doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 1 Project: IEEE P Working Group for Wireless Personal Area Networks (WPANs) Submission Title: [UWB Channel Modeling Contribution from Intel] Date Submitted: [1 July, 2002] Source: [Jeff Foerster and Qinghua Li ] Company [Intel Research and Development] Company [Intel Corporation] Address [JF3-212, 2111 N.E. 25 th Ave., Hillsboro, OR, 97124] Voice [ ], FAX: [ ] Re: [In response to the Call for Contributions on Ultra-wideband Channel Models (IEEE P /208r1- SG3a).] Abstract:[This contribution proposes a UWB path loss and multipath model for assisting in the evaluation of possible UWB physical layer submissions for a high-rate extension to IEEE ] Purpose:[In this presentation, we propose a method for standardizing link budgets to use in comparing different UWB PHY proposals for achieving the desired throughputs and ranges for the standard. In addition, we present some multipath channel measurements that were preformed by Intel and compare these measurements with different channel models that have been considered by the industry for indoor channels. The results suggest a possible UWB multipath channel model that could be used to compare different UWB PHYs.] 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 doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 2 UWB Channel Modeling Contribution from Intel Jeff Foerster and Qinghua Li Intel Research and Development Intel Corporation doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 3 Overview of Talk Summary of channel measurements Link budget and path loss model Multipath model comparisons Path amplitude distribution Discrete time multipath models Proposed multipath model for comparing UWB PHYs doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 4 Channel Measurements Performed by: Robbie Adler, David Cheung, Evan Green, Minnie Ho, Qinghua Li, Cliff Prettie, Leslie Rusch, Keith Tinsley 870 channel realizations between 1-20 m. Pulse based and frequency based measurements doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 5 Channel Measurements Direct sounding of the time-varying channel transfer function Various filters were used with the goal of providing measurements equivalent to pulse measurement using the oscilloscope doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 6 Channel Measurements doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 7 Channel Measurements doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 8 Link Budget and Path Loss Goal: Provide a consistent way to compare the link budget of different PHYs to ensure the throughput and range requirements are properly met. Issues How to make this flexible to accommodate different UWB waveforms (bandwidths, etc.)? How to make this generic that is somewhat independent of implementation (antenna patters, etc.)? doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 9 Link Budget and Path Loss Approach: Is the Friis formula applicable (in any sense) for UWB waveforms [QCOM FCC comment]? Received power spectral density: Received average power doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 10 Link Budget and Path Loss Assume fixed antenna gain with frequency At most 1.5 dB from Friis narrowband model Assume increasing antenna gain with frequency (fixed effective aperature) doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 11 Link Budget and Path Loss Comparisons with measurement loss at 1 meter AT&T results: 47 dB LOS ( GHz, 5 GHz center freq.) Intel results: ~47 dB LOS (2-8 GHz, 5 GHz center freq.) 46.4 dB predicted by Friis at 5 GHz center freq. Results suggest Friis formula can be used to predict 1 meter path loss. Path loss model: Free space + link margin Proposers to provide link margin What should minimum link margin be? doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 12 Link Budget and Path Loss doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 13 Multipath Model Comparisons We compared the following models to the measurements: Path amplitude distributions Rayleigh Log-normal IEEE model with Rayleigh fading Saleh-Valenzuela (S-V) Model -K Model doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 14 Amplitude Distribution Found the ML estimate of the model parameters to match the measurement data at particular delays Used the Kolmogorov-Smirnov test with 1% significance level (one figure of merit) doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 15 Amplitude Distribution Comparison of distributions (1-20m) LOSNLOS doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 16 Amplitude Distribution Standard deviation of log-normal variable Mean of 4.8 dB (relatively constant with excess delay) Similar to other indoor channel models (Hashemi93) doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 17 Amplitude Distribution Observations Rayleigh distribution of each multipath arrival does not appear to match measurement data as well as Log- normaldid not test other distributions Std of log-normal random variable similar to other published results Seems to agree with intuition for UWB signals Smaller number of paths arriving in small bin time (167 psec), so central limit theorem no longer valid to justify complex Gaussian amplitudes and Rayleigh envelopes Paths experiencing several reflections will be multiplied by several reflection coefficients, so log of path amplitude will be sum of independent random variables and ~ Gaussian via central limit theorem doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 18 Multipath Model Comparison Selected three important channel characteristics to use for comparisons between model and measurements Mean excess delay ( m (ns) ) RMS delay spread ( rms (ns) ) Mean number of significant multipath arrivals within 10 dB of peak arrival (Mean NP 10dB ) For S-V and -K Models, used brute force search of model parameters to match channel characteristics doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 19 Multipath Model Comparison IEEE model TDL with 167 psec sample time spacing + Rayleigh fading Simulations of model resulted in > 2x number of paths for fixed RMS delay spread compared to measurements doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 20 Multipath Model Comparison Saleh-Valenzuela (S-V) Model Based on clustering of arrivals with double exponential multipath intensity profile (MIP) Assumes Rayleigh amplitude distribution doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 21 Multipath Model Comparison -K Model Single exponential MIP model with two-state Marchov process for random arrival probability Log-normal fading with 4.8 dB std. NLOS+LOSLOS only doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 22 Multipath Model Comparison Observations S-V Models appears to have the ability to model both LOS and NLOS channels with proper model parameters (flexibility of double exponential MIP model) -K Model can accurately model NLOS channels, but difficult to match LOS characteristics IEEE model is not appropriate for UWB channels doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 23 Proposed Multipath Model Observations of channel characteristics from measurements Channel appears to have clusters doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 24 Proposed Multipath Model Propose a model based on the S-V model with Log-normal amplitudes (not Rayleigh) Key model parameters = cluster arrival rate = ray arrival rate, i.e., the arrival rate of path within each cluster = cluster decay factor = ray decay factor = standard deviation of lognormal fading term (dB) doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 25 Proposed Multipath Model Discrete time model of channel impulse response Should amplitude be real or complex? What does complex mean for UWB pulses? Just corresponds to fine time delay Suggest using real coefficients Assumes Rx is synchronized (same as phase coherent) Paths will be independent with 167 psec min. path spacings (corresponding to inverse of measurement BW) doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 26 Proposed Multipath Model Model parameters for different channel conditions For channels not captured or reflected by measurement data doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 27 Proposed Multipath Model Procedure for generating channel realizations 1. 2. doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 28 Proposed Multipath Model Procedure for generating channel realizations doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 29 Proposed Multipath Model Procedure for generating channel realizations 8. Normalize the total energy of the channel to unity: doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 30 Proposed Multipath Model Possible model options Changing minimum path resolution (changing BW) For T > 167 psec, re-sample impulse response at the lower rate (LPF followed by re-sampler will account for overlap of paths) For T < 167 psec (down to 133 psec =1 / (7.5 GHz)): Replace arrival times in impulse response procedure with times that are multiples of desired T (will not account for possible parameter changes as BW increases). Keep the 167 psec based channel sample rate, and interpolate to lower sample times (may not account for possible larger number of resolvable paths) doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 31 Proposed Multipath Model Possible model options Normalization of multipath energy Rather than normalize each impulse response realizationcould normalize the average over all realizations generated by the model Use complex taps rather than real Propose to use independent, uniform phase with same amplitude envelope generated by model doc.: IEEE /287 Submission July 2002 Intel Research and Development Slide 32 Summary