| Wireless sensor networks (WSNs) are capable of a myriad of tasks, from monitoring critical infrastructure such as bridges to monitoring a person's vital signs in biomedical applications. However, their deployment is impractical for many applications due to their limited power budget. Sleep states are one method used to conserve power in resource-constrained systems, but they necessitate a wake-up circuit for detecting unpredictable events. In conventional wake-up-based systems, all information preceding a wake-up event will be forfeited. To avoid this data loss, it is necessary to include a buffer that can record prelude information without sacrificing the power savings garnered by the active use of sleep states.;Unfortunately, traditional memory buffer systems utilize digital electronics which are costly in terms of power. Instead of operating in the target signal's native analog environment, a digital buffer must first expend a great deal of energy to convert the signal into a digital signal. This issue is further compounded by the use of traditional Nyquist sampling which does not adapt to the characteristics of a dynamically changing signal. These characteristics reveal why a digital buffer is not an appropriate choice for a WSN or other resource-constrained system.;This thesis documents the development of an analog pre-processing block that buffers an incoming signal using a new method of sampling. This method requires sampling only local maxima and minima (both amplitude and time), effectively approximating the instantaneous Nyquist rate throughout a time-varying signal. The use of this sampling method along with ultra-low-power analog electronics enables the entire system to operate in the muW power levels. In addition to these power saving techniques, a reconfigurable architecture will be explored as infrastructure for this system. This reconfigurable architecture will also be leveraged to explore wake-up circuits that can be used in parallel with the buffer system. |
