Echo (communications protocol)

Abstract: While echo questions (EcQs) are often said to be identified by their prosodic properties, there is no empirical study actually supporting such claim. Focusing on wh-utterances we provide results from a production study, a classifier, and a perception study to argue that prosody is not a reliable cue to identify an inquisitive utterance as EcQ. We also offer a model that unifies the semantics of utterances inquiring about what has just been said (EcQs) and utterances inquiring about ‘non-discursive’ facts, information seeking questions (InfQs), while keeping the interpretation of the utterance true to form.

Abstract: Full forwarding dense false target jamming signals correlate highly with real target echoes, and their training samples are difficult to obtain, constraining radars from effectively identifying real and false target echoes. To overcome this challenging problem, this article systematically analyzes and studies the frequency response characteristics of the radar and jammer and models their influence on the amplitude–frequency features of the real and false target echoes. Then, positive-unlabeled learning (PU learning)-based algorithm is proposed to solve the jamming signal recognition problem of missing label information. The core idea of this algorithm is to obtain the amplitude–frequency response features of the two signal types for initial dataset construction and then use the support vector machine (SVM) to estimate the class prior probabilities of each echo to reconstruct a new training dataset. After that, a dual-channel feature fusion network (1DCNN-LSTM) is introduced, comprising a 1-D convolutional neural network (1DCNN) and a long short-term memory (LSTM) network to improve further recognition accuracy. The effectiveness of the proposed features and the PU-1DCNN-LSTM algorithm is demonstrated through simulated and measured experiments, revealing that the proposed method can guarantee a recognition accuracy of 98.4% on the measured data.