Protecting practical FFT implementations that share common processing elements

Functional faults are discussed. Before any reconfiguration can be done, the error must first be detected and, then, the malfunctioning unit must be identified. Fault detection is considered. The authors assume there is at most one malfunctioning in the processing array at a time. They describe a general method of detecting any fault occurrence, either permanent or transient, in a practical radix-2 N-point discrete Fourier transform system independent of the technology base and for any reasonable interconnection scheme among these PEs. An algorithm-based fault tolerant (ABFT) concurrent error detection (CED) scheme is derived by scheduling all the PEs with certain characteristics. Two fault detection techniques with these attributes are demonstrated. The general use of gain spaces for ABFT CED scheduling is examined.<>