In this paper, a novel fully digital modulator for a switch mode power amplifier (SMPA) is presented. The modulator converts the input baseband amplitude and phase signals into a two-level pulse train for driving an RF SMPA. The simulation results demonstrated a proof of the concept. The proposed architecture is analyzed in terms of accuracy by measuring EVM. The impact of the modulator parameters on the SMPA performance is investigated and corresponding results are shown. The modulator was simulated with a 20 MHz LTE signal.

In this paper, we propose a novel approach for measuring similarity between web objects. Our similarity measure is defined based on the representation of a web object as a collection of tags. Precisely, we first construct a vector space in which multiple terms are mapped into a single dimension by using information available from Open Directory Project and Delicious.com. Then we position web objects in the vector space and apply the traditional cosine measure for similarity computation. We demonstrate that the proposed similarity computation method is able to overcome the limitation of traditional vector space approach while at the same time require less computational cost compares to LSI (Latent Semantic Indexing).

The authors describe an extended backpropagation algorithm for fully connected recurrent neural networks applied to speech recognition. The extended delta rule is approximated by excluding some of the past activities of the dynamic neurons to reduce computational complexity without performance degradation. In speaker-dependent recognition of a confusable syllable set, the fully recurrent neural network with the approximated backpropagation algorithm showed better performance than the multilayer perceptron and the self-recurrent network with comparable time complexity. In addition, it is found that most self-recurrent connections become excitatory and most mutual recurrent connections become inhibitory.>