Section: New Results

Self-describing objects and tangible data structures

Participants : Nebil Ben Mabrouk, Paul Couderc [contact] .

A development in the line of the composite objects (see section 3.3 ) are self-describing objects. While previous works enabled integrity checking over a set of physical objects, these mechanisms were limited in two aspects: expressiveness and autonomy. More precisely, objects support the detection of special conditions (such as a missing element), but not the characterization of these conditions (such as describing the problem, identifying the missing element). Moreover, this compromises the autonomous feature of coupled objects, which would depend on external systems for analysing these special conditions. Self-describing objects are an attempt to overcome these limitations, and to broaden the application perspectives of autonomous RFID systems.

The principle is to implement distributed data structure over a set of RFID tags, enabling a complex object (made of various parts) or a set of objects belonging to a given logical group to "self-describe" itself and the relation between the various physical elements. Some applications examples includes waste management, assembling and repair assistance, prevention of hazards in situations where various products / materials are combined etc. The key property of self-describing objects is, like for coupled objects, that the vital data are self-hosted by the physical element themselves (typically in RFID chips), not an external infrastructure like most RFID systems. This property provides the same advantages as in coupled objects, namely high scalability, easy deployment (no interoperability dependence/interference), and limited risk for privacy.

However, given the extreme storage limitation of RFID chips, designing such systems is difficult:

  • Data structures must be very frugal in terms of space requirements, both for the structure and for the coding.

  • Data structures must be robust and able to survive missing or corrupted elements if we want to ensure the self-describing property for a damaged or incorrect object.

In the context of RFID system, the resiliency property of such data structures enables new information architecture and autonomous (offline) operation, which is very important for some RFID applications. We previously applied the self-describing objects approach to the waste management domain [1] , which has shown to be a specially challenging situation for RFID. This challenge is found more generally in pervasive computing scenarios involving RFID reading in uncontrolled environments (see section 4.3 ).

Pervasive support for RFIDs.

We propose to apply our approach to improve the robustness of RFID inventories / batch checking: when many objects are read at once by an RFID reader, miss read are common and raise reliability and operational issues for applications. An innovative solution to this problem is to take advantage of the multiplicity of tags by leveraging them as a distributed memory shared by a logical group. In this way, it is possible to support error detection as well as information recovery. We proposed a flexible protocol to support robust EPC retrieval in adverse reading conditions. The proposed protocol uses erasure correcting techniques to enable error-free recovery of misread EPCs [2] . It is further customizable with respect to the rate of misread tags and application requirements. This work was the object of an Inria patent (Patent filed in April 2015 - Inria 179). Fine-tuning the protocol parameters is still the object of on going experimentation in the context of the Pervasive_RFID project (see section 9.1.1 ).

At the software level, RFID inventory reliability issue is usually addressed by anti-collisions mechanisms and redundancy mechanisms. Anti-collisions protocols limit the risk of data corruption when multiples tags have to reply to an inventory request. Redundancy is often implemented in RFID readers by aggregating the results of multiple inventory requests over a time frame, to give the tags multiple opportunities to reply. While useful, these strategies cannot ensure that a given inventory is valid or not (in other words, one or more tags may be missing without being noticed). In situations where we have to read large collection of objects of various types, the performance is difficult to predict but may still be adequate for a given application. For example, some application can tolerate missing some tags, provided that miss read probability could be characterized. In some cases, read reliability could be improved using mechanical approaches, such as introducing movements in objects or antenna to introduce radio diversity during read. Finally, distributed data structure can be used over a set of tags to be used to mitigate the impact of misread (by using data redundancy) and to help the reading protocol by integrating hints about the tag set collection being read.

We studied extensively by experimentation the behaviour of existing RFID solutions in the context of uncontrolled environment (meaning, random placement of tags on objects mixing various materials) in order to characterize their real-world performance regarding the parameters of such as tags numbers, density, frequencies, reader antenna design, dynamicity of objects (movements), etc. From these experimentations, we would like to identify the conditions that are favorable to acceptable performance, and the way where there are hopes of improvement with specific design for these difficult environments. These results should also allow improving the performance: high level integrity checks can guide low level operations by determining whether inventories are complete or not. This cross layer strategy enables faster and more efficient inventory protocols.