On Noise Models in PLC. A.J. Han Vinck, F. Rouissi, T.Shongwe, G. Colen and L. Oliveira

Transcription

1 On Noise Models in PLC A.J. Han Vinck, F. Rouissi, T.Shongwe, G. Colen and L. Oliveira

2 We want to discuss the influence of impulse noise on OFDM Random impulse noise Periodic impulse noise (time) Han Vinck et al. WSPLC 2015 Klagenfurt 2

3 Modem designs are based on wireless experience i.e. OFDM based modulation DBPSK, DQPSK, CODING Reed-Solomon Convolutional Repetition with interleaving Why re-inventing the wheel? Han Vinck et al. WSPLC 2015 Klagenfurt 3

4 OFDM parameters for CENELEC and FCC band Parameters CENELEC band FCC band Frequency band [35.2 KHz, 91.4 KHz] [152.3, KHz] FFT(used subcarriers) 256(72) 256(72) Time OFDM frame 695 μs 231,7 μs Sample duration 2.5 μs μs Sampling frequency 400 KHz 1.2 MHz Max. bitrate 33.4 Kbit/s 303 Kbit/s Han Vinck et al. WSPLC 2015 Klagenfurt 4

5 main characteristics needed for the modeling (1) Temporal: duration: time to extinction of noise; Inter Arrival Time (IAT): time interval between two impulses Estimated probability density function of inter-arrival time Estimated Probability density function of pulses durations Probability density X: Y: inter-arrival time (µs) Probability density X: Y: Pulse duration (µs) Han Vinck et al. WSPLC 2015 Klagenfurt 5

6 main characteristics needed for the modeling (2) Spectral: energy per pulse in the relevant spectrum Mean(PSD) /pulse (dbw/hz) The average value of the PSD in the band up to 500 KHz is : dbw/hz dbm/hz frequency (KHz) Han Vinck et al. WSPLC 2015 Klagenfurt 6

7 But there are many different types Examples: FP7 project: PLC Channel Characterization and Modelling, Cortes, Tonello, Zimmermann and Doster, et al. We take an averaging approach(with all disadvantages) -at least wegetan impression! Han Vinck et al. WSPLC 2015 Klagenfurt 7

8 Summary of the performance of the OFDM modulation for the measured data (no attenuation) CENELEC: FCC: E s = J; σ 2 G= J (-80 dbm); SNR = 30 db E s = J; σ 2 G= ;(-120 dbm) SNR = 70 db Parameters Average Highest Worst (10%) Best( 10%) probability CENELEC σ 2 I J J J J E s /(σ 2 G + σ 2 I) 17 db 21 db 12 db 30 db σ 2 I / σ 2 G FCC σ 2 I J J J J E s /(σ 2 G + σ 2 I) 16 db 29 db 9.3 db 47 db σ 2 I / σ 2 G The averageburstlength: fort-symbol = 2.5 μsis 14 symbols = μsis 43 symbols Han Vinck et al. WSPLC 2015 Klagenfurt 8

9 We need(future?): for an entire OFDM frame of duration T(OFDM) the distribution on: Pulse duration interarrival time psd Han Vinck et al. WSPLC 2015 Klagenfurt 9

10 Our assumption: OFDM randomizes the noise The inputenergy(channeloutput) tothefft (OFDM) is converted to Gaussian with the same energy(assumption) The capacity for a completely randomized Gaussian channel would then be: P/2B capacity( -,-) = Blog 2 ( σ +σ G I ) BUT: thequestioniswhatisthelosscomparedwith: known channel state at the receiver? known channel state at the receiver and transmitter(upper bound)? Han Vinck et al. WSPLC 2015 Klagenfurt 10

11 Two traditional memoryless noise models: Middleton Class-A and mixed Gaussian mixed Gaussian prob: 1-A => Gaussian -2 channelstates: 2 Average = 2 σi prob: A => σg + A 2 σ G 2 2 σ G + σi We need A:= time that the frame is disturbed by impulse noise Estimate: A= pulse duration x # ofpulses/ OFDM frameduration Parameters Average Highest Worst (10%) Best( 10%) probability Pulse duration 36 μs 6 μs > 82 μs < 2.1 μs Interarrival time 127.μs 25 μs > 14 μs < 270 μs A Han Vinck et al. WSPLC 2015 Klagenfurt 11

12 Whatcanwegainbyusingthechannelstate? (memory of the noise) Using waterfilling argument(high P) G σ I + P/2B σg +σ I + P/2B capacity( + +) = (1- A)Blog2(1+ 2 ) + ABlog2( 2 2 ) σ σ +σ / A P/2B(1- A) (low power) capacity(+ +) = (1- A)Blog2 (1+ 2 σ G I G ) Using Gaussian input with average power P 2 2 P/2B σ G +σ I / A + P/2B capacity( - +) = (1 - A)Blog 2 (1 + 2 ) + ABlog 2 ( 2 2 ) σ σ +σ / A G G I The randomized channel P/2B capacity( -,-) = Blog 2 ( σ +σ G I ) 2 σ I gain 10log 10 (1 + 2 σ G ) db Han Vinck et al. WSPLC 2015 Klagenfurt 12

13 Example for FCC and CENELEC capacities FCC B = 337 khz; A = 0.3; E s = J; σ 2 G= ; σ 2 I= J C(+,+)= 7.8 Mbit/s => C(-,+) = 7.8 Mbit/s => C(-,-) = 1.8 Mbit/s CENELEC B = 56.2 khz; A = 0.3; E s = J; σ 2 G= J; σ 2 I= J C(+,+)= 560 Kbit/s => C(-,+) = 500 kbit/s => C(-,-) = 240 kbit/s Han Vinck et al. WSPLC 2015 Klagenfurt 13

14 Noise mitigation Nulling and clipping(zhykov) Reducesthenoiseenergyin theofdm S. Zhidkov, Analysis and comparison of several simple impulsive noise mitigation schemes for OFDM receivers, IEEE Trans. Commun. vol. 56, no. 1, pp. 5 9, Jan Compressed sensing(lampe, Mengi) Uses the strong impulse to solve equations Successive impulsive noise suppression in OFDM, Mengi, A. ; Vinck, A.J.H. Publication Year: 2010, Page(s): 33 37, ISPLC Iterative detection(häring, Papilaya) Use the nulled positions to cancel impulse noise ImprovingPerformance of the MH-Iterative IN Mitigation Scheme in PLC SystemsIEEE Transactions on Power Delivery, April 2014, V. Papilaya and A.J. Han Vinck Han Vinck et al. WSPLC 2015 Klagenfurt 14

15 Noise mitigation Nulling and clipping(zhykov) Reducesthenoiseenergyin theofdm S. Zhidkov, Analysis and comparison of several simple impulsive noise mitigation schemes for OFDM receivers, IEEE Trans. Commun. vol. 56, no. 1, pp. 5 9, Jan Compressed sensing(lampe, Mengi) Uses the strong impulse to solve equations Successive impulsive noise suppression in OFDM, Mengi, A. ; Vinck, A.J.H. Publication Year: 2010, Page(s): 33 37, ISPLC Iterative detection(häring, Papilaya) Use the nulled positions to cancel impulse noise ImprovingPerformance of the MH-Iterative IN Mitigation Scheme in PLC SystemsIEEE Transactions on Power Delivery, April 2014, V. Papilaya and A.J. Han Vinck Han Vinck et al. WSPLC 2015 Klagenfurt 15

16 Frequency domain output for time periodic input interleaving phase + position Perfect period fitting interleaving Position only (different scale) Leaking due to imperfect period fitting Shongwe, T.; Han Vinck, A.J., "Broadband and Narrow-band Noise Modelling in Powerline Communications," Wiley Encyclopedia of Electrical and Electronics Engineering, to be published. ISPLC 2015 Austin, Han Vinck 16

17 Possible mitigation solutions A SOLUTION USING NOTCH FILTERS PERIODIC IMPULSIVE NOISE REDUCTION IN OFDM BASED POWER LINE COMMUNICATION Sumi Mathew andprasanth Murukan A SOLUTION USING INTERLEAVING IN THE TIME DOMAIN Non-parametric Mitigation of Periodic Impulsive Noise in Narrowband PLC Jing Lin and Brian L. Evans Impulsive Noise Mitigation in Powerline Communications Using Sparse Bayesian Learning Jing Lin, Marcel Nassar, and Brian L. Evans, MODELS: Analysis of the Periodic Impulsive Noise Asynchronous with the Mains in PLC Jose Antonio Cortes, Luis Dıez, Francisco Javier Canete and Jesus Lopez Han Vinck et al. WSPLC 2015 Klagenfurt 17

18 Results for time-domain Interleaving Time and phase interleaving Error floor: Pe=> Q E σ b 2 Spreading of the impulse noise over the N subcarriers Conclusion: - Error floor determined by the noise variance Han Vinck et al. WSPLC 2015 Klagenfurt 18

19 Nulling gives error floor depending on f p The introduced noise depends on the signal energy NullinggivesError floor: Pe=> Q E p b fe b = Q 1 f p Conclusion: - Error floor determined by the period Han Vinck et al. WSPLC 2015 Klagenfurt 19

20 an example of an OFDM modem in the G3-PLC standard for burst error correction -Onlyforrandomerrors - Soft decision not possible - Produces burst error at decoder output Interleaver to randomize errors Han Vinck et al. WSPLC 2015 Klagenfurt 20

21 We suggest to replace the convolutional code + interleaver Short Block code G = R = ½ for4-ary symbols Errors in frequency domain occur in bursts: block code words are symbols for the RS code We need narrow band disturbance properties! Han Vinck et al. WSPLC 2015 Klagenfurt 21

22 Concatenated Code example: block code Corrects 1 QAM error G = α α 2 2 QAM symbols 3 QAM symbols Shortened Reed-Solomon code length n < 2 6 Example: n = 14, k = 8 to correct 3 symbols Transmits 96 bits with 70 OFDM carriers ( 250 Kbit/s) David Forney

23 Simulation comparison (no nulling, no π) RS+CC+Interleaver versus RS+ShortBlockCode RS+CC+Interleaver RS+ShortBlockCode There is a 1-2 db difference in performance for this example Han Vinck et al. WSPLC 2015 Klagenfurt 23

24 conclusions OFDM spreads the impulse noise ( signal attenuations should be looked at!) In Cenelec band, background noise is dominating, degradation about 10 db In FCC band, impulse noise degrades performance by db Time Periodic noise and narrow band interference can be mitigated by short block code + RS codeasan alternative tothecomplexviterbidecoding Many details need a closer look Han Vinck et al. WSPLC 2015 Klagenfurt 24

25 Sources of periodic noise switched-mode power supplies. switching actions of rectifier diodes found in many electrical appliances you can clearly see peaks separated by about 120 khz, the fundamental frequency of the switching power supply used for this experiment. Han Vinck et al. WSPLC 2015 Klagenfurt 25

26 More details on the codes Reed-Solomon SBC kx 8 bits Symbol (8 bits) n x 8 bits 4 x 2 bits Short block code 8x 2 bits 8 QAM R = ½ symbols error correcting (8,4), d min = 4) code The Reed-Solomon code hasdimensions: n = 64, k = 56, 4 symbol error correcting theshortblock codehasr = ½ for4-ary symbols G = Block decoding errors are correctedbythers code ora (8,4) codefor4-ary symbolswith d min = 4 Han Vinck et al. WSPLC 2015 Klagenfurt 26

27 Another approch: noise mitigation by nulling Transmitted symbol Periodic noise with constant phase and constant amplitude) nulling Nulling mitigates the influence of impulse noise Han Vinck et al. WSPLC 2015 Klagenfurt 27