A new approach for block-floating-point arithmetic | IEEE Conference Publication | IEEE Xplore