Abstract: In recent years, networked systems such as wireless sensor networks (WSNs) have gained popularity in the research community due to their broad potential military and civilian applications. WSNs are generally composed of a large number of low-quality sensors equipped with limited communication capabilities and limited energy. However, a collective of these nodes can form a powerful network for information gathering and processing. Due to limited communication capacity and also for the sake of energy saving, the number of bits of information to be transmitted between nodes is quite restrictive. Therefore, signals such as sensor measurements are to be severely quantized before their transmissions. This introduces interesting and challenging problems such as what information is to be transmitted and how many bits are needed to represent the information in order to achieve a given performance. Quantized estimation has been extensively investigated in literature such as [4, 6, 9, 11– 13, 16] where various quantization schemes have been addressed. Luo studied the static parameter estimation under severe bandwidth constraints with each sensor’s observation quantized to one or a few bits in . The resulting estimators turn out to exhibit comparable variances with that of the estimator relying on un-quantized observations. Note that one of the major challenges of quantized estimation is that quantization is highly nonlinear and there lack of efficient filtering methods for nonlinear systems. By applying the particle filter to quantized measurements, a quantized version of particle filter is proposed by reconstructing the required probability density . Unfortunately, the filtering performance is poor under a low number of quantization levels. With severe quantization, e.g., binary quantization, a dynamic quantization scheme based on feedback from the estimation center is designed for the state estimation of a hidden Markov model in . The main disadvantage is that the solution involves a rather complicated on-line optimization and does not lead to a recursive filter.