Report

Cyclostationary Noise Mitigation in Narrowband Powerline Communications Jing Lin and Brian L. Evans Department of Electrical and Computer Engineering The University of Texas at Austin Dec. 4, 2012 Local Utility Smart Grid Communications Communication backhauls: carry traffic between concentrator and utility Local utility Data concentrator Last mile communications: between smart meters and data concentrators Transformer Smart meters Home area networks: interconnect smart appliances, line transducers and smart meters 1 Local Utility Powerline Communications Category Band Bit Rate (bps) Coverage Applications Ultra Narrowband (UNB) 0.3-3 kHz ~100 >150 km Last mile comm. Narrowband (NB) Broadband (BB) 3-500 kHz 1.8-250 MHz ~500k ~200M Multikilometer <1500 m Last mile comm. Home area networks Standards • TWACS • PRIME, G3 • ITU-T G.hnem • IEEE P1901.2 • HomePlug • ITU-T G.hn • IEEE P1901 2 Non-Gaussian Noise in NB-PLC • Non-Gaussian noise is the most performance limiting factor in NB-PLC o Performance of conventional system degrades in non-AWGN o Non-Gaussian noise reaches 30-50 dB/Hz above background noise in PLC o Typical maximum transmit power of a commercial PLC modem is below 40W o Significant path loss Power Lines 100 kHz LV 1.5-3 dB/km MV (Overhead) 0.5-1 dB/km MV (Underground) 1-2 dB/km 3 Cyclostationary Noise: Dominant in NB-PLC • Noise statistics vary periodically with half the AC cycle o Caused by switching mode power supplies (e.g. DC-DC converter, light dimmer) Data collected at an outdoor low-voltage site 4 Statistical Modeling of Cyclostationary Noise • Linear periodically time varying(LPTV) system model [Nassar12, IEEE P1901.2] 5 Model Parameterization • Periodically switching linear autoregressive (AR) process o Introduce a state sequence , o Parameterize each LTI filter by an order-r AR filter AR coefficients at time k: … State sequence AR parameters … Observation 6 Nonparametric Bayesian Learning of Switching AR Model • Hidden Markov Model (HMM) assumption on the state sequence o HMM with infinite number of states o Transition probability matrix should be sparse vectors (clustering) Self transition is more likely than inter-state transitions o Sticky hierarchical Dirichlet Process (HDP) prior on [Fox11] 7 Nonparametric Bayesian Learning of Switching AR Model • Learning AR coefficients conditioned on the state sequence … … o Partition into M groups corresponding to states 1 to M o Form M independent linear regression problems o Solve for using Bayesian linear regression [Fox11] 8 Cyclostationary Noise Mitigation Approach • Estimate switching AR model parameters o Receiver can listen to the noise during no-transmission intervals o Estimate the switching AR model parameters • Noise whitening at the receiver o , 9 Simulation Settings • An OFDM system FFT Size # of Tones Data Tones Sampling Frequency Modulation FEC Code 256 128 #23 - #58 400 kHz QPSK Rate-1/2 Convolutional • Cyclostationary noise is synthesized from the LPTV system model 10 Communication Performance Uncoded Coded 11 Reference • [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 Signal Proc, 2012. • [IEEE P1901.2] A. Dabak, B. Varadrajan, I. H. Kim, M. Nassar, and G. Gregg, Appendix for noise channel modeling for IEEE P1901.2, IEEE P1901.2 Std., June 2011, doc: 2wg-11-013405-PHM5. • [Fox11] E. B. Fox, E. B. Sudderth, M. I. Jordan, A. S. Willsky, “Bayesian Nonparametric Inference of Switching Dynamic Linear Models,” IEEE Trans. on Signal Proc, vol. 59, pp. 1569–1585, 2011. 12 Thank you 13