Page 338 - DSP Integrated Circuits

7.7 FFT Processor, Cont. 323

under certain conditions, find the global minimum and not get stuck in local
minima [10,16, 21].
Simulated annealing can be applied to optimize a cyclic schedule. We can find
an initial schedule with almost any of the methods presented earlier. The only
restriction is that the schedule must be feasible. Further, we can reduce the search
space by performing both an ASAP and an ALAP schedule to determine the life-
span of the operations. The simulated annealing algorithm can be directly applied
to the cyclic scheduling problems just discussed. The operations are randomly
moved within their feasible time frame so that the cost function is minimized.
Generally, for any optimization method it is difficult to find a good cost function
that has both local and global properties such that the global minimum is reached.

7.7 FFT PROCESSOR, Cont.

In this section we will schedule the processes of the FFT to satisfy speed con-
straints while respecting constraints on the available resources. We have already
decided to use two butterfly PEs and two logical memories.
Figure 7.52 shows the global schedule for the processes in the FFT processor.
The time estimates included in the figure will be explained later. The FFT processor

Figure 7.52 Global schedule for the FFT processor