I’m not sure that the debugging i did show me all things i want. I constructed a simple communication channel that consists of one node, first ensemble, and second ensemble. In these case, I selected a LIF neuron. I looked at the neurons of ensemble generate spike trains and the output of the neurons of ensemble is referenced by the output of the ensemble. Next, I looked at an ‘DotInc’ operator in build_connection method. These operator take the output of ensemble and weights that built by build_decoder method as input and make a weighted signal as output. In build_decoder method, the decoders are optimized in population representation. It means that it samples evaluation points over [-1,1] and determine targets to find an optimal decoder to decode the points to targets. In the spiking LIF, however, the output of the ensemble is spike trains. I don’t understand why Nengo DotInc perform with spike trains in temporal representation and decoder optimized in population representation. In population-temporal representation, decoding is calculated with the the decoder and the impulse response. Do I miss some principles to understand it?