Propagation Channel Interpolation for Fingerprint-Based Localization of Illegal Radios

A Guide of Fingerprint Based Radio Emitter Localization Using Multiple Sensors

“…In the online phase, the training system was placed at 10 random locations in the targeted area at 1297MHz frequency, 1MHz bandwidth and 0.5W transmit power in order to mimic the operation of unknown illegal radios. More details can be found in [35]. Figure 22 shows a comparison of the CDF of distance error between several localization algorithms.…”

“…Here, the data across the spatial domain is assumed to be generated by a Gaussian process, and regression is performed by learning the spatial covariance from measured data. Another example is regression along the frequency domain, where a log-linear model with respect to frequency is employed to predict fingerprints at any frequency [35]. In these given examples, the Gaussian distribution function and the log-linear function can be regarded as kernel functions.…”

“…Thirdly, to improve the pattern matching results, we require regression and interpolation of the training fingerprints in multiple domains to match the parameters of the illegal radio. In this section, we explain techniques which can overcome the problems above, which were previously introduced [47] [48], including a hybrid algorithm to combine the strengths of multiple fingerprints [35]. Simulation results in a dense urban environment are presented, followed by preliminary measurements in an open field conducted using the developed hardware prototype.…”

“…A maximum likelihood approach was employed to derive a hybrid algorithm which combines the strengths of both fingerprints [35]. Since the Xcorr and Phasediff fingerprints are statistically independent, localization can be performed by maximizing their joint log-likelihood as the following equation:…”

“…The offline collection of fingerprints can be considered as a supervised learning process to develop the fingerprint database with indices of known locations. Since it is impossible to measure the fingerprints in all the space and frequency, statistical regression algorithms interpolate the data sets by using the property of propagation channel in space and frequency [21]. In the phase of pattern matching, statistical tracking algorithms such as the Kalman filter, Particle filter, and Recurrent Neural Network (RNN) will enhance the performance by exploiting prior knowledge of users [22].…”