UWB Double-Directional Channel Sounding

Transcription

1 2005/09/23 Oulu, Finland UWB Double-Directional Channel Sounding - Why and how? - Jun-ichi Takada Tokyo Institute of Technology, Japan takada@ide.titech.ac.jp

2 Table of Contents Background and motivation Antennas and propagation in UWB UWB double directional channel sounding system Parametric multipath modeling for UWB ML-based parameter estimation Examples

3 UWB Systems Low power Short range Location awareness High resolution in time domain Example applications IEEE a : high speed PAN IEEE a : low speed and location aware Ground penetrating radar

4 Impulse radio < 2ns Simple hardware Low power consumption t

5 Indoor Multipath Environment Rx Tx Rx

6 Transmission in Multipath Environment Tx pulse (500MHz BW) ~ 2ns Rx Rx Tx t Rx pulse ~ 2ns Multipath components can be distinct t ~ 14ns

7 Free Space Transfer Function Friis transmission formula d Tx ΩTx ΩRx Rx H ( f ) = H ( f, d ) H ( f, Ω ) H ( f Ω ) Friis Free Space Tx Tx Rx, 1 f Normalized by isotropic antenna Rx

8 Ideal Antenna Cases Constant aperture size Example : Pyramidal horn H Friis f H Ant f Constant gain Example : Biconical Both are too idealized H Ant = const. H Friis 1 f

9 Frequency Characteristics of Antenna 4.8cm Dipole (resonant at 3.1GHz) Transfer function Frequency dependent Angular dependent

10 Directional Transfer Function of Antenna Drastically changed by direction

11 Directional Impulse Response of Antenna 0.2ns

12 Conventional System vs UWB Antenna and propagation issues Antenna Conventional systems Gain (frequency flat) UWB-IR Distortion Multipath Distortion Distinction

13 Conventional Channel Model IEEE a Model Impulse response realizations Channel includes antennas and propagation Valid only for test antennas (omni)! Time (ns) 20ns

14 Channel Modeling Approach of UWB Rx Tx Rx Antennas and Propagation shell be separated in the model

15 Antenna Model Parameters Directive Polarimetric Frequency Transfer Function H Ant ˆ ( f, θ, ϕ) = θ( θ, ϕ) H ( f, θ, ϕ) + ˆ ϕ θ θ,ant (, ϕ) H ( f, θ, ϕ) ϕ,ant Solid angle Ω = ( θ,ϕ) x z ϕ θ θˆ ϕˆ y

16 How to Get Antenna Model Parameters Electromagnetic (EM) wave simulator MoM (NEC, FEKO, ) FEM (HFSS, ) FDTD (XFDTD, ) Spherical polarimetric measurement Three antenna method for testing antenna calibration

17 Propagation Modeling Rx Tx Rx Double-directional model Direction of departure (DoD) Direction of arrival (DoA) Delay time (DT) Magnitude (polarimetric, frequency dependent)

18 Double Directional Ray Model Rx Tx Rx H Multipath L l= 1 a l ( f, Ω, Ω ) Tx = ( f ) δ ( Ω Ω ) δ ( Ω Ω ) exp( j2πfτ ) Tx Rx Tx,l Rx Rx,l l

19 Double Directional Channel Model has been studied for MIMO systems Tx rich multipath Rx S/P conv. Coding Separation Decoding P/S conv.

20 MIMO Antennas Design of array antenna is a key issue of MIMO channel capacity.

21 MIMO Channel Matrix Rx Tx Rx H ( f ) = Tx Rx Rx antenna array vector H Rx d Ω ( f, Ω ) H ( f, Ω, Ω ) H ( ) Tx f, Ω Rx d Ω Rx Tx Multipath Tx Rx H Tx Tx antenna array vector

22 MIMO vs UWB Antenna and propagation issues Antenna MIMO Array configuration UWB-IR Frequency distortion Multipath Double directional Magnitude Frequency flat Frequency dispersive Propagation modeling approaches are the same.

23 Two different aspects of propagation model Transmission system design Stochastic, site generic Equipment design and installation More deterministic, site specific

24 UWB Channel Sounding Time domain vs Frequency domain Time domain (Pulse) Frequency domain (VNA) Tx Power Large Small Calibration Difficult Easy Data processing Raw data Deconvolution Fourier transform Superresolution (subspace/ml) Resolution Fourier High resolution

25 UWB Channel Sounding Directive antenna vs Array antenna Tx Power Sync. Data processing Resolution Directive antenna Small Timing Raw data Deconvolution Fourier Array antenna Large Timing and phase Fourier transform Superresolution (subspace/ml) High resolution

26 UWB Channel Sounding Real array vs Synthetic array Realization Measurement time Mutual coupling Antenna spacing Real array Multiple antennas RF switch Short To be compensated Limited by antenna size Synthetic array Scanning Long None No restriction

27 UWB Channel Sounding System Vector network analyzer + antenna positioner Measurement of spatial transfer function automatically

28 UWB Channel Sounding System Architecture Frequency domain Synthetic array VNA XY positioner Pros and Cons Short range ~ low power handling Output power Cable loss Antenna scanning Static environment No array calibration

29 Double Directional Channel Model Discrete path model Channel consists of discrete ray paths Ω Ω h f h f f D f D f l( ) = 0(, τl) γββ ( ) r tl rβ (, r rl) tβ (, t tl ), β = ψ, φ β = ψ, φ r t (21.3) Path transfer function Free space path loss Sum with respect to polarizations Excess path loss Rx complex directivity Tx complex directivity Multipath model L H( f) = hl ( f ) l= 1

30 Model of Synthetic Array Complex gain changes due to position 2π f Dr βm ( f, Ωr) = Dr ( f, Ωr)exp j rr ωˆ c r β m r r r = xˆx + yˆy + z ˆz, r r r r mr mr mr m r Position vector ωˆ (21.5). (21.6) = xˆcosψ cosφ + yˆcosψ sinφ + zˆsin ψ. r r r r r r Propagation vector z (21.7) DOA or DOD y ψ φ x

31 Subband Model Ω Ω h( f) = h ( f, τ ) γ ( f) D ( f, ) D ( f, ), l 0 l ββ r tl rβr rl tβt tl β = ψ, φ β = ψ, φ r t (21.9) γ can not be considered as constant over UWB bandwidth. Piecewise constant f

32 Spherical Wave Model For short range paths, plane wave approximation is not appropriate. Spherical wave model Scattering center Spherical wavefront R rl ω rl Coordinates origin Direction of propagation can not be treated as constant.

33 Spherical Wave Model at Rx Array h ( f) = h ( f, τ ) γ ( f) D ( f, Ω ) D ( f, Ω ) lm m 0 l l r rl t tl t r 2π f 2π f exp j Rr l rr m R t r rl exp j tm ˆ l t c r ω. c (21.9) Phase delay correction wrt origin Scattering center Spherical wavefront R rl ω rl Coordinates origin

34 Issue on Spherical Wave Model Not always compatible with doubledirectional model Scattering center seen from Rx Wavefront at Rx Tx array Ray path Rx array Wavefront at Tx Scattering center seen from Tx Compatible case

35 Issue on Spherical Wave Model Not always compatible with doubledirectional model Tx array Wavefront from #1 Rx array #1 Wavefront from #2 Mirror image of Tx array #2 Incompatible case

36 Issue on Spherical Wave Model SIMO and MISO (single-directional) processing Matching by using ray-tracing Accurate time delay due to UWB τ 1 τ 2

37 Channel Parameter Estimation Parametric channel model Free from antenna geometry Resolution still influenced by measurement configuration h ( f) = h ( f, τ ) γ ( f) D ( f, Ω ) D ( f, Ω ) lm m 0 l l r rl t tl t r Two major approaches Subspace based ML based 2π f 2π f exp j Rr l rr m R t r rl exp j tm ˆ l t c r ω. (21.9) c Parameters to be estimated

38 Parametric Channel Model Measured data contaminated by Gaussian noise y = H + n, (21.11) Parameters to be estimated μ mk mk mk r r r I { } r r l li i 1 l l R = l l = γ, ψ, φ,, τ, L μ = μ. l= 1 var( n ) l mk r 2 = σ (21.13) (21.12) z DOA or DOD y ψ φ x τ 1 τ 2

39 Likelihood Function Conditional probability of the observation data assuming parameter set Likelihood function K M 1 p( y μ) exp k= 1 mr = 1 πσ Observed data y = { y 1 m M, 1 k K} ML estimate r mk r mk r =. mk r r r μ maximizing p for given y y H ( μ) 2 σ 2 (21.15) (21.14)

40 Maximum Likelihood Estimation Exhaustive joint search of μ K M 1 p( y μ) exp k= 1 mr = 1 πσ y H ( μ) r mk r mk r =. 2 σ 2 (21.15)

41 Expectation Maximization (EM) Algorithm Estimate of complete data x from incomplete data y (E-step) x = h + b ( y H ). (21.17) l l l ML applied to complete data (M-step) arg max p ( xl μ) = arg min xl hl( μl). μ μ l 2 (21.19) Least square problem to be solved by matched filtering

42 EM Algorithm and Matched Filtering Matched filter detection μ l H a ( μl) xl = arg max. μ H l a ( μ ) a( μ ) l l (21.22) H ai ( μl) xli ˆ γ li =, H a ( μ ) a ( μ ) i l i l (21.23)

43 Space Alternating EM (SAGE) Algorithm Sequential search of parameters ψˆ rl a ( ψ, φ, R, τ ) x H rl rl l l l = arg max, ψ H rl a ( ψ rl, φr l, Rl, τl) a( ψr l, φr l, Rl, τl) H a ( ψˆ rl, φr l, Rl, τl) xl ˆ φ = arg max, rl φ H rl a ( ψˆ ˆ rl, φr l, Rl, τl) a( ψ rl, φr l, Rl, τl) H a ( ψˆ ˆ rl, φ, R ) r l, τ l xl l Rˆ l = arg max, R H l a ( ψˆ ˆ ˆ ˆ rl, φ, R ) ( r l, τ l aψ rl, R ) l φ, rl l, τ l H a ( ψˆ ˆ ˆ rl, φ, ) rl Rl, τ l xl ˆ τ l = arg max, τl H a ( ψˆ ˆ ˆ ˆ ˆ ˆ rl, φ, Rl, τ l) a( ψ rl, φ, Rl, τ l) rl rl (21.24) (21.25) (21.26) (21.27) Good initial estimate is necessary.

44 Successive Cancellation Approach l = 1 Rough global search of l-th path Fine local search of l-th path by SAGE Subtraction of l-th path from observation Convergence? l = l + 1 end No model order estimation In advance.

45 Experiment in an Indoor Environment (1) Measurement site: an empty room Rx Tx Rx Tx (1) X-Y scanner

46 Experiment in an Indoor Environment (2) Floor plan of the room

47 Experiment in an Indoor Environment (3) Estimated parameters : DoA (Az, El), DT Measured data : Spatially 10 by 10 points at Rx 801 points frequency sweeping from 3.1 to 10.6 [GHz] (sweeping interval: 10 [MHz]) Antennas : Biconical antennas for Tx and Rx Calibration : Function of VNA, back-to-back IF Bandwidth of VNA : 100 [Hz] Wave polarization : Vertical - Vertical Bandwidth of each subband : 800 [MHz]

48 Measurement Result (1) The result of ray path identification There 6 waves detected and are almost specular waves.

49 Measurement Result (2) #2 Rx #1 Tx #6 Rx Tx (1) 6 specular waves were observed. Frequency range: 3.1 ~ 10.6 [GHz] Tx, Rx: Biconical antennas Spatial scanning: horizontal plane, points whose element spacing is 48 [mm]

50 Measurement Result (3) Tx Rx Tx #3 Tx #4 #5 Rx #1 Rx #1 Tx (1) Rx #4 is a reflection from the back of Rx

51 Measurement Result (4) Extracted spectrum of direct wave Transfer functions of antennas are already deconvolved. The phase component is the deviation from free space phase rotation (ideally flat).

52 Experiment in an Indoor Environment (4) Comparison of the measurement result in 9 different Rx position The path type detected in each measurement was almost same.

53 Measurement Result (5) Estimated source position for direct wave Maximum deviation is 17cm from source point. Estimated by measurement

54 Measurement Result (6) Estimated reflection points in back wall reflection All the reflection points are above those predicted by GO. Predicted by GO Estimated by measurement

55 Discussion Some problems have been appeared. 2 ~ 4 spurious waves detected during the estimation of 6 waves Residual components after removing dominant paths Signal model error (plane or spherical) Estimation error based on inherent resolution of the algorithm implementation Many distributed source points (diffuse scattering) Further investigation in simple environment

56 Performance Evaluation in Anechoic Chamber Tx1 Tx2 Anechoic chamber Synthesized URA VNA X-Y Scanner 3-dB power splitter GPIB PC GPIB

57 Specifications of Experiment Frequency : 3.1 ~ 10.6 GHz 0.13 ns Fourier resolution Antenna scanning plane : 432 mm square in horizontal plane 10 deg Fourier resolution 48 mm element spacing (less than half 3.1 GHz) Wideband monopole antennas were used Variation of group delay < 0.1 ns within the considered bandwidth SNR at receiver: About 25 db

58 Aim of Anechoic Chamber Test Evaluation of spatio-temporal resolution Separation and detection of two waves that Spatially 10 deg different and same DT Temporally 0.67 ns ( = 20 cm ) different and same DoA

59 Setup of Experiment Rx X-Y scanner Tx

60 Spatial Resolution Test (1) Tx1 Tx2 10 deg

61 Spatial Resolution Test (2) 10 deg separated waves are accurately separated. Parameters and spectra are accurately estimated. The estimated phase denotes a deviation from free space phase rotation (~ 3 mm). Antenna characteristics are already deconvolved.

62 Temporal Resolution Test (1) Tx1 Tx2 20 cm

63 Temporal Resolution Test (2) 0.67 ns separated waves are accurately resolved. Subband width:1.5 GHz Spectrum estimation is impossible in the higher and lower frequency region of Δτ = [ns] 2 = 0.75[GHz]

64 Subband Processing (1) relieves a bias of parameter estimation due to amplitude and phase fluctuation within the band Tradeoff between the resolution and accuracy of parameter estimation: some optimization is needed!! Log-likelihood π Cancel? Frequency

65 Subband Processing (2) How to choose the optimum bandwidth of subband? Suppose two waves are Δθ and Δτ separated θres < Δθ Angle resolution : θ res θres > Δθ Delay resolution τ res τ res < Δτ τ Δτ > res Bandwidth within which deviation of antennas and propagation characteristics is sufficiently small 1 Δτ Impossible to resolve

66 Subband Processing (3) Behavior for the detection of two waves closer than the inherent resolution of the algorithm Regard two waves as one wave (ex. same incident angle) Two separated waves, but biased estimation of power (ex. 5 deg different incident angles)

67 Deconvolution of Antenna Patterns Deconvolution of antennas Construction of channel models independent of antenna type and antenna configuration Deconvolution is post-processing (from the estimated spectrum by SAGE) Simple implementation rather than the deconvolution during the search

68 Spherical vs Plane Wave Models (1) Plane wave incidence (far field incidence) Spherical wave incidence (radiation from point source) How these models affect for the accurate estimation? Spurious (ghost path) and detection of weak paths Empirical evaluation of model accuracy

69 Spherical vs Plane Wave Models (2) Detection of 20 db different two waves Is a weaker source correctly detected? #2 = 15 deg 20 db weaker #1 = 0 deg

70 Spherical vs Plane Wave Models (3) Log-likelihood spectrum in the detection of weaker path True Value 15 deg Plane 5 deg Spurious!! #1 0 deg Spherical 15 deg Correct!!

71 Summary of Evaluation Works (1) Evaluation of the proposed UWB channel sounding system in an anechoic chamber Resolved spatially 10 deg, temporally 0.67 ns separated waves Spectrum estimation is partly impossible in the highest 1 2Δτ and lowest frequency regions of. The algorithm treats two waves closer than inherent resolution as one wave, or results in biased power estimation even if they are separated.

72 Summary of Evaluation Works (2) For reliable UWB channel estimation with SAGE algorithm An optimum way to choose the bandwidth of subband The number of waves estimation is done by SIC- type procedure Deconvolution of antennas effects from the results of SAGE For channel models independent of antennas

73 Summary of Evaluation Works (3) Spherical incident wave model is more robust than plane wave incident model Spurious reduction is expected Effective in the detection of weaker path

74 Indoor Double Directional Measurement (1)

75 Indoor Double Directional Measurement (2) Azimuth-Delay spectrum Tx side Rx side

76 Indoor Double Directional Measurement (3)

77 Indoor Double Directional Measurement (4)

78 Indoor Double Directional Measurement (5)

79 Indoor Double Directional Measurement (6)

80 Summary Background and motivation of double directional sounding Antennas and propagation in UWB UWB double directional channel sounding system Parametric multipath modeling for UWB ML-based parameter estimation Examples

81 References Jun-ichi Takada, Katsuyuki Haneda, and Hiroaki Tsuchiya, "Joint DOA/DOD/DTOA estimation system for UWB double directional channel modeling," to be published in S. Chandran (eds), "Advances in Direction of Arrival Estimation," to be published from Artech House, Norwood, MA, USA. Katsuyuki Haneda, Jun-ichi Takada, and Takehiko Kobayashi, "Experimental Evaluation of a SAGE Algorithm for Ultra Wideband Channel Sounding in an Anechoic Chamber," joint UWBST & IWUWBS 2004 International Workshop on Ultra Wideband Systems Joint with Conference on Ultra Wideband Systems and Technologies (Joint UWBST & IWUWBS 2004), May 2004 (Kyoto, Japan). Hiroaki Tsuchiya, Katsuyuki Haneda, and Jun-ichi Takada, "UWB Indoor Double-Directional Channel Sounding for Understanding the Microscopic Propagation Mechanisms," 7th International Symposium on Wireless Personal Multimedia Communications (WPMC 2004), pp , Sept (Abano Terme, Italy).