Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs) Submission Title: FPP-SUN Bad Urban GFSK vs OFDM Date Submitted: July 30, 2009 Source: Rishi Mohindra, MAXIM Integrated Products Contact: Rishi Mohindra, MAXIM Integrated Products Voice: +1 408 331 4123 , E-Mail: Rishi.Mohindra@maxim-ic.com Re: TG4g Call for proposals Abstract: PHY proposal towards TG4g Purpose: PHY proposal for the TG4g PHY amendment Notice: This document has been prepared to assist the IEEE P802.15. It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein. Release: The contributor acknowledges and accepts that this contribution becomes the property of IEEE and may be made publicly available by P802.15. Slide 1 Rishi Mohindra, MAXIM

FPP-SUN Bad Urban GFSK vs OFDM Rishi Mohindra

Contents Bad Urban Channel Model GFSK simulation set up OFDM simulation details PER simulation results Channel Statistics PER with FHSS for GFSK Conclusions Appendix Channel models in Matlab and ADS

Bad Urban Channel Model Channel Type Tap Number Tap Relative Delay (us) Average relative power (dB) Bad Urban 1   2 5 -3 Obstacle (Building etc) Tx Rx Transmitted signal Received signal Slide 4

GFSK Simulation set up Modulation etc Data Rates Channels: 100 kb/s 500 kb/s Modulation etc GFSK, h = BT = 0.5 Receiver filter Noise Bandwidth = 1.5x Data Rate = approximate signal bandwidth Channels: AWGN Bad Urban Payload Packet Length for PER measurement 100 bytes (appended after 24-bit preamble) 1000 bytes (appended after 24-bit preamble) Up to 8000 packets simulated, each with a random channel snap shot to keep PER variance small Synchronization for Integrate & Dump & data sampling Cross correlation with 24-bit preamble pattern for maximum peak, for every channel snap shot. Used 8 samples per bit after post-demodulator down sampling. Demodulation & Data estimation Non-coherent FM demodulation followed by down sampled synchronized Integrate & Dump over each bit period. Slide 5

Data Rates used for Bad Urban OFDM simulations = 188 kb/s and 750 kb/s OFDM Simulation set up   OFDM Option 1 OFDM Option 2 OFDM Option 3 OFDM Option 4 OFDM Option 5 Unit FFT size 128 64 32 16 8 Active Tones 108 52 26 14 6 # Pilots tones 4 2 # Data Tones 100 48 22 12 Approximate Signal BW 1064 518 264 146 68 kHz BPSK 1/2 rate coded and 4x repetition 98 kbps BPSK 1/2 rate coded and 2x repetition 195 94 43 BPSK 1/2 rate coded 391 188 86 47 BPSK 3/4 rate coded 586 281 129 70 QPSK 1/2 rate coded 781 375 172 QPSK 3/4 rate coded 562 258 141 16-QAM 1/2 rate coded 750 344 187 62 16-QAM 3/4 rate coded 516 Data Rates used for Bad Urban OFDM simulations = 188 kb/s and 750 kb/s Slide 6

PER Simulation Results GFSK OFDM GFSK GFSK in AWGN OFDM GFSK simulations performed by Rishi Mohindra, MAXIM OFDM simulations performed by Partha Murali, Redpine Signals Slide 7

PER Simulation Results OFDM GFSK GFSK in AWGN GFSK in AWGN GFSK OFDM GFSK simulations performed by Rishi Mohindra, MAXIM OFDM simulations performed by Partha Murali, Redpine Signals Slide 8

Bad Urban Channel Statistics for 100 kb/s GFSK Slide 9

Bad Urban Channel Statistics for 500 kb/s GFSK Slide 10

PER with FHSS for GFSK 500 kb/s Looked at channel snap shots (seed values) that produced bit errors in a packet. Then swept the carrier frequency for the above seed values of the channel model, in order to check if frequency diversity (or FHSS) will help for GFSK It was observed that frequency diversity (or FHSS) did not help i.e. PER = 100% over frequency once a bad multipath channel occurs even for very high SNR values of 45 dB This is because in the simple 2-ray model with Rayleigh fading, the channel frequency response is repetitive, producing frequency selective fading for 500 kb/s GFSK signal A more elaborate channel model has to be used for depicting “Frequency Correlation Function” over the operating band, and for determining the improvement with frequency diversity (i.e. FHSS) However, even with FHSS, the expected PER will still be worse than 20% for each channel for the Bad Urban multipath case, and this will require multiple re-transmissions leading to more battery drain compared to OFDM. Slide 11

Conclusions for Bad Urban channels 188 kb/s OFDM has at least 20 to 30 dB SNR advantage over 100 kb/s GFSK in Bad Urban channel conditions for PERs from 10% down to 1 %. The lower the required PER, more the advantage of OFDM over GFSK. 188 kb/s OFDM can operate at 10 dB lower power than 100 kb/s GFSK considering actual receiver power levels Even 750 kb/s OFDM has 12 dB SNR advantage over 100 kb/s GFSK at 1% PER 750 kb/s OFDM can operate at 6 dB lower power than 100 kb/s GFSK considering actual receiver power levels 500 kb/s GFSK will not work efficiently in Bad Urban channels for Gas and Water meters due to PER in excess of 20% that would lead to higher battery drain due to multiple re-transmissions at different hop channels Receiver power level where OFDM operates at 1-10% PER, the GFSK systems will operate at 30-100% PER for similar data rates. GFSK systems will require a very high density of repeaters to cover the range in a mesh network compared to OFDM Slide 12

Appendix Bad Urban Channel models in Matlab (used for OFDM) ADS (used for GFSK) Slide 13

Bad Urban Channel Model in Matlab for OFDM simulations function [y]=two_ray_multipath_channel(x,amp1,amp2,n_FsDelay) % x is the input to the Multipath channel % y is the output of the Multipath channel %amp1 = Average amplitude of 1'st path in dB %amp2 = Average amplitude of 2'nd path in dB %n_FsDelay = Delay in Number of Samples at the sampling rate of x. % If x is sampled at 3125KHz (5x) and n_FsDelay=16 then sample delay of second tap is 5us. RV1=(randn(1,1)+j*randn(1,1))/sqrt(2)*10^(amp1/20); RV2=(randn(1,1)+j*randn(1,1))/sqrt(2)*10^(amp2/20); y=(x*RV1 + [zeros(1,n_FsDelay) x(1:(length(x)-n_FsDelay))]*RV2)/sqrt(10^(amp1/20) + 10^(amp2/20)); OFDM simulations performed by Partha Murali, Redpine Signals (refer to document IEEE 802. 15-09-0486-04-004g, FPP SUN Simulation Results – Redpine Signals) Slide 14

Bad Urban Channel Model in ADS for GFSK simulations Above ADS channel model replicates the Matlab code in the previous slide. Power levels and SNR are set in the top-level simulation after obtaining the ensemble average power Slide 15