Overview. Lecture 7: Smart Antennas. Part I. Overview (cont d) What is a Smart Antenna. Motivation. Smart Antennas in Software Radios

Transcription

1 Overview Lecture 7: Smart Antennas (Introduction to Smart Antennas for Software Radio) Part I 7-1 Introduction Definition, benefits, comparison with diversity technique Fundamental of smart antennas Array steering vector, array calibration, multiple fixed beam antenna Diversity combining techniques MRC, EGC, SC Beamforming concepts Array geometry, multipath channel model, types of beamforming Beamforming algorithms Array fundamentals, MMSE, MSINR, MVAR, Least Square Blind adaptive algorithms 7-2 Overview (cont d) What is a Smart Antenna Transmit diversity Classification, applications in practical communication system MIMO (Multiple Input Multiple Output) Spatial multiplexing, BLAST Space-time processing 2-D rake receiver, interference cancellation Channel model Geometrical models, statistical models Network level issues Impact of diversity/beamforming on the capacity, coverage Hardware considerations 7-3 Definition Antenna array system aided by some smart algorithm to combine the signals and designed to adapt to different signal environments Algorithms can be implemented in DSP The antenna can automatically adjust to a dynamic signal environment The gain of the antenna for a given direction of arrival is adjustable Some antennas are smarter than others 7-4 Smart Antennas in Software Radios Motivation Software radios and smart antennas complement each other Software radios need to adapt to different protocols, systems and channel environments Smart antennas aid software radios in attaining this flexibility through the use of signal processing algorithms to combine the received signals in an optimum manner Smart antennas provide the benefits that motivate the adoption of software radios Implementation of smart antenna algorithms require flexibility in the infrastructure which is provided by software radios 7-5 Recent years have seen a great proliferation of wireless systems for higher data rates and more capacity PCS: IS-95, GSM, IS-136 ( kbps) 2.5G: GPRS, EDGE ( kbps) 3G: HDR, HSDPA (2 Mbps) WLAN: IEEE , Bluetooth, Home RF (30 kbps 50 Mbps+) Interference environment becomes increasingly hostile as the number of active devices and standards increases Capacity limitation Interference is a big issue in migration of technology Advanced methods are necessary to overcome these problems 7-6 1

2 Benefits of Smart Antennas Benefits of Smart Antenna Co-channel (jamming) and adjacent channel interference reduction Multiple access interference reduction for capacity improvement Robustness against multipath, fading, and noise to improve coverage and range Reduced power consumption Lower probability of interception and detection Enhance location estimates Interference Reduction Omni Antennas Smart Antennas By reducing co-channel and adjacent channel interference, frequency reuse factor can be reduced and more channels can be used at a cell site thus increasing capacity Spatial Division Multiple Access (SDMA) Benefits of Smart Antenna for Base Stations Some channels might be able to be reused even in the same cell Intercell Interference Multipath Uplink Base Station Multipath Downlink JAMMER Mobile Signal Fading JAMMER 7-9 Benefits Co-channel (jamming) and adjacent channel interference reduction Multiple access interference reduction for capacity improvement Robustness against multipath, fading, and noise to improve coverage and range Reduced power consumption for the handset Lower probability of interception and detection 7-10 Enhance location estimates Benefits of Handheld Smart Antenna Basic Types of Diversity Anti-jam capability Extended range Reduced fading Lower transmit power Increased battery life Low probability of intercept Increased spectral efficiency Increased capacity 7-11 Time Multiple versions of the same signal can be combined to produce a higher-fidelity estimate of the original signal This is the basic concept behind Rake receivers Space/polarization Receive elements (antennas) placed in different locations receive the same signal Different copies of the same signal can be combined to produce a higher-fidelity estimate of the original signal This is the basic concept behind smart antenna Frequency Multicarrier systems such as Orthogonal Frequency Division Multiplexing (OFDM) Useful for high data rate systems in multipath environments

3 Spatial Diversity: Smart Antenna Categories Antenna Diversity Principles Switched beam systems Multiple fixed beams using array of antennas Tx/Rx picks beam offering greatest signal enhancement and interference reduction/rejection Used to improve widely deployed cellular systems; increase base station range Adaptive antenna systems Array of antenna elements that can change gain pattern Adjust to noise, interference, multipath and to mobile users More fully integrated approach Requires additional BS station equipment and most Multiple antenna sensors provide diversity signals Independent copies of the same signal that experience different fading increase the probability of having a usable signal at any instant Diversity is effective if the two signals are uncorrelated to a level where the correlation coefficient is approximately 0.7 or less Space Possible diversity dimensions: Polarization Pattern (amplitude and/or phase) effective when the commun. std supports adaptive arrays Different Spacing in Different Channels Diversity vs. Beamforming Diversity combining Adaptive beamforming Transmitter Free Space Receiver Diversity does not help and is not needed d Antenna spacing must be several wavelengths Narrow Angle Spread e.g., mobile to base station d Spacings on the order of 0.25 λ works well Wide Angle Spread e.g., mobile, handheld, peerto-peer, indoor 7-15 Combine signals from different antenna elements using various algorithms Signal from each element is processed separately Signals have to be uncorrelated for maximum performance Mitigates fading Increases gain Can improve polarization match Focus the antenna s gain in the direction of the desired signal Achieved by (complex) scaling each antenna input Antenna elements have to be separated by λ/2 Signals are correlated All advantages of diversity combining 7-16 Basics of Antenna Arrays Fundamental of Smart Antennas Steering vector Response of an array to a single plane wave Narrowband array model Used to describe difference in signal received at one antenna relative to the other Small time delay between received signals at different antenna elements modeled as phase shift Delay should be small compared to the inverse of the signal bandwidth

4 Array Steering Vectors Array Steering Vector Incident wave Incident wave Incident wave fronts d k sin Incident wave fronts d k sin 1 k d k Received bandpass signal at antenna 1 1 k d k Received bandpass signal at antenna k : Antenna gain : Measured noise : Incident signal with channel distortion Complex envelope of 7-19 Narrowband array approximation: - Assume complex envelope is varying slowly relative to the carrier period - Small time delay (τ k ) can be modeled as simple phase shift (ψ k ) if delay is small compared to the inverse bandwidth of the signal Complex envelope of 7-20 Array Steering Vector Array Calibration Complex baseband envelope (of the received signal) at the kth element In vector notation Array steering vector a(φ) Describes mapping between Angle Of Arrival (AOA) and the array response Defined completely by array geometry and gain patterns of individual antenna elements In practice, can vary from the theoretical value RF front end inconsistencies between different Array Calibration Process of determining a(φ) as a function of φ Needed to ensure good performance of signal processing and beamforming algorithms Transmitter is placed at a known angle φ and a(φ) is estimated from received data Sources of error Steering vectors can vary with carrier frequency Drift with time, temperature or other environmental factors branches Array Calibration Techniques Remote Transmitter Approach Analytical methods Based on mathematical equations Numerical methods Based on computer simulation Experimental methods Based on measurement data Most accurate and reliable approach Remote transmitter approach and test tone approach Rotate array relative to a fixed source at closely spaced angles Array response at each angle is recorded and normalized to estimate steering vectors φ i Array

5 Test Tone Approach Test Tone Approach RF signal input directly into each array element for calibration Differences in array hardware causes variation in the phase of the output signal Signal injected into the kth array branch = x k (t) Output y k (t) = c k x k (t) Ideal splitter, x 1 (t) = x 2 (t)= x M (t) Splitter x 1 x 2 x m array hardware c 1 y 1 c 2 y 2 c m array outputs y m Only response of each branch relative to the response of a reference element is needed AOA estimation depends on relative phase responses, not on absolute phases Set k =1 as reference element Response of branch k relative to branch 1, < > denotes time averaging Relative phase differences can be compensated for by multiplying by r k w RF signal generator Array Ambiguity Spatial Sampling Theorem Incident wave fronts Array ambiguity exist if d k sin 1 k d k Ambiguity occurs when φ 2 = π - φ 1 since Incident wave 1 Incident wave 2 Ambiguity resolved by shaping the gain patterns of individual elements Ambiguities will not exist if every View spatial samples as discrete time samples of the received signal with Sampling interval: Spatial sampling frequency: Minimum sample rate occurs when or From Nyquist s sampling theorem,, where is the carrier frequency Spatial Nyquist theorem: Multiple Fixed-Beam Antenna Array Multiple Fixed-Beam Antenna Array Different elements experience different fading envelopes Diversity systems attempt to select or combine the signals at the antenna elements to increase the received signal power Common diversity algorithms Selection Maximum ratio combining Equal gain combining Optimum combining Multipath mitigation and range extension is the main goal Exploit low correlation between temporal signal fades at different antenna elements Provide range extension and enhanced coverage Reduction in multipath delay spread possible due to discrete directivity reducing multipath signals Complexity in signal processing is less and hence it is cost-efficient At close angle of arrivals the system will not be able to discriminate multipath

6 Multiple Fixed-Beam Antenna Array Multiple Fixed-Beam Antenna Array Selection diversity Selection of one of M antenna elements per-user basis The selection criterion is usually the largest SNR Requires beamforming network or directional antennas, RF switch and switch control logic to implement selection diversity Equal gain combining Co-phase each element output and add the resulting signals Maximal ratio combining Weigh each sensor output by its SNR before combining ~1 db improvement relative to equal combining, however it is more complex Optimum combining Optimize the weighting based on some criterion (cost function) such as minimum mean squared error or constant modulus 7-31 Advantages Provides range extension and enhanced coverage Some reduction in multipath delay spread may be possible due to discrete directivity Less complex signal processing Cost-efficient Disadvantages Cannot discriminate multipath with close angle of arrivals Cannot coherently combine multipath to take advantage of spatial diversity Received power level sensitive to the user terminal movement 7-32 Summary Diversity vs. Beamforming Introduce the general concept of smart antennas Definition, benefits, comparison with diversity techniques Review the fundamental of smart antennas Array steering vector, array calibration, multiple fixed beam antenna Next topic Diversity combing techniques Beamforming concepts 7-33 Diversity combining Combine signals from different antenna elements using various algorithms Signal from each element is processed separately Signals have to be uncorrelated for maximum performance Mitigates fading Increases gain Can improve polarization match Adaptive beamforming Focus the antenna s gain in the direction of the desired signal Achieved by (complex) scaling each antenna input Antenna elements have to be separated by λ/2 Signals are correlated All advantages of diversity combining 7-34 Diversity vs. Beamforming Linear combination of each branch Diversity Combining Techniques MRC 60 EGC SC Beam (MMSE) 60 Desired signal 0.8 Interference Selection Diversity - Equal Gain Combining - Maximal Ratio Combining

7 Selection Diversity Switched Diversity Uses only the strongest decision statistic Shows greatest SNR improvement when Desired signal undergoes uncorrelated Rayleigh fading at each antenna element Background noise process are AWGN and uncorrelated across antenna elements SNR improvement with M branches Receiver does not switch antenna elements until current antenna element undergoes deep fade Requires only single RF chain to serve all antennas Mean SNR Scanning Diversity This type of diversity combining is popular in switched beam antenna systems Equal Gain Combining Maximal Ratio Combining Equal Gain Combining This is a good choice only if we know that all components are equal in magnitude Signals have to be co-phased before summing Otherwise, weak components can contribute a lot of noise but little signal to the overall decision statistic Mean SNR In the presence of uncorrelated Rayleigh fading Achieved better performance than selection diversity SNR increases almost linearly with M 7-39 Select {ω 1, ω 2,, ω L } to maximize the SNR of the combined decision statistic Each sensor output is co-phased and weighted by its SNR weight each decision statistic in direct proportion to the relative strength of the component Mean(γ M ) = MΓ Combiner output varies linearly with M MRC is usually used in Rake Receivers The Rake receiver works so well that some have proposed intentionally introducing multipath in indoor environment Average SNR improvements Performances Uncorrelated Rayleigh fading channel (L: # of Antennas) mrc egc sc L=1 10 log (<γ>/γ), db Bit Error Probability L=2 L= No. of branches, M SC MRC EGC Eb/No (db)

8 Assumptions for Array Processing Beamforming Concept Far-field signal source Plane wave Homogeneous medium No wave-front distortion Narrowband signal Time delay α Phase shift All antenna elements (sensors) have known locations and transfer characteristics Calibrated array Signal propagation parallel to the plane of the array Array Geometry Array Geometry The signal received at each antenna is identical except for a time delay This time delay is dependent on the Angle of Arrival (AOA) θ The figure below shows a plane wave received at two antennas d is the separation of the antenna #1 and antenna #k The time delay (τ) in a line of sight environment is d k sin θ y θ 1 k d k x 7-45 Uniform Linear Array Geometry y θ incident plane wave, r(t) 1 2 M d Planar Array Geometry y θ incident plane wave, r(t) (x 1, y 1 ) x (x 2, y 2 ) (x M, y M ) x Delay at k th element ( k 1) d sin( θ ) τ k = c Delay at k th element xk cos( θ ) + yk sin( θ ) τ k = c 7-46 Multipath Channel Models Spatial Signature Received signal x(t) in single element antenna L = number of multipaths α l (t) = complex quantity that models received phase and amplitude Specular multipath channel model Each multipath arrives at a distinct AOA φ l If delays are very small in comparison with the inverse signal bandwidth Spatial Signature Channel can be modeled as flat fading (τ l 0, l = 1, 2,, L) Parameters L, α l, τ l, φ l, modeled through extensive empirical measurements 7-47 Effect of all arriving signal vectors lumped into a single vector quantity Measurements indicate steering vectors appear to be smeared over a range of AOA angle spread

9 Spatial Signature Multiuser Environment With D simultaneous users, the array output is where signal from d th user is Defining Beamforming algorithm for user 1 selects sensor weights, w(t), such that the output is Beamforming Beam Patterns Discriminate between signals according to their AOA Beam pattern controlled by complex weights in each RF chain at the receiver Antenna elements are generally separated by λ/2 The beamformer output is given by A beampattern describes the gain versus AOA of the beamformer reflects array geometry Antenna gain pattern The beampattern is dependent on - array geometry (number of antennas, physical extent of the array) - carrier frequency - gain pattern of each individual antenna - antenna weights Adaptive Beamforming Systems Analog Beamforming x 1 (t) Narrowband adaptive array or linear combiner w 1 x 2 (t) w y(t) 60 interferer x M (t) w M 180 The weight vector is adjusted to improve 0 desired the reception of some desired signal signal Useful for interference rejection, multipath fading mitigation, and increased antenna gain 7-53 Complex weights at the receiver can be implemented through analog means Add arbitrary phase and amplitude to the received signal Can achieve continuous beamforming Low power consumption by reducing multiple RF chains Bank of complex weight sets can be used to select beam Adaptive analog beamforming

10 Operating Differences Example of Beam Pattern 4 element Uniform Linear Array (ULA) with λ/2 spacing Antenna weights equal to unity 10 0 Beamforming Lobes and Nulls that Switched Beam (Red) and Adaptive Array (Blue) Systems Might Choose for Identical User Signals (Green Line) and Co-channel Interferers (Yellow Lines) Note the similarity with a rectangular time-domain window function Angle of Arrival (degrees) 7-56 Example of Beam Pattern Example of Beam Pattern 16 element ULA with λ/2 spacing Antenna weights equal to unity element ULA with λ/2 spacing Antenna weights tapered with a Hamming window The taper reduces side lobe height The drawback is a wider main lobe Angle of Arrival (degrees) Angle of Arrival (degrees) 7-58 Beamforming Smart Antenna Principle (1) Weight and sum signals from array elements to optimize reception of desired signal and null interfering signals Moving Target Optimum beamforming: equal-power transmitters, SNR in =30 db, SINR opt =39 db Optimum beamforming: equal-power transmitters, SNR in =30 db, SINR opt =35.3 db array elements desired transmitter normalized array pattern (linear scale) array elements desired transmitter interfering transmitters normalized array pattern (linear scale) 4 element linear array. Constant Modulus Algorithm working in three environments. Note gain changes as a function of angle No Interference Six Interferers Example: Eight element array, equal-power transmitters, 30 db SNR

11 Smart Antenna Principle (2) Smart Antenna Principle (3) Moving Interference Three Interference 4 element linear array. Constant Modulus Algorithm working in three environments. Note gain changes as a function of angle. 4 element linear array. Constant Modulus Algorithm working in three environments. Note gain changes as a function of angle Adaptive Beamforming Systems Adaptive Narrowband Beamforming Systems Narrowband beamforming is equivalent to spatial filtering By choosing appropriate sensor coefficients, it is possible to steer the beam in the desired direction By varying the sensor coefficients (spatial filter taps) adaptively, the interference is reduced Wideband beamforming requires joint space-time processing Phase shift at the antennas is frequency dependent Down Converter Down Converter Down Converter ADC W 1 ADC W 2 ADC W M Generate Error Signal Demodulator Desired Signal or Signal Property Frequency-dependent response is required (filter) Adaptive Algorithm Adaptive Wideband Beamforming Systems Summary Z -1 Z -1 Z -1 W 1,1 W 1,2 W 1,p Z -1 Z -1 Z -1 W 2,1 W 2,2 W 2,p Z -1 Z -1 Z -1 W 2,1 W 2,2 W 2,p Structure must be frequency selective since the appropriate phase shift at one extreme of the band is not the appropriate phase shift for the other extreme Receiver A Rake is possible alternative to the FIR filter Smart Antennas Diversity combining technique Linear combining technique (SC, EGC, MRC) Basic concept of beamforming technique Array geometry, channel model, adaptive beamforming concept Next topic Adaptive beamforming algorithms

12 Optimum Performance Criteria Beamforming Algorithms Similar to time-domain optimum filtering, choose spatial filter weights to optimize a performance criterion Common optimum performance criteria Minimum mean square error Maximum SINR Minimum variance All criteria lead (under the right circumstances) to maximum SINR solution 7-67 σ s2 : Desired user s signal power V 1 : Desired user s signature vector R i : Interference and noise covariance matrix 7-68 Two Antenna System Minimum Mean Square Error (MMSE) Received signal Minimize the cost function: the mean of squared error Steering vector Setting the gradient of E{ e 2 }, relative to w H, to zero Angle of arrival : known as Wiener-Hopf or optimum Wiener solution R xx and r xd can be calculated analytically, assuming perfect knowledge of the transmitted signal, and channel parameters MMSE Maximum Signal to Interference and Noise Ratio (MSINR) For two antenna system without noise, let a d = [a 1 a 2 ] T, a i = [a 3 a 4 ] T Beam pattern Desired signal Interference The filtered output of the beamformer can be re-written as The average output SINR is given by Taking the gradient of the average output SNR, : generalized eigen equation if R n is invertible AOA (φ) 7-71 scalar values

13 MSINR Minimum Variance Distortionless Response (MVDR) For two antenna system without noise let a d = [a 1 a 2 ] T, a i = [a 3 a 4 ] T R nn singular matrix Minimize the output noise variance while preserving the desired signal Filtered output: Constraint: Variance: Minimizing Lagrange multiplier Taking the derivative of L w.r.t w* and setting it to zero Beam (MMSE) Beam (MSINR) Desired signal Interference The optimum weight vector is which is the same as the MSINR s optimum weight vector AOA (φ) Adaptive Algorithms for Beamforming Least Mean Square (LMS) Algorithm Solutions for the optimum criteria Require complex matrix inversion and perfect channel information Practical implementation Recursively optimize the cost function Avoid matrix inversions which can be computationally challenging to implement Channel statistics are estimated based on a training sequence Least Mean Square (LMS) Least Squares (LS) Recursive Least Squares (RLS) Gradient of cost function (mean square error) w.r.t w*n Method of steepest descent one point estimate MSE μ = 0.1 μ= 0.01 μ= Number of iteration 7-76 LMS algorithm Least Square (LS) Method Beam pattern without noise and with noise Practical implementation of MMSE estimator Statistical average is replaced by the sample or time average Beam (MMSE) Beam (LMS) Desired signal Interference Beam (MMSE) Beam (LMS) Desired signal Interference AOA (φ) AOA (φ) Also called the direct matrix inversion (DMI) method due to the matrix inversion operation Performance depends on the number of independent samples N = (n 2 -n 1 ) > 2 # of antennas: training sequences

14 LS algorithm Recursive Least Square (RLS) Method Beam pattern according to the length of training sequence N = 11 (left), N=16 (right) Provide the recursive solution to a weighted LS problem Beam (MMSE) Beam (LS) Desired signal Interference Beam (MMSE) Beam (LS) Desired signal Interference forgetting factor between 0 and 1 Weighting vector of LS algorithm Recursive calculation AOA (φ) AOA (φ) Blind Adaptive Algorithms Blind Techniques for Beamforming Blind adaptive algorithms attempt to restore a known property, such as constant modulus (for FM type signals), finite alphabet property of digital signals, cyclostationarity of digitally modulated signals, etc. Requires determination of w such that y=w H x has the desired property 7-81 Power Method Maximize output power subject to a gain constraint Decision Directed Algorithm (DDA) Make a best guess of the output state Use the guess as the training signal Constant Modulus Algorithm (CMA) Adapts to restore the envelop property Time- or Frequency- Code Gated Algorithms (CGA) A practical maximum SINR solution Feasible for W-CDMA Spectral Self Coherence Restoral (SCORE) Restores spectral correlation properties 7-82 Lagrange Multiplier Method for Blind Adaptive Algorithm Lagrange Multiplier Method Eigen-based blind adaptive algorithm maximize the output power of array the same as maximizing SINR when interference and desired signal is uncorrelated and background noise is spatially white Array output power P Beam pattern according to several SIR SIR = 0 db (left), SIR > 0 db (right) Beam (MMSE) Beam (LMMP) Desired signal Interference S/I = 3 db S/I = 6 db S/I = 9 db MMSE (ideal) Applying Lagrange multiplier with constraint of w 2 =1 Adaptive weight update using the gradient of the cost function (gradient ascent algorithm) AOA (φ) AOA (φ)

15 Constant Modulus Algorithms (CMA) For constant modulus (FM, PM, CPFSK) signals, change weight vector to reduce envelope fluctuations at output The general form of CMA error (or cost) function Can be applied to the signal of non-constant envelope CMA via Stochastic Gradient Descent Least Square CMA Static LS CMA Dynamic LS CMA CMA Capture effect: CMA extract the strongest signal Depends on the initial weight vector set around interference (left) and around desired signal (right) 90 1 Beam (MMSE) Beam (CMA) Desired signal Interference Beam (MMSE) Beam (CMA) Desired signal Interference AOA (φ) AOA (φ) Stochastic Gradient Descent CMA Least Square (LS) CMA Use the cost function with δ = 1, p = 2 and q = 2 Least squares CMA uses extension of the method of non-linear least squares (Gauss s method) LS CMA cost function:,where Partial Taylor Series Expansion of the cost function Estimate with one-point average CMA: 7-87 where is the offset vector Setting J n (Δ n ) relative to Δ n to zero, Δ n which minimize J n (Δ n ) is The LS CMA weight update: The weight update can be performed either on sample-by-sample basis or on block basis 7-88 LSCMA LSCMA Implementation Two antenna system (σ d2 =1, σ i2 =0.5: SIR = 3dB) Dynamic LS CMA (Block size = 32) Static LSCMA: the weight vector is obtained by several iteration over the same data block Dynamic LSCMA: the weights computed over one data block applied to the next block MSE Beam (MMSE) 0.3 Beam (LSCMA) Desired signal 0.2 Interference AOA (φ) Sample index (n)

16 Code Gated Algorithms (CGA) Concepts of CGA Based on a class of time- or frequency-gated adaptive beamforming algorithms Frequency-gated algorithm LPF R yy Exploits the absence of the desired signal in a known frequency band to derive the statistical properties of the interference and the noise C * s (t) HPF R nn In CDMA LPF the desired signal collapses to a narrow-band after multiplication with desired user s spreading code A low-pass filter can be used to extract the narrow-band desired signal HPF HPF A high-pass filter can be used to extract the interference Find the eigen vector and noise corresponding to the dominant R The name CGA accentuates the use of the spreading code to nn R yy =R ss +R nn R eigen value nn separate the desired signal from interference and noise f Optimum and suboptimum CGA CGA Output SINR Different User Distributions With inverse R nn (R nn ) -1 R yy w = λw, ( y is despread signal, n is MAI plus noise). Optimum method. Spatial matched filter with spatially non-white interference and noise. Without inverse R nn R yy w= λw. Sub-optimum method. Spatial matched filter only with spatially white interference and noise. BER CGA Performance W ith inverse Rnn, uniform user spread W ithout inverse Rnn, uniform user spread W ith inverse Rnn, non-uniform user spread W ithout inverse Rnn, non-uniform user spread Lower Bound (BPSK) Eb/No in db compensated for processing gain and antenna gain BER CGA Bit Error Rate Different User Distributions CGA Performance W ith inverse Rnn, uniform user spread W ithout inverse Rnn, uniform user spread W ith inverse Rnn, non-uniform user spread W ithout inverse Rnn, non-uniform user spread Lower Bound (BPSK) Eb/No in db compensated for processing gain and antenna gain 7-95 Summary Beamforming algorithms Beamforming criteria: MMSE, MSINR, MVAR Adaptive beamforming algorithms with training sequence LMS algorithm LS algorithm RLS algorithm Blind adaptive algorithms Lagrange Multiplier Max Power Constant Modulus Algorithm Code Gated Algorithm Next topic Transmit diversity MIMO (multiple input multiple output)

17 Transmit Diversity Transmit Diversity Diversity / beamforming gains are provided by multiple transmit antenna elements Transmit diversity architecture (vs. receive diversity) Generally, transmitter needs to know the channel information to determine weight vector determined by various algorithms Transmit Diversity at Base Station [1] Motivation Traditional receive diversity is not yet practical at the handset Computational load required to implement algorithms consume additional power reducing talk time Recent interest: Inclusion of transmit diversity in 3G standard Known as downlink beamforming in mobile communications Objective of downlink beamforming: Maximize received signal to interference ratio at the desired mobile Minimize interference to other mobiles and adjacent base station 7-99 Transmit Diversity at Mobile Station [2] Motivation Emerging technology by low power signal processing technology and small RF components Possible to equip a mobile station with multiple antennas Alternative plan for the system performance improvements When BS cannot increase # of receive antennas Can provide additional system performance improvements Multiple input multiple output (MIMO) system Interference reduction by transmit beamforming Transmit Diversity Benefits Anti-jam capability Extended range Low probability of intercept Reduced fading Lower transmit power Interference reduction Increased battery life Transmit Diversity and MIMO Techniques

18 Transmit Diversity (1) Weight Reuse Technique Reuse of uplink weight if f c is the same in both uplink and downlink Applicable to TDD (time division duplexing) systems Downlink weights are a scaled version of uplink weights demodulator shares the uplink weights with the tx Relatively static channel during reuse of the weights Transmit Diversity (2) Feedback Approach Channel estimation is fundamental to downlink beamforming Probing signal (training sequence) from BS to MS Channel response measured at each mobile and feedback to the base station Pros & Cons Simple implementation of transmitter Overhaul of protocols and signaling Useful when channel is relatively static Interruption of normal information transmission Transmit Diversity (3) Angle-of of-arrival (AOA) Based Approach o The base station estimates AOA information of a mobile by direction finding (DF) algorithms AOA is assumed to be relatively the same although channel transfer functions are different at f c,ul and f c,dl From AOA information, weights are calculated based on maximizing SINR Transmit Diversity (4) Fixed Multiple Beams Approach Fixed multiple beams for both reception and transmission Strongest beam in the uplink is used in the downlink Various combining techniques (MRC, EGC, SC, etc.) can be used Can be extended to steerable beams Not an optimal solution but SINR is improved at the mobile station Transmit Diversity (5) Antenna Pre-coding Approach Use of pre-coding at the transmit antenna elements [3] Use of decoding at the receiver to regenerate the original symbol Decoding in the form of equalization Pre-coding: Linear Time Invariant (LTI) filter Higher filter delay Unrealizable for practical purposes Linearly Periodic Time Varying (LPTV) filter: Smaller filter delay Realizable filter order Transmit Diversity (6) Space-Time Coding Approach Space time processing at transmitter Introduce spatial and temporal correlation into transmit signal using space time coding Space Time Trellis Coding (STTC) Space Time Block Coding (STBC) Diversity gain from multiple antennas Coding gain from spatial and temporal correlation No channel information required at transmitter

19 Transmit Diversity (7) Hybrid Diversity/Beamforming Intelligent Antenna System [4] Exploit both diversity gain and beamforming gain Beamforming is performed in each diversity branch Transmit Diversity (8) Opportunistic Beamforming Proposed for multi user diversity system (HDR, HSDPA) [5] Does not require the direction of the desired user α(t) and θ(t) are determined by pseudo random number if any user is accidentally on the direction of random beam, optimum beamforming gain can be obtained from that beam Transmit Diversity (8) Opportunistic Beamforming Fade distribution (Rayleigh fading channel is assumed) (a) Static channel Before opportunistic beamforming (b) Artificial fading fluctuation After opportunistic beamforming increase achievable data rate (determined by channel status between BS and MS) and fairness Transmit Diversity Application (a) (b) Closed Loop Transmit Diversity in W-CDMAW Base station transmit diversity Latency and feedback overhead issues Weight vector Maximize the received SNR at mobile station Quantize the optimal weight vector to a finite set of magnitude and phase: Mode 1: phase information only; Mode 2: magnitude and phase information Open Loop Transmit Diversity in W-CDMAW Base station transmit diversity Does not require any feedback from mobile station Time Switched Transmit Diversity (switched diversity for coding gain from independent channel experienced symbol-by-symbol) Space Time Transmit Diversity (space time coding diversity)

20 Open Loop Transmit Diversity in CDMA2000 Base station transmit diversity Orthogonal Transmit Diversity Similar to switched diversity exploit coding gain Space Time Spreading (Space Time Coding Diversity) Multiple Input Multiple Output MIMO Systems Definition Systems utilizing multiple transmit and multiple receive antennas are commonly known as Multiple Input Multiple Output (MIMO) systems Performance improvements from the multiple antennas at both transmitter and receiver Array Gain SNR improvement Diversity Gain fading mitigation Spatial Multiplexing Gain channel capacity Interference Reduction / Avoidance MIMO techniques Beamforming Array gain Interference reduction Spatial multiplexing High data rate Examples: D-BLAST, V-BLAST Space-time coding Diversity gain through space-time coding Space-time block coding / trellis coding Spatial Multiplexing In rich scattering environments, independent data signals transmitted from different antennas can be uniquely decoded to yield an increase in channel capacity Optimization Channel decomposition via singular value decomposition Diagonal matrix whose diagonal term is the square root of eigenvalue of HH + Water filling problem in parallel Gaussian channel ~ ~ ~ ~ 7-20

21 Optimization Information capacity MIMO with Rayleigh fading channel AWGN Fading(1x1) Fading(2x2) Fading(4x4) Optimal System Practical consideration Channel information at transmitter is typically impossible without any feedback from the receiver Channel information at receiver must be measured/estimated using training sequence Complexity of matrix calculation exponentially increases as the number of antennas increases Capacity (bits/s/hz) SNR(dB) Optimal system structure BLAST (Bell( Laboratories Layered L Space Time) Layered space time architecture [6] Proposed by Foschini High bit rate using multiple antenna elements Input stream is demultiplexed into several layers encoded and modulated independent of other layers D-BLAST, V-BLAST BLAST D-BLAST (Diagonal BLAST) Theoretically achieve log-det capacity with appropriate Rx processing Lost triangles / High complexity / Coding constraints n=1: detect a without interference from other antenna n=2: detect a after nulling b out by projecting the receiving vector onto the null space of H 3,n BLAST D-BLAST (Diagonal BLAST) Theoretically achieve log-det capacity with appropriate Rx processing Lost triangles / High complexity / Coding constraints n=3: detect a after nulling b and c out by projecting the receiving vector onto the null space of H 3,n and H 2,n BLAST V-BLAST (Vertical BLAST) Achieves far lower capacity than the log-det bound No lost triangles / Lower complexity / Simple 1-D codecs Optimum detection ML detection Suboptimum detection Ordered successive interference cancelling n=4: subtracting the decoded layer 3 from the receiving vectors

22 Application of BLAST in W-CDMAW Orthogonalized spatial multiplexing [7] V-BLAST architecture Multiple channels spread by orthogonal codes are superposed and transmitted through each antenna References [1] Jeffrey H. Reed and R. Michael Buehrer, Smart Antenna Overview, MPRG presentations [2] Raqibul Mostafa, Feasibility of Smart Antennas for the Small Wireless Terminals, Ph.D dissertation, VirginiaTech., [3] Gregory W. Wornell, Mitchell D. Trott, Efficient Signal Processing Techniques for Exploiting Transmit Antenna Diversity on Fading Channels, IEEE Trans. on Signal Processing, Jan [4] Robert A. Soni, R. Michael Buehrer, and Roger D. Benning, Intelligent Antenna System for cdma2000, IEEE Signal Processing Magazine, July [5] Pramod Viswanath, David N. C. Tse, and Rajiv Laroia, Opportunistic Beamforming Using Dumb Antennas, IEEE Trans. Information Theory, June [6] Gerard J. Foschini, Layered Space Time Architecture for Wireless Communication in a Fading Environment When Using Multi-Elelment Antennas, Bell Labs Technical Journal, Autumn [7] 3GPP TR25.848, Physical layer aspects of UTRA High Speed Downlink Packet Access Summary Transmit diversity Introduction Transmission diversity technique Applications in practical communication system MIMO Definition Optimum MIMO system and information channel capacity D-BLAST, V-BLAST Orthogonalized spatial multiplexing Next Topic Space-time processing Space-Time Processing Channel model Motivation Two separate approaches on diversity Space (multiple antennas) Time (multipath components) Additional gains are possible through the combination of both spatial and temporal diversity Spatial diversity Temporal diversity Space-Time Adaptive Processing Adaptive signal processing in both space and time domain Maximize the suppression of CCI, thereby maximizing the Signal to Interference plus Noise Ratio (SINR) Resolve and combine multipath signals, thereby providing dramatic improvements in diversity gain and SINR

23 STAP Algorithms Space-Time Processing for CDMA Cascaded spatio-temporal processing Spatial processor output to a temporal processor (or reverse) MRC, EGC, SC - diversity combining techniques MMSE, Max SINR - optimum combining techniques Sub-optimum performance Joint spatio-temporal processing Signals in the temporal and spatial domain are processed simultaneously Processing may be diversity or optimum combining techniques Computationally complex though optimum performance CDMA system Subject to large multiple access interference (MAI) in addition to multipath fading Conventional RAKE receiver Not very robust against MAI: Depends on cross-correlation properties of PN sequence Cannot exploit spatial dimension to mitigate fading Antenna Arrays Offer array gain; Exploit spatial diversity; Reduce cochannel and adjacent channel interference; Enhance system capacity Space-time processing (combined antenna array and RAKE) Exploit both spatial and temporal diversity Can significantly outperform the conventional RAKE ST-MLSE, ST-MMSE receiver Cascaded Spatio-Temporal Processing 2-D rake receiver Spatial combiner followed by temporal combiner Any combining method can be used for spatial combining and temporal combining ex) Beamforming for spatial combining, diversity or optimum combining for temporal combining Possible to form a beam toward each multipath separately Cascaded Spatio-Temporal Processing 2-D rake receiver Temporal combiner followed by spatial combiner Any combining method can be used for spatial combining and temporal combining, but beamforming is not appropriate for spatial processor in this structure since the spatial property of each multipath can be lost in temporal processor Space-Time Maximum Likelihood Sequence Estimation (ST-MLSE) Space-time extension of conventional MLSE Received sequences at multiple antenna X = H S + N L samples of the received sequence at M branches X i,j : j th received sequence at the i th antenna h i,k :channel response during [k(n-1)t,knt] Finding a sequence S that minimizes X-H S F 2 Implemented by Viterbi algorithm L samples of Q consecutive elements of sequence Toeplitz matrix form ST-MLSE (Cont d) Effective in removing ISI due to the time varying multipath channel Degradation of ST-MLSE performance due to channel estimation errors accurate and computationally efficient channel estimation required Improved performance over temporal processing; higher complexity with delay spread & average channel delay In the presence of CCI with delay spread, the implementation of Viterbi algorithm is complicated ST- MMSE equalizer can be an alternative solution

24 Space-Time Minimum Mean Square Error (ST-MMSE) Equalizer Space-time extension of temporal MMSE equalizer Use the following matrix weight for the space-time equalizer M branches x N samples Weight vector is determined by achieving the following MMSE criteria : reshaped w, x MN x 1, δ : delay chosen to center the ST filter Space-Time Coding Effective in canceling CCI primarily in spatial dimension and ISI in the space or time dimension Space-Time Coding Space time processing at transmitter Introduce spatial and temporal correlation into transmit signal using space time codes Space Time Trellis Coding (STTC) Space Time Block Coding (STBC) Diversity gain from multiple antennas Coding gain from correlation among multiple antenna and multiple symbols Transmit diversity technique, but no channel information required at transmitter Space-Time Trellis Coding Convolutional code applied to space and time domain Each antenna output is mapped into modulation symbol Maximum likelihood sequence estimator ( Viterbi algorithm) Example) Delay Diversity (by Wittneben [4]) Encoder structure for two antennas Generator matrix form QPSK mapping [a 1 a 2 a 3 a 4 ] Space-Time Trellis Coding Space-Time Trellis Coding Trellis diagram Current state is defined by the previous input value or encoder state value (a 3 a 4 ) of the encoder in the previous slide Example) Delay Diversity (by Wittneben [4]) Trellis diagram Input sequence: Output sequence: c 1 : c 2 : Trellis coded modulation (TCM) technique Transmit diversity gain Maximum performance at frequency flat fading channel Performance improves in fast fading channel Requires accurate estimation of channel gains at receiver Research issues Finding optimum trellis code Decoding technique for frequency selective fading channels Space-time turbo coding Implementation of receiver architectures

25 Space-Time Block Coding Space time block codes Proposed by Alamouti for two transmit antennas [5] Simple decoding but MRC-like diversity gain Orthogonal form of channel makes it possible to transmit symbols at full rate Decoding MRC-like performance but 3dB penalty from equal power distribution of original Tx power