doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 1Rishi Mohindra, MAXIM Project: IEEE P Working Group for Wireless Personal Area Networks (WPANs) Submission Title: FPP-SUN Implementation Considerations Date Submitted: July 5, 2009 Source: Rishi Mohindra, MAXIM Integrated Products Contact: Rishi Mohindra, MAXIM Integrated Products Voice: , 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 P 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 P Slide 1 doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 2 Future Proof Platform for SUN: Implementation Considerations Rishi Mohindra IEEE 802 Plenary Session San Francisco July 2009 Supporters: Emmanuel Monnerie [Landis & Gyr], Partha Murali [Redpine], Steve Shearer [Independent], Shusaku Shimada [Yokogawa Electric Co.], Bob Fishette [Trilliant], Sangsung Choi, [ETRI], Roberto Aiello [Independent], Kendall Smith [Aclara], David Howard [On-Ramp] doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 3 Contents OFDM as Future Proof Technology Path loss OFDM implementation considerations Zero-IF Transceiver Architecture advantages Low-IF receivers for GFSK Phase Noise Receiver dynamic range Crystal Tolerance and Coarse + Fine Frequency Error Correction Spectral mask & PA backoff Battery Life Calculations Legacy Meter Reading example use cases Calculations Gate count and relative cost Conclusions doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 4 Legacy Meter Reading technology issues: Different frequency bands, modulations, network topologies, interference conditions, MACs, Network Layers etc. No single existing flavor can work in all situations Data rates & modulation techniques are old and there is lack of advantages that are offered by todays modern VLSI digital technologies. Digital gate count and signal processing capabilities reflect 15 year old 0.35 um CMOS technology. No existing deployment technology offers the possibility of future proofing through improved data rates and range Poor spectral efficiency (bits/s/Hz) Possible misconception that simple is cheap without a broader vision of modern technology advantages relating to cost and performance The g FSK proposals do not fully resolve the above issues, and will largely provide a new face to old technology with commonality and legacy compatibility as the driver instead of Future Proofing. doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 5 OFDM as Future Proof Technology Physical Layer feature requirements 40kb/s to 1Mb/s through put data rates at nodes 20 year battery life for all non-AC powered nodes Cost Requirements Solution cost to be comparable to legacy solutions while supporting significantly higher data rates 1Mb/s solution cost should only be a small increment to the 40kb/s BOM Special Network Requirements for Gas & Water meters Not depend on AC powered nodes for majority of repeaters or Hops. Support max # nodes (e.g nodes) per BS in a MESH or STAR network 140 dB desirable link budget for up to a few miles range Should not be limited in range due to multipath conditions i.e. work well in rms delay spreads in excess of 10 usec. Special Requirements for MESH networks with AC powered hops Support 10x increase in network throughput with 100 kb/s link data rates at battery operated nodes. Allow upgrade paths for up to 1 Mb/s link data rates between AC powered nodes or hops for improved network capacity in the future Only OFDM offers a common future-proof technology that is scalable according the above needs doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 6 A look at Path Loss Path Loss analysis set up: Freq = 900 MHz MESH: Transmitter and Receiver antenna heights = 0.5m STAR: Base Station antenna height = 8.9m, nodes antenna height = 0.5m Path Loss models: Free Space 2-ray LOS with loss-less ground reflection for STAR and MESH Hata urban for STAR and MESH Hata suburban for STAR Hata rural for STAR doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 7 Path Loss Models doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 8 Hata Path Loss data Freq = 1.8GHz, MS height = 2m, BS height = 1m Typical urban in Southern England X-axis: 3 = 1000m doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 9 Path Loss Conclusions Path Loss at 1km range based on Hata models: Urban MESH = 154 dB, while Urban STAR = 137 dB due to antenna height difference Suburban STAR = 127 dB Rural STAR = 108 dB Rural STAR (Hata) matches the 2-ray LOS model for larger distances. Allow >20dB extra loss for gas/water meter at below ground level Hata model uses narrow band measurements & is prone to more spatial fading fluctuations Using wider modulation bandwidths (up to 1MHz for OFDM) will drastically reduce the deep spatial fades and reduce link budget margin requirements Implementation considerations to combat path loss: use a modulation that requires lowest S/N, e.g. OFDM BPSK requires 1dB S/N versus 11 dB for GFSK. Minimize receiver noise figure & Modem Implementation loss. OFDM has only 0.5dB modem implementation loss (in addition to receiver noise figure). GFSK has 2-3 dB implementation loss (strongly phase noise and filter ISI dependent) without time- domain equalizer. Total Receiver Loss = Noise Figure + Implementation Loss. Total OFDM Receiver Loss can easily be kept below 4dB in 0.13 um RF CMOS technology at very low battery current doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 10 OFDM Data Rates Scalable Option #12345 Occupied Bandwidth 1.06 MHz 518 kHz 264 kHz 146 kHz 68.4 kHz Base FFT Modulation/CodingSNRData Rate, kb/s BPSK r=1/2 rate, 4x repetition BPSK r=1/2, 2x repetition BPSK r=1/ BPSK r=3/ QPSK r=1/ QPSK r=3/ QAM r=1/ QAM r=3/ SNR shown for 1000 bytes 10% PER in AWGN doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 11 Receiver Sensitivity for 4 dB total loss Scalable Option #12345 Occupied Bandwidth 1.06 MHz 518 kHz 264 kHz 146 kHz 68.4 kHz Base FFT Modulation/CodingSNRReceiver Sensitivity, dBm BPSK r=1/2 rate, 4x repetition BPSK r=1/2, 2x repetition BPSK r=1/ BPSK r=3/ QPSK r=1/ QPSK r=3/ QAM r=1/ QAM r=3/ doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 12 Receiver Sensitivity for 4 dB total loss doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 13 Link Budget for 30 dBm transmitter & 4 dB receiver loss Scalable Option #12345 Occupied Bandwidth 1.06 MHz 518 kHz 264 kHz 146 kHz 68.4 kHz Base FFT Modulation/CodingSNRLink Budget, dB BPSK r=1/2, 4x repetition BPSK r=1/2, 2x repetition BPSK r=1/ BPSK r=3/ QPSK r=1/ QPSK r=3/ QAM r=1/ QAM r=3/ doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 14 Zero-IF Architecture for OFDM Key Features Direct conversion between RF and IQ baseband Same RF Synthesizer frequency between Transmit & Receive: allows fast turn-around for ACK etc Sharing of IQ Low-pass between transmitter & receiver (due to Time Division Duplexing) means smaller die area doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 15 Zero-IF Receiver Architecture for OFDM Advantage of Zero-IF No Image rejection issues as far as Blockers are concerned. Low-pass filter at 0 Hz IF frequency consumes significantly lower current than Low-IF Polyphase filters, and also has fewer active stages in filter DC notch has no impact on OFDM since there is no DC-subcarrier. Notch is accomplished by digital DC cancellation every packet (OFDM STF field is DC balanced in IQ, unlike GFSK that requires a blank period for DC calibration) Architecture proven to consume lowest current Uncoupled I & Q signal paths allows unlimited Digital Image Rejection calibration Possible disadvantages 1/f flicker noise in IQ base band signal paths can slightly increase noise figure of subcarriers close to DC dB f0 Blockers Low-pass Channel Filter response Spectrum at Base Band DC doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 16 Analog Zero-IF + Digital Low-IF Receiver Architecture for OFDM dB f0 Analog Low-pass Anti-aliasing Filter response (before A/D) Spectrum at Base Band and Analog+Digital Channel Filtering DC 0 >80 dB Digital Image Rejection in real-time over Temperature Digital Polyphase Channel Filter response (after A/D) Analog DC notch Not required Blockers doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 17 Detailed Zero-IF OFDM Receiver Core example doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 18 Low-IF Architecture for FSK & ASK Disadvantage of Low-IF Poor Image rejection over temperature even after calibration (may not be an issue in ISM bands) Complex Polyphase filter consumes significantly higher current than low-pass filters due to higher operating IF frequency, and more active stages in filter DC notch can have an impact on Group Delay flatness that can degrade ISI for GFSK Significantly larger die area for a given data rate compared to OFDM zero-IF receiver Slow Transmit-Receive switching due to RF Synthizer frequency programming & settling Why used because FSK receivers cant work well with Zero-IF IQ DC offsets Mitigate close-to-DC 1/f flicker noise in IQ base band signal paths doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 19 Low-IF Architecture for FSK, ASK & DSSS dB frequency 0 Image Rejection Blocker at Image Spectrum at Base Band Polyphase Channel Filter response DC DC notch doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 20 Local Oscillator Phase noise impact on OFDM EVM, with Symbol Time Scaling Conclusion: 16-QAM OFDM tolerates Phase Noise that is 15 dB relaxed compared to GFSK transceivers Results in significant Battery current reduction Out-of-band blocking will largely influence phase noise BPSK OFDM requires10dB smaller C/I compared to GFSK BPSK OFDM tolerates 10dB higher phase noise for same blocker and Rx sensitivity Option 2 Option a Such good EVM is Only needed for 256-QAM doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 21 Receiver Dynamic range for Active Channel Filter Output Non-coherent GFSK receiver channel filters require 8 13 dB higher filter dynamic range compared to OFDM receiver Extra current savings possible for OFDM receiver filter Noise Floor OFDM BPSK r=1/2 OFDM QPSK r=1/2 GFSK BT=0.5 f=0.25R b Non-coherent dB Receiver Saturation Level 1 dB >2 dB 4 dB >3 dB Receiver Saturation Level >2 dB DR > 3 dB DR > 7 dB 13 dB DR > 15 dB doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 22 Complete OFDM Receiver Dynamic range & Noise Figure based on FSK receiver specifications (16-FFT case) Noise Figure (3.2 dB), IP3 (-23 dBm), Phase Noise and blocking specs kept same as FSK receivers Equivalent receiver performance is shown above It is able to support even 64-QAM OFDM ! Over-kill for SUN applications with BPSK to 16QAM Total effective S/(N+D) curve Performance looks good for 64-QAM OFDM ! doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Slide 23 Crystal Tolerance and Short Training Field Relative Frequency error due to crystal Local Oscillator Only Null STF Subcarriers lost due to frequency error prior to frequency error correction These are used in the LTF and Symbols, and are shifted back inside channel filter after digital frequency correction Receiver digital channel filter before frequency correction Carrier Leakage Receiver digital channel filter after frequency correction Crystal tolerance is not an issue since many STF null subcarriers can be lost due to frequency error before the Freq Error Estimation algorithm is affected. 40 ppm tolerance is OK for 500kHz OFDM bandwidth. doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 24 Integer and Fractional Frequency Errors freq freq1 freq2 Local Oscillator Short Training Field Subcarriers Null subcarriers Center Null subcarrier Carrier Leakage Total frequency error freq = freq1 (integer error) + freq2 (fractional error) doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 25 Frequency Error Estimation sequence Correlation + AGC Coarse Freq error correction + DC cancellation Short Training Field (10x repetition)Long Training Field (2x repetition)Symbols Fine Freq error correction + Channel Estimation time Delay A* conjugate X Mean STF Fractional freq error FFT Subcarrier Correlation Integer freq error doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide kHz OFDM Modulation bandwidth meets FCC & ETSI Mask Compliance for non-FHSS ETSI allows 25mW transmit power in a 600kHz bandwidth at 868.3MHz Zigbee operates in this band due to the allowed power and bandwidth Spectral plot overlaid with the ETSI emission mask is shown next slide About +-43ppm frequency error can be tolerated at 868MHz before violating the ETSI mask that is fixed at a channel center frequency of 868.3MHz Allowed ETSI transmitted power level in 515kHz bandwidth is +13.3dBm. FCC requires 500kHz min modulation bandwidth for non-FHSS. Up to +30dBm transmit power allowed. Should meet -41dBm/MHz emission limit in Restricted Bands at 2390MHz & 2483MHz. Plot shows full compliance doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 27 ETSI Spectral Mask and PA backoff +13 dBm output power PA = Rapps model with Rho = 3 doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 28 Rapps PA model. 6 dB back off from saturation power for PA, rho = 1. At MHz channel, meets requirement for 2390MHz restricted band (10 MHz from carrier). At 2477 MHz channel, meets requirement for 2483 MHz restricted band (6 MHz offset from carrier). dBm/MHz dBm/10kHz doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 29 EVM versus PA Backoff 16-QAM, r= QAM, r=0.5 PA = Rapps model with Rho = 3 PA efficiency can be 25% or higher for BPSK & 16-QAM OFDM due to small back off from saturation to meet EVM requirement doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 30 ETSI Emission Mask at MHz (fe = 300 kHz) doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 31 ETSI allowed transmitted power level (1% duty cycle alowed with LBT) doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 32 ARIB Emission Mask for 954 MHz doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 33 GFSK requires 20 dBm output for the same link performance in AWGN as a 10 dBm OFDM transmitter. doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 34 Legacy Meter Reading example use cases for battery life calculation STAR topology cases: 50000 nodes per BS, 12.5 kHz channels in Licensed bands, GFSK, 30 dBm Tx, expensive crystal with TBD ppm life time accuracy, slow real time frequency error calibration at nodes for sub-ppm timing accuracy, slotted time structure with no collision, 6 hour transmission gaps per node. Lower PA efficiency due to ramp-time for spectral compliance. Light MESH topology cases: 25000 nodes per network with one concentrator point, up to 3 hops from end-nodes to concentrator point, ISM band FHSS GFSK, hops over AC powered devices only (1 per sq mile). IEEE802.11n WLAN at 2.4GHz with 20 year battery life, 2 mile network radius per concentration point, hops over AC powered devices. MESH topology cases: 50000 nodes per network including up to AC powered hop-nodes, FHSS GFSK or MSK. 2000 nodes per concentrator, multiple hops. Multiple nodes per home. 2.4GHz ISM band, IEEE OQPSK DSSS, 250 kb/s, 30dBm Tx at BS, 24 dBm Tx at gas & water nodes. Rake receiver. 40 ppm crystal tolerance. 2 mile network radius. doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIM Battery Life Assumptions for STAR network seconds Frame Interval kb/s data rate for OFDM and GFSK 250 ms Receiver ON time per Frame (in addition to payload receive bursts) One transmit and one receive payload burst per frame 50 bytes Transmit and Receive payload data per burst (excluding PHY overheads) OFDM transmit power +24 dBm, GFSK transmit power +30dBm for similar AWGN link budget 25 years battery self-discharge time Background application programs and synchronization timers running Current consumptions accounted for: Transmitter (analog + digital) Receiver (analog + digital) MAC Application (running continuously with 10% duty cycle) Synchronization timer (running continuously) Battery self discharge Slide 35 doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 36 OFDM Battery Life Calculations (based on STAR network) Symbol time128usec Non-data symbols per packet6 # total Tx symbols =75=9.6ms # total Rx symbols =75=9.6ms Total Rx current =19.6mA Total Tx current =365mA Data Rate46.875kb/s Battery capacity2400mAH Battery voltage3V # Tx data bytes50 Tx power250mW PA efficiency25% # Rx data bytes50 Rx search time250ms Repetition interval22000s Application program10% duty cycle # Bursts per interval1 Computed Battery life = 20years doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 37 OFDM mAH contributions (based on STAR network) On time per interval (sec) Average Icc (mA) % mAH BlocksIcc (mA)VccRemarks Analog Phy: Xtal Osc + bandgap0.532ms settle E RF PLL+VCO1030.1ms settle Rx path + IQ ADC E Tx path + IQ DAC E PA Digital Phy: Dig: Rx STF correlator + AGC13During Rx search time E Dig: Rx demod13During Rx burst time E Dig: Tx13During Tx time E MAC + timer + Application processes: MAC: Rx mode E MAC: Tx mode E Timer Application Total silicon current Battery Self discharge Self discharge current0.011For 25 years shelf life Net available capacity over life21% Total battery discharge + silicon current doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 38 GFSK Battery Life Calculations (based on STAR network) Data Rate46.875kb/s Battery capacity2400mAH Battery voltage3V # Tx data bytes50 Tx power1000mW PA efficiency48% # Rx data bytes50 Rx search time250ms Repetition interval22000s Application program10% duty cycle # Bursts per interval1 Computed Battery life = 19years Bit time21.3usec # training + preamble (non data) bits88 # total Tx bits =488=10.411ms # total Rx bits =488=10.411ms Total Rx current =19.8mA Total Tx current =720mA doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 39 GFSK mAH contributions (based on STAR network) On time per interval (sec) Average Icc (mA) % mAH BlocksIcc (mA)VccRemarks Analog Phy: Xtal Osc + bandgap0.532ms settle E RF PLL+VCO1430.1ms settle Rx path E Tx path E PA Digital Phy: Dig: Rx correlator + AGC0.23During Rx search time E Dig: Rx demod0.23During Rx burst time E Dig: Tx0.53During Tx time E MAC + timer + Application processes: MAC: Rx mode E MAC: Tx mode E Timer Application Total silicon current Battery Self discharge Self discharge current0.011For 25 years shelf life Net available capacity over life22% Total battery discharge + silicon current doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 40 OFDM data rate = kb/s using BPSK r=1/2, 24dBm transmit power GFSK data rate = kb/s, BT=h=0.5, 30dBm transmit power Same AWGN link budget for OFDM and GFSK X-axis = #Tx = #Rx bursts per seconds with 50byte payloads doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 41 OFDM data rate = kb/s using 16-QAM, r=1/2, 24dBm transmit power GFSK data rate = kb/s, BT=h=0.5, 30dBm transmit power Assuming OFDM has 8 dB multipath advantage over GFSK as an example X-axis = #Tx = #Rx bursts per seconds with 50byte payloads doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 42 SoC Gate Count, Die Area & cost Note: OFDM digital gate count provided by Redpine Signals Cost GFSK SoC costs 1.2x more than OFDM Soc RF Power Amplifier + associated electronics cost: GFSK 30 dBm transmitter costs 5x more than OFDM 20 dBm transmitter doc.: IEEE g Submission July 6, 2009 Rishi Mohindra, MAXIMSlide 43 Conclusions Advantages of OFDM over GFSK implementations for similar data rates Total chip set costs less Tolerates significantly higher phase noise Potentially use much cheaper crystals Requires simpler RF Transceiver architecture SoC has smaller die area Battery life is better and potentially 2x longer 10 dB link budget advantage in AWGN Potentially 20 to 30 dB link budget advantage in frequency selective fading conditions where GFSK may still be operable