While the positioning error increases as AP numbers drop for the FP method, the number of APs does not affect the positioning performance so much for the FPM method. FP is highly dependent on the stability and the number of existing APs. FPM shows more resilience to a changing wireless network environment. Results in Section 3.6.1 show that averaging more k neighbours for fingerprinting give better performance, FPM also shows a similar pattern. The particles in the FP method are weighted by a single fingerprint solution (the mean of k-NN) thus if the solution at a certain epoch is biased from the truth, the FP solution would be contaminated as well. Due to the large fluctuation shown in Wi-Fi signals, this may occur quite often during the FP method. On the other hand, the particles in FPM are weighted by all potential fingerprints, therefore the positioning solution would not be affected too much if only very few fingerprints of the total potential fingerprints are biased. Hence a larger number of fingerprints should be counted as potential locations. Although this may mean a large ambiguous area of fingerprints alone, at least the fingerprints around the true location would not be discarded. FPM proves to be more appropriate for the DR/Wi-Fi integrated navigation solution as it averages out the error and proves to be more resilient to Wi-Fi signal variation.
However it must be remembered that this result is based on simulated Wi-Fi RSS and real data tend to be much more noisier. Thus to tackle the complexity of a real environment and the potential failure of Wi-Fi network, collaborative algorithms are developed by bringing in ranging measurements from a number of collaborative users in a local network.
3.7
Summary
This chapter gives details to some popular indoor positioning methods, including PDR using foot mounted inertial sensors, Wi-Fi fingerprinting and indoor map matching. To reduce fingerprint database training time, Gaussian Process regression is applied to generate the database. Trials show that GPR reduces training time by reducing the number of required training points and the time for training each point. A particle filtering based PDR and Wi-Fi integrated pedestrian navigation algorithm is also introduced here for more stable positioning results.
Simulations of the basic Wi-Fi fingerprinting procedure is presented in this chapter to develop understandings of positioning performance under different conditions, i.e. setting different measurement error and different number of nearest neighbour, k. The PDR and Wi-Fi integration navigation
is further developed into a fingerprint mapping navigation (FPM) solution to reduce positioning error and noise. The performance of FPM simulation is analysed with different number of APs. Its performance is compared to PDR solutions and conventional fingerprinting solutions and obvious improvement can be seen in FPM, especially when the number of APs reduce and conventional fingerprinting becomes less reliable.
Chapter 4
Collaborative positioning with ranging
constraint
4.1
Introduction
The first aspect of collaborative positioning has already been considered in the previous chapter, i.e. the integration of inertial measurements and Wi-Fi signal measurements into a single system. This chapter will take a step further and look into the other aspect of collaborative positioning which involves the integration of multiple systems, or users, through ranging measurements between multiple users and transmitters.
A typical collaborative network consists of a number of fixed transmit- ter nodes, known as anchors (denoted as Tx), and a number of unknown moving nodes, known as rovers (denoted as Rx). In collaborative posi- tioning, the heading drift of each rover can be constrained by integrating ranging to other rovers and anchors. Accurate ranging measurements can push the state estimation of rovers towards the true position by providing information on the geometry of the network. This fixes the rover and other nodes into the geometry with a certain distance between each other (i.e. the ranging measurement). By sharing this collaborative information between each other, the positioning results of all rovers within the network are improved.
Signals of opportunity provides a major opportunity for collaborative positioning. Our environment is filled with a variety of opportunistic signals, e.g. GNSS, Wi-Fi, cellular signals, radio signals etc. Usually, GNSS signals would not be considered opportunistic, however different signals behave differently in different environments and each is suitable for positioning in different environments. While GNSS provides very accurate positioning outdoors, they are not reliable inside, where Wi-Fi signals work best. In this rapidly developing modern era where we are constantly facing a mass of information, it is more about selecting the right and valid information than simply searching for information. In collaborative
positioning, the selection of signals should be aimed at seamless transfer between different positioning environments, achieving high positioning accuracy with relatively low computation cost. The authors in Yang et al. (2009) demonstrate that while a number of signals of opportunity are available, not all of them improve the positioning accuracy. The authors search for an optimal collaborative network among users and signal sources based on differential ranges.
As already discussed in Chapter 3, each navigation method has its own strengths and weaknesses. Wi-Fi and IMU integration has been introduced to compensate the drift of inertial sensors as well as the unstable signals from Wi-Fi sensors. Yet low-cost inertial sensors used for pedestrian navig- ation can have a very large gyro drift that leads to errors of hundreds of meters in a few seconds. Even with corrections from Wi-Fi signals, such positioning instability cannot be easily overcome. Relative ranging, i.e. the implementation of P2P ranging, can restrict such measurement bias when integrated efficiently.