HEIR: Heterogeneous interference recognition for wireless sensor networks

Wireless Technology Recognition Based on RSSI Distribution at Sub-Nyquist Sampling Rate for Constrained Devices

“…For instance, the presence of a microwave oven is identified when the RSSI of the chip CC2420 drops below the noise floor, which is likely caused by saturation in the receiver chain. The intrinsic periodicity extracted by a simplified version of the spectral correlation function proposed by [ 30 ],and the characteristics of the spectrum trace obtained by a commercial Wi-Fi card proposed by [ 31 ] are used in combination with features of RSSI series (such as idle and transmission time) to distinguish technologies in the 2.4-GHz ISM band. Although each of these works has its own strength in certain application scenarios, the common interdependency on features from the time domain of RSSI series, such as packet duration and inter-packet gap, implies that they lack the capability to distinguish streaming technologies, such as LTE and DVB-T.…”

“…Given that prior works [ 27 , 29 , 30 , 31 ] were dedicated to recognizing ISM band (non-streaming) technologies using time domain features, the remainder of this work aims to automate technology recognition using a different feature space established from the RSSI distribution, to identify a different set of technologies, including both non-streaming technologies (represented by Wi-Fi) and streaming technologies beyond the ISM bands (i.e., DVB-T, LTE). The usage of the feature space is illustrated via a sample algorithm in Section 5.1 and evaluated in Section 5.2 ; then, the algorithm and feature space are extended for recognition of the mixed LTE-U and Wi-Fi signal in Section 5.3 .…”

“…The analysis for variable N is important, as developers on constrained devices usually do not have direct access to the raw IQ samples, so RSSI is provided by accessing a register or other types of interface towards hardware modules, which have a limited access rate. For instance, the TelosB sensor platform allows the RSSI register to be read at the rate of 32.15 kHz after customization [ 29 ], while the solution in [ 30 ] only refreshes RSSI at 11.3 kHz. Note that this is the rate at which the RSSI register can be accessed, it is not the sample rate used by the ADC on the RF front-end.…”

“…These limitations indicate the importance of sufficient diversities in the training data and eventually the need for more dynamic algorithms. The analysis of the true positive rate with respect to N and T unveils the conditions for the features to stay reliable, which is when the average interval of RSSI is kept below 200 s and the observation time is above 50 ms. Recall that the sensor platforms used by [ 29 , 30 ] allow RSSI to be collected at 11.3 kHz and 31.25 kHz, respectively. The 200 s ( kHz) requirement can clearly be supported on these devices.…”

“…The work in [ 29 ] states a minimum observation time of 2.9 ms, which suffices for the purpose of identifying only the ZigBee signal (the minimum packet interval of ZigBee is 2.8 ms). Solutions aiming for classification of multiple technologies generally rely on longer observation periods, ranging from 100 ms [ 31 ] to several seconds [ 30 ], or multiple days and months if the solution is intended for non-real-time applications [ 18 ]. Therefore, the period of 50 ms is considerably shorter than most existing solutions and hence suitable for making real-time decisions in the context of dynamic spectrum sharing.…”