March 2003 doc IEEE 802 15 03095 r

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Project: IEEE P 802. 15 Working Group for Wireless Personal Area Networks (WPANs) Submission Title: Ultra Wide-Band Modulation Schemes: A Communications Theory Perspective Date Submitted: March 3, 2003 Source: Eric Ojard and Jeyhan Karaoguz Company: Broadcom Corporation Address: 190 Mathilda Place, Sunnyvale, CA 94086 Voice: 408 543 3320 E-Mail: eo@broadcom. com, jeyhan@broadcom. com Re: [802. 15. 3 a Call for proposal] Abstract: Ultra Wide-Band Modulation Schemes: A Communications Theory Perspective Purpose: [TG 3 a-Broadcom-CFP-Presentation. ] Notice: This document has been prepared to assist the IEEE P 802. 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 P 802. 15. Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Ultra-Wide-Band Modulation Schemes A Communications Theory Perspective Eric Ojard Broadcom Corporation Submission 2 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Introduction • This presentation is a tutorial on some of the options available in designing a PHY protocol for UWB • Key questions: – Channelization for uncoordinated piconets: Frequency Division or Code Division? – How much bandwidth should we use? – Coding Scheme – Modulation & Spreading Options • Various options are considered and the trade-offs are analyzed Submission 3 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Constraints and Requirements • FCC allows use of 3. 1 -10. 6 GHz at – 41 d. Bm/MHz • 802. 15. 3 SG 3 a Target Rates – 110 Mbps @ 10 m required – 200 Mbps @ 4 m required – 480 Mbps @ 1 m desired • Should operate in the presence of 3 other uncoordinated piconets (requires some type of channelization) Submission 4 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Outline • Theoretical Capacity in Low-SNR regime • Coding & Spreading Examples • Uncoordinated Piconets: Frequency Division vs Code Division • Bandwidth Usage • Code Division Spreading Options Submission 5 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Theoretical Capacity slope: 1 bit/s/Hz per 3 d. B log 2(1+SNR) SNR log 2(e) slope: 2 X per 3 d. B 2(1 -H(Q(sqrt(SNR)))) Submission 6 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Theoretical Capacity (cont’d) • The graph can be divided into two distinct regions: – High SNR regime (SNR > 0 d. B or C > 1 bit/s/Hz) e. g. 802. 11 a/b/g, 802. 15. 3 – Low SNR regime (SNR < 0 d. B or C < 1 bit/s/Hz) e. g. UWB • The curves behave differently in the low-SNR and high-SNR regimes: – At high SNR, capacity increases by 1 bit/s/Hz for every 3 d. B – At low SNR, capacity increases by a factor of 2 for every 3 d. B • • • Coding has a much bigger payoff in the low-SNR regime than in the high-SNR regime. At low SNR, binary modulation is optimal – nothing to be gained from higher-order constellations. In theory, very high rates are achievable at very low SNR: – @ -10 d. B, 7 GHz * 0. 15 bits/s/Hz = 1 Gbit/s – @ -13 d. B, 7 GHz * 0. 075 bits/s/Hz = 500 Mbps – @ -16 d. B, 7 GHz * 0. 037 bits/s/Hz = 250 Mbps Submission 7 Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/low_snr_cap_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Theoretical Capacity (cont’d) ½ log 2(1+SNR) 1 -H(Q(sqrt(SNR))) 10 log 10(p/2)=1. 96 d. B 10*log 10(ln(2))=-1. 59 d. B Another way of viewing the same curves: Eb/N 0=SNR/2 R Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Coding & Spreading • The previous slides showed only the Theoretical Capacity. • At low SNR, reduce rate by 2 X for every 3 d. B from the Shannon limit. • It is straightforward to combine well-known binary codes with spreading sequences, as shown in the following slides. – easy to get within 6 d. B of Shannon limit using convolutional codes – possible to get much closer to Shannon limit using concatenated codes and/or iterative decoding. Submission 9 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Coding & Spreading Examples of well-known codes combined with spreading no spreading *code plots assume optimal soft-input decoding Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Coding & Spreading Examples of well-known codes combined with spreading (another way of viewing the same data) no spreading *code plots assume optimal soft-input decoding Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Channelization Options • The solution should support 4 uncoordinated piconets (02/104 r 15) • Channelization Options: – Code Division Multiplexing (CDM) – Frequency Division Multiplexing (FDM) • FDM: 4 non-overlapping frequency bands: – 6 d. B penalty in transmitted power – Additional path loss penalty for high-frequency channels: 20 log 10(4 pfc/c) d. B – Potential for better performance compared to CDM in cases where uncoordinated piconets are very close – Lost immunity to frequency-selective fading is minor (see slides on fading vs bandwidth) • CDM: Each piconet has a different spreading code – allows use of maximum transmitted power – maximum immunity to frequency-selective fading Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Link Margin: FDM vs. CDM Reference: IEEE P 802. 15 -02/490 r 0 -SG 3 a • In the following slides, we consider two cases: – (1) FDM: fmin=8. 725 MHz, fmax=10. 6 MHz (highest-freq channel in a 4 -channel system) • • PT = -8. 27 d. Bm L 1 = 52. 10 d. B – (2) CDM: fmin=3. 1 MHz, fmax=10. 6 MHz • • Submission PT = -2. 25 d. Bm L 1 = 47. 61 d. B 13 Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/link_margin_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Link Margin • From a coding perspective. . . – 480 Mbps @ 1 m is very easy – 200 Mbps @ 4 m is harder – 110 Mbps @ 10 m is the hardest • The FDM system* requires ~10. 4 d. B more coding gain than CDM. *highest frequency band: 8. 725 -10. 6 GHz Submission 14 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Link Margin: FDM vs CDM • (1) FDM: fmin=8. 725 GHz, fmax =10. 6 GHz (highest frequency channel) – – requires S+I <= 6. 2 d. B for 110 Mbps @ 10 m requires strong code & near-optimal receiver very little margin not impossible, but very demanding. • (2) CDM: fmin=3. 1 GHz, fmax =10. 6 GHz – requires S+I <= 16. 7 d. B for 110 Mbps @ 10 m – Very easy, even with weak code & high implementation loss. Submission Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/piconet_interference_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Uncoordinated Piconets w/ CDM dref Desired Transmitter • dint Receiver Interfering Transmitter Assume uncoordinated piconets use the same frequency band with different spreading codes. – Assume true orthogonality isn’t practical due to random multipath & lack of synchronization. – Treat interference as uncorrelated noise with same PSD as desired signal. Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Uncoordinated Piconets w/ CDM • For a CDM system operating within 9 d. B of the Shannon Limit, the target rate of 110 Mbps can be met @ dref/dint=3. 3 • Although strong codes aren’t needed to meet the basic requirements in an interference-free environment, coding gain & receiver performance (S+I) will have a large impact on performance in self-interference environments. – dref/dint increases by 2 X per 6 d. B of coding gain • Regardless of coding gain, CDM systems will never allow uncoordinated piconets “on top” of each other (“near-far problem”). • In theory, FDM could perform much better when uncoordinated piconets are very close. * *FDM isn’t the only way to achieve this. This can be achieved by any method that results in truly orthogonal signals (including multipath effects) at the receiver. Submission 17 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 How Much Bandwidth Should We Use? • Pros of wider bandwidth: – Under the FCC’s UWB regulations, transmit power is proportional to the bandwidth. • But note that the net benefit on link margin is reduced at high frequencies due to L 1=20*log 10(sqrt(4 pfminfmax)) – In a self-interference environment (e. g. uncoordinated piconets using same frequency band). . . • the achievable rate is proportional to the bandwidth. • for a given target rate, dref/dint ~ sqrt(BW) – Better Immunity to Frequency-Selective Fading • Cons of wider bandwidth: – typically results in higher cost, power. Submission 18 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Bandwidth & Fading • One of the key advantages to Ultra-Wide-Band technology is its inherent immunity to frequency-selective fading. • Narrowband signals cannot resolve multipath components; the entire frequency band could fall in a deep spectral null. • The immunity to fading is a function of the ratio of bandwidth to center frequency. ~18 d. B ~20 MHz Example Channel: 20 MHz channel can have 15 -20 d. B fade. Submission Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/bw_fade_test. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Fading Probability vs. Bandwidth CM 1 Submission CM 2 Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/bw_fade_test. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Fading Probability vs. Bandwidth (cont’d) CM 3 Submission CM 4 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Bandwidth & Fading (cont’d) • 15 MHz -> 5 GHz: reduction in 1% worst-case fade: – – CM 1: 15 d. B CM 2: 12 d. B CM 3: 12 d. B CM 4: 9 d. B • Not much difference between 1. 5 GHz curves and 5 GHz curves: anything with BW > ~1 GHz has good fading immunity. • The accuracy of these results is highly dependent on the accuracy of these channel models. Submission 22 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Spreading & Modulation Options • Here we consider modulation schemes where uncoordinated piconets share the same frequency band (CDM) • Several possible variations on DSSS* (not an exhaustive list) – – Long PN Spreading Sequence Symbol-Length Spreading Sequence Multi-Symbol-Length Spreading Sequence Symbol-Length Spreading with Short Time Hopping *DSSS = Direct Sequence Spread Spectrum Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Long Sequence Spreading • Chip sequence is a long (effectively infinite length) pseudo-noise sequence. – Every symbol has a different spreading sequence. – Every uncoordinated piconet has a different spreading sequence. • Advantages – Perfect autocorrelation properties (flat PSD) – Perfect cross-correlation properties with uncoordinated piconets. • Disadvantages – Near-optimal detection requires a high complexity receiver for a large number of multipath components. – Any additional ISI mitigation requires a more sophisticated receiver design Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Symbol-Length Spreading (SLS) Sequence • The chip sequence is the same for every symbol. – Linear Time Invariant (LTI) modulation • Advantages – Lower-Complexity receiver • Disadvantages – Imperfect Autocorrelation: PSD has ripple (assuming binary spreading sequences) – Imperfect Cross-correlation with uncoordinated piconets. • Random multipath tends to provide low correlation, but this breaks down in free-space. – Trade-off between autocorrelation and cross-correlation becomes harder to manage at higher symbol rates. Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Symbol-Length Spreading w/ Time-Hopping • The chip sequence is the same for every symbol. • To reduce the correlation with uncoordinated piconets, symbol positions are dithered by a pseudo-random hopping pattern. • Better Cross-correlation properties compared to plain Symbol-Length Spreading. • Time-varying ISI makes optimal detection more complex. Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Multi-Symbol-Length Spreading Sequence • The chip sequence repeats every N symbols, where N is a small integer. • Better auto-correlation and cross-correlation properties compared to symbol-length spreadspectrum. • Higher Complexity Detection than Symbol-Length Spreading • Time-varying ISI makes optimal detection more complex. Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Comparison of DSSS variations LTI? Constant Baud? Flatness of PSD Decorrelation of Piconets Receiver Complexity Equalzation (if desired) Long Sequence Spreading No Yes Perfect High Very Difficult Symbol-Length Spreading Sequence Yes Fair Lowest Easy SLS w/ Time Hopping Yes No Fair Good to Perfect Low Difficult Multi-Symbol Spreading Sequence No Yes Good Medium Difficult Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Conclusions • In Theory, the UWB environment enables rates far in excess of the target rates. • Supporting 4 uncoordinated piconets is the biggest challenge. – FDM would require very strong coding to meet the target rates, but could perform better when uncoordinated piconets are very close. – CDM could meet target rates with weaker coding, but performance would be limited when uncoordinated piconets are close. • Wider bandwidth always enables higher performance. . . – especially when uncoordinated piconets share the same frequency band – but at the expense of higher complexity. Submission 29 Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Backup Slides Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Channel Models • The proposed channel model for simulations is described in 802. 15 -02/368 r 5. • 3 parts: – Path loss Model – Multipath Model – Shadowing Model Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Path Loss Model L = 20*log 10(d) d. B, where d is in meters Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Multipath & Shadowing • The Multipath Model is a Saleh-Valenzuela model, modified so that multipath gains have a lognormal distribution rather than a Rayleigh distribution. • 4 Multipath Parameter Sets: – – CM 1: 0 -4 m LOS CM 2: 0 -4 m NLOS CM 3: 4 -10 m LOS CM 4: 4 -10 m NLOS • Shadowing: log-normal shadowing with 3 d. B standard deviation. Submission Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/channel_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Example Channels (CM 1 & CM 2) CM 1: 0 -4 m LOS Submission CM 2: 0 -4 m NLOS Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/channel_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Example Channels (CM 3 & CM 4) CM 3: 4 -10 m LOS Submission CM 4: extreme NLOS Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/channel_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Multipath Model Power-Delay Profiles Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Coding & Spreading @ Pe=1 e-8 Examples of well-known codes combined with spreading no spreading *for Pe=1 e-5, shift points left by ~2 d. B Submission Eric Ojard, Broadcom Corp.

*plots generated by function ~/research/uwb/piconet_interference_plots. m March 2003 doc. : IEEE 802. 15 -03/095 r 1 Uncoordinated Piconets w/ CDM dref Desired Transmitter • dint Receiver Interfering Transmitter Assume uncoordinated piconets use the same frequency band with different spreading codes. – Assume true orthogonality isn’t practical due to random multipath. – Treat interference as uncorrelated noise with same PSD as desired signal. Submission Eric Ojard, Broadcom Corp.

March 2003 doc. : IEEE 802. 15 -03/095 r 1 Uncoordinated Piconets w/ CDM • For a CDM system operating within 9 d. B of the Shannon Limit, the target rate of 110 Mbps can be met @ dref/(dref+dint)=0. 77 • Although strong codes aren’t needed to meet the basic requirements in an interference-free environment, coding gain & receiver performance (S+I) will have a large impact on performance in self-interference environments. • Regardless of coding gain, CDM systems will never allow uncoordinated piconets “on top” of each other (near-far problem). • In theory, FDM could perform much better when uncoordinated piconets are very close. * *FDM isn’t the only way to achieve this. This can be achieved by any method that results in truly orthogonal signals (including multipath effects) at the receiver. Submission 39 Eric Ojard, Broadcom Corp.