Abstract
This thesis combines the RSSI fingerprinting technique in MIMO system with the particle filter to perform positioning. The traditional RSSI technique suffers from effects of NLOS and multi-paths. The rsearchers proposed a method called fingerprinting technique to improve the disadvantages of the RSSI technique, but it still suffers the multi-modal effects. Therefore, this study use the particle filter to optimize the output. However, the tradtional particle filter is too complex to implement in ASIC architecture. Therefore, this work is to modify the particle filter architecture. In the system domain, the MIMO fingerprinting technique can deal with the effects of multi-paths and shadowing; In the signal processing domain, the particle filter can handle the the non-linearity and non-Gaussian properties. The Spacing and Run is the key points while building the fingerprinting map. The Spacing is the resolution of the map, and the Run is the ability against the time-varying channel. The Spacing is smaller, the accuracy is better. The Run is larger, the ability is better. However, there is also a trade-off between the performance and cost. For the particle filter, this work modify the sampling and resampling strategy. The traditional sampling strategy is to add random noise to scatter the particles, and this work proposed a method called selective sampling which offers a ThresholdS to decide which particle should be added by random noise. The simulation shows that this strategy can accelerate the convergence time and improve the accuracy. Because the efficiency of the traditional systematic resampling strategy is too low, researcher proposed the independent metropolis hasting(IMH) resample which can be pipelined. However the performance fo the IMH resample is not good enough. Therefore, based on the IMH resample, this work proposed the threshold IMH(T-IMH) resample which offers a ThresholdR to be compared to the weights of the particles. When the ThresholdR is larger, it means that the present particle is not reliable, so we replace the present particle with the previous particle; Otherwise, we keep the present particle. This architecture can be still pipelined, and the simulation shows that the accuracy is better. For the mobile positioning, the processing period is smaller, the accuracy is better. Moreover for the base station(BS), the processing period is smaller, the more number of users that the BS can afford. Therefore, this work propose the distributed particle filter which is a kind of parallel processing to increase the processing speed. This work perform the sample and weight in each processing element(PE), and the resample in a central unit(CU). This work uses the ARM-3.2 TSMC 0.18 um cell library for logic synthesis and the Artisan memory compiler for the memory elements. The simulation shows that this proposed distributed architecture has smaller area throughput (AT) product. To conclude this work, this work modify the particle filter architecture so that we can implement the particle filter and shorten the processing period.