In frequency-division duplex (FDD) systems, channel-state information (CSI) is estimated by the receiver and then fed back to the transmitter through a feedback link, which inevitably requires additional bandwidth and power. In this letter, we jointly study optimal bandwidth allocation between the data channel, modeled as a flat-fading multiple-input single-output (MISO) channel, and the feedback channel for maximum average throughput in the data channel using a beamforming scheme. We consider two models of the partial CSI at the transmitter (CSIT): the noisy CSIT, modeled as jointly Gaussian with the actual channel state, and the quantized CSIT. In the first model, we use distortion-rate theory to relate the CSIT accuracy to the feedback-link bandwidth. In the second model, we derive a lower bound on the achievable rate of the data channel based on the ensemble of a set of random quantization codebooks. We show that in the MISO flat-fading channel case, beamforming based on feedback CSI can achieve an average rate larger than the capacity without CSIT under a wide range of mobility conditions.

We consider the performance of a forward link (base-to-mobile) shared channel with and without a multiple-input multiple-output (MIMO) antenna system. A CDMA shared channel is analyzed and simulated for which users are served in a time division multiplexed (TDM) manner. The slot structure and chip rate are based on the wideband CDMA standard. Link-level simulations are presented to compare the performances of two different receivers: linear minimum mean-square-error (MMSE), and nonlinear (turbo/iterative) MMSE. In addition, MIMO performance is compared with and without antenna array weight feedback.

Our focus in this paper is on applying multiple-input multiple-output (MIMO) antenna systems to the time division multiplexed (TDM) shared channel on the WCDMA forward link. We describe cellular system-level simulations with and without MIMO. These simulations include multipath channel models and they account for other-cell interference. We compute the SINR at the output of a linear MMSE receiver and map these values to throughput using simulated turbo code results. Simulation results are presented with various numbers of antennas at the transmitter and receiver. We show that with two transmit and two receive antennas, system capacity can be doubled relative to one transmit and one receive antenna. Two transmit and four receive antennas can provide three times the capacity of a one transmit and one receive antenna system. Also, results are presented that compare three sector systems to six sector systems.

Neural image processing filters provide an efficient means for synthesis of high-order nonlinear systems. Parallel hardware implementations are feasible. A modular design methodology for analysis and synthesis of nonlinear neural network filters is described. This method is based on a building block approach: subnetworks are designed separately and assembled via linear connecting layers. Efficient neural filters are designed by removing the connecting layers. Classical image processing filters such as Volterra, order statistic, and morphological filters are synthesized using large multilayer networks.>

In a recent paper by M. Chen and M. Maury (1990), it was shown that multilayer perceptron neural networks can be used to form products of any number of inputs, thereby constructively proving universal approximation. This result is extended, and a method for the analysis and synthesis of single-input, single-output neural subnetworks is described. Given training samples of a function to be approximated, a feedforward neural network is designed which implements a polynomial approximation of the function with arbitrary accuracy. For comparison, example subnets are designed by classical backpropagation training and by mapping. The examples illustrate that the mapped subnets avoid local minima which backpropagation-trained subnets get trapped in and that the mapping approach is much faster.