GRETSI 2009 Signal Processing and Communications for Sensor Networks Martin Vetterli, EPFL and UC Berkeley joint work with T. Ajdler, G. Barrenetxea, H. Dubois-Ferriere, F.Ingelrest, M. Kolundzija, R. Konsbruck, Y. Lu, O. Roy, T. Schmid, L. Sbaiz, E.Telatar, M.Parlange (EPFL), P.L.Dragotti (Imperial), M.Gastpar (UCBerkeley) Support: Swiss NSF National Center on Mobile Information and Communication Systems http://www.mics.org Audiovisual Communications Laboratory Acknowledgements • To the organizers! • Swiss and US NSF, our good friends and sponsors • The National Competence Center on Research ‘’Mobile Information and Communication Systems’’ (MICS) • The SwissExperiment, a large scale environmental monitoring effort in the Swiss Alps • K.Ramchandran and his group at UC Berkeley, for sharing pioneering work on distributed source coding • Colleagues at EPFL and ETHZ involved in MICS - M.Grossglauser, for making things move - E.Telatar, for wisdom and figures! - J.Bovay, for NCCR matters Spring 2009 - 2 Outline 1. Introduction Wireless sensor networks From “one to one” to “many to many”, or from Shannon to now! 2. The structure of distributed signals and their sampling Sensor networks as sampling devices of the real world Distributed image processing: The plenoptic function Spatial sound processing: The plenacoustic function Acquiring the diffusion equation: trade spatial for temporal super-resolution 3. Distributed source coding Source coding, Slepian-Wolf and Wyner-Ziv Distributed R(D) for sounds fields 4. On the interaction of source and channel coding To separate or not to separate... That is the question! The world is analog, why go digital? Gaussian sensor networks 5. Environmental monitoring Environmental monitoring for scientific purposes and sensor tomography SensorScope: Real life environmental monitoringin the Swiss Alps 6. Conclusions Spring 2009 - 3 From centralized to “self-organized” • Classic solutions (e.g. GSM, UMTS): characterized by heavy fixed infrastructures • Evolution of wireless communication equipment: computational power , size , price , ~ transmit power • 110 Billion US$ for UMTS licenses: is there another way? Ad-hoc networking solution: - multihop, collaborative - reinvented many times - self-organization cute but tricky ; ) Current practice -> hybrid solution: multihop access to backbone -> Sensor Networks Spring 2009 - 4 The Change of Paradigm Old view: one source, one channel, one receiver (Shannon 1948) Source Channel Receiver Current view: distributed sources, many sensors/sources, distributed communication medium, many receivers! sources channels receivers Note: still many questions open! Spring 2009 - 5 Wireless Sensor Networks as Signal Processing Devices Signals exist everywhere...they just need to be sensed! – distributed signal acquisition: many cameras, microphones etc – these signals are not independent: more sensors, more correlation – there can be some substantial structure in the data, due to the physics of the processes involved Computation is cheap – local computation – complex algorithms to retrieve data are possible Communication is everywhere – this is the archetypical multi-terminal challenge – mobile ad hoc networks, dense, self-organized sensor networks are built – the cost of mobile communications is still the main constraint Cross-disciplinarity – fundamental bounds (what can be sensed?) – algorithms (what is feasible?) – systems (what and how to build?) Spring 2009 - 6 The swiss version of homeland security :) Distributed sensor network for avalanche monitoring: Method: drop sensors, self-organized triangulation, monitoring of location/distance changes, download when critical situation Challenges: extreme low power, high precision, asleep most of the time, when waking up, quick download ... and all self-organized! Legacy technology: build a chalet, see if it stands after 50 years! Spring 2009 - 7 Outline 1. Introduction 2. The structure of distributed signals and their sampling Sensor networks as sampling devices of the real world PDEs are the name of the game Temporal sampling is easy…. Spatial sampling without filtering! Distributed image processing: The plenoptic function Handle with care: not bandlimited! Spatial sound processing: The plenacoustic function Non-separable, but essentially bandlimited Sampling theorem, interpolation, and applications Acquiring the diffusion equation Trade spatial for temporal super-resolution Applications 3. Distributed source coding 4. On the interaction of source and channel coding 5. Environmental monitoring 6. Conclusions Spring 2009 - 8 2. The Structure of Distributed Signals and Sampling A sensor network is a distributed sampling device Physical phenomena – distributed signals are governed by laws of physics – partial differential equation at work: heat and wave equation… – spatio-temporal distribution: evolution over time and space Sampling – regular/irregular, density – in time: easy – in space: no filtering before sampling – spatial aliasing is key phenomena Note: here we assume that we are interested by the ‘’true’’ phenomena, decision/control: can be different! Spring 2009 - 9 2. The Structure of Distributed Signals and Sampling Analog signals: Different dimensions have physical meanings (e.g. space and time). The analog signals are governed by certain physical law. sound field: wave equation pollution plume: diffusion equation plenoptic field: ray model (far field); wave equation (near field) Spring 2009 - 10 2. Sampling versus sampling physics (1/4) 1. Classic sampling – Key: Spaces V, W, bijection f(t) fk 2. Spatio-temporal sampling t – Temporal filtering easy – No spatial filtering possible! x Spring 2009 - 11 2. Sampling versus sampling physics (2/4) 3. Sampling physics – Space V can be partly parametric (e.g. point source, FRI, sparse) – PDE is given by the physics of the problem – No spatial filtering in (t,x) but PDE does spatial filtering for us! Goals – From samples find field – From samples find sources Spring 2009 - 12 2. Challenges of sampling physics (3/4) Sampling physical fields given by PDEs and driven by sources Good news: • PDEs are known, and well understood • PDE often regularize the problem (e.g. spatial smoothing) • Some sources are in subspaces Challenges: • • • • • • • Inhomogeneous dimensions: t and x are indeed different Cost of sampling in x much higher than in t Multidimensional sampling, possible non-separable Regular sampling in time, regular/irregular in space Sources are in manifolds Aliasing and undersampling, especially in space, are a real problem Some events are not bandlimited, and will never be Spring 2009 - 13 2. Challenges of sampling physics (4/4) Key physical phenomenas: The wave equation: • In far field: ray tracing is a good approximation The diffusion or heat equation: Navier-Stokes (turbulence): • When averaged: diffusion or heat equation Random walks: • When averaged: diffusion or heat equation Spring 2009 - 14 2. Sampling the real world We consider 3 ‘’real’’ cases, and follow: – – – – what is the physical phenomena what can be said on the ‘’discretization” in time and space is there a sampling theorem what is the structure of the sampled signal 1. Light fields – wave equation for near field – ray tracing for far field – plenoptic function and its sampling 2. Sound fields – wave equation for sounds – plenacoustic function and its sampling 3. Diffusion fields – heat equation – diffusion equation and sampling Spring 2009 - 15 2.1 The Plenoptic Function [Adelson91] Multiple camera systems – physical world (e.g. landscape, room) – distributed signal acquisition – possible images: plenoptic function, 7-dim! Background: – pinhole camera & epipolar geometry – multidimensional sampling Implications on communications – camera sources are correlated in a particular way – limits on number on ‘’independent’’ cameras – different BW requirements at different locations Spring 2009 - 16 Examples 3D 3D 2D 4D [Stanford multi-camera array] 5D [Imperial College multi-camera array] Spring 2009 - 17 2.2 The Plenacoustic Function [AjdlerSV:06] Multiple microphones/loudspeakers – physical world (e.g. free field, room) – distributed signal acquisition of sound with “many” microphones – sound rendering with many loudspeakers (wavefield synthesis) This is for real! – – – – sound recording special effects movie theaters (wavefield synthesis) MP3 surround etc Wave equation: – Source: BL in time, sparse in space – PDE: essentially BL in (time,space) MIT1020 mics Spring 2009 - 18 Plenacoustic function and its sampling Setup Questions: – Sample with “few” microphones and hear any location? – Solve the wave equation? In general, it is much simpler to sample the plenacoustic function – Dual question also of interest for synthesis (moving sources) – Implication on acoustic localization problems – Application for acoustic echo cancellation Spring 2009 - 19 Examples: PAF in free field and in a room for a given point source • • • We plot: p(x,t), that is, the spatio-temporal impulse response The key question for sampling is: , that is, the Fourier transform A precise characterization of for large and will allow sampling and reconstruction error analysis Spring 2009 - 20 Plenacoustic function in Fourier domain (approx.): :: temporal frequency : spatial frequency Sampled Version: Thus: Spatio-temporal soundfield can be reconstructed up to 0 Spring 2009 - 21 Computed and Measured Plenacoustic Functions • • Almost bandlimited! Measurement includes noise and temperature fluctuations Spring 2009 - 22 A sampling theorem for the plenacoustic function Theorem [ASV:06]: • • • Assume a max temporal frequency Pick a spatial sampling frequency Spatio-temporal signal interpolated from samples taken at Argument: • • • Take a cut through PAF Use exp. decay away from central triangle to bound aliasing Improvement using quincunx lattice Spring 2009 - 23 Plenacoustic function: Application Application to wavefield synthesis [M. Kolundzija:09]: • • Sound field reconstruction Wide space equalization Spring 2009 - 24 Some generalizations: The EM case Electromagnetic waves and UWB • Wave equation • 3 to 6 GHz temp. frequency • And a triangle! Spring 2009 - 25 The heat equation and diffusion processes The diffusion-advection process (Fick’s law): where a,b: wind, s: unknown source Model for: temperature, chemical plumes, smoke from forest fires, radioactive materials ... Example: heat diffusion in time and frequency Spring 2009 - 26 A sampling theory for diffusion processes [Y.Lu:08-09] Model: diffusion of unknown instantaneous sources (e.g. sudden release of pollutants) Goal: sample the field using a sensor network, and estimate and . Assumptions: is a Poisson process, with average time is (approximately) bandlimited, with bandwidth Problem Statement: What is the minimum total sampling density? (At least ) What is the trade-off between spatial and temporal sampling rates? Spring 2009 - 27 A sampling theory for diffusion processes [Y.Lu:08-09] Spring 2009 - 28 The heat equation and diffusion processes Theorem: Sampling a homogeneous diffusion process with Nyquist density fs: temporal samples achievable unachievable condition number spatial density We have to place the sensors at the right locations! Spring 2009 - 29 On sampling and representation of distributed signals We saw a few examples: – Plenoptic function and light fields – Plenacoustic function and sound fields – Heat equation and diffusion processes It is a general phenomena – Random walks and the heat equation – Electromagnetic fields and wave equation – Diffusion processes and averages of turbulence This has implications on: – Sampling: where, how many sensors, how much information is to be sensed – Gap between simple (separate) and joint coding – Spatio-temporal waterpouring Spring 2009 - 30 Outline 1. Introduction 2. The structure of distributed signals and sampling 3. Distributed source coding Introduction Source coding, sampling, and Slepian-Wolf Distributed rate-distortion function for acoustic fields 4. On the interaction of source and channel coding 5. Environmental monitoring 6. Conclusions Spring 2009 - 31 Correlated source coding and transmission Dense sources = correlated sources – physical world (e.g. landscape, room) – degrees of freedom ‘’limited’’ – denser sampling: sources are more correlated Background: – Slepian- Wolf (lossless correlated source coding with binning) – Wyner-Ziv (lossy source coding with side information) Implications on communications – such results are starting to be used... – many open problems (e.g. general lossy case is still an open problem...) – separation might not be the way... are there limiting results? Below, specific results: – Distributed rate-distortion for acoustic fields based on plenacoustic function – Also: Distributed compression: a distributed Karhunen-Loeve transform Optimal data gathering using Slepian-Wolf Spring 2009 - 32 Slepian-Wolf (1973…) Given – X, Y i.i.d with p(x,y) X Then: encode separately, decode jointly, without coders communicating Achievable rate region – R1 ¸ H(X/Y) – R2 ¸ H(Y/X) – R1 + R2 ¸ H(X,Y) R Y R2 H(Y) H(Y/X) H(X/Y) H(X) R1 • For many sources…. rather complex (binning) • Lossy case: mostly open! • Example of result: SW based data gathering [CristescuBV:03] Spring 2009 - 33 The plenacoustic function as a model, Konsbruck (1/4) Stationary spatio-temporal source on a line, measured by a microphone array Greens’ function – Fourier Transform essentially supported on a triangle! Spring 2009 - 34 The plenacoustic function as a model (2/4) Quincunx sampling lattice Spring 2009 - 35 The plenacoustic function as a model (3/4) Quincunx sampling lattice Key insight: discrete spatio-temporal process is white! Spring 2009 - 36 The plenacoustic function as a model (4/4) Distributed rate-distortion functions for white sound field – Centralized – Quincunx sampling based – Rectangular sampling based – Thus: the distributed R(D) is determined for this case! – For white source, some loss Spring 2009 - 37 On distributed source coding… Three cases studied: – Data gathering with Slepian-Wolf (Cristescu et al) – Distributed versions of the KLT (Gastpar, Dragotti et al) – Distributed rate-distortion for acoustic fields (above) These are difficult problems.... – lossy distributed compression partly open – high rate case: Quantization + Slepian-Wolf – low rate case: mostly open In many case – Strong interaction of “source” and ‘’channel’’ – Large gains possible but we are only seeing the beginning of fully taking advantage of the sources structures and the communication medium... The leads us to revisit the separation principle! Spring 2009 - 38 Outline 1. Introduction 2. The structure of distributed signals and sampling 3. Distributed source coding 4. On the interaction of source and channel coding To separate or not to separate... The world is analog, why go digital? To code or not to code... Gaussian sensor networks 5. Environmental monitoring 6. Conclusions Spring 2009 - 39 4. On the interaction of source and channel coding Going digital is tightly linked to the separation principle: – in the point to point case, separation allows to use “bits” as a universal currency – but this is a miracle! (or a lucky coincidence) There is no reason that in multipoint source-channel transmission the same currency will hold (M.Gastpar) Multi-source, multi-sink case: – correlated source coding – uncoded transmission can be optimal – source-channel coding for sensor networks Spring 2009 - 40 4.1 To separate or not to separate… In point to point, if R < C, all is well in Shannon land. In multipoint communication, things are trickier (or more interesting) Famous textbook counter example (e.g. Cover-Thomas) R2 C2 Source Channel Y binary erasure multiaccess log2 3 1/3 1/3 0 1/3 X H(Y) 1 H(Y/X) H(X/Y) H(X) log2 3 R1 1 C1 No intersection, but communication possible! Spring 2009 - 41 Sensor networks and source channel coding [GastparV:03/04] Consider the problem of sensing – one source of analog information but many sensors – reconstruct an estimate at the base station Model: The CEO problem [Berger et al], Gaussian case W1 U1 W2 Source U2 F1 F2 X1 X2 S Z Y G S WM UM FM XM Question: distributed source compression and MIMO transmission or uncoded transmission? Spring 2009 - 42 Example: Gaussian Source, Gaussian Noise Performance (growing power shared among sensors): – with uncoded transmission: – with separation: Exponential suboptimality! Condition for optimality: measure matching! – Can be generalized to many sources Spring 2009 - 43 It is the best one can do: Communication between sensors does not help as M grows! Intriguing remark: – by going to ‘’bits’’, MSE went from 1/M to 1/Log(M) – ‘’bits’’ might not be a good idea for distributed sensing and communications If not ‘’bits’’, what is information in networks? [Gastpar:02] Spring 2009 - 44 Outline 1. Introduction 2. The structure of distributed signals and sampling 3. Distributed source coding 4. On the interaction of source and channel coding 5. Environmental monitoring Monitoring for scientific purposes Environmental monitoring The SensorScope project The CommonSense project 6. Conclusions Spring 2009 - 45 Environmental Monitoring: Technological Paradigm Change Today, one of the primary limitations in environmental research is the lack of simultaneous high-density spatial and temporal observations Monitoring for scientific purposes – “create” a new instrument for critical data – most current acquisitions are undersampled – verification of theory, simulations Environmental data – unstable terrain, glaciers – watershed monitoring – pollutant monitoring, forest monitoring Orders of magnitude of difference – price – size – power We expect this will have a transformational effect on – what is monitored 1K$ “each” – how it is monitored – what is understood 100K$ Spring 2009 - 46 The SensorScope Project (2005-…) Team: G. Barrenetxea, H.Dubois-Ferriere,T.Schmid,F.Ingelrest, G.Schaeffer + M. Parlange & EFLUM http://sensorscope.epfl.ch What are we trying to accomplish? SensorScope: distributed sensing instrument relevant datasets with clear documentation all data on-line, real-time anybody can compute/analyze with Sensor nodes: many possible platforms inc. low power (Berkeley motes, tinynode, tmote) many types of sensing (e.g. cyclops) First Step: SensorScope I a few dozen nodes self-organized network up for 9 months large dataset collected fun platform and testbed Spring 2009 - 47 SensorScope II [w. M.Parlange] SensorScope II collaboration with EFLUM (Laboratory of Environmental Fluid Mechanics and Hydrology) 10 real-world deployments from build to high mountain environments hundreds of Megabytes of sensing data publicly available Genepi Rock glacier, 2600 m Genepi Rock glacier, your computer very interesting theoretical (physics) and practical problems! we need reliable and meaningful data! Improved networking packet combining, routing without routes more power efficient platforms (tinynodes) Data analysis signals are far from....Gaussian! Spring 2009 - 48 The core of SensorScope: WeatherStation WeatherStation Centered around Tinynode (lowest-power sensor node, with medium range) Solar energy subsystem: Energy autonomous Sensors are daisy-chained to a single connector: No limit on the type and number of sensors Automatic sensor recognition: No configuration required Local storage: SD card (2 GB) GPS & GPRS module Fast and easy installation on all types of terrain: Spring 2009 - 49 SensorScope Front End Features: Centralized data access and administration Real-time monitoring Data visualization and download Network health and battery status Organize stations into sets Set up alerts for out-of-range conditions Security and account management User friendly Spring 2009 - 50 Network architecture Sensor network with ad hoc data gathering protocols (10 to 100’s) Basestation with available wide area communication (e.g. GPRS) Web server with data online Spring 2009 - 51 Networking Ad Hoc Networking: We use a custom communication stack: Keep it as simple as possible (robustness) Works by overhearing (minimizes traffic) Written for TinyOS 2.x Main features: Routing tables are updated dynamically (allows to add/remove stations) Radio duty-cycle < 10% (low energy consumption) Stations are synchronized (all “on” at the same time, consistent time stamps) Shortest path routing with random selection (among the “shortest path high quality link neighbours”) Spring 2009 - 52 Networking: Random, biased selection of next hop Spring 2009 - 53 Power is the basic problem! Power usage in a Tinynode (a) Off (b) Listening (c)-(g) various sending power Communications is power hungry Careful management of power Power gathering (e.g. solar panels) Energy efficient protocols for data gathering and GPRS connection Spring 2009 - 54 From Theory to Practice! All the tools are there (in theory): Routing algorithms, data correlation, time synchronization, But ... Make theory work in practice is hard ... The Theory … The Practice… Spring 2009 - 55 Application Example: Risk Analysis Real problem: land slides, infrastructure damage etc: Understanding the changing environment, effects of warming, loss of permafrost etc Spring 2009 - 56 Application Example: Genepi Location: Rock glacier above Martini (VS) Spring 2009 - 57 Spring 2009 - 58 A day in the life of Genepi! Fully autonomous camera, GPRS based, Onboard image processing, Open platform, Linux based Spring 2009 - 59 Results from Genepi Spring 2009 - 60 6. Conclusions There are some good questions on the interaction of – – – – – physics of the process: space of possible values sensing: analog/digital representation & compression: local/global transmission: separate/joint decoding & reconstruction: applications From joint source-channel coding to source-channel communication – This goes back to Shannon’s original question, but multi-source multi-point communication is hard... On-going basic questions: – are there some fundamental bounds on certain data sets? – are there practical schemes to approach the bounds? – what is observable and what is not? Applications: – environmental monitoring has many interesting, high impact questions – technology amazingly mature – datasets very far from ‘’usual’’ models Spring 2009 - 61 Thank you for your attention! Questions? “Would you like to see the top on Google Earth?” © New Yorker Spring 2009 - 62 References • On sampling – M. Vetterli, P. Marziliano, T. Blu. Sampling signals with finite rate of innovation. IEEE Tr. on SP, Jun. 2002. – T. Ajdler, L. Sbaiz and M. Vetterli, The plenacoustic function and its sampling, IEEE Transactions on Signal Processing, Oct. 2006. – T. Blu, P.L. Dragotti, M. Vetterli, P. Marziliano and L. Coulot, Sparse Sampling of Signal Innovations, IEEE Signal Processing Magazine, Vol. 25, Nr. 2, 2008. – M.N. Do, D.Marchand-Maillet, M. Vetterli, On the Bandwidth of the Plenoptic Function, IEEE Tr.IP, submitted, 2008. – Y.M. Lu and M. Vetterli, Spatial Super-Resolution of a Diffusion Field by Temporal Oversampling in Sensor Networks, IEEE ICASSP 2009. • Correlated distributed source coding – R.Cristescu, B.Beferull and M.Vetterli, Correlated data gathering, Infocom2004. – M. Gastpar, P. L. Dragotti, and M. Vetterli. The distributed Karhunen-Loeve transform. IEEE Tr. on IT, Dec. 06. – R.Konsbruck, E.Telatar, M.Vetterli, The distributed rate-distortion function of sounds fields, ICASSP06. Spring 2009 - 63 References • On sensor networks, separation uncoded transmission – M.Gastpar, M.Vetterli, PL Dragotti, Sensing reality and communicating bits: A dangerous liaison - Is digital communication sufficient for sensor networks? IEEE Signal Processing Mag.,July 2006 – M. Gastpar, B. Rimoldi, M. Vetterli. To code or not to code: lossy source-channel communication revisited, IEEE Tr. on IT, 2003 – M.Gastpar, M..Vetterli, The capacity of large Gaussian relay networks, IEEE Tr on IT, March 2005. • SensorScope – See http://sensorscope.epfl.ch – G. Barrenetxea, F. Ingelrest, G. Schaefer and M. Vetterli,The Hitchhiker's Guide to Successful Wireless Sensor Network Deployments.,. ACM SenSys2008. – F. Ingelrest, G. Barrenetxea, G. Schaefer, M. Vetterli, O. Couach and M. Parlange, SensorScope: Application Specific Sensor Network for Environmental Monitoring, to appear in ACM Transactions on Sensor Networks. Spring 2009 - 64