THIS PAPER IS ELIGIBLE FOR THE STUDENT PAPER AWARD. An enhanced on-off division multiple access (ODMA) transmission scheme is proposed for unsourced multiple access network. The message of each active user is divided into two parts, where the first part is used to determine an on-off pattern, and the second part is encoded and transmitted in a time-hopping manner according to an on-off pattern. Leveraging the super sparse property of ODMA, the users' on-off pattern and pattern collisions are blindly detected based on the received signal without the help of pilot. Moreover, the on-off pattern detection, data decoding and collision recovery are performed iteratively over one sparse graph to enhance the overall system reliablity. We propose a finite-length performance analysis to the on-off pattern detection and iterative multi-user decoding, based on which both the user access sparsity, and channel code are optimized. Numerical result shows that with a rate 1/3 low-density parity-check code over GF(2^6), the gap between the proposed scheme and the random coding bound is less than 1.2 dB for up to 300 active users.