b].