One research emphasis at the Signal
and Information Processing Laboratory (ISI) is signal processing
with nonlinear analog electronic networks. While there are many
excellent reasons for prefering digital over analog processing
in most cases, we believe nonetheless that novel analog techniques
may enable the use of advanced nonlinear signal processing techniques
in applications that require very high speed or very low power
consumption. We are guided by the idea to
exploit, rather than to fight, the nonlinearities of transistor
physics and to address the problem of component variations by
suitable redundancy and adaptivity on the system level.
As an illustrative example of the above stated signal processing
with nonlinear analog electronic networks, we are implementing
analog decoders for error correcting codes. A key discovery here
is that the factor graph of many codes (including trellis codes,
turbo codes, and low-density parity check codes) can be translated
directly into analog transistor circuits. |
