Low Power Programmable Architecture for PeriodicActivity MonitoringMohammad-Mahdi Bidmeshki, Roozbeh Jafari{bidmeshki, rjafari}@utdallas.eduDepartment of Electrical EngineeringThe University of Texas at DallasRichardson, TX 75080-3021ABSTRACTBody sensor networks (BSNs) are considered a great exam-ple for cyber-physical systems due to their close couplingwith human body. Activity monitoring is one of the numer-ous applications of BSNs. Continuous and real-time moni-toring of human activities has many applications in health-care and wellness domains. BSNs utilizing light-weight wear-able computers and equipped with inertial sensors are highlysuitable for real-time activity monitoring. However, powerrequirement is a major obstacle for miniaturization of thesewearable systems, due to the need for sizable batteries, andalso limits the life time of the system. In this paper, wepropose a low-power programmable signal processing archi-tecture for dynamic and periodic activity monitoring appli-cations which utilizes the properties of the physical world(i.e., human body movements) to reduce the power con-sumption of the system. The signif i cant power reduction isachieved by performing signal processing in a tiered-fashionand removing the signals that are not of interest as earlyas possible. Our proposed architecture uses wavelet decom-position and is favorable for the discrimination of periodicactivities. The experimental results show that our architec-ture achieves 75.7% power saving while maintaining 96.9%sensitivity in the detection of target actions, compared withthe scenario where the signal processing is not performed intiered-fashion. This creates opportunities to enable the nextgeneration of self-powered wearable computers.Categories and Subject DescriptorsC.3 [Computer Systems Organization]: Special-Purposeand Application-Based Systems—Signal processing systems,Real-time and embedded systemsGeneral TermsDesign, Algorithms, ExperimentationPermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for prof i t or commercial advantage and that copiesbear this notice and the full citation on the f i rst page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specif i cpermission and/or a fee.ICCPS ’13, April 8-11, 2013, Philadelphia, PA, USACopyright 2013 ACM 978-1-4503-1996-6/13/04 ...$15.00.KeywordsWearable Computing, Body Sensor Networks, Signal Pro-cessing, Power Optimization, Activity Monitoring1. INTRODUCTIONLong-term sensing and real-time monitoring of humanbody movements has numerous applications in healthcareand wellness assessment. This monitoring reveals impor-tant information about the quality of life, specially in thosewho suf f er from diseases such as Parkinson [17], or goingthrough rehabilitation, for example after a knee surgery [2],and provides the basis for home telecare paradigms. Usingthe real-time activity recognition and classif i cation, specialevents or activities can be captured and the care providercan be notif i ed to take the appropriate action.Advances in technology introduce light-weight wearablecomputers which provide the essential sensing, computingand communication platform needed for these monitoringapplications which are called Body Sensor Networks (BSNs).BSNs are excellent examples of cyber-physical systems inwhich BSN nodes are closely coupled with the physical world,i.e., the human body. Several of these tiny nodes can beplaced on dif f erent parts of the body and closely monitorand register every health related event on the patient aswell as the patient’s movements and activities. Sensor nodesequipped with inertial sensors provide a mechanism for thenatural capturing of human body movements, with mini-mum external intervention [3]. However, the size of batteryneeded to power these nodes during the monitoring periodis one of the major obstacles which prevents the miniatur-ization of these sensor nodes and limits their life time andwearability. Moreover, the limited power source restrictsthe processing power of the sensor nodes and imposes theuse of signal processing algorithms with low complexity, forreal-time classif i cation.Our ultimate objective in the design of BSNs is to createbatteryless units which can use body heat or body move-ments as the source of energy. However, the power budgetof these sources is in the order of µW. Current state of theart low-power microcontrollers still require the power bud-get of a few mW or a few hundreds of µW and are notsuitable for this purpose. ASIC design can satisfy the powerrequirement, but it is not programmable and is restricted toa specif i c application. BSN applications, on the other hand,have specif i c properties that guide through a more ef f i cientapproach for signal processing.Despite the necessity of continuous monitoring, many BSNapplications are interested in specif i c events during the mon-81
