System Overview
The AlarmNet system integrates heterogeneous devices, some wearable on the patient and some placed inside the living space. Together they perform a health-mission specified by a healthcare provider. Data is collected, aggregated, pre-processed, stored, and acted upon, according to a set of system requirements we have identified. This page gives an overview of the system architecture and components.
Figure 1: Illustration of a smart assisted-living space, instrumented with sensors and devices.
Figure 2: System architecture, showing roles of components from front-end to back-end, and physical associations among them.
Figure 1 graphically depicts the variety of sensors and devices in the architecture. Multiple body networks may be present in a single system. Traditional healthcare provider networks may connect to the system by a gateway, or directly to its database.
Some elements of the network are mobile, while others are stationary. Some can use line power, but others depend on batteries. If any fixed computing or communication infrastructure is present it can be used, but the system can be deployed into existing structures without the need to retrofit.
Figure 2 shows the components of the architecture, dividing devices into strata based on their roles and physical interconnect. Each tier of the architecture is described in the following sections.
Mobile Body Networks
This network comprises tiny portable devices equipped with a variety of sensors (such as heart-rate (see Figure 3), heart-rhythm, temperature, oximeter (see Figure 4), accelerometer (see Figure 5)), and performs biophysical monitoring, patient identification, location detection, and other desired tasks.
Figure 5: Wearable location and activity-classifying body network.
Figure 4: Telos-based pulse-oximeter sensor (Harvard design).
Figure 3: Telos-based EKG sensor (Harvard design).
These devices are small enough to be worn comfortably for a long time. Their energy consumption should also be optimized so that the battery is not required to be changed regularly. They may use “kinetic” recharging.
Actuators notify the wearer of important messages from an external entity. For example, an actuator can remind an early Alzheimer patient to check the oven because sensors detect an abnormally high temperature. Or, a tone may indicate that it is time to take medication.
The sensors and actuators in the body network are able to communicate among themselves. A node in the body network is designated as the gateway to the emplaced sensor network. Due to size and energy constraints, nodes in this network have little processing and storage capabilities.
More details about the particular body networks we have developed are available.
Emplaced Sensor Network
Figure 6: Bed sub-system.
This network includes sensor devices deployed in the environment (rooms, hallways, furniture) to support sensing and monitoring, including: temperature, humidity, motion, acoustic, camera, etc. It provides environmental sensing and control, interfaces for other devices, and a spatial context for operation.
Some of the emplaced sensors may form a semi-autonomous sub-system. For example, the sensors shown in Figure 6 are placed in/by a bed and can detect breathing rate, pulse, movement, and bed-exit falls. These are unobtrusive sensors that do not require interaction with the resident.
All devices are connected to a more resourceful backbone, potentially by multiple hops. Sensors communicate wirelessly and may use either wired or battery power. Nodes in this network may vary in their capabilities, but generally do not perform extensive calculation or store much data. The sensor network also interfaces to multiple body networks, seamlessly managing hand-off of reported data and maintaining patient presence information.
AlarmGate Backbone
A backbone network connects traditional systems, such as PDAs, PCs, and databases, to the emplaced sensor network. It also connects discontiguous sensor nodes by a high-speed relay for efficient routing.
AlarmGate nodes possess significant storage and computation capability, for query processing and location services. Yet, their number is minimized to reduce cost. The backbone may communicate wirelessly or may overlay onto existing wired infrastructure.
Back-end Databases and Analysis
One or more nodes connected to the backbone are dedicated databases for long-term archiving and data mining. If unavailable, nodes on the backbone may serve as in-network databases.
Figure 7: Circadian Rhythms (example).
We postulate that a person's behavior during the daily cycle of life at home, tends to fall into basic and regular patterns, called "Circadian Activity Rhythms" (CAR), that are heavily influenced by their long-term choices, basic necessities, and overall lifestyle. CAR depends on social rhythms and also interacts with the biological rhythms of organisms. They entail wake/sleep cycles regulated by melatonin secretion, activities of daily living (ADL), core temperature changes, and are measurable by observing residents' cyclic physical activity within the living space. (VG03)
Back-end analysis programs monitor CAR deviations over long periods of time, helping to identify behavioral changes which might signal the advance of degenerative disease and the need for increased care frequency. An example is Alzheimers disease, in which a person may spend as much as 40% of night-time awake, and sleep frequently during the day (MRAI+00).
Human Interfaces
Humans interface with the network through handhelds (PDAs), personal computers (PCs), or through wearable devices. These are used for data management, querying, object location, memory aids, and configuration, depending on who is accessing the system and for what purpose.
Limited interactions are supported with the on-body sensors and control aids. These may provide memory aids, alert delivery, and an emergency communication channel.
PDAs and PCs provide rich interfaces to real-time and historical data. Doctors use them to interface with the network to specify medical sensing tasks and to view real-time and historical patient data.