The subje t of this paper is a Smart Spell Che ker System (SSCS) that an adapt to a parti ular user by using the user's feedba k for adjusting its behavior. The result of the adjustment is manifested in a di#erent ordering of the suggestions to the user on how a parti ular spelling mistake should be orre ted. The SSCS uses the Adaptive Software Ar hite ture (ASA). The ASA onsists of a hierar hy of layers, ea h ontaining a number of omponents alledKnowledge Sources.

In this paper we consider the problem of monitoring illuminations of multiple emitters by one electronic receiver located on a moving platform. The emitters exhibit a quasi-periodic radiation pattern, each in a different frequency band and with a different illumination period and illumination time. The goal is to tune the receiver to the appropriate frequency band at the appropriate time so as to maximize the probability of detection of each illumination of each radar (expressed as a quality of service metric). This problem can be viewed as two subproblems: (1) generating tuning requests by particular frequency bands and (2) scheduling the receiver to satisfy the requests. In our scenario the number of requests from all the bands exceeds the physical capability of one receiver (overload) and thus selection is needed. The goal is to push the limits on the overload while still maintaining a required level of quality of service metric. This can be achieved by controlling both the generation of the tuning requests and the scheduling. According to the control theory metaphor of software development, we map this problem onto a control architecture with one system-level controller and a collection of emitterlevel controllers (one per emitter). We show that our control based sensor manager has significant advantages over a scheduler without feedback in terms of both the overload and the quality of service metric. We also discuss design issues of such sensor management systems.