首页>
外国专利>
Time-varying data communication using neural networks supplied with parameters
Time-varying data communication using neural networks supplied with parameters
展开▼
机译:使用带有参数的神经网络进行时变数据通信
展开▼
页面导航
摘要
著录项
相似文献
摘要
Time-dependent data is coded and modulated by a neural network comprising multi-layer perceptrons (MLP). For each extended neuron (120) in a layer (302. fig.8), serial in parallel out (SIPO) buffers (122) allow previous inputs and outputs to model the function as well as present ones, and these gates are cyclically blocked or opened according to a switching control word. The output from the entire extended neuron is similarly controlled by a timing control waveform (350, fig.8). These control signals specifying which neurons are on or off at a given time determine whether the system is in training mode or running mode. Different channel coding, compression coding, modulation (eg. QPSK) or demodulation (eg. Viterbi) schemes can be emulated or "learned" by different parts of the neural network by having new parameters sent to different neuron layers, allowing a transceiver comprising such a network to act as a "software radio".
展开▼