Sensory information coming from natural sensors and being propagated on afferent nerve fibers could be used as feedback for a more efficient closed-loop control of a Functional Electrical Stimulation system. In order to extract and separate these signals according to their nerve fascicule origins, we propose a new architecture of a multipolar cuff electrode and an optimized integrated acquisition circuit. Concerning the electrode, we propose a specific configuration using a large number of poles in order to both reject parasitic signals, such as electromyogram and provide a maximum of recording channels in order to help the signal localization inside the nerve. Moreover, specific low-level analog signal processing was designed to extract the expected low-amplitude signal from its noisy environment. This signal processing is implemented in an ASIC that has to be implanted close to the electrode to achieve the best signal-to-noise ratio.