In SpecSense we envision a large-scale RF spectrum monitoring system that will feed into multitudes of spectrum-aware applications forming an entire ecosystem of spectrum data, analytics, and apps. Our goal is to investigate what it takes to develop an end-to-end enabling platform to support this vision. The SpecSense system (i) crowdsources spectrum monitoring using low-cost, low-power custom-designed hardware, and (ii) provides necessary API support for spectrum-aware apps via a central spectrum server/database platform. The project addresses various algorithmic and systems-level challenges for SpecSense.
A SpecSense prototype has already been shown to work well in lab prototypes. Efforts are ongoing to make it more robust and scalable. The following is description of various project activities and outcomes.
Spatial interpolation of RF signal is the key enabler for many applications of SpecSense. The confluence of highly granular data gathering ability in SpecSense and the need for accurately answering the spectrum occupancy queries by applications require revisiting of spatial interpolation techniques. While prior work has indeed considered spatial interpolation of spectrum data, they either used synthetic data or had access to much less granular data than what SpecSense is able to gather. Thus, some of the inefficiencies have not been exposed. Our specific work showcases improved spatial interpolation of RF signals by data collected via distributed sensors. Specifically, we have extended the well-known Ordinary Kriging (OK) interpolation technique by (i) “detrending” the signal by averaging the pathloss exponent, and (ii) partitioning the given spatial region based on pathloss characteristics. For experimental evaluation, we have used real dataset collected using RTLSDR sensor with Nexus 5 and Samsung Galaxy S4 phones, while the third dataset is synthetic.
During SpecSense operation, there are three main events: (i) availability (arrival) of new spectrum sensors, (ii) spectrum occupancy query requests by user apps, (iii) actually sensing/measurement of the spectrum signal by some of the spectrum sensors (when instructed by the system, in response to the queries). The first two steps are user driven, while the third step involves an optimization problem. We have addressed the problem of selecting a minimum number of sensors to report signal at their locations in order to best predict the values at the given query locations. This saves the battery power and backhaul communication costs for the sensors that contribute little to answering spectrum queries. The above optimization problem may be executed at the arrival of each new query, or at regular intervals for the set of queries arriving in the previous interval. In the latter approach, the interval may depend upon certain factors such as the arrival frequency of queries, tolerable prediction delay, etc. We have formulated the above sensor selection problem and developed heuristic algorithms.
For both these problems, our developed techniques provide superior performance over baseline techniques. Specifically, our extension of OK reduces the prediction error by 1/3 to 1/2 relative to regular OK. The sensor selection problem is shown to be NP hard. We have developed multiple greedy heuristics and evaluated their performances. More details on this are available in our Infocom 2017 paper. A somewhat related spatio-temporal spectrum map generation problem is addressed in our work in IEEE WCNC 2020 .
We have explored the spectrum patrolling problem in SpecSense in terms of detecting unauthorized RF transmissions in a collaborative fashion while consuming only a limited amount of resources on the sensors. We have posed this as a collaborative signal detection problem where the individual sensors’ detection performances may vary widely, but are hard to model using traditional approaches. Still, an optimal subset of sensors and their configurations must be chosen to maximize the overall detection performance subject to given resource (cost) limitations. The resource here might be energy of the sensors or backhaul bandwidth.
The broad goal of this work is to develop mechanisms to select the right set of sensors that optimizes the performance of detection task for a given cost. There are two sub-problems that arise: 1) modeling individual sensor performance and cost for given configurations, 2) fusing data from multiple sensors and selecting the optimal subset to maximize detection performance subject to cost limitations (or, minimizing cost subject to a given detection performance). While these problems are not entirely new in a general sense, the specific nature of crowdsourced spectrum patrolling problem makes them challenging.
Our modeling approach for the sensor performance provides significant improvement over state-of-the-art analytically-based ‘whitebox’ models. Using this approach, we have addressed the problem of sensor selection and fusion of heterogeneous sensors deployed over a region of interest to improve intrusion detection performance within a cost budget. We have investigated different scenarios of homogeneous, heterogeneous and reconfigurable sensors. Our sensor selection algorithms perform significantly better than reasonable baseline heuristics and is able to account for correlated sensing. In our work we are able to highlight challenges of the patrolling problem in a cost-effective fashion using crowdsourced sensors and develop mechanisms to address them.
Parts of this work have been published in IEEE Infocom 2018, IEEE Transactions on Cognitive Communications and Networking 2019 and ACM/IEEE IPSN 2020 . A solution approach for localizing multiple intruders from time-skewed observations has appeared in IEEE DySpan 2019. .
In many spectrum patrolling problem it is of interest to localize the (unauthorized) transmitter in addition to detecting it. One needs to do this using minimal resources, i.e., using the minimal number of sensors. In our ongoing work, we have investigated two versions of this problem - 1) Offline sensor selection: We have posed the offline sensor selection for transmitter localization as an optimization problem, where the objective is to maximize the accuracy of localization. We have defined each possible location of the transmitter as a distinct hypothesis, and posed localization as a classification problem. We have developed preliminary algorithms and their performance bounds. 2) Online sensor selection: The sensor selection for transmitter localization can be posed as an online problem as well. In this version, our algorithm picks a sensor for probing, and its output is used to select the next sensors. This allows for more accurate sensing while keeping the cost of running the sensors low, since they are switched on only intermittently. We are working on algorithms and analyzing their performance bounds.
Our general goal is to develop algorithms with low complexity that has a constant factor approximation with the optimal and then also experimental evalutate the algorithms using data from the SpecSense testbed. A part of this work has been published in IEEE Infocom 2020.
We have done a series of benchmarkng exercises of low-cost spectrum sensors used in SpecSense in their ability to detect an unauthorized transmitter that transmits intermittently for very small durations (micro-transmissions). As an example hardware we have used several embedded boards (e.g., Odroid, Raspberry Pi) as the compute host and USRP-B210 and RTL-SDR as the RF sensors. We have run a benchmarking study quantifying the impact of device hardware (e.g., processor, clock) and crucial sensing parameters such as sampling rate, integration size and frequency resolution in detecting such transmissions. We have discovered that the detection performance deteriorates significantly even with the best possible sensing parameter settings with lower performing device hardware, specifically the when transmit durations are very short (in 10s of microseconds or shorter). Our interest here is understanding the limits of performance of the low-cost hardware for spectrum sensing, for specific targeted spectrum applications, and exploit distributed sensing to improve the sensing performance. We have published multiple papers based on these benchmarked performances in IEEE DCOSS 2016, ACM Hotwireless 2016 and PAM 2019.
Field-programmable gate arrays (FPGAs) represent a promising alternative due to their high computation speed and low energy cost. Using the setup we developed in the first year, we have completed a set of systematic measurement-based benchmarking of the energy consumption and latency of FPGA-based spectrum sensing. Our setup consists of a Xilinx ZedBoard FPGA development board connected to a Myriad-RF1 transceiver board. The signal detection unit can be pre-configured to run a detection algorithm based on energy detection, waveform detection, or autocorrelation specifically for DTV signals. For comparison we have chosen Raspberry Pi-based and smartphone-based spectrum sensors based on our earlier work. We focus on two distinct metrics of resource usage —(i) compute latency and (ii) energy consumption. We have perform separate experiments on our Raspberry Pi, smartphone and FPGA-based sensors. We find that the FPGA results in significant latency improvements – for energy-based detection, the FPGA has around 73× lower latency than the Raspberry Pi and 69× lower than the smartphone. Regarding power consumption, we observe that the Zynq FPGA consumes about 8 times less power than the Raspberry Pi. A paper describing this work now has appeared in IEEE DySpan 2018 Symposium.
We are in the process of releasing codes used in the project. Preliminary versions are availble in the following repositories:
The project contributes to the research education of 5 PhD students and several MS students. Two of the PhD students have now graduated and employed as post-docs/research scientists. The scientific results have been disseminated via publications and conference presentations. The lab participates in the lab rotation for Women in Science and Engineering (WISE) program in Stony Brook that targets improving diversity in science and engineering.
Salman posted a demo of SpecSense system on Youtube.10. January 2020
Paper accepted at ACM/IEEE IPSN 20206. December 2019
Paper accepted at IEEE Infocom 202011. November 2019
Two papers presented in IEEE DySPAN22. August 2019
Arani defends his dissertation and graduates19. August 2019
Paper accepted at IEEE Transactions on Cognitive Communications and Networking29. March 2019
Mallesh presents our paper at PAM 2019 in Patagonia, Chile13. December 2018
Paper on detection of intermittent transmission accepted at PAM 2019