Hi Subodh ;)

I guess it would be a good exercise in distributed processing at the embedded level
What if u...oops ... you used a bus-based system to distribute data between 8051s connected in a master-slave architecture?

For an FFT lets assume, the master uC only has the job of acquiring sample data, distributing it to the slaves and collating the processed data. The slaves (say 4 of them) will do the actual FFT processing for 2,4,8 etc samples parallely, return the results and wait for the new data

It seems to possible- How feasible is it? Can it actually be a way to make reasonably fast but cheaper DSP hardware for low-end applications? Due to the overhead of distribution, I think it is only worthwhile if the amount of parallelism is high... but how much parallelism would that be, and how much is possible?

Is this the same as pipelining? Isn't pipelining more like vector processing rather than array processing? How lost am I over here?

kundi