I want to use subbands resulted from wavelet decomposition of EEG signal segment (segment has a length of 512 sample) as raw data to a 1d-CNN network and compare the performance of the network respect to subbands but the problem is, subbands length are getting smaller when going from subband 1 to subband 2. What should I do? The last subband has only 32 samples and if I use it as input for my network, then surely performance of short subbands will be worse than long ones.
I tried to upsample the low length subbands but I think its not correct to do this.