Perception of situations and acting in context is for humans a major part when interacting and communicating in everyday life. Taking into account that context is related to artefacts and also the observations how humans perceive context in space and time the following basic properties for context are stated. These basic qualities of context foster a systematic foundation for a system that supports nature-like context in a Ubiquitous Computing environment. The properties are extended by basic design criteria for complex and distributed systems.
6.2.1 Locality and Proximity
Situation and context can be seen as phenomenon that is related and bound to a particular place or region. The place or region where context information emerges – or that is assigned to this context information – plays an important role, especially in mobile and embedded systems. The place or region must not be seen isolated, it is always an attribute assigned to an identity, a process, a device, a task, an application, or to data. In mobile location-aware systems the position is an attribute of the device and implicitly of the user who is carrying the device.
Collecting data from the environment and acquiring context out of this data is inherently bound to a location. The readings are collected at a particular position and therefore they represent the context for this particular position or the area related to this position. The information is fully relevant at this position. Generally the relevance
of the data as well as the certainty on the correctness of the data declines with the distance from its point of origin. An example is measuring the temperature at a certain point. At this point the temperature is correct and relevant, however when interested in the temperature at a point nearby it is observable that with an increasing distance between the points the uncertainty whether or not the temperature is also valid for the new point gets larger. And when the distance between the point of origin and the point of interest is too large the reading is meaningless. This leads to the conclusion that if several sensors of the same type with similar quality exist the one that is closest to the point of interest is the most relevant to look at.
As seen from these observations locality of context is quite important and should therefore be included in the model as one of the basic principles. For the model the following aspects and requirements should be taken into account:
• context information has a point or region of origin
• at the point or region of origin the relevance of the context information is maximal
• the relevance decreases with an increasing distance from the point or region of origin
• if several sensors of the same type are available the one which is spatially closest has the highest relevance
6.2.2 Time
Time is for the human understanding and classification of situations a vital aspect. It is also highly relevant for sensor data that is acquired from the environment. Using concurrency or exploiting the fact that events take place coincidental or within close timely boundaries are a basic way of relating different aspects of complex situations. Regarding the time aspects when acquiring sensor data and contexts a similar semantic as for location is observable. Values are created at a certain point in time; and these values are in general more relevant to an event that happens roughly at the same time than to an event that takes place much later or earlier.
The concept of time should also be included in the model, exploiting the following basic observations:
• context information has a time of origin • the relevance is maximal at the time of origin
• relevance decreases with an increasing time distance from time of origin • if several sensors of the same type are available the one which provides the
most timely reading has the highest relevance
In certain application areas it may be useful and beneficial to relate the relevance to issues that are specific to the application rather than to the temporal and spatial distance. In the model and platform described in this chapter this case will not be further regarded, as for many application areas in Ubiquitous Computing time and space are a prime concern.
6.2.3 Independence Between Acquisition and Use
The context, regarded as the type of situation that surrounds something else, is widely independent of the way it is used. From everyday experience we know that the perception of a situation does not change the current situation. Similarly in a context- aware system, acquisition of context does not influence the context. Again as we can recall from everyday experience after perceiving a situation the action taken may have an influence on the future situation. This again is similar in a context-aware system, knowing a context and changing the behaviour of an application according to the current context may change future contexts.
For modelling and designing a distributed platform it is desirable to reduce the number of dependencies as much as possible. To realise a loosely coupled system identifying issues that can be dealt with separately is of major importance.
There is a great variety of methods and technologies available for the acquisition of context in Ubiquitous Computing environments. The process of inferring context from data collected in the environment is itself usually a multi-level process that is conducted by components that are independent to some extent [Schmidt,99c],
[Golding,99], e.g. sensors, feature extraction, and perception of context should be independent.
It is also desirable that the algorithms and methods that supply context do this in a most general way abstracting from supplying context to a specific application. If context is delivered in a general way and independent of a specific application it becomes also feasible to develop context-aware applications without a particular sensing environment in mind. In the area of location aware systems this is already the case. Everyone can develop an application making use of location, e.g. by using a general description such as geographic coordinates, without knowing what the actual sensing system will be.
Having independence between context acquisition and context use also makes it possible to simulate either side to test and debug the other.
This compiles into the following wish list summarising the requirements: • context acquisition and context use is highly independent
• all levels in context acquisition are highly independent
• it is possible that more entities that supply context exist independently (even for the same context)
• it is possible that more entities that uses context exist independently (even for the same context)
• applications are modelled independent of time and location but using these properties implicitly
The property of independence between context acquisition and context use is not necessarily general. In some application areas where acquisition is specifically designed for an application it may advantageous not to insist on independence. However this has usually the price of less flexibility and greater complexity.
6.2.4 Distribution and Scalability
In everyday life humans have a great ability to filter information and shift to information that is relevant, one prominent example is the so called ‘cocktail party effect’ [Handel,89]. This basic mechanism protects our information processing system from information overload, because the processing system is limited. Similarly the question of scalability arises in systems that deal with many stimuli from the environment and potentially a vast number of contexts.
As artefacts and infrastructures are inherently distributed it is one obvious approach to exploit the spatial and temporal properties of context to achieve scalability. This follows the concepts described earlier on the basic properties of locality and time for context. Scalability over time can be realised by modelling data in a way that it disappears after a certain time.
The concept of spatial and temporal scalability can be illustrated by the following examples describing human perception. The perception of sound scales spatially, as with an increasing distance to the source of sound the volume as well as the quality decreases. From a certain distance the sound not audible at all. The scalability over time can be illustrated when considering the human perception of smells. At the moment they are created the smell is at the maximum (e.g. while cooking). Over time the smell fades till it is not recognizable anymore. If this scalability would not be ‘built-in’ to our world life as we know it would be hardly possible; just imaging sound would not be locally limited.
The following aspects related to scalability and distribution are taken into account for building the model:
• the distribution of information is locally restricted – localised scalability
• the lifetime of information is restricted – scalability over time. (In the case where is no new information added the amount of data decreased over time.) • the spatial distribution of individual components is a basic property
6.2.5 Transparency
Again looking at the human way of perceiving context it becomes apparent that the spatial and temporal distribution is transparent. It happens to be around without further considerations. The reference to a person’s position and to the current time is implicit and usually unnoticed.
When designing a system that supports the use of context in applications it seams desirable to offer a similar degree of transparency about the distribution of context. The challenge is to create a system that provides context information for applications dependent on the temporal-spatial relationship between the application and its environment. The underlying mechanism for distribution and spreading of context should however be transparent for the context user as well as for the one who produces context information. The components that offer context should be able to influence the spreading of the context they are creating.
In summary this results in the following demands for the model and architecture: • context is always bound to the current location and time
• distribution and fading mechanisms are built into the model and the platform • context is transparently distributed for context creators and users
The description of the nature of context in a Ubiquitous Computing environment – seen from a human perspective – led to the extraction of basic properties. These observations become the foundation for the distribution model and platform described in the reminder of this chapter.