The term ‘neuromorphic feedback loop’ may denote a feedback loop comprising a machine-learning unit. The input of the neuromorphic feedback loop may be directly or indirectly derived from the read-out signal, converted into a series of electrical spikes which may be used as input signals to a neural network. The neural network may be trained according to different characteristics in different frequency ranges. It may be used to reduce noise (attenuation) and amplify one or more specific frequency ranges, e.g., in order to capture spoken language of a specific person in a noisy environment (if, e.g., many people speaking in parallel)
The term ‘spike event circuit’ may denote an electronic circuit converting electrical signals into a series of electrical spike events of different repetition frequency. It may be used as input signals to a neural network for further analysis and feedback loop and to tune the signal processing output of the signal generation. The neural network may also be used to generate a preprocessed output of the artificial cochlea.
According to a possible embodiment of the artificial cochlea, the neuromorphic feedback loop may comprise a spike event circuit—which may mainly operate in an analog mode—connected to a neural network. Thereby, the spike event circuit may be connected via an AER (address event representation) interface which is a neuromorphic inter-chip communication protocol designed allowing real-time connectivity between artificial neurons. Thus, the signal spikes, generated by the analog spike event circuit, may be received by the input layer of the artificial neural network. The output of the trained artificial neural network may then be used to directly influence the functioning and electrical characteristics of the frequency filter and/or the amplitude forming circuit.
patents.justia.com
