Èíôîðìàöèîííûå ïðîöåññû, Òîì 6, 2, 2006, ñòð. 114123.
c© 2006 Osipov.
INFORMATION THEORY AND INFORMATION PROCESSING
A study of a Code Division Multiple Access detector based
on a recurrent neural network
Institute for Information Transmission Problems, Russian Academy of Sciences
B. Karetnyi per. 19, GSP-4, 101447 Moscow, Russia
email: d firstname.lastname@example.org
Received April 11,2006
Abstract—In this paper a Code Division Multiple Access detector based on a neural network,
which is proposed in , is considered. It is shown how the choice of an updating scheme and/or
the network parameters influences the detector performance. A weight updating scheme that
might be used instead of the classical ones is proposed.
Code Division Multiple Access (CDMA) is one of the most powerful random multiple access
techniques. A number of CDMA methods that differ both in the performance manner and tasks
set were proposed recently. Direct Sequence CDMA (DS-CDMA) is a simple method of the kind,
which enables one to overcome certain imperfections of other multiple access methods as well as
those of other CDMA methods. In a DS-CDMA system each information sequence is multiplied by
a code sequence, all resulting encoded sequences being transmitted simultaneously. That is why it
is necessary to obtain an information sequence from the received one, the process itself being called
detection and the device, which is to implement it a detector.
An optimal detector for asynchronous additive multiple access channel white Gaussian noise
(AWGN), based on multiuser maximum likelihood sequence estimation (MU-MLSE), has been
proposed by Verdu in . The complexity of MU-MLSE, which is based on the Viterbi algorithm,
grows exponentially as the number of users increases. Thus, a low complexity suboptimal detector is
needed. One can design such a detector using either the maximum likelihood (ML) principle or the
maximum a posteriori (MAP) principle. Our goal is to maximize the balance between computational
complexity and dete