The neural recording arrays used in brain-machine interfaces (BMIs) provide volumes of data that are much larger than realistic wireless bandwidth arguably limiting the possibility of using usable neuroprosthetics untethered. Current compression strategies focus on improving signal reconstruction fidelity and not motor decoding performance downstream, and therefore tend to be inefficient at bandwidth usage. This paper is a proposal of SemNeural-Comm a semantic communication model in wireless BMIs which utilizes bandwidth efficient transmission of neural signals. The suggested architecture trains a neural semantic encoder conditioned on neuroscience-grounded goals which learns control-relevant neural features and removes irrelevant patterns of activity. In a generative diffusion model in the decoder, the semantic transmissions compressed into a highly compressed form are reconstructed to give enough neural information to decode the motor intent correctly. The framework builds a neural-semantic common graph of knowledge that represents the connections between neural recording regions, planned actions, and the possibility of semantic compression in order to make decisions of adaptive encoding. Predicted channels 2 Twin prediction determines neural semantic rate allocation, making signals of motor cortex more important under bad channel conditions and wider neural records more important when bandwidth allows. Semantic-neural co-training procedure encodes, transmits, and decodes jointly sequence the end-to-end BMI control performance. Detailed assessments of SemNeural-Comm, through the publicly available Competition IV Dataset 2a (https://www.bbci.de/competition/iv/) and nine subjects show that SemNeural-Comm treats motor imagery decoding with an accuracy of 92.37% at a reduced transmission cost of 97%, a substantial improvement over conventional compression and the current semantic communication baselines, which have statistical significance . Keywords: Brain-machine interface, semantic communication, neural signal compression, bandwidth-efficient transmission, motor imagery decoding, diffusion model, common knowledge graph.
Yasser et al. (Sat,) studied this question.