Peak Power Ratio Suppression in OFDM System
Orthogonal frequency-division multiplexing (OFDM) modulation can reduce the influence of inter-symbol interference and enable high-quality communication. However an OFDM signal has a large instantaneous peak power, which is measured as peak-to-average power ratio (PAPR), since the subcarrier signals are modulated independently. Our research area is to develop technique of PAPR reduction method. In our method, the PAPR reduction problem is formulated as a combinatorial optimization problem, and Hopfield neural network (HNN) and chaotic neural network (CNN) are applied to solve the problem. In order to design a practical chip, we also reduce computational complexity of our neural network and implement it with 8,192 neurons using an FPGA device, which is for practical OFDM transmitter of the terrestrial digital broadcast.
Cluster Map Based Bayesian Blind Equalizer
The equalization in digital communication systems is well known as combating the ISI and additive noise to reconstruct the transmitted symbols. In contrast to the general view point of inverse filtering, the equalization can be viewed as a classification problem and the optimal symbol-decision solution can be achieved by Bayes decision theory, which leads to the Bayesian equalizer. Due to the ability of forming nonlinear decision boundary, Bayesian equalizers present superior performance over the linear equalizers. The research of blind equalization, which is carried out in the absence of known channel and training sequence, began from the pioneering work of Y.Sato in 1975. So far, a huge number of approaches have been proposed, such as CMA, HOS based algorithm, SOS based algorithm, blind MLSE, and recently proposed received signal constellation (RSC) geometry based method, etc.
In this research, we try to develop a blind Bayesian equalizer, which is baed on the cluster map generated from the known equalizer structure. The key idea is first to obtain the recevied signal vectors by unsupervised clustering algorihtm then match these clusters to the cluster map, finally forms the equalizer. Clearly, this method can be attributed to the RSC based approach.
Space-Time Linear Precoder
MIMO technique, which exploits the spatial diversity, is known as a cost-effective solution for high data rate in wireless communication systems. In the MIMO systems, the space-time linear precoding, which is the joint processing at the transmitter and the receiver based on the MIMO channel matrix eigenvalue characterized eigenmode, is introduced to improve the transmission performance. Under the assumption of konwn channel state information, the space-time linear precoder is designed in accordance with various criterias, such as maximum the output capacity, MSE and weighted MSE, minimum BER.
Particularly, for the minimum BER criteria, ML decision rule based space-time linear precoder (space-time linear ML precoder) minimizes BER for any constellation size, and is applicable to some schemes of wireless LAN (e.g., HIPERLAN/2, IEEE 802.11x). The potential of this precoder is determined by the eigenmode SNR. In this research, we aim at discovering an eigenmode SNR increasing method with the space-time linear ML precoder.