Report

Semiconductor Research Corporation Presentation Texas Analog Center of Excellence, The University of Texas at Dallas Smart Grid Communications Prof. Brian L. Evans Dept. of Electrical & Computer Engineering Wireless Networking & Communications Group The University of Texas at Austin 14 December 2012 In collaboration with UT Austin PhD students Ms. Jing Lin, Mr. Yousof Mortazavi, Mr. Marcel Nassar and Mr. Karl Nieman; Freescale engineers Mr. Mike Dow and Dr. Khurram Waheed; and TI engineers Dr. Anand Dabak and Dr. Il Han Kim http://users.ece.utexas.edu/~bevans/projects/plc/index.html ISTOCKPHOTO.COM/© SIGAL SUHLER MORAN Outline • Research group • Smart power grids • Powerline noise Cyclostationary Gaussian mixture • Testbeds • Conclusion IEEE Signal Processing Magazine Special Issue on Signal Processing Techniques for the Smart Grid, September 2012. 1 Embedded Signal Processing Laboratory • Present: 9 PhD, 0 MS, 5 BS • Alumni: 20 PhD, 9 MS, 140 BS • Communication systems Hugo Powerline communication systems (design tradeoffs) Wi-Fi (interference modeling & mitigation for ISM bands) Cloud Radio Access Networks (LTE basestation coordination) Mixed-signal IC design (mostly digital ADCs and synthesizers) Marcus Jing Chao Debarati Yousof Marcel Karl Kyle • Video processing (rolling shutter artifact reduction) • Electronic design automation (EDA) tools/methods • Part of Wireless Networking & Communications Group 160 grad students, 20 faculty members, 13 affiliate companies 2 Research Group – Completed Projects 20 PhD and 9 MS alumni System SW release Prototype Companies Matlab DSP/C Freescale, TI MIMO testbed LabVIEW LabVIEW/PXI Oil&Gas resource allocation LabVIEW DSP/C Freescale, TI Underwater space-time proc.; comm. MIMO testbed Matlab Lake Travis testbed Navy Camera image acquisition Matlab DSP/C Intel, Ricoh Display image halftoning Matlab C HP, Xerox video halftoning Matlab fixed point conv. Matlab FPGA Intel, NI distributed comp. Linux/C++ Navy sonar Navy, NI ADSL Wimax/LTE EDA tools Contribution equalization DSP Digital Signal Processor MIMO Multi-Input Multi-Output LTE PXI Qualcomm Long-Term Evolution (cellular) PCI Extensions for Instrumentation 3 Research Group – Current Projects 9 PhD students System Contributions Powerline comm. noise reduction; MIMO testbed Wi-Fi interference reduction SW release Prototype Companies LabVIEW LabVIEW / PXI chassis Freescale, IBM, TI Matlab FPGA Intel, NI time-based ADC Cellular cloud radio access network architecture Handheld camera reducing rolling shutter artifacts EDA tools reliability patterns MIMO Multi-Input Multi-Output IBM 45nm Huawei Matlab Android TI NI PXI PCI Extensions for Instrumentation 4 ISTOCKPHOTO.COM/© SIGAL SUHLER MORAN Outline • Research group • Smart power grids • Powerline noise Cyclostationary Gaussian mixture • Testbeds • Conclusion IEEE Signal Processing Magazine Special Issue on Signal Processing Techniques for the Smart Grid, September 2012. 5 Today’s Power Grids in USA • 7 large-scale power grids each managed by a regional utility company 700 GW generation capacity in total for long-haul high-voltage power transmission Synchronized independently, and exchange power via DC transfer • 130+ medium-scale power grids each managed by a local utility Local power distribution to residential, commercial and industrial customers • Heavy penalties in US for blackouts (2003 legislation) Utilities generate expected energy demand plus 12% Energy demand correlated with time of day Effect of plug-in electric vehicles (EVs) on energy demand uncertain Generation cost 30x higher during peak times vs. normal load • Traditional ways to increase capacity to meet peak demand increase Build new large-scale power generation plant at cost of $1-10B if permit issued Build new transmission line at $0.6M/km which will take 5-10 years to complete Source: Jerry Melcher, IEEE Smart Grid Short Course, 22 Oct. 2011, Austin TX USA 6 Smart Grid Goals • Accommodate all generation types Renewable energy sources Energy storage options • Improve asset utilization and operating efficiencies Scale voltage with energy demand Reduce peak demand Analyze customer load profiles and system load snapshots • Improve system reliability Power quality monitoring Remote disconnect/reconnect Outage/restoration event notification Enabled by smart meter communications • Enable informed customer participation Source: Jerry Melcher, IEEE Smart Grid Short Course, 22 Oct. 2011, Austin TX USA 7 Smart Grid Wind farm HV-MV Transformer Central power plant Grid status monitoring Utility control center Smart meters Integrating distributed energy resources Houses Offices Device-specific billing Automated control for smart appliances Medium Voltage (MV) 1 kV – 33 kV Industrial plant High Voltage (HV) 33 kV – 765 kV 8 Smart Grid Communications Local utility Communication backhaul carries traffic between concentrator and utility on wired or wireless links Data concentrator MV-LV transformer Smart meter communications between smart meters and data concentrator via powerline or wireless links Smart meters Home area data networks connect appliances, EV charger and smart meter via powerline or wireless links Low voltage (LV) under 1 kV 9 Powerline Communications (PLC) Categories Band Bit Rates Coverage Narrowband 3-500 kHz ~500 kbps • (ITU) PRIME, G3 MultiSmart meter • ITU-T G.hnem kilometer communication • IEEE P1901.2 Broadband 1.8-250 MHz ~200 Mbps <1500 m Enables Standards • HomePlug Home area • ITU-T G.hn data networks • IEEE P1901 • Use orthogonal frequency division multiplexing (OFDM) • Communication challenges o Channel distortion o Non-Gaussian noise 10 Comparison Between Wireless and PLC Systems Wireless Communications Narrowband PLC (3-500 kHz) Time-selective fading and Doppler shift (cellular) Periodic with period of half AC main freq. plus lognormal time-selective fading Power loss vs. distance d d –n/2 where n is propagation constant e – a(f) d plus additional attenuation when passing through transformers Propagation Dynamically changing Determinism from fixed grid topology Synchronization Varies AC main power frequency Additive noise/ interference Assumed stationary and Gaussian Gaussian plus non-Gaussian noise dominated by cyclostationary component Time selectivity Asynchronous interference MIMO Uncoordinated users in Due to power electronics and Wi-Fi bands; uncoordinated users using other standards Frequency reuse in cellular Standardized for Wi-Fi and cellular Number of wires minus 1; G.9964 standard for broadband PLC 11 ISTOCKPHOTO.COM/© SIGAL SUHLER MORAN Outline • Research group • Smart power grids • Powerline noise Cyclostationary Gaussian mixture • Testbeds • Conclusion IEEE Signal Processing Magazine Special Issue on Signal Processing Techniques for the Smart Grid, September 2012. 12 Types of Powerline Noise Background Noise Cyclostationary Noise Impulsive Noise -50 -100 -150 0 100 200 300 Frequency (kHz) 400 time 500 Spectrally shaped noise with 1/f spectral decay Period is synchronous to half of the AC cycle Random impulsive bursts Superposition of low intensity noise sources Switching power supplies and rectifiers Circuit transient noise and uncoordinated interference Present in all PLC Dominant in Narrowband PLC Dominant in Broadband PLC 13 Cyclostationary Noise in Narrowband PLC Medium Voltage Site Low Voltage Site Field measurements collected jointly with Aclara and Texas Instruments near St. Louis, MO USA 14 Cyclostationary Noise Modeling • Linear periodically time-varying system model H1 v R H2 N n R N … HM Hi - Linear time invariant filter N - Period in samples o Period (half of the AC cycle) is partitioned into M segments o Noise within each segment is stationary, i.e. modeled by an LTI system Segment: 1 23 15 Cyclostationary Noise Model Fitting • M = 3 segments captures temporal-spectral cyclostationarity Measurement data Noise synthesized from model Proposed TI-Aclara-UT model adopted in IEEE P1901.2 narrowband PLC standard 16 Asynchronous Noise Modeling Wireless Emissions Uncoordinated Meters (coexistence) Total interference at receiver: Interference from source i 17 Asynchronous Noise Modeling • Aggregate interference from multiple sources Dominant interference source Impulse rate l Impulse duration m Ex. Rural areas, industrial areas with heavy machinery Middleton class A Ex. Semi-urban areas, apartment complexes Middleton class A Ex. Dense urban and commercial settings Gaussian mixture model Homogeneous network li = l, mi = m, g(di) = g General (heterogeneous) network li, mi, g(di) = gi 18 Asynchronous Noise Model Fitting Homogeneous PLC Network General PLC Network Tail probabilities (which direct relate to communication performance) Middleton Class A is special case of Gaussian mixture model (GMM) 19 ISTOCKPHOTO.COM/© SIGAL SUHLER MORAN Outline • Research group • Smart power grids • Powerline noise Cyclostationary Gaussian mixture • Testbeds • Conclusion IEEE Signal Processing Magazine Special Issue on Signal Processing Techniques for the Smart Grid, September 2012. 20 Our PLC Testbeds • Quantify application performance vs. complexity tradeoffs Provide suite of user-configurable algorithms and system settings Display statistics of communication performance • 1x1 PLC testbeds (completed) TI PRIME modems (testbed #1) and Freescale G3 modems (testbed #2) Adaptive signal processing algorithms for bit loading and interference mitigation Goal: Improve communication performance 2-3x on indoor power lines • 2x2 PLC testbed (on-going) Use one phase, neutral and ground for 2 x 2 differential signaling Extend our 2 x 2 real-time DSL testbed (deployed in field by oil & gas company) Adaptive signal processing algorithms for crosstalk cancellation Goal: Improve communication performance by another 2x on indoor power lines 21 1 x1 PLC Testbed #1 Hardware Software • National Instruments (NI) controllers stream • NI LabVIEW Real-Time system runs data transceiver algorithms • NI cards generates/receives analog signals • Desktop PC running LabVIEW is used • Texas Instruments (TI) analog front end as an input and visualization tool to couples to power line display important system parameters. 1x1 Testbed 22 OFDM Systems in Impulsive Noise • FFT spreads impulsive energy over all tones SNR in each tone is decreased which increases symbol error rate • Many narrowband PLC systems operate over -5 dB to 5 dB in SNR Data subchannels/tones carry same number of bits (1-4) in current standards 3 dB SNR gain could increase one bit/subchannel for same symbol error rate 23 Parametric vs. Nonparametric Noise Mitigation Parametric Nonparametric Must build a statistical model of the noise Yes No Requires training data to compute model parameters Yes No Degrades in performance due to model mismatch Yes No Has high complexity when receiving message data No Yes 24 Proposed Non-Parametric Methods • Exploit sparsity of impulsive noise in time domain time Build statistical model each OFDM symbol using sparse Bayesian learning (SBL) At receiver, null tones contain only additive noise (Gaussian + impulsive) • SNR gain vs. conventional OFDM systems at symbol error rate 10-4 Complex OFDM, 128-point FFT, QPSK, data tones 33-104, rate ½ conv. code Gaussian mixture model w/ 3 terms; Middleton Class A with A = 0.1 and = 0.01 6 dB SNR gain could mean +2 bits/tone System Uncoded Coded Noise SBL w/ null tones SBL w/ all tones SBL w/ decision feedback GMM 8 dB 10 dB - MCA 6 dB 7 dB - GMM 2 dB 7 dB 9 dB MCA 1.75 dB 6.75 dB 8.75 dB 25 Communication Performance w/o Error Correction Gaussian mixture model noise Non-parametric methods in blue Parametric methods in red Proposed NSI CS+LS: [Caire08] MMSE: [Haring02] SBL: [Lin11] 26 Communication Performance w/ Error Correction Proposed NSI NSI Non-parametric methods in blue Parametric methods in red Gaussian mixture model noise 27 Exploiting Sparsity in Time Domain Reprise • Time-domain block interleaved OFDM (TDI-OFDM Bursts span consecutive OFDM symbols Coded performance in cyclostationary noise Interleave Bursts spread over many OFDM symbols Complex OFDM, 128-point FFT, QPSK, data tones 33-104, rate ½ conv. code 28 FPGA Test System for G3 PLC Algorithms NI PXIe-7965R (Virtex 5) NI PXIe-1082 Real-time host tone map: data tone 0 f (kHz) 0 23 58 127 35.94 90.63 199.2 FPGA Timing/Resource Utilization Parametric Approximate Message Passing (AMP) mitigation method Base logic clock = 40 MHz, most data streams 16 bits wide Execution time: 5 iterations × 4776 cycles/iteration = 23880 cycles without AMP with AMP Supports streaming operation at 400 kS/s (G3 sample rate) • Can recover up to 8 dB SNR in impulsive noise environments • 100x reduction in average bit error rate using QPSK and 40 dB impulses with 3% probability • Preliminary resource utilization: • • • • Possible to exploit more parallelism for higher throughput (steps 1-4 of AMP) Conclusion • PLC systems are interference limited • Statistical models for interference Cyclostationary models synchronous with zero crossings of AC cycle Gaussian mixture model for asynchronous noise • Interference mitigation Non-parametric sparse Bayesian learning algorithms do not map well to FPGAs Parametric distributed approximate message algorithms map well to FPGAs • Testbeds • Project Web site: http://users.ece.utexas.edu/~bevans/projects/plc/index.html 31 References • • • • • • • • • [Caire08] G. Caire, T.Y. Al-Naffouri, and A.K. Narayanan. Impulse noise cancellation in OFDM: an application of compressed sensing. Proc. IEEE Int. Symp. Information Theory, pages 1293–1297, 2008. [Cho04] J. H. Cho. Joint transmitter and receiver optimization in additive cyclostationary noise. IEEE Trans. on Information Theory, 50(12), 2004. [Garcia07] R. Garcia, L. Diez, J.A. Cortes, and F.J. Canete. Mitigation of cyclic short-time noise in indoor power-line channels. Proc. IEEE Int. Symp. Power Line Comm. and Its Applications, pp. 396–400, 2007. [Haring02] J. Haring. Error Tolerant Communication over the Compound Channel. Aachen, 2002. [Haring03] J. Haring and A. J. H. Vinck. Iterative decoding of codes over complex numbers for impulsive noise channels. IEEE Trans. on Information Theory, 49(5):1251–1260, 2003. [Lampe11] L. Lampe. Bursty impulse noise detection by compressed sensing. Proc. IEEE Int. Symp. Power Line Commun. and Appl., pages 29–34, 2011 [Liano11] A. Liano, A. Sendin, A. Arzuaga, and S. Santos. Quasi-synchronous noise interference cancellation techniques applied in low voltage PLC. Proc. IEEE Int. Symp. Power Line Comm. and Its Applications, 2011. [Lin11] J. Lin, M. Nassar, and B. L. Evans, “Non-Parametric Impulsive Noise Mitigation in OFDM Systems Using Sparse Bayesian Learning”, Proc. IEEE Int. Global Comm. Conf., 2011. [Lin12] J. Lin and B. L. Evans, “Cyclostationary Noise Mitigation in Narrowband Powerline Communications”, Proc. APSIPA Annual Summit and Conf., 2012. 32 References • • • • • • • • • [Nassar09] M. Nassar, K. Gulati, M. DeYoung, B.L. Evans, and K. Tinsley. Mitigating near-field interference in laptop embedded wireless transceivers. Journal of Signal Proc. Systems, pages 1–12, 2009. [Nassar11] M. Nassar and B.L. Evans. Low Complexity EM-based Decoding for OFDM Systems with Impulsive Noise. In Proc. Asilomar Conf. on Sig., Systems, and Computers, 2011. . [Nassar12] M. Nassar, A. Dabak, I.H. Kim, T. Pande, and B.L. Evans. Cyclostationary noise modeling in narrowband powerline communication for smart grid applications. Proc. IEEE Int. Conf. on Acoustics, Speech and Sig. Proc., pages 3089–3092, 2012. [Nassar12mag] M.Nassar, J.Lin, Y. Mortazavi, A.Dabak, I.H.Kim and B.L.Evans, “Local Utility Powerline Communications in the 3-500 kHz Band: Channel Impairments, Noise, and Standards”, IEEE Signal Processing Magazine, vol. 29, no. 5, pp. 116-127, Sep. 2012. [Nieman13] K. Nieman, J. Lin, M. Nassar, K. Waheed and B. L. Evans, “Cyclic Spectral Analysis of Power Line Noise in the 3-200 kHz Band”, Proc. IEEE Int. Sym. on Power Line Communications and Its Applications, Mar. 24-27, 2012, submitted. [Pauli06] V. Pauli, L. Lampe, and R. Schober. ”turbo dpsk” using soft multiple-symbol differential sphere decoding. IEEE Trans. on Information Theory, 52(4):1385–1398, 2006. [Raphaeli96] D. Raphaeli. Noncoherent coded modulation. IEEE Trans. on Comm., 44(2):172–183, 1996. [Tipping01] M.E. Tipping. Sparse Bayesian learning and the relevance vector machine. Journal of Machine Learning Research, 1:211–244, 2001. [Umehara01] D. Umehara, M. Kawai, and Y. Morihiro. Performance analysis of noncoherent coded modulation for power line communications. Proc. Int. Symp. Power Line Commun. and Its Appl., pages 291–298, 2001. 33 Backup Slides 34 Simulated Performance • Symbol error rate in different noise scenarios ~10dB ~6dB ~6dB ~8dB ~4dB Gaussian mixture model Middleton class A model • MMSE w/ (w/o) CSI: Parametric estimator assuming known (unknown) statistical parameters of noise • CS+LS: A compressed sensing and least squares based algorithm 35 A Smart Grid Communication to isolated area Power generation optimization Integrating alternative energy sources Load balancing Disturbance monitoring Smart metering Electric car charging & smart billing Source: ETSI 36 Power Lines • Built for unidirectional energy flow • Bidirectional information flow throughout smart grid will occur Low Voltage (LV) under 1 kV High Voltage (HV) 33 kV – 765 kV Medium Voltage (MV) 1 kV – 33 kV Transformer Source: ERDF 37 Local Utility Powerline Communications (PLC) • PLC modems (PRIME, etc.) use carrier sensed multiple access to determine when the medium is available for transmission • MV router plays similar role as a Wi-Fi access point 38 Sources of Powerline Noise Uncoordinated transmission Power line disturbance Electronic devices Taken from a local utility point of view 39 PLC In Different Frequency Bands Category Band Ultra Narrowband Narrowband Broadband Bit Rate Applications Standards 0.3 – 3 kHz ~100 bps • Automatic meter reading • Outage detection • Load control N/A 3 – 500 kHz • Smart metering ~500 kbps • Real-time energy management 1.8 – 250 ~200 Mbps • Home area networks MHz • PRIME, G3 • ITU-T G.hnem • IEEE P1901.2 • HomePlug • ITU-T G.hn • IEEE P1901 All of the above standards are based on multicarrier communications using orthogonal frequency division multiplexing (OFDM). 40 Physical Layer Parameters for OFDM Narrowband PLC Standards CENELEC A band is from 3 to 95 kHz. FCC band is from 34.375 to 487.5 kHz. PRIME and G3 use real-valued baseband OFDM. Others are complex-valued. 41 Smart Power Meters at Customer Site • Enable local utilities to improve Operating efficiency System reliability Customer participation • Automatic metering infrastructure functions Interval reads (every 1/15/30/60 minutes) and on-demand reads and pings Transmit customer load profiles and system load snapshots Power quality monitoring Remote disconnect/reconnect and outage/restoration event notification • Need low-delay highly-reliable communication link to local utility • 75M smart meters sold in 2011 (20% increase vs. 2010) Source: Jerry Melcher, IEEE Smart Grid Short Course, 22 Oct. 2011, Austin TX USA 42 Non-Gaussian Noise: Challenge to PLC • Performance of conventional communication system degrades in non-AWGN environment • Statistical modeling of powerline noise • Noise mitigation exploiting the noise model or structure Listen to the environment Estimate noise model Use model or structure to mitigate noise 43 Cyclostationary Noise Modeling in Narrowband PLC (3-500 kHz) 1. M. Nassar, A. Dabak, I. H. Kim, T. Pande and B. L. Evans, “Cyclostationary Noise Modeling In Narrowband Powerline Communication For Smart Grid Applications”, Proc. IEEE Int. Conf. on Acoustics, Speech, and Signal Proc., Mar. 25-30, 2012, Kyoto, Japan. 2. M. Nassar, J. Lin, Y. Mortazavi, A. Dabak, I. H. Kim and B. L. Evans, “Local Utility Powerline Communications in the 3-500 kHz Band: Channel Impairments, Noise, and Standards”, IEEE Signal Processing Magazine, Special Issue on Signal Processing Techniques for the Smart Grid, Sep. 2012, 14 pages. 44 Impulsive Noise in Broadband PLC: Modeling and Mitigation 3. M. Nassar, K. Gulati, Y. Mortazavi, and B. L. Evans, “Statistical Modeling of Asynchronous Impulsive Noise in Powerline Communication Networks”, Proc. IEEE Int. Global Communications Conf., Dec. 5-9, 2011, Houston, TX USA. 4. J. Lin, M. Nassar and B. L. Evans, “Non-Parametric Impulsive Noise Mitigation in OFDM Systems Using Sparse Bayesian Learning”, Proc. IEEE Int. Global Communications Conf., Dec. 5-9, 2011, Houston, TX USA. 45 Statistical-Physical Modeling • Interference from a single source Noise envelope k pulses in a window of duration T (k) (j) Tk Pulse emission duration (1) (2) τj Pulse arrival time t=0 Emission duration: geometrically distributed with mean μ Pulse arrivals: homogeneous Poisson point process with rate λ Assuming channel between interference source and receiver has flat fading 46 Impulsive Noise Mitigation in OFDM Systems • A linear system with Gaussian disturbance v y = Fe FHF x Fn = Fe v , * v ~ C N (x, I ) 2 g Estimate the impulsive noise and remove it from the received signal yˆ = y F eˆ x g Apply standard OFDM decoder as if only AWGN were present 47 Parametric Vs. Non-Parametric Methods • Noise in different PLC networks has different statistical models • Mitigation algorithms need to be robust in different noise scenarios Parametric Methods Non-Parametric Methods Assume parameterized noise statistics Yes No Performance degradation due to model mismatch Yes No Training needed Yes No 48 Non-Parametric Mitigation Using Null Tones J : Index set of null tones FJ : DFT sub-matrix e: Impulsive noise in time domain g: AWGN with unknown variance • A compressed sensing problem Exploiting the sparse structure of the time-domain impulsive noise • Sparse Bayesian learning (SBL) Proposed initially by M. L. Tipping A Bayesian inference framework with sparsity promoting prior 49 Sparse Bayesian Learning • Bayesian inference Sparsity promoting prior: Likelihood: Posterior probability: e | g ~ C N (0, ), = diag ( g ) yJ | g , 2 ~ C N (0, F F I ) * 2 e | yJ ; g , ~ CN (m , e ) 2 • Iterative algorithm Step 1: Maximum likelihood estimation of hyper-parameters (γ, σ2) Solved by expectation maximization (EM) algorithm (e is latent variable) Step 2: Estimate e from the mean of the posterior probability, go to Step 1 50 Non-Parametric Mitigation Using All Tones • Joint estimation of data and noise : Index set of data tones z : Received signal in frequency domain J Treat the received signal in data tones as additional hyper-parameters Estimate of z J is sent to standard OFDM equalizer and symbol detector 51