Abstract
We study a cooperative uplink CDMA network with multiple sources and multiple relays. The sources simultaneously access the channel using different spreading sequences and compete for the bandwidth at the relays, and the bandwidth-limited relays then forward the information from the sources to the destination. Under the assumption that the transmission dimension at relays is limited and less than the number of sources, we propose a reduced-rank multiuser relaying (RR-MUR) scheme where the data received from sources are first compressed into lower dimensions before being retransmitted. Specifically, we employ linear precoders at relays and a decoder at the destination to minimize the mean square error (MSE) of the estimate at the destination. In this paper, we propose a centralized methods to devise the precoders and decoder with global channel state information (CSI). Moreover, by taking into consideration the channel estimation error at both relays and the destination, we devise the robust precoders and decoder that minimize the MSE under the presence of channel estimation error. For practical considerations, a decentralized scheme is then proposed: the Minimum Mean Square Error Principle Component Analysis (MMSE-PCA) method. We show that the MMSE-PCA scheme is equivalent to the optimal decentralized design where each relay's precoder is computed separately by neglecting the presence of other relays. Through the numerical comparison, we show that the proposed iterative procedure outperforms the other existing relaying methods for uplink CDMA systems. When the relays and the destination have channel estimation error, the presence of error dramatically degrades the detection accuracy at the destination.