By using the AGU, the size of additional logic is 75K (in NAND2). When the memory accessed with the logic generated by a memory compiler without an AGU, the size of the memory is 294K (in NAND2). When the memory is accessed by the proposed AGU, the size of the required memory becomes 7K (in NAND2). Thus, the total memory requirement with the size of the AGU overhead becomes 82K (in NAND2), which about a reduction of 72%. The AGU takes 1 cycle to execute, and the clock speed remains the same because the AGU logic is not on the critical path but the check node processing is on the critical path. Thus in case of the LDPC without an AGU, the throughput for rate-1/2 is 15.94Mbps and that for the rate-3/4 decoder is 28.80Mbps. We have designed decoders for several different group sizes, and we found that grouping of 16 nodes shows the best trade-off in terms of both the throughput and the area. When we increase the size of grouping further, the throughput increases, but the size becomes unacceptably large. So we conclude that for the CMMB standard, grouping of 16 nodes is satisfactory.