To panda
I recommended FPGA as better solution.
Really, hundreads of DSP microprocessors could support the
GHz data flow, sharing the common input data buffer.
But in this situation the FPGA based solution will really
occupy less hardware and consume less energy.
DSP microprocessor is out of questions for applications
with very heavy algorithms like CELP voice coding.
Because there is not any useful technology to map such
an algorithm into FPGA.
May be Celoxica and Handel C?
But it seems to be rather stiff because
the programmer has to parallel the algorithm by hand.
But it is the question of the behavioral synthesis.