1. Energy-Efficient Signal Processing Techniques For Smart Grid Heterogeneous Communication Networks Task ID: 1836.133 Prof. Naofal Al-Dhahir, Univ. of Texas at Dallas Prof. Brian L. Evans, Univ. of Texas at Austin

2. Task Summary  Task Description: 1) Low-complexity interference cancellation methods that exploit PLC interference characteristics 2) Efficient wireless coexistence mechanisms in the unlicensed 902-928 MHz frequency band 3) Improve reliability and energy efficiency of two-way wireless and power line communications (PLC) between smart meters & data concentrators  Anticipated Results:  Signal processing algorithms and real-time prototypes to demonstrate enhanced performance of wireless and PLC transceivers for smart grid applications 2

3. Task Deliverables Date Task June 2015 1. Algorithms/software for low-complexity interference cancellation methods that exploit interference characteristics to reduce bit error rate by 10x (delivered) August 2015 3.Architecture/algorithm for PLC-wireless diversity combining method with at least 2x improvement in energy efficiency over state of the art (delivered) January 2017 2.Efficient wireless coexistence mechanisms in the unlicensed 902-928 MHz frequency band January 2017 4.Demonstrations on UT Austin wireless and PLC test beds

4. Students & Liaisons  Graduated Students and Current Affiliation  Dr. Jing Lin, Qualcomm, May 2014  Dr. Karl Nieman, National Instruments, Dec. 2014  Current Students and Anticipated Graduation Date  Mostafa Ibrahim, December 2017  Ghadi Sebaali, May 2020  Internships  Mostafa Ibrahim, TI, Summer’14 and Summer’15  Ghadi Sebaali, Freescale, Summer’15  Liaisons  Dr. Il-Han Kim (TI), Dr. Anand Dabak (TI), Dr. Wenxun Qiu (TI), and Dr. Khurram Waheed (Freescale) 4

5. Smart Grid Goals  Accommodate all generation types  Improve operating efficiencies Scale voltage with energy demand Bill customer using real-time rates Reduce peak demand (duty cycling) Analyze customer load profiles Analyze system load snapshots  Improve system reliability Monitor power quality D isconnect/reconnect remotely Notify outage/restoration event  I nform customer Source: Jerry Melcher, IEEE Smart Grid Short Course, Oct. 2011, Austin TX USA Enabled by two-way smart meter communications ISTOCKPHOTO.COM/© SIGAL SUHLER MORAN 5

6. Smart Meter Communications 6 Unlicensed 900 MHz Wireless Communications Narrowband Powerline Communications (3-500 kHz) Power loss vs. distance d d – /2 propagation constant e –  (f) d plus attenuation from transformers PropagationDynamicStatic (fixed grid topology) Additive noise/ interference model Gaussian mixtureCyclostationary interference and Gaussian mixture Asynchronous interference Uncoordinated users and electronic emissions Power electronics and uncoordinated users Multi-Input Multi- Output (MIMO) Enhance data rate (spatial multiplexing) & reliability (diversity); Wi-Fi standards Enhance data rate (spatial multiplexing) using single or three phase; not much diversity Noise and interference will be used interchangeably.

7. Research Summary  Accomplishments during the past year  PLC-wireless diversity combining scheme with at least 2x improvement in energy efficiency over using one link  Organized ISPLC in March 2015 in Austin, Texas  Future directions  Efficient wireless coexistence mechanisms in the sub-1GHz unlicensed 902-928 MHz frequency band between the IEEE 802.15.4g and the IEEE 802.11ah standards  Demonstrations of the achieved PLC/Wireless combining performance gains on UT Austin wireless and PLC testbeds  Technology transfer & industrial interaction  Regular conference calls are being held with TI and Freescale to discuss the work progress and get feedback 7

8. Task #1: PLC Interference Mitigation  Cyclostationary interference  Period is half AC power cycle  Spectrum varies with time  Modeled as Gaussian noise feeding three different filters  Receiver-based methods  Sparse Bayesian learning  Highly parallel algorithms  Transmitter-receiver methods  Adapt modulation to match cyclostationary noise  Exploit time-frequency sparsity 8

9. Task #1: PLC Interference Mitigation  Sparse Bayesian Learning (SBL)  Model impulsive noise as a sparse vector in time domain  Estimate and mitigate interference without training  Parallel approximate message passing algorithm  Real-time implementation fills one Xilinx Vertex-5 FPGA 9 SystemNoiseSBL w/ null tones SBL w/ all tones SBL w/ dec. feedback UncodedGaussian Mixture8 dB10 dB-- Middleton Class A6 dB7 dB-- CodedGaussian Mixture2 dB7 dB9 dB Middleton Class A1.5 dB6.3 dB9.3 dB Periodic0.8 dB4.8 dB6.8 dB

10. Task #1: PLC Interference Mitigation  Time-frequency modulation diversity  Allocate codeword to time-frequency slots  Estimate interference bandwidth and duration  Listen between transmissions to model noise states  Refine during transmission by estimating noise power 10 Transmitter Methods Throughput Reduction Channel/Noise Info at Transmitter Previous Adaptive modulation [Nieman13] ✗ Full Concatenated error correction coding (PLC standards) ✔ None Proposed Time-frequency modulation diversity ✗ Partial

11. Task #1: PLC Interference Mitigation 11 Parameters Values Sampling Rate400 kHz FFT Size256 CP Length30 # Data Tones72 Convolutional Code Rate 1/2, length 7 Interleaver Size72 bits Packet Size256 Bytes >100x >2dB Length-2 code Length-3 code Subcarriers OFDM symbols … … … … Subcarriers … … … … …

12. Task #2: Coexistence in 900MHz Band  Goal: Coexistence of IEEE 802.11ah & 802.15.4g smart utility networks in unlicensed 900 MHz band  Interference avoidance and/or management  Our simulation results show 13m physical separation needed between interferers and victims (not practical)  Receiver: channel sensing  Exploit signal waveform properties  Transmitter: dynamic spectrum management  Adjust transmit power/BW to reduce mutual interference  Long successful track record in DSL and other standards 12

13. Task # 3 : PLC/Wireless Diversity Simultaneous PLC/wireless transmissions using low-voltage power lines in 3-500 kHz band and unlicensed 902-928 MHz wireless band  Goal : Improve reliability of smart grid communications using PLC/wireless receive diversity combining methods

14. Task #3 :Symmetric Diversity Combining  Same channel, noise, and interference statistics  Same Average SNR Old: Combining of two wireless links

15. Task # 3 : Asymmetric Diversity Combining  Different channel, noise and interference statistics  PLC and wireless might have different average SNR ! New : PLC/Wireless combining for Smart Grid Comm.

16. Task # 3 : Noise Models Gaussian mixture Noise R1R2R3

17. Task # 3 : Applying Conventional MRC 17

18. Task # 3 : Impulsive Noise in PLC and Wireless Noise power over frequency sub-channels across multiple OFDM blocks PLC PAR = 21 dB AWGN PAR = 10 dB Wireless PAR = 14 dB

19. Task # 3 : Proposed PLC/Wireless Combining 19

20. Task # 3 : Proposed PLC/Wireless Combining PSD Combining Instantaneous SNR Combining

21. Task # 3 : Combining Metrics Comparison 21 Noise Power Ratio = PLC Noise Power/ Wireless Noise Power One OFDM Block 36 Active Sub-Channels out of 256 Noise power over frequency sub-channels across multiple OFDM blocks

22. Task # 3 : Simulation Parameters 22 Region 1Region 2Region 3 Time Percentage60 %30 %10 % Power (dB)-6.591.935.15

23. Task # 3 : Performance Results  Average BER vs Eb/No of both links (equal Eb/No) - Fading channels Average SNR Combining 4 dB PSD Combining 5.5 dB Instantaneous SNR Combining 7 dB

24. Task # 3 : Performance Results  Average BER vs Eb/No of the PLC link at Eb/No = 2 dB for the wireless link - Fading channels Average SNR Combining 4.5 dB PSD Combining 8 dB Instantaneous SNR Combining 10 dB

25. Task # 3 Conclusions  PSD combining provides the best performance /complexity tradeoff - better performance than average-SNR combining at lower complexity than instantaneous-SNR combining  Our proposed PSD estimation method does not require pilot overhead while instantaneous-SNR combining requires high pilot overhead (resulting in data rate loss) 25

26. Back-up Slides 26

27. Task #1: Interference Mitigation  Contribution  Model impulsive noise as a sparse vector in time domain  Apply sparse Bayesian learning (SBL) methods for estimation and mitigation without training  Three SBL algorithms proposed  Estimate and subtract the noise impulses by using the noise projection onto null and pilot tones  Perform joint noise estimation and OFDM detection using information in the date tones  Embed the algorithm into a decision feedback structure 27

28. Task #1: SBL System Overview 28 A time-domain interleaved OFDM system  Demodulated OFDM:  Assumptions:  Select interleaver size such that e_π is a sparse vector  Perfect channel estimation  New decision metric:

29. Task #1: Noise Estimation Methods  How to estimate  Apply Sparse Bayesian Learning  the global optimum is always the sparsest solution  all local optimal solutions are sparse  the number of local optima is the smallest  Propose three non-parametric algorithms:  Estimation Using Null and Pilot Tones  Estimation Using All Tones  Decision Feedback Estimation 29

30. Task #1: SBL Using Null & Pilot Tones  Apply SBL technique to the impulsive noise estimation using null and pilot tones  Use EM algorithm to obtain the MAP estimate of the time-domain impulsive noise  Transform to the frequency domain and subtract it from the received signal 30

31. Task #1: Estimation Using All Tones  Motivation:  High number of non-data tones is a tradeoff between improved performance and reduced throughput  Goal:  Exploit information available in all tones to estimate the impulsive noise given limited number of non-data tones  Apply EM algorithm:  Three hyperparameters 31

32. Task #1: SBL w/ Decision Feedback  Goal:  Exploit redundancy in the coded data tones as side information to provide a second estimate of eˆ′.  Transfer information back-and-forth between  impulsive noise estimator using non-data tones  the decoder using data tones 32

33. Task #1: Low Complexity Algorithms  Sequential SBL: performs a sequential addition and deletion of candidate basis functions 33 EstimatorOperationComplexity Using null and pilot tonesMatrix MultiplyO(N 2 M) Matrix InversionO(M 3 ) Using all tonesMatrix MultiplyO(N 3 ) Matrix InversionO(N 3 ) Sequential SBL w/ unknown background noise power Matrix MultiplyO(N 2 K) Matrix InversionO(K 3 ) Sequential SBL w/ known background noise power Matrix MultiplyO(N 2 K)

34. Task #1: Performance Analysis  SNR gains in asynchronous impulsive noise  Up to 9 dB in coded systems  Up to 10 dB in uncoded systems  SNR gains in periodic impulsive noise  Up to 6 dB in coded systems 34 SystemNoiseSBL w/ null tones SBL w/ all tones SBL w/DF UncodedGM8 dB10 dB-- MCA6 dB7 dB-- CodedGM2 dB7 dB9 dB MCA1.5 dB6.3 dB9.3 dB Periodic0.8 dB4.8 dB6.8 dB

35. Task #1: Time-frequency Modulation  Allocate components of a codeword to time- frequency slots  Require partial noise information  Narrowband interference width  Burst duration 35 Time-domain noise Subcarriers OFDM symbols … … … …

36. Task #1: Noise Power Estimation  Offline estimation  Utilize silent intervals between transmissions  Semi-online estimation  Between transmissions: Estimate start/end instances of all stationary intervals  In transmissions: Estimate noise power spectrums 36 Time Offline Semi- online Transmission Workload at the noise power estimator Low Med High

37. Task #1: Semi Online Approach  Measure noise using cyclic prefix  Formulate a compressed sensing problem  Define  Collect multiple measurements in the same stationary interval 37 Cyclic Prefix OFDM symbol + - Noise NBI AWGN

38. Task #1: Proposed OFDM PLC System 38  Using new noise model, add: 1.Impulsive noise mitigation 2.Cyclic adaptive modulation and coding

39. Task #1: Implementation on FPGA 39 Determine static schedule, map to fixed-point data and arithmetic Translate to hardware Floating- point algorithm

40. Task #1: Real-Time Implementation  Up to 8 dB of impulsive noise mitigated in real-time testbed 40 uncoded bit-error-rate (BER) signal-to-noise ratio (SNR) [dB] target BER = 10 -2 4 dB gain for 20 dB impulse power 8 dB gain for 30 dB impulse power

41. Task #1: Adaptive Modulation/Coding  Motivation: spectral and temporal variations of noise power over an AC cycle and the cyclic nature of the noise  G3-PLC currently supports static modulation with tone mask over an OFDM frame but only allows for a fixed group of six subcarriers to be masked over the duration of the OFDM frame   Extended mechanisms to a cyclic adaptive modulation and coding scheme (MCS) (changes to the green traditional system are shown in red in the above). 41

42. Task #1: Results for Adaptive Mod. Throughput of NB-OFDM PLC system employing cyclic adaptive MCS scheme over the G3-PLC CENELEC-A band is boosted 2× versus the conventional G3-PLC operation using the next-best rate-optimal choice DQPSK with tone map. 42

43. Task # 3 : Proposed PLC/Wireless Combining

44. Task # 3 : Instantaneous Noise Power Estimation  As a simple technique to estimate the instantaneous noise power, we employ comb-type pilots inserted periodically within the data symbols  We estimate the noise power in the pilot locations followed by linear interpolation to compute estimates over all symbols

45. Task # 3 : Noise PSD Estimation  The noise PSD can be estimated by averaging the received signal power  Estimated PSD and actual PSD vs the active sub-channel indices (36 sub-channels in the CENELEC A band [35-91]kHz  Averaging is performed over 512 OFDM Symbols

46. Task #4: PLC Testbed Problem Noise, interference, frequency selectivity, fading, and cross-talk Goal Increase communication rate and reliability Solution System-level design exploration tool 46 Quantify communication performance vs. complexity tradeoffs Determine achievable communication delay, throughput and reliability Broadband powerline communication transceivers

47. Task #4: PLC Testbed Some of the algorithms:  Power/Bit Loading  Echo/Near End Cancellation (NEXT)  FEQ and FEXT Cancellation  Time-domain Equalizer (TEQ) 47 HardwareSoftware NI PXI 1045 Embedded ControllerGraphical User Interface (GUI) in LabVIEW NI PXI-5122 for analog-to-digital (A/D) conversion Real-time target in LabVIEW RT NI PXI-5421 for digital-to-analog (D/A) conversion C++ Dynamically Linked Library (DLL)

48. References  Jing Lin; Pande, T.; Il Han Kim; Batra, A.; Evans, B.L., "Robust transceiver to combat periodic impulsive noise in narrowband powerline communications," in Communications (ICC), 2015 IEEE International Conference on, vol., no., pp.752-757, 8-12 June 2015. doi: 10.1109/ICC.2015.7248412  J. Lin, M. Nassar and B. L. Evans, "Impulsive Noise Mitigation in Powerline Communications using Sparse Bayesian Learning", IEEE Journal on Selected Areas in Communications, vol. 31, no. 7, Jul. 2013, pp. 1172-1183.  Karl Neiman, “Space-Time-Frequency Methods for Interference-Limited Communication Systems “, PhD. Dissertaition,The University of Texas at Austin, 2014.  K.F. 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 ISPLC, 2013. Won best paper award  K.F. Nieman, M. Nassar, J. Lin, and B.L. Evans, "FPGA implementation of a message-passing OFDM receiver for impulsive noise channels. Proc. IEEE Asilomar Conf. on Signals, Systems, and Computers, 2013. Won best student paper Architecture and Implementation Track  K. Waheen, K. F. Nieman, Adaptive cyclic channel coding for orthogonal frequency division multiplexed (OFDM) systems, US patent pending, 2014. 48