Commercial wireless communication microwave links as an opportunistic sensor network for environmental monitoring Hagit Messer messer@eng.tau.ac.il 7 Nov. 2014 Empirical plot (ITU) The Physics: EM attenuation vs. Rain-rate relation AdB aR LEff b R [mm/h Leff [km] a, b= function of (frequency, rain temp., DSD) In microwave frequency bands, a major source of signal interruptions is precipitation (rainfall) The opportunity: Microwave use in Cellular Communications (10-40 GHz) Base Station 1 User 1 Microwave Link (~75% in EU) Base Station 2 User 2 Wired Line, Fiber Optics (~25%) About 75% of European backhaul links are microwave links The Idea: high resolution, low cost meteorological monitoring => Measurements from EXISTING wireless networks can be used for environmental monitoring. The Result: The available measurements (Israel) Israel - area: 20,770 / 22,072 km2 (151st) 8,019 / 8,522 sq mi Data Availability in Israel (2013) – RSL time series from thousands of microwave links New in 2012: 2014 Thanks! Links’ Frequency Distribution Red markers were used for one direction (antenna A to antenna B) and blue markers were used for the second one (antenna B to antenna A). The plot included 3515 pair links totalling in 7030 links (Cellcom). Typical Data: Measured RSL links of different length, geometry and frequency link1,2:ramle-ramot_meir link3,4:Ramle-Azarya -36 -38 18.01.2010 Cellcom RSL 4:48 -37 -38 -38 -39 -39 0:00 -42 -40 -45 -50 -44 -46 Min -55 -48 Max -60 -50 -65 -52 -70 -41 -42 -40 -40 RSL [dBm] 9:36 RSL [dBm] 14:24 -40 RSL [dBm] 19:12 link7,8:Tachlifi-Ramle -37 -40 RSL [dBm] 0:00 link5,6:ramle-neve_nof_lod -39 -41 -42 -44 50 100 150 Time [14:27+..min] 0 -43 -44 -44 -45 -45 50 100 150 Time [14:27+..min] -42 -43 -43 0 -41 0 -46 50 100 150 Time [14:27+..min] 0 50 100 150 Time [14:27+..min] -75 link9,10:ramle-rasko_lod -38 -40 link11,12:Eged_Ramle-Ramle -42 -35 link13:ramle-bayt_arya2 -43 -40 -36 -45 -44 -44 -46 -48 -50 0 50 100 150 Time [14:27+..min] -50 -45 RSL [dBm] RSL [dBm] Features: • Quantization errors • Variable dynamic range • Variable zero level • Non linear pre-processing RSL [dBm] -42 -46 -47 -37 -55 -60 -65 -48 -70 -49 -75 -50 link14:ramle-ramat_dan -35 RSL [dBm] -80 -38 -39 -40 0 50 100 150 Time [14:27+..min] -80 0 20 40 Time [14:27+..min] 60 -41 0 50 100 150 Time [14:27+..min] Commercial Microwave Network (CMN) as an opportunistic sensor network for environmental monitoring Wireless Sensor Network – main characteristics 1. low cost sensors; 2. spatially distributed sensors; 3. autonomous sensors. RSL measurements of Microwave Links from communication systems: 1. low cost sensors - YES 2. spatially distributed sensors - YES 3. autonomous sensors YES Problems: 1. Installation costs 2. Energy source 3. Communication and data processing 4. Maintenance 5. Reliability Our technique DOES NOT suffer from these difficulties Two Parallel Research Directions: o o o o Meteorological tools and Theoretical Signal Processing applications (mostly ad-hoc) Questions (mostly related to sensor networks) Accurate, high tempo-spatial resolution rainfall mapping o Conditions for precipitation reconstructability Flood Prediction o Optimal fusion of different Study of other-than-rain (in quality of measurements, meteorological phenomena sampling rate and spatial (fog, moisture) distribution) sensor networks Precipitation classification and robust estimation (rain, snow, o Multimodality: Optimal use of co/prior information (e.g., sleet) radar) Rainfall measurements Graham Upton Department of Mathematical Sciences, University of Essex Why? MANTISSA: To improve temporal resolution & to get better access to specific areas of interest. US: all the above + to improve tempo-spatial coverage and resolution and to improve accuracy. Instantaneous Mapping – Radar vs Links (synthetic example) Estimation of rainfall fields using commercial microwave communication networks of variable density Zinevich, A; Alpert, P; Messer, H ADVANCES IN WATER RESOURCES Volume: 31 Issue: 11 Pages: 1470-1480, 2008 High Resolution Instantaneous Mapping – Radar vs Links (real data) Rain Rate Estimation Using Measurements From Commercial Telecommunications Links Goldshtein, O; Messer, H; Zinevich, A IEEE TRANSACTIONS ON SIGNAL PROCESSING Volume: 57 Issue: 4 Pages: 1616-1625, 2009 Dynamic reconstruction of rainfall – results r x, y, t t υr x, y, t 0 Frontal Rainfall Observation by a Commercial Microwave Communication Network Zinevich, A; Messer, H; Alpert, P JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY Volume: 48 Issue: 7 Pages: 1317-1334, 2009 Other mapping techniques Space–time dynamics of 15-min rainfall depths (two panels per time step) from links (Left) and radars plus gauges (Right) for September 10, 2011, 2030–2045 and 2230–2245 hours UTC (validation). Overeem A et al. PNAS 2013;110:2741-2745 ©2013 by National Academy of Sciences ACCURATE RECONSTRUCTION OF RAIN FIELD MAPS FROM COMMERCIAL MICROWAVE NETWORKS USING SPARSE FIELD MODELING Yoav Liberman, Hagit Messer Tel Aviv University School Of Electrical-Engineering Tel Aviv, Israel Still open problems with rain fields reconstruction o Coverage o Accuracy (accumulated vs. “instantaneous”): min/max, quantization o Multimodality o Variability of rain rate with height o Challenging areas (e.g. urban, slopes) o ….. Multimodality – preliminary results Dr. Rana Samuels Other than rain phenomena The SP challenge: Attenuation is caused by many different environmental factors o The opportunity: DIVESITY! Many spatially and temporally distributed sensors of different sensitivity (ML length), frequency. o Some interesting results o Rain + other phenomena: ML signal attenuation model 𝐴 𝑛, 𝐿 = 𝐴𝑝 𝑛, 𝐿 + 𝐴𝑤 𝑛 + 𝐴𝑣 𝑛, 𝐿 + 𝐴0 + 𝑁 𝑛 + 𝑞[𝑛] [𝑑𝐵] Detect and classify these atmosphere phenomena ”Tree Type” Classification Algorithm Rain Other methods RSL Statistical Test: MFLRT Precipitation Sleet Snow Other atmosphere phenomena Statistical Test: GLRT Moist antenna or fog Water vapor “Dew” Fog Characterization of link attenuation Example: Fog detection Predominant Fog and Visibility Monitoring Techniques Satellite Systems Visibility Meters Transmissometers Human Observer Fog Induced attenuation in the microwave region: F - Fog induced attenuation (DB/Km) LWC- Fog Liquid Water Content (gr/m3) - Dependent on the temperature- T (K) and link frequency- f (GHz) Units: ((dB/km)/(gr/m3) Source: ITU-R P.840-6 (2013): Attenuation due to clouds and fog N David, P Alpert, H Messer, The potential of commercial microwave networks to monitor dense fog‐feasibility study Journal of Geophysical Research: Atmospheres 118 (20), 11,750-11,761, Oct. 2013 Fog induced attenuation up to 100 GHz 2D Simulation test • Goal: calculating the minimum detectable LWC using measurements from multiple MLs at 20, 38, 80 GHz. • How ? • A fog modeled patch was swept across an area of the whole map we wished to create. • 696 microwave links (existing deployment) • During each of the three simulations (20, 38, 80 GHz) the algorithm was applied over an area of about 14,000 square km Sensitivity tests at 20, 38 and 80 GHz (simulations) Gultepe, I., Müller, M. D., and Boybeyi, Z.: A new visibility parameterization for warm-fog applications in numerical weather prediction models, J. Appl. Meteor. Climatol., 45, 1469–1480, 2006. Publications: review papers O Sendik, H. Messer, A New Paradigm for Precipitation Monitoring, IEEE Signal Processing Magazine, in press (to be published on May 2015) o Messer, H. ; Zinevich, A. ; Alpert, P., Environmental sensor networks using existing wireless communication systems for rainfall and wind velocity measurements, IEEE Instrumentation & Measurement Magazine, Volume: 15, Issue: 2 , pp 32 – 38, April 2012 o Messer, H.; Goldshtein, O.; Rayitsfeld, A.; Alpert, P.; Recent results of rainfall mapping from cellular network measurements, Acoustics, Speech and Signal Processing, 2008. ICASSP 2008. IEEE International Conference on March 31 2008-April 4 2008 Page(s):5157 – 5160 o Messer, H.; Rainfall Monitoring Using Cellular Networks, Signal Processing Magazine, IEEE, Volume 24, Issue 3, 2007 Page(s):144 – 142 o Messer, H; Zinevich, A; Alpert, P: Environmental monitoring by wireless communication networks, SCIENCE, 312 (5774): 713-713 MAY 5 2006 o Publications: Rainfall Mapping o o o o o o Y. Liberman, Object Tracking Extensions for Accurate Recovery of Rainfall Maps Using Microwave Sensor Network, EUSIPCO 2014 Y. Liberman, R. Samuels, P. Alpert, and H. Messer, New algorithm for integration between wireless microwave sensor network and radar for improved rainfall measurement and mapping, Atmos. Meas. Tech., 2014 Yoav Liberman, Hagit Messer H,, Accurate Reconstruction of Rain Field Maps from Commercial Microwave Networks Using Sparse Field Modeling, ICASSP 2014 Zinevich, A; Messer, H; Alpert, P: Frontal Rainfall Observation by a Commercial Microwave Communication Network, JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY Volume: 48 Issue: 7 Pages: 1317- 1334, 2009 Goldshtein, O; Messer, H; Zinevich, A: Rain Rate Estimation Using Measurements From Commercial Telecommunications Links, IEEE TRANSACTIONS ON SIGNAL PROCESSING Volume: 57 Issue: 4 Pages: 1616-1625, 2009 Zinevich, A; Alpert, P; Messer, H: Estimation of rainfall fields using commercial microwave communication networks of variable density, ADVANCES IN WATER RESOURCES Volume: 31 Issue: 11 Pages: 1470-1480, 2008 Publications: Accuracy assessment o o Asaf Rayitsfeld, Rana Samuels, Artem Zinevich, Uri Hadar, Pinhas Alpert: Comparison of two methodologies for long term rainfall monitoring using a commercial microwave communication system, Atmospheric Research. Volumes 104–105, February 2012, Pages 119–127 Zinevich, A; Messer, H; Alpert, P: Prediction of rainfall intensity measurement errors using commercial microwave communication links , Atmospheric Measurement Techniques Atmos. Meas. Tech., 3, Pages: 1385-1402, 2010 Publications: Other Meteorological Applications o o o o Ostrometzky J. , Cherkasky, D. and Messer, Accumulated Mixed Precipitation Estimation Using Measurements From Multiple Microwave Links, Advances in Meteorology, Sept. 2014 Cherkasky, D. Ostrometzky J. and Messer, Precipitation Classification Using Measurements from Commercial Microwave Links , IEEE Transactions on Geoscience and Remote Sensing, Volume: 52 , Issue: 5, May 2014 N David, P Alpert, H Messer, The potential of commercial microwave networks to monitor dense fog - feasibility study, Journal of Geophysical Research: Atmospheres 118 (20), 11,750-11,761, Oct. 2013 David, N; Alpert, P; Messer, H, The potential of cellular network infrastructures for sudden rainfall monitoring in dry climate regions, Atmospheric Research Volume 131, Pages 13–21, September 2013 More publications: Other Meteorological Applications o o o David, N; Alpert, P; Messer, H: Humidity Measurements using Commercial Microwave Links In Advanced Trends in Wireless Communications, Editor: Mutamed Khatib. InTech, February 2011 David N.; Alpert P.; Messer H: Flash floods warning technique based on wireless communication networks data, Geophysical Research Abstracts Vol. 12, EGU2010-1169-3, 2010 David, N; Alpert, P; Messer, H: Technical Note: Novel method for water vapour monitoring using wireless communication networks measurements, ATMOSPHERIC CHEMISTRY AND PHYSICS Volume: 9 Issue: 7 Pages: 2413-2418, 2009 Publications: Signal Processing o o o o o J. Ostrometzky, H. Messer, Accumulated Rainfall Estimation Using Maximum Attenuation of Microwave Radio Signal, SAM 2014 Y Broyde, M Livshitz, H Messer, ADAPTIVE STATISTICAL LEARNING OF CELLULAR USERS BEHAVIOR, Signal Processing, Volume 93, Issue 11, Pages 3151–3158, November 2013 O Harel, H Messer, Extension of the MFLRT to Detect an Unknown Deterministic Signal Using Multiple Sensors, Applied for Precipitation Detection, Signal Processing Letters, IEEE Volume: 20 , Issue: 10 , Oct. 2013 Sendik, O; Messer, H; On the reconstructability of images sampled by random line projections, IEEE 27th Convention, ISRAEL 2012 Dani Cherkassky, Jonatan Ostrometzky and Hagit Messer: The Use of Linear Feature Projection for Precipitation Classification Using Measurements from Commercial Microwave Links, in Latent Variable Analysis and Signal Separation Lecture Notes in Computer Science, 2012, Volume 7191/2012, pp 511-519 Publications: Signal Processing o o Elad Heiman and Hagit Messer, Parameter Estimation from Multiple Sensors with Mixed Resolution of Quantization, Electrical & Electronics Engineers in Israel (IEEEI), 2014 IEEE 28th Convention of Omry Sendik and Hagit Messer, On the achievable coverage of rain field mapping using measurements from a given set of microwave links, Electrical & Electronics Engineers in Israel (IEEEI), 2014 IEEE 28th Convention of International Impact Africa! Doumounia, Ali, et al. "Rainfall monitoring based on microwave links from cellular telecommunication networks: First results from a West African test bed ". sretteL hcraeseR lacisyhpoeG (2014) http://raincell01.sciencesconf.org/ Rain Cell Africa first Workshop 11-14 Nov 2014 Ouagadougou, Burkina Faso Examples of Signal Processing Challenges o Classification of atmospheric phenomena o Parameter estimation from sensor networks of different quantization levels o Optimal rain rate estimation from Min/Max measurements Test statistic : MFLRT max 0≤𝐿<𝑁 𝑙𝑔𝑙 𝑋 − 𝑙 𝑙𝑛 𝑙𝑔𝑙 𝑋 = 𝑙 𝑛 𝑙𝑔𝑙 𝑋 𝑙 +1 𝑢 𝑙𝑔𝑙 𝑋 𝑙 −1 𝐻𝑙 > <𝛾 𝐻0 𝑃 𝑋 ; 𝜃𝑙 , ℋ𝑙 𝑃 𝑋 ; 𝜃0 , ℋ0 The motivation for this test is that it asymptotically minimizes the Kullback-Leibler distance (divergence) between the true probability density function (pdf) and the estimated one. The Model : 𝑥𝑖 [𝑛] = 𝐿𝑖 𝐴𝑙 [𝑛] + 𝜇𝑖 + 𝑤𝑖 [𝑛] 𝐴𝑙 𝑛 - Unknown deterministic signal of unknown waveform ( 𝑙 – sparse signal ) Extension of the MFLRT to Detect an Unknown Deterministic Signal Using Multiple Sensors, Applied for Precipitation Detection O Harel, H Messer Signal Processing Letters, IEEE Volume:20 , Issue: 10 , Oct. 2013 Precipitation classification Experimental setup Where Link Frequency [GHz] Distance [KM] Antenna Height [m] Height from sea level [m] Nachla Sde-Moshe Sde-Moshe – K.Gat Glikson 18.958/17.948 6 22/20 126/154 18.820/17.810 4 20/25 154 Precipitation Detection - MFLRT VS GLRT we can see that while the GLRT can serve for detection of precipitation in a given time period, it is unable to estimate the signal order 𝑙 because, as mentioned before, the GLRT always chooses the highest order of the signal (i.e. assigns precipitation to all samples). Waveform estimation The usage of the MFLRT overcomes this limitation, and this detector, in addition, can operate as an estimator for wet/dry samples Precipitation coverage estimation Problem definition The MFLRT can also be used for deriving the precipitation coverage • We define the ”precipitation coverage” problem as a problem of identifying, in a given network, what are the MLs that are affected by precipitation at a particular time. Frequency diversity • The MFLRT decides, in a given network (region), whether or not precipitation is present at a particular time, and in addition, which are the MLs that are affected by precipitation (precipitation coverage problem) Experimental setup Where Results Examples of Signal Processing Challenges o Classification of atmospheric phenomena o Parameter estimation from sensor networks of different quantization levels o Optimal rain rate estimation from Min/Max measurements Our proposed approach – assume a parametric model 𝑓 𝑥, 𝑦; 𝜽 𝑧𝑖 = 𝑄𝑖 (𝑠𝑖 (𝑓 𝑥, 𝑦; 𝜽 ) + 𝑛𝑖 ) 𝒛 = 𝑧1 , 𝑧2 , . . . , 𝑧𝑁 𝑠1 (𝑓 𝑥, 𝑦; 𝜽 ) 𝑠2 (𝑓 𝑥, 𝑦; 𝜽 ) 𝑇 𝑠3 (𝑓 𝑥, 𝑦; 𝜽 ) 53 𝑠4 (𝑓 𝑥, 𝑦; 𝜽 ) 27 July 2016 𝜽? 𝑧𝑖 = 𝑄𝑖 (𝑠𝑖 𝑓 𝑥, 𝑦; 𝜽 + 𝑛𝑖 ) θml = arg max 𝐿(𝒛; 𝜽) 𝜽 𝑁 𝐿 𝒛; 𝜽 = log 𝑃(𝒛; 𝜽) = log(𝑃 𝑧𝑖 ; 𝜽 ) 𝑖=1 𝑚𝑖 = 𝑠𝑖 𝑓 𝑥, 𝑦; 𝜽 Δi Δi 𝑃 𝑧𝑖 ; 𝜽 = 𝑃(𝑚𝑖 ∈ 𝑧𝑖 − , 𝑧𝑖 + ) 2 2 + ni ) 𝑁 θml = arg max 𝜽 log( 𝑖=1 ∆ 𝑧𝑖 + 2𝑖 ∆ 𝑧𝑖 − 2𝑖 54 − 𝑚−𝑠𝑖 𝑓 𝑥,𝑦;𝜽 𝑒 2𝜎𝑖 2 2 𝑑𝑚) 27 July 2016 Real data for rain mapping 1 − 2 1−𝜌2 𝑓 𝑥, 𝑦; 𝜽 =R ∙ e ∗[ 2 2𝜌 x−μx y−μy x−μx 2 y−μy + − 2 2 𝜎𝑥 𝜎𝑦 σx σy 55 ] 27 July 2016 Real data 18/01/2010, 10 events every 5 min 56 27 July 2016 13:30 13:35 57 27 July 2016 Pelephone only Pelephone+Cellcom 𝛒 SD 𝛒 SD 13:20 0.4408 3.501 0.5238 3.58 NML 0.4421 3.54 0.5268 3.54 ML Pelephone only 𝝆𝒄𝒐𝒓𝒓 SD Pelephone+Cellcom 𝝆𝒄𝒐𝒓𝒓 SD 13:40 0.65 2.64 0.7 1.67 NML 0.66 3.31 0.75 1.43 ML Pelephone only 𝝆𝒄𝒐𝒓𝒓 SD Pelephone+Cellcom 𝝆𝒄𝒐𝒓𝒓 SD 14:00 0.452 1.5 0.512 1.4 NML 0.452 1.5 0.512 1.4 ML Examples of Signal Processing Challenges o Classification of atmospheric phenomena o Parameter estimation from sensor networks of different quantization levels o Optimal rain rate estimation from Min/Max measurements 60 The Rain Rate Statistics ; Best for heavy storms (>50[mm/hr]). ; Best for light to moderate storms. Example of a Log-Normal (-0.5,1.4) random process Rain-gauge measurements Tel Aviv, December 2013 The IMS rain-gauges cannot detect rainrate intensity lower than 0.6 [mm/hr]! 61 Generalized Extreme Value Distribution Gamma maxima asymptotically converge into Type I Log-Normal maxima asymptotically converge into Type II 62 ACCUMULATED RAIN ESTIMATION FROM MIN/MAX RSL DATA SERIES: DIRECT MLE FROM THE EXTREME STATISTICS 63 The Estimation Process yi Max RainRate Calculation Maximum Likelihood Estimation Trans. to rain-rate statistics Accumulated rainfall calculation • The available set of measurements is the minimum and maximum RSL data series: 64 The Estimation Process yi Max RainRate Calculation Maximum Likelihood Estimation Trans. to rain-rate statistics Accumulated rainfall calculation • The minimum and maximum rain-rates are calculated from the maximum and minimum RSL’s: • A(tn) is the minimum or maximum attenuation, and is being calculated directly from the maximum and minimum RSL data series. 65 The Estimation Process yi Min/Max Rain-Rate Calculation zi Maximum Likelihood Estimation Trans. to rain-rate statistics Accumulated rainfall calculation • The set of min/max rain rates {zi} continue to the maximum likelihood estimation, which finds the GEV parameters: 66 The Estimation Process yi Min/Max Rain-Rate Calculation zi Maximum Likelihood Estimation ^ ^^ ε,μ,σ Trans. to rain-rate statistics Accumulated rainfall calculation • Then, the GEV parameters are transformed into the rain- rate Probability Distribution Function parameters: 67 The Estimation Process yi Max RainRate Calculation zi Maximum Likelihood Estimation ^ ^^ ε,μ,σ Trans. to rain-rate statistics ^^ κ,Θ Accumulated rainfall calculation • From The rain-rate PDF parameters, the accumulated rainfall can be found simply by using the known expected value formula: ; Gamma PDF expected value ; Log-Normal PDF expected value 68 The Experimental Setup: 69 Results 70 Comparison with other method The Research Team (2013) o Prof. Hagit Messer-Yaron – o Prof. Pinhas Alpert – meteorology signal processing Graduate Students (2013): o Yoni Ostromtzky Ori Auslender Yoav Liberman Elad Heiman Oz Harel o o o o Noam David o Ronen Radian Post-doc: Dr. Rana Samuels o Graduated: Dr. Atrem Zinevich (Ph.D) Oren Goldstein, Asaf Rayitsfeld, Uri Hadar, Yoni Broyde, Lior Palti, Dani Cherkassky, Yoni Ostromtzky, Omry Sendik (M.Sc.) Dionis Teshler (B.Sc.), (Anat Neumann) Acknowledgments This project could not be done without the collaboration of Cellcom & Pelephone, and Partner who provided the data. o o o In Pelephone, we would like to thank: A. Shilo, N. Dvela, A. Hival and Y. Shachar. In Cellcom, we would like to thank: Y. Eisenberg, Y. Dagan, Y. Koriat, I. Inbar. Shahar Shilian, Eli Levi In Partner we would like to thank H. Ben Shabat, A Shor, H Mushvilli and Y. Bar Asher.