3.2. Positioning and Tracking 39states are possible predecessor states although most of them will have a p( s| s)= 0,depending on the mode controlled variance. Also, it is a HMM of first order.The evaluation was conducted on an indoor office style scene with an area of 600 m2.The WAF model is configured with 6 different material coefficients. 3 APs wereplaced to lead to NLOS conditions at the evaluation site. The 2D grid resolution of0. 5 mx 0. 5 m leads to a state space with cardinality 2400. An averaged LE of around3 m for this setup is reported3.The HMM based localization system reported in[16] also uses a variant of the WAFmodel, dubbed RSSI delay profile, for simulating the radio propagation. Similar to[27] the variance of the emission probabilities p( x| s) is influenced by assumed noiseon the attenuation factors. For the transition probabilities p( s| s), different discretetwo-dimensional PDFs are possible. All states are possible predecessor states. It isa first order Markov model.Synthetic measurements were generated with the WAF model for an office like scenewith an area of 1200 m2. The noise level of the sythetic data, given by differentvariances, should represent natural conditions. 4 APs were evenly placed to form asquare. p(s| s') is modelled as a conic shaped two-dimensional PDF with maximumprobability at p( s| s) that is rapidly decreasing with increasing distance. On thissetup, a RMSE-LE of 5 m is reported.3.2.2 Particle FiltersParticle Filters(PF) are one of the most recent approaches to indoor localization.Historically, they originate from the field of localization for robotics. In this context,they have the typical sensor information of vehicles in the form of speed available.Two of the presented implementations can use the information of the step length andthe step heading and/or RSSI readings for pedestrian navigation. The techniquesto process this additional information source in the PF approach seems to originatein the research of robot navigation.Four implementations will be presented by focusing on the following properties:1. What is the available information source? If its a radio propagation model,what is the type of it?2. How is the environment modelled, is it a 2D or 3D model?3. How is the weighting of the particles defined? How is the state-conditional,the emission probability p( x| s) applied?4. How are the transitions p( s| s) modelled and how are they used during thesampling of new particles?5. How is the problem of sample impoverishment handled?6. On what scene was the evaluation conducted, how many particles were usedand what error rates were reached?3The value is derived from the CDF plot as only a median of 2 m is reported and there seem tobe a large number of outliers in range of> 8 m