Introduction to Beamforming and Space-Time Adaptive Processing

G.V.K.Sharma Associate Professor Department of ECE GITAM Institute of Technology GITAM University

In this Presentation

Fundamentals of Antenna Arrays

Linear Arrays Planar Arrays Phased Arrays Adaptive Arrays Signal Models Conventional Beamforming Optimal Beamforming MATLAB Illustration

Beamforming (Spatial Filtering)

Space-Time Adaptive Processing

Fundamentals of Antenna Arrays

Many applications require radiation characteristics (like gain, directivity) that may not be achievable by a single element Antenna array is a geometric arrangement of antenna elements Resulting radiation pattern is a vector sum of individual patterns Antenna arrays provide more directivity by the phenomena of wave interference Directive Gain in a given direction is a measure of ability of an antenna/array to radiate power in that given direction

Fundamentals of Antenna Arrays

As the length of the antenna aperture increases, beamwidth decreases As the number of antenna elements increase, directive gain increases

Linear Arrays

Linear array is a linear arrangement of antenna elements with equal spacing d between successive elements Uniform LA is LA with equal current excitation and uniform progressive phased shift between elements The electric field at a far observation point is (assuming isotropic elements) is independent of

Radiation Pattern of an 8-Element Uniform Linear Array with d = 0.5

Polar Plot

E(sin) vs direction cosine (sin)

Array Controls

Increasing Array Size

by Adding Elements

Increasing Array Size

by Separating Elements

Amplitude Tapering (Windowing) and Phase Quantization

In Antenna Arrays the current exitations (aperture distribution) are tapered i.e multiplied by a window sequence to reduce the side lobe level. Reduction of Side Lobe Level is obtained at the cost of increase in beamwidth In practice phase shifters are implemented as part of TR modules, using finite number of bits Due to quantization error (difference between desired phase and actual quantized phase) the sidelobe levels are affected

Planar Arrays

Planar Arrays have antenna elements placed on a plane according to some grid configuration (rectangular, circular) Planar arrays can control the beamshape in both planes (, ) and form pencil beams whereas linear array only controls the pattern in one plane Total electric field at a far field observation point is given by

Radiation pattern of an 8 8-element uniformamplitude and spaced square planar array

Radiation Pattern Translation from Spherical coordinates into U,V space Spherical Radiation Pattern

Phased Array Antennas

Array antennas synthesize narrow directive beams that may be steered mechanically or electronically Electronic steering is achieved by controlling the phase of the electric current feeding the array elements, thus the name phased array Phase relation is maintained using a network of power dividers and phase shifters Direction is selected by adjusting the phase difference provided by each phase shifter (usually done using a microprocessor)

Scanned Every 30deg, N=15, d=/4

Linear Phased Array

Adaptive Arrays

An adaptive array not only steers the beams but also the nulls Nulls are steered towards the direction of jammers and nullify their detrimental effects Adaptive arrays first sample the environment to estimate the interferences Next a weight vector is calculated to modify the sidelobes for effective null steering and supression of interferences

Beamforming (Spatial Filtering)

Beamforming refers to coherent combination of data from multiple antenna elements to provide selectivity in the angle of arrival i.e to form and steer an antenna beam Temporal Filters retain desired frequency components and block undesired frequency components. Similarly beamformers receive signals from desired directions and steer nulls in the undesired directions hence also called spatial filters

Signal Models

In spatial array processing there are three types of signals

desired target signal jammer signal noise signal

Thus the total signal received by the array is x = xs + xj + xn Jammer and Noise are classified as interference. The undesired interference signal is xu = xj + xn Both jammer and noise are characterized as zero mean normally distributed. Hence the covariance matrix of this undesired signal would be Ru = E[xuxuH] = Rn + Rj

Signal Models

Desired Target Signal

Signal is narrowband

x(t) = e

j2ft

Antenna Array has receiver behind each element. These receivers digitize the received signal The combined output of the receivers is a N-dimensional signal

x is complex baseband signal received at left most element V is spatial array vector

Signal Models

Noise Signal

In the receiver array each element produces thermal noise Modelled as zero mean Gaussian random process The noise covariance matrix is Rn = n2I, n2 =kTnB

Jamming Signal

Jammers are modeled as spatial point sources that constantly transmit high power omni-directional interference signal The signal covariance matrix is equal to Where 2 jammer noise power j Vj is array manifold vector associated with jammer direction of arrival If we are dealing with N jammers, then the covariance matrices would add, because we assume jammers are mutually uncorrelated

Conventional Beamformer

The beamformer output y of our array is

The complex chosen

weights

wi

are

to control the sidelobe level to steer the main beam towards an angle 0

However this data independent beamformer may not provide nulls in the direction of interferers and hence suboptimal SINR

Optimal Beamformer (MVDR)

A minimum variance distortionless response beamformer (MVDR) also called an optimal beamformer accomplishes two objectives

Minimize the array output interference power Get the target desired signal without any distortion

Controllable parameters are array weights wi (weight vector w) Array output in vector notation is Output is a combination of desired signal and interference components The interference output is the sum of noise and jammer outputs or

If we want to minimize the interference output power, then we must minimize

subject to the constraint

Optimal Beamformer (MVDR)

The constrained minimum variance distortionless response can be achieved with the weight vector

where

is the desired target steering vector

The desired weight vector can be calculated to be

with

Matlab Illustration (Both Jammers in Sidelobes, with No Weighting)

Unadapted Array Pattern 0 0 Distortionless Beamformer Array Pattern -10 -10

Normalized Array Response (dB)

-20

Normalized Array Response (dB) -80 -60 -40 -20 0 20 Angle of Arrival (degrees) 40 60 80

-20

-30

-30

-40

-40

-50

-50

-60

-60

-80

-60

-40

-20 0 20 Angle of Arrival (degrees)

40

60

80

N=16 antennas elements are used Desired target located at 00 Jammers located at 180 (with SNR of 50dB) and -330 (with SNR of 30dB) Two dotted vertical lines indicate the angles of arrival of the two jammers

Adaptive Beamforming

Optimal Beamforming only sounds good only in theory Obtaining Rin (interference covariance matrix) requires infinite number of samples. Hence we can only estimate it. A sampled interference covariance estimation is Various adaptive beamforming methods are based on collecting data from which correlation matrix is estimated

Block Adaptive Implementation

Uses block of data to estimate the adaptive beamforming weight vector and is known as Sample Matrix Inversion (SMI) RLS Algorithm Steepest Descent Method

Sample by Sample Adaptive Implementation

Training Data

Before the beamforming system can be used, it must be trained with target- free samples Training Data are of two types

Target Free training data xin = xn + xj Target in training data xin = xn + xj + xs

In applications like radar, target free training data is always available by taking measurements at ranges shorter or longer than the target

Space-Time Adaptive Processing (STAP)

STAP is concerned with the two-dimensional processing of signals in both the spatial and temporal domains to optimally discriminate targets from both clutter and jamming Detection of slowly moving targets by air- and spaceborne MTI radar (moving target indication) is heavily degraded by the motion induced Doppler spread of clutter returns. Space-time adaptive processing (STAP) can achieve optimum clutter rejection via implicit platform motion compensation.

General STAP Architecture

Space Time Environment is sampled in spatial domain by using array of antenna elements Also sampled in temporal domain by transmitting a series of pulses to obtain doppler information

Space-Time vs Spatial Processing

Degrees of Freedom

In spatial array processing, the degrees of freedom equals number of antenna elements N In STAP, every antenna transmits a train of M pulse and applies a complex weight on each echo after receiving them. Hence the degrees of freedom equals NM. In spatial array processing only progressive phase shift between antenna elements is used In STAP, progressive phase shift between antenna elements and between successive pulses received from each antenna element is exploited

Progressive Phase Shift

Signal Behavior in Space-Time Environment

Receiver Noise has no structure in space/frequency and therefore appears as a uniform noise floor Broadband Noise Jammers are localized in AOA but spread across the entire doppler spectrum Appears as ridge of energy localized in AOA but spread across all doppler shifts Scatterers at an angle of w.r.t antenna boresight will have a doppler shift of

Space Time Signal Environment

Space Time Adaptive Processing

Angle doppler characteristics of the echo from a moving point target depend on both the radar platform motion and the target motion If the target is stationary and directly on the boresight, the doppler shift will be zero and will fold into the clutter However if the target is moving it will separate from the clutter on the doppler axis and shown in the figure

Optimal STAP Processing

The spatial steering vector for a ULA of M sensors is

The temporal frequency steering vector assuming L transmitted/received pulses is

The two dimensional LM X 1 space time steering vector is given by The optimal STAP weight vector is given by

Thank You

Matlab Source Listing (beamformer.m) Input Section

close all clear all %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % User Input Section lambda = 0.03; % wavelength d = lambda/2; % element spacing dl = d/lambda; N = 16; % # of array elements aoa_max = asin(1/2/dl); % maximum "real space" AOA (radians) window_on = false; % true or false Nangle = 1000; % # of angles for evaluating beam pattern Nm1 = Nangle-1; t_aoa = pi/180*(0); % target AOA (radians) j_aoa1 = pi/180*(18); % jammer #1 AOA (radians) j_aoa2 = pi/180*(-33); % jammer #2 AOA (radians) SNR = 0; % signal to noise ratio (dB) JSR1 = +50; % jammer #1 to noise ratio (dB) JSR2 = +30; % jammer #2 to noise ratio (dB) % End user input section %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Matlab Source Listing (beamformer.m) Computing the Optimal Weight Vector

p_n = 1; % noise power p_t = p_n*(10^(SNR/10)); % target power p_j1 = p_t*(10^(JSR1/10)) ; % jammer power p_j2 = p_t*(10^(JSR2/10)) ; % jammer power % compute signal vectors target = sqrt(p_t)*exp(j*2*pi*(0:N-1)'*d*sin(t_aoa)/lambda); if (window_on) t = target.*taylorwin(length(target),4,-30); else t = target; end j1 = sqrt(p_j1)*exp(j*2*pi*(0:N-1)'*d*sin(j_aoa1)/lambda); j2 = sqrt(p_j2)*exp(j*2*pi*(0:N-1)'*d*sin(j_aoa2)/lambda); % compute covariance matrix with and without jammers R = p_n*eye(N); Rj = p_n*eye(N) + p_j1*j1*(j1') + p_j2*j2*(j2'); disp(['Covariance matrix rank with jammers = ',num2str(rank(R))]); % compute beamformer weight vector with and without jammers w = R\conj(t); wj = Rj\conj(t);

Matlab Source Listing (beamformer.m) Plotting the Output Patterns

% compute and display beampattern theta = -aoa_max + 2*aoa_max/Nm1*(0:Nm1); W = zeros(Nangle,1); Wj = zeros(Nangle,1); for p = 1:Nangle W(p) = w.'*exp(-j*2*pi*(0:N-1)'*dl*sin(theta(p))); Wj(p) = wj.'*exp(-j*2*pi*(0:N1)'*dl*sin(theta(p))); end Wp = db(abs(W),'voltage'); scale = 10*log10(N^2); % scale = 0; Wp = Wp - scale; Wjp = db(abs(Wj),'voltage') - scale; figure(1) plot(180/pi*theta,Wp) axis([-180*aoa_max/pi +180*aoa_max/pi -60 0]) % grid xlabel('Angle of Arrival (degrees)'); ylabel('Normalized Array Response (dB)') title('Unadapted Array Pattern') vline(180/pi*[j_aoa1, j_aoa2]) kappa = t'*transpose(inv(Rj))*conj(t); wj1 = wj/kappa; Wj1 = zeros(Nangle,1); for p = 1:Nangle Wj1(p) = wj1.'*exp(-j*2*pi*(0:N1)'*dl*sin(theta(p))); end Wjp1 = db(abs(Wj1),'voltage'); figure(3) plot(180/pi*theta,Wjp1) axis([-180*aoa_max/pi +180*aoa_max/pi -60 0]) % grid xlabel('Angle of Arrival (degrees)'); ylabel('Normalized Array Response (dB)') title('Distortionless Beamformer Array Pattern') vline(180/pi*[j_aoa1, j_aoa2])