This paper presents a real-time and memory-efficient arrhythmia detection system with binary classification that uses antidictionary coding for the analysis and classification of electrocardiograms (ECGs). The measured ECG signals are encoded using a lossless antidictionary encoder, and the system subsequently uses the compression rate to distinguish between normal beats and arrhythmia. An automated training data procedure is used to construct the automatons, which are probabilistic models used to compress the ECG signals, and to determine the threshold value for detecting the arrhythmia. Real-time computer simulations with samples from the MIT-BIH arrhythmia database show that the averages of sensitivity and specificity of the proposed system are 97.8% and 96.4% for premature ventricular contraction detection, respectively. The automatons are constructed using training data and comprise only 11 kilobytes on average. The low complexity and low memory requirements make the system particularly suitable for implementation in portable ECG monitors.

In this paper, we propose a memory-efficient structure for a pulse Doppler radar in order to reduce the hardware's complexity. The conventional pulse Doppler radar is computed by fast frequency transform (FFT) of all range cells in order to extract the velocity of targets. We observed that this method requires a huge amount of memory to perform the FFT processes for all of the range cells. Therefore, instead of detecting the velocity of all range cells, the proposed architecture extracts the velocity of the targets by using the cells related to the moving targets. According to our simulations and experiments, the detection performance of this proposed architecture is 93.5%, and the proposed structure can reduce the hardware's complexity by up to 66.2% compared with the conventional structure.

This paper presents our developed decoder which adopts the idea of statically optimizing part of the knowledge sources while handling the others dynamically. The lexicon, phonetic contexts and acoustic model are statically integrated to form a memory-efficient state network, while the language model (LM) is dynamically incorporated on the fly by means of extended tokens. The novelties of our approach for constructing the state network are (1) introducing two layers of dummy nodes to cluster the cross-word (CW) context dependent fan-in and fan-out triphones, (2) introducing a so-called "WI layer" to store the word identities and putting the nodes of this layer in the non-shared mid-part of the network, (3) optimizing the network at state level by a sufficient forward and backward node-merge process. The state network is organized as a multi-layer structure for distinct token propagation at each layer. By exploiting the characteristics of the state network, several techniques including LM look-ahead, LM cache and beam pruning are specially designed for search efficiency. Especially in beam pruning, a layer-dependent pruning method is proposed to further reduce the search space. The layer-dependent pruning takes account of the neck-like characteristics of WI layer and the reduced variety of word endings, which enables tighter beam without introducing much search errors. In addition, other techniques including LM compression, lattice-based bookkeeping and lattice garbage collection are also employed to reduce the memory requirements. Experiments are carried out on a Mandarin spontaneous speech recognition task where the decoder involves a trigram LM and CW triphone models. A comparison with HDecode of HTK toolkits shows that, within 1% performance deviation, our decoder can run 5 times faster with half of the memory footprint.