METIS Second Training & Seminar Smart antenna: Source localization and beamforming Faculté des sciences de Tunis Unité de traitement et analyse des systèmes haute fréquences Ali Gharsallah Email:ali.gharsallah@fst.rnu.tn Page 1
ArrayComm (USA) - installations in WLL - tests for GSM 18 Smart Antennas in Mobile Communications on the Globe GigabitWireless(USA) WLL Ericsson (SW) first system system solution with SA GSM (commercially available) Radio Design AB (SW) NMT-45 IntelliWave Wireless Local Loop System NTT DoCoMo (Japan) Testbed for UTRA Raytheon (USA) Commercially available Fully Adaptive Smart Antenna System UMTS? ARPA (USA )/GloMo project TSUNAMI-SUNBEAM- SATURN/METRA Projects (EU) - Wide range of R&D activity - Recommendations for standardization Metawave (USA) - Field Trials GSM/DCS 18 system Commercially available IntelliCell Switched Beam System
Others applications - Transportation and Navigation - Emergency services - location Sensitive Billing
Summary General introduction of smart antenna Propagation Channel characteristic Smart antenna Architecture applications in indoor and outdoor channels Direction of arrival estimation The data model and Methods classification new methods for the DOA estimation New Beamforming techniques Conclusions and perspectives 4
Radio Channel The signal propagation is corrupted by different obstacles that generate multiple replicas of the transmitted signal at the receiver: MULTIPATH The multipath components have different delay, attenuation, phase shift and angle of arrival. At the receiver the multipath components are summed (incoherently) and originate variation of the signal amplitude: FADING 5
Smart antenna architecture Plane wave x 1 (n) x 2 (n) w* 1 (n) w* 2 (n) w* 3 (n) Σ y(n) output x M (n) M elements w* M (n) DOA estimation adaptive Algorithm METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 6
Smart antennas (indoor) Base Station reflection "2" Direct wave"1" Direct wave "2" Emitter 1 emitter 2 METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 7
Smart antennas (outdoor channel) Cellules Mobile Desired Signal Interfering Signal Base station METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 8
MODEL Antenna array of M elements at BS K tx users L multipaths from different directions for each user The received signal is: s k (n) transmitted signals A kl a( ϑ kl ) x( n) K 1 L 1 = s( n) A a( ϑ) + A s ( n τ ) a( ϑ ) + v( n) k= l= if l= 1 if k k= complex channel attenuation for l-th path and k-th user steering vector for l-th path and k-th user v (n) complex Gaussian noise Desired signal kl k kl kl Interferers + noise= u(n)
directions of arrival Estimation (spectral approaches) METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 1
data model : DOAs Plane waves S 1 S2 Θ 1 Θ 2 Capteurs positions Directions of arrival of Plane Waves emitted by two Source S1 and S2 on an uniform linear array with Five Sensors METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 11
Classification of the methods Methods of estimation DOA Classic Methods Methods with high resolution -Beamformermethods -Capon -MUSIC - Propagateur - ESPRIT 12
Observations data Period of Observation T x 1 (t). x n (t). x N (t) Observations covariance matrix R 1 = T N n= 1 x n x H n x n L ( t ) = a( θ k ) s k ( t ) + η n ( t ) k =1 a( θk ) depends on the position and the geometry of the antenna array. 13
Spectral-based Estimation These methods estimate the DOA by computing the spatial spectrum and then determining the local maxima. MVDR Estimator (Minimum variance distortionless response) This is the Minimum Variance method of spectrum estimation: it assumes the DoA estimate of a point source (assuming all other sources as interferences) as the value ϑ that maximizes the following expression a(ϑ) 1 ( ϑ) = 1 a ( ϑ) R a P H ( ϑ ) is the scanning vector for varying values of angle (it has the same form as the steering vector) π ϑ 2
MUSIC algorithm (Multiple Signal Classification) - Very simple (conceptually) - Based on eigenstructure method - This method relies on the properties of the correlation matrix, R : the space spanned by its eigenvectors may be partitioned into two orthogonal subspaces, named the signals subspace and the noise subspace the signals subspace is spanned by the eigenvectors associated to the L largest eigenvalues and the noise subspace is spanned by the eigenvectors associated with the M-L smallest eigenvalues of the correlation matrix (L directions estimation). the steering vectors corresponding to the directional sources are orthogonal to the noise subspace and are contained in the signal subspace.
MUSIC Complexity Autocorrelation matrix Computed on N snapshots Singular Values Decomposition of correlation matrix Calculate the M-L eigenvectors associated to the noise subspace Frequency Vector scanning the DOA value ~ NM2 6.63 (N) 3 (on NxN matrix) the size increases with the number of snapshots Look up table MUSIC spectrum (M-L) 2 +(M-L) compl. Mul. +(M-L) add
SOME REMARKS A key issue for adaptive arrays in wireless is their behavior in LOS and multipath environments LOS number of antennas (M) greater than total number of signals possible to cancel M-1 interferers, if they are enough separated (!) Multipath impossible to form enough beams and nulls since total number of arriving signals exceeds number of antennas performance of the array depends on the total number of signals
6 Simulation results MUSIC 3 x 111 MUSIC-E 5 2.5 4 3 2 spectogramme 2 1.5 1 1.5 1 2 3 4 5 6 7 8 9 1 angle(deg.) 1 2 3 4 5 6 7 8 9 angle(deg.) Pseudospectre of the elevation angles estimation of two sources located at (8,4 ) et (9,4 ), respectively, with 5 elements on each axes array with SNR =1 db. 18
Simulation results 15 x 14 proposed method MUSIC method 1 spectrogramm 5 3 4 5 6 7 8 elevation angles(deg.) Spectrogram of elevation angle estimation for 4 sources located at (1,5 ),(1,55 ),(1,6 ) and (1,65 ), respectively, with 1 snapshots, SNR=1 db and 8 elements at the z axes subarray 19
A Neural Network-Based on Back-Propagation Algorithm for the Synthesis of Phased Antenna Array METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 2
The objective The objectives of this part are: A general approach only by phase adjustment, to synthesis desired beam and multibeam with zero steered. Neural network based on back-propagation technique to the problem of finding the weights of antenna array. 1 2 1 2 3 4 i K J I 21
Numerical Outils of Synthesis with the phase only Synthesis It is the research of the excitation currents which make it possible as well as possible to approach a given form (desired radiation pattern) Desired radiation pattern In general, definite starting from a gauge specified in module Directing Beam Directing Beam with Zero (Interference) Directing MultiBeam 22
Numerical Outils of Synthesis with the phase only Three possibilities of synthesis: 1- Synthesis in amplitude 2- Synthesis in amplitude and phase 3- Synthesis in phase METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 23
Numerical Outils of Synthesis with the phase only Position of the problem Gauge Systems of nonlinear equations E c ( x, θ ) = E ( θ ) j j d j = 1,..., M. lobe principal (1) (2) (3) (4) c j N i= 1 i ( k. x. ( θ ) + ) E ( θ ) = 2 I cos sin ϕ i j i lobes secondaires lobes secondaires Function of Error : w 1 w 2 w 3 w 4 w 5-9 -6-3 3 6 9 θ en degré max ERR( x) = Min ERR( x, θ j ) j j = 1,..., M METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 24
Numerical Outils of Synthesis with the phase only Radiation pattern (1 beam) Synthesied excitations -4-1 25 35 ϕn ϕn 1 58 15.3-38.2-51.9 2 173.2 46.8-113.8-154.5 3 288.6 77.8-189.8 12.7 4 43.9 19.3 94.5 -.6 5 159.3 14.8 18.9-13.2 6 274.7 172.3-56.7-25.8 7 3.1 23.8-132.3-38.4 8 145.5 235.3-27.9 51 ϕ n ϕn ϕ n ϕ n 25
Numerical Outils of Synthesis with the phase only Time computing A news method based on the neural networks to reduce time How to make? 1 2 1 2 3 4 i K J I METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 26
Synthesis of Antenna array by the Neural Networks To design a model (data-processing or electronic) who can simulate a specified function, with an aim of having a faster method - Structure of a neural network 1 2 1 2 Input 3 Output 4 i J I Input layer Output layer Hidden layer 27
Synthesis of Antenna array by the Neural Networks Type of Neural Network MLP (Perceptron) whose training is based on the back-propagation algorithm Back propagation algorithm the adjustment of the weights of the neural network by the back-propagation algorithm being based on the law of gradient is given by the following iterative equation: w ij ( k +1) = w ( k) 1 N J = 2 k = 1 ij dj ε dw ( ) 2 E ( k ) E d ( k ) The exit of the neural network is consisted the various phases (φi) applied to each element of the network of antennas. ij METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 28
Synthesis of Antenna array by the Neural Networks Procedure of development of a neural network: 1- Collect and analyzes data by (SMART)( 2- Choice of a neural network 3-Base of training and data preparation for a neural network. 4-the phase of training of the network 5- performance METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 29
Synthesis of Antenna array by the Neural Networks Process of data preparation Case of a beam and zero Example of synthesis of radiation pattern of a directing beam (-6 )) and Zero (-3 ) Input 1-1 ϕ 1 ϕ ϕ 2 3 ϕ 4 ϕ ϕ ϕ 5 6 7 ϕ 8 65..9 21.4.7 183.3 294.1 158.6 359.3 176.7 Output 3
Synthesis of Antenna array by the Neural Networks To fix an architecture To fix an architecture Training on basis 1 Training on basis 1 Training on basis 2 Training on basis 2 To To change change architecture architecture No Performance Performance Yes Save Save the the parameters parameters METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 End End 31
Results of simulation Excitations Synthetized by neural networks -28-3 13 3 ϕ n ϕ n ϕ n ϕ n 1 64 33 147 314 2 156.5 338 1 214 3 237 346.5 6 128.8 4 319.5 355.5 19.5 42.5 5 4.5 4.5 319.5 317.5 6 123 13.5 3 231.2 7 23.5 22 26 146 8 296 3 213 46 METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 32
Realization of a neuronal processor Antenna array s n ( t) = I cos( wt) Carte de commande ϕ 1 8 bits Input ϕ 1 DAC + Amp Processeur Neuronal Xilinx Spartan-3 ϕ 8 ϕ 8 Output In ( t) = I cos( wt + ϕ ) n
Design and Realization of an antenna array Antenna (Bande 2.45 GHz) Antenna Array Plan E 51.3cm ε r = 2.5 ±,2 tgδ = 2 2.1 7.2cm 72 mm 9 mm 54 mm 9 mm Module de S11 72 mm 36 mm 36 mm 12 mm 18 mm S11 en db -5-1 -15-2 -25 1,5 2,5 Fréquence en GHz S11 en db -5-1 -15-2 -25-3 Module de S11 2 3 Fréquence en GHz
Design and Realization of an antenna array h = 4 mm.5 mm 72 mm 9 mm 36 mm 72 mm 54 mm9 mm 36 mm 18 mm 12 mm 36 mm Traditional excitation 54 mm 54 mm 72 mm Antenna Array
Design and Realization of an antenna array 2 beams ( -5 et -2 )
Design and Realization of an antenna array 3 beams ( -4,2 et 5 )
Conclusions and perspectives METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 38
Conclusions Smart Antennas advantages: Minimization of the multipath effect Control the radiation of the array in the desired direction and minimize the used Power Implantation of new methods of directions of arrival estimation that give ameliorations to the quality of estimation Presentation of a new Synthesis of Antenna array pattern by a Neural Network METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27 39
Thank You! Page 4 METIS Second Training & Seminar, Cairo (Egypt), 6-7.11.27