Signal Processing Tools for Co-operative Planet Exploration Employing Ad-hoc Sensors Networks

New missions to explore the surface of unknown planets need the use of more sensors that analyse the characteristic of the planet. For these Missions have been employed a plethora of instruments and sensors to monitor the surface under inspection to obtain measurements about the quantity of interest as for example electromagnetic properties or the presence of organic substances like water. Different scenarios could be considered for the sensors/instruments deployment. In this context, a wireless sensor network can be established where each sensor processes the data and transmits the results the remote station for further studies. This station can be a base station placed on the planets surface or a satellite/orbiting spacecraft around the planet able to collect data sent from the sensors. Therefore in this kind of application it appears to be interesting to consider a sensors network as possible solution for the communication problem. In general, a wireless sensor network consisting of a number of sensors spread across a geographic area. Each sensor has communication capability and sufficient intelligence for signal processing. The network topology may change rapidly and unpredictability over time. An ad-hoc network of sensors can be employed in planet explorations to make available and analysable the data during the study. A network infrastructure with small physical nodes and low power consumption is necessary. In fact, the sensors, which have to be carried on space mission, are difficult to charge. The lifetime of a sensor is determined by the battery life, thereby, the minimization of the energy is required. Another problem in the use of a lot of different sensors are determining the coverage area, with an optimisation of the coverage, the number of sensors can be reduced. To this end multiple aspects must be taken into account related to desired precision of estimate, energy available to the sensors for processing and for moving, communication bandwidth among sensors and base stations, resources available, signal processing techniques for sensor network to take decision/estimation after collection of a certain number of observations, etc. In this last context, the theory of Distributed Decision making is a clear example about how the theoretical study of Statistical Decision Theory can greatly benefit from a data fusion viewpoint when dealing with cooperative decisions taken by different modules observing the same event under different perspective (i.e. monitoring different observed variables related to the same phenomenon). Another aspect is the possibility to develop an intelligent module that can optimally fuse in an adaptive way measurements coming from different autonomous sensors. The intelligence of such a module can be associated not only with its capability to produce integrated data about a observed physical phenomenon, but also to consider that such data are more useful and significant if they are acquired in such a way as to maximize the contextual potentiality of the information such data can carry. Thanks to the intelligence of the sensors different fusion strategies depending on possible failures of some of the sensors involved can be developing. Adaptability in setting the strategies necessary to maximize precision of a multisensor estimate is a key issue for remote space missions. This is especially true if such adaptability comes out by the autonomous procedures for decision and estimation adopted by the pool of sensors without need of messages from remote base stations that can arrive with a not tolerable delay. Remote reconfigurability through local planetary space communication links among intelligent sensor units will be also taken into account by keeping possibility of reconfiguring fusion behaviour by higher priority commands from base station links. Introduction. Planet exploration is not new, and in fact, The first experiments of exploration of the planets of the solar system begun between 1962 and 1973 with the Mariner spacecraft; these vehicles were designed and built by the NASA’s Jet Propulsion Laboratory and was relatively small robotic rangers. Now, thirty years after the first experiments, is more and more necessary to study planets with unmanned spacecrafts before sending there astronauts to visit. In particular, new missions to explore the surface of Mars have foreseen the use of more sensors that analyse the characteristic of the planet. In general sensors with extreme miniaturization, availability, accuracy, reliability and power savings can be used. For these Missions have been employed a plethora of instruments and sensors to monitor the surface under inspection to obtain measurements about the quantity of interest as for example electromagnetic properties or the presence of organic substances like water. Different scenarios could be considered for the sensors/instruments deployment. First of all, the sensors are inserted inside a robot or into a similar structure that could have movement capability or not. This structure is put on the planet soil and the sensors start to collect measurements and data. A second possibility consists of using more than one robot for example two or three to cover a large zone of the soil to obtain different measurements in different places. Each structure hosts different sensors according to the quantity under monitoring. Moreover, in this heterogeneous context sensors could be employed for remote sensing (for example inserted in satellite or orbiting spacecraft around the planet). In particular, the main application could be devoted to detect the presence of water or other organic substances on the soil. Otherwise, for imaging application the use of remote sensing techniques allow to obtain image of the soil where the organic substances are detected by the sensors network. In both cases, a sensor network is established where each sensor processes the data and transmits the results the remote station for further studies. Therefore in this kind of application it appears to be interesting to consider a sensors network as possible solution for the communication problem. Data Sensor Networks. In general, a smart sensor network consisting of a number of sensors spread across a geographic area. Each sensor has wireless communication capability and sufficient intelligence for signal processing. In general, the smart sensor networks are used for different problems: • Determine the value of some parameter at a given location; the temperature, the atmospheric pressure, the relative humidity can be analysed • Detect the occurrence of events of interest and estimate parameters of observed events A mobile ad hoc network (MANET) [3] is an autonomous collection of mobile users (nodes) that communicate over wireless links. The network topology may change rapidly and unpredictability over time. An ad-hoc network of sensors can be employed in Mars explorations to make available and analysable the data during the study. A networks infrastructure with small physical nodes and low power consumption is necessary. In fact, the sensors that have to be carried on a planet are difficult to charge. The lifetime of a sensor is determined by the battery life, thereby, the minimization of the energy is required. Another problem in the use of a lot of different sensors are determining the coverage area, with an optimisation of the coverage, the number of sensors can be reduced. The most important studies about the coverage have been made for the analysis of an initially unknown environment for mobile robots. There are some techniques to solve this problem, one of these is the generalized Voronoi diagrams, where a set of discrete sites (points) partitions the plane into a set of convex polygons such that all points inside a polygon are closest to only one site. Another study of coverage is to determine the optimum number of base stations required to achieve the system operator's services in communication networks. An important requirement in a sensor network is the self-organization; in fact, the placement of the sensors in different hostile locations makes unfeasible the manual configuration. The network has to reconfigure itself and continue to transmit. To improve the overall performance it is necessary to fuse data from multiple sensors in an optimum manner. This data fusion steps require the transmission of control messages and so it puts limit on the network architecture. The basic idea of distributed detection is to have a number of independent sensors each making a local decision and combining these decisions at a remote centre to generate a global decision. Reconfigurable Data Fusion. On the basis of the studies carried out during planet exploration and plethora of scientist’s interest in the case of Mars planet for further research comes out the idea to design a reconfigurable module to insert in an ad hoc sensors network deployed on the Mars surface and devoted to gather measurements about object of interest (for example dielectric constant) or presence of water or other organic substances. This module should be based on accomplishments obtained in the study of statistical data fusion theory and techniques in conjunction with the quite novel concept of ad-hoc sensor networks, leveraging the research carried worldwide that has produced a significant set of tools which are able to exploit the synergies and the complementarities offered by measurements provided by different sensors in order to monitor different aspects of the same phenomenon. At the same time it seems to be understood only recently the necessity of exploiting a coherent framework that helps when designing optimal strategies for cooperative sensing behaviours of pools of autonomous sensors. To this end multiple aspects must be taken into account related to desired precision of estimate, energy available to the sensors for processing and for moving, communication bandwidth among sensors and base stations, resources available, etc. The theory of Distributed Decision making (by Varshney [4]) is a clear example about how the theoretical study of Statistical Decision Theory can greatly benefit from a data fusion viewpoint when dealing with cooperative decisions taken by different modules observing the same event under different perspective (i.e. monitoring different observed variables related to the same phenomenon). A space mission involving unknown planet exploration such as the one foreseen for Mars is a clear domain where a module just described could be necessary. In particular, the possibility seems attractive of developing a intelligent module that can optimally fuse in an adaptive way the measurements coming from the different autonomous sensors. The intelligence of such a module can be associated not only with its capability to produce integrated data about a observed physical phenomenon (e.g. water presence under Mars soil), but also to consider that such data are more useful and significant if they are acquired in such a way to maximize the contextual potentiality of the information such data can carry. For example, integrating measurements that proof the presence of water on Mars can be more significant if the information is related to an extended Mars surface and if success occurs in communicating with Earth the sufficiently certain information about the location where the pool of robot provided by sensing found water evidence. Furthermore, the intelligence of such a module should be devoted to make available different fusion strategies depending on possible failures of some of the sensors involved. Adaptability in setting the strategies, necessary to maximize the precision of a multisensor estimate is a key issue for remote space missions. This is specially true if such adaptability comes out by the autonomous procedures for decision and estimation adopted by the pool of sensors without need of messages from remote base stations that can arrive with a not tolerable delay. Remote reconfigurability through local planetary space communication links among intelligent sensor units will be also taken into account by keeping possibility of reconfiguring fusion behaviour by higher priority commands from base station links. Distribution of Estimation and Decision toolkits, Sensor Measurements Another example is to collect measurements provided by different electromagnetic sensors placed on the surface under monitoring at investigating the electromagnetic properties of a planetary soil. In order to do this, three sensors are considered: Soil Dielectric Spectroscopy Probe (SDSP); Ground Penetrating Radar (GPR); Time Domain Electromagnetic Measurement (TDEM) system. All these quantities can be provided at the sensors level and further transmitted the remote station for data storage and collection. The above sensors can be thought inside a robot structure that can be fixed or with mobile capabilities. In such kind of context some problems could arise at different level. The first domain concerns with the communication network infrastructure able to support such kind of applications. In the last years great interest has been shown towards the use of the ad hoc wireless sensors networks as communication paradigm. This motivated the rapid deployment of such kind of topology which allows to establish a rapid network infrastructure without necessity to use base station or intermediate radio nodes. Communication is direct between the elements constitute the ad hoc network. Such kind of transmission paradigm is well suited especially in the case where it is necessary to arrange a communication infrastructure which do not have pre-existent network. Therefore the usage of an ad hoc wireless sensors network will be taken in account. The nodes could be thought as robots equipped with the different sensors used to obtain the useful measurements. A robot can be equipped with one or more sensors inside it and it could be fixed/move around the zone of soil under inspection. The robot will also be equipped with a certain processing capacity and storage, witch could arise power consumption problems and processing complexity. In fact, a good strategy in order to avoid great energy consumption should be following and it is linked with the processing capacity available on the sensors. In fact, the design of signal processing techniques and tools having the main characteristics to be optimised to be supported in a good way on the sensors embedded in the robots have to be considered. Each sensor measures some quantities related to the quantities of interest and it can send the obtained data directly to the remote station.. All the signal processing techniques can be centralized or distributed [4]. During the pre-processing phase great attention is devoted on the precision of the obtained measurement and how to link it to the quantities of interest. In fact, often, the measured quantities are not directly linked with those of interest, therefore before to perform data fusion among the resulting data set some transformations are required to link the measured quantity to the quantity of interest. Such kind of transformations allow one to perform the further multisensor data fusion linking the measurements to the quantity of interests. After measurements data collection and after having obtained the quantity of interest the following phase is devoted to determine the data fusion methodology. In this case could be followed two possible strategies. In the first, each sensor sends towards the remote station the obtained measurements for further fusion. The second strategy, could involve a cooperative exchange of information about the measured quantities between the sensors of the same type involved in the measurement process. Such kind of process could be helpful for a refinement in the measurement process because each sensor shares and updates its measurement information on the basis of that one received by the other sensors. In this context it could be possible to use fusion rules inside the network itself. For example, one robot (that could be named on the master of the sensors network) can use a fusion rule to fuse, on site, the information. This strategy could be useful in the case when the communication channel between the remote station and the sensors network is not available. It is possible to keep the system functioning and storing measurements while waiting for the re-establishment of the channel. Both the two strategies should employ fusion strategies for the considered context. Another aspect is to formulate the problem of fusing data from independent sensors from a theoretical point of view. Each sensor provides an estimate of the quantity of interest employing the transformation mentioned before independent from each other. In particular, each quantity estimate is associated with a confidence value. The fusion problem can be formulated as the hypothesis-testing problem related to the more likely presence of a monitored quantity in the considered zone of the planet soil, given the estimates provided by the available sensors. In this context the use of the distributed Bayesian decision theory will be used as possible reference model for the development of a theoretical framework for decision-level fusion. The choice of this model is motivated by its extensive use in the literature to derive a number of decisionlevel fusion. It will be also investigated the choice of other distributed decision strategy as for example distributed sequential decision [4]. In fact, in many practical situations observations are collected sequentially, and more information becomes available as time progresses. In such cases, we may wish to process the observations sequentially and make a final decision as soon as we are satisfied with the decision quality or detection performance. Different kind of strategies allows exploiting different kind of distributed sensors network topologies. For example will be taken in account parallel structure but also serial structure with the possibility to choose other different kind of topologies on the basis of the sensors scenario. It is expected that a large number of sensors with limited sensing, communication and computing power will be distributed over Mars to develop situational awareness. So the challenge will be to determine near-optimal estimation and decision-making strategies. In addition, statistical signal processing will be performed in the presence of large uncertainties and appropriate algorithms will be developed. Sensor networking issues including communication amongst sensors, network architecture, energy-efficient protocols, and data fusion will be addressed. The choice of a robust data fusion methodology is of primal importance to warrant the success of such mission (or at least minimize the fall risks). Although many fusion methodologies exist already in literature based on several different theoretic frameworks (Bayesian, Fuzzy Logic, Possibility theory, DempsterShafer theory, etc), one of the most difficult challenge for the Data Fusion community has not been investigated until now: How to deal with paradoxical/conflicting sources of information? What and how decision can be taken in such situations? How to handle properly the available knowledge and how to adapt the fusion methodology and algorithms with the evolution of parameters of the systems and environment? All of these difficult questions haven't been answered in a unique theoretical framework for data fusion. All existing methods have mainly been developed and improved for many particular problems known beforehand. But deep theoretical research works still remain in the context of planet exploration due to the difficulty of mission planning and all the uncertainties (possible sensor defect or biases, algorithms failures, unknown environment, etc) related to the mission itself. Recently, new theoretical investigations on robust data fusion methodology dealing with uncertain and conflicting/paradoxical information have been made and a new Theory of Uncertain, Plausible and Paradoxical reasoning (called DSmT) has been proposed by Dezert and Smarandache in 2002 [5]. DSmT appears to be very promising for the generation of high-level fusion and expert systems and a new strong alternative to classical approaches managing uncertainty A theoretical framework has to be developed with reference to some real-world scenarios where the number and nature of the sensors that can provide reliable information about the monitored quantities. In particular, the validity of the statistical independence assumption among estimations from different sensors has to be investigated. The independence assumption allows for the use of well-known decision fusion techniques based on the Bayesian decision theory. On the other hand, if such an assumption cannot be deemed to be valid, other techniques should be devised. Data fusion is aimed to produce reliable estimation of the quantity of interest by exploiting information produced by different (homogeneous or heterogeneous) sensors. The choice of the architecture for the data fusion process should take into account the following key factors: efficiency, reliability, required transmission band, data synchronization, data redundancy, and costs. These factors will be also taken into account when designing the fusion techniques in order to limit their complexity. Algorithms for data fusion will be further subdivided into different groups according to the level of abstraction of the information involved signal-level fusion, feature-level fusion, decision-level fusion. At each level the following problems will be addressed and suitable techniques developed which are synchronization, data alignment, feature selection, data classification and decision