Non-Uniform Linear Nested Array Implementation for Improved Beamforming and

Optimization

Nethravathi K A Deepika Anantharaman Bhishm Tripathi

Assistant Professor, Department of ECE Student, Electrical Engineering Scientist E
RV College of Engineering University of Pennsylvania Electronics and Radar Development Establishment
Bengaluru, India USA Ministry of Defence(MoD), DRDO
[email protected] [email protected] [email protected]

Ashish K Adiga
Student, Department of ECE
RV College of Engineering
Bengaluru, India
[email protected]

Abstract—Beamforming has been an area of interest for a well as nested linear arrays for different beamforming
lot of research work as the multiantenna system has become an techniques is performed. There is a detailed explanation
important topic in wireless communication. While both theo- of the theoretical and optimum techniques as well as real
retical, as well as real-time practical implementation techniques
for beamforming have been extensively researched, the existing time (adaptive) beamforming techniques which derive
techniques mainly focus on obtaining peak performance in the their functionality depending on the dynamically varying
fastest and most accurate way possible and usually ignore environmental conditions in which they are required to
applications providing improved performance with limited operate. Different beamforming techniques using different
resource utilization. This paper further explores the concept constraints and algorithms will be the primary focus of this
of nonuniform spacing in such multiantenna receivers and
shows how beamforming equivalent to regular uniform sensor paper.
arrays can be achieved and analyses many trade-offs that
designers and users must consider for a minimalistic approach.
The objective of this paper is to analyze various beamforming A. Linear Arrays
techniques on Uniform Linear Arrays (ULA) and the perfor- Uniform Linear Arrays (ULA) are computationally sim-
mance comparison between various techniques along with the
analysis of output signal reconstruction and the array response. pler, but not good for spatial resolution. Uniform Planar
It also contains a detailed analysis of the uniformly spaced Arrays (UPA) are good in Angle of Arrival (AoA) es-
arrays and nested arrays using optimization techniques for timation, but not optimum computationally or cost wise.
error minimization. The construction of Non-uniform Linear Non-uniform (nested) arrays offer a great blend of the
Nested Arrays(NULNA) with optimum element spacing will be advantages of both ULA and UPA with optimized spacing
analyzed.
and customizable beam patterns. Nested arrays are like fixed
Keywords-Non-Uniform Linear Nested Array, Beamforming, subarray randomization in a planar array, except it does not
Uniform Linear Array, Non-Uniform Linear Array, Array of require as many elements at all to begin with and all the
antennas
skewed elements are still on one line. It is also cost effective
and can be optimized to provide a desired solution in terms
I. I NTRODUCTION
of performance for any application.
Antenna array receiver system requires a mechanism In [1] the authors talk about nested arrays in terms of
to be able to process the multiple versions of the same their mathematical tractability and their suitability for ap-
incident signal across its many elements. And for this, it is plications requiring increased degrees of freedom from a
necessary to find a way to combine them all into a single linear array which still retains the computational simplicity
composite signal after ensuring they are filtered and in of a ULA while providing better angle-range estimation, a
complete sync with one another. Beamforming plays an valuable feature of nested arrays. In [2], [3], the authors
important role in such situations. There are many types of talk about the degrees of freedom of a nested array and
antenna arrays which have been designed, namely, linear, how a nested array with N elements can have greater
planar, nested linear etc. In this paper, a comprehensive degrees of freedom than a ULA with N elements. In [4],
study and demonstration of the performance of linear as the authors talk about how these increased degrees of
freedom help in better resolution capability of the arrays 7. Compute vs using us , the steering vector for the acquired
and thus a better performance than ULA with respect to steering angle, which is the same as the DoA of the signal
a few parameters, albeit with its own trade-offs. In [5], here.
the authors lay a solid mathematical groundwork and offer 8. Model the SoI using the steering angle and SNR values.
many solutions and variations to the tweaking of the beam 9. Model random noise across the N samples.
pattern responses of array receivers and how they can be 10. Create the received signal x = SoI + noise.
manipulated by manipulating the beamforming weights of 11. Create the beamformer weights by the matched filter
the system. In [6], the array processing and the mathematical technique cmf = vs .
models behind designing and simulating them in a coding 12. The output signal from the beamformer y = cmf × x .
environment along with extensive analyses of different non- 13. Beam pattern response B = cmf × vs .
adaptive techniques have been illustrated. In [7], the authors 14. Steered Response of the received signal R = cmf × x .
explained why the estimation of a second order statistical The flow for every other kind of beamformer is also done
quantity such as covariance is essential for adaptive weights in a similar manner, albeit with modifications done to the
construction and how the results obtained vis-a-vis such a weights of the beamformer, which is how the different
technique can be improved. In [8], the authors explained beamformers vary from one another.
about Diagonal Loading for correlation matrix estimation
which gave an enhanced adaptive beamformer performance.
In [9], the author explains how the concept of estimated B. Uniform Linear Array - Adaptive Beamforming Algo-
covariance matrix can make or break the performance of rithm Design
a beamformer. In [10], the author explains that covariance Estimation algorithms are used to estimate the DoA of
matrix is what limits the exploitation of what can otherwise the SoI, with care taken to make sure that jammer signals
prove to be a valuable application of N-level nesting. Special are not incorrectly taken to be SoI and then the beam
emphasis is given to the difficulties faced in the computation former weights are gradually corrected using adaptive
of Nth-order statistical quantities (like the covariance matrix techniques to resemble that of the optimum beamformer
for 2-level nesting) required for N-level nesting. In [11] and weights to achieve as good a performance as is possible.
[12] the authors use a Standard Capon Beamformer and The estimation of the weights done in adaptive techniques
elaborate on the signal model for a nearly real-time signal requires knowledge of the received signal spectrum.
environment, which was used to make the simulation results
seem as close to the practical outputs as possible. Correlation matrix of the signal x is the required second
This paper is organised as follows. Section II presents order term which is used to estimate the adaptive weights
the design methodology of linear as well as nested arrays of the beamformer in practical applications.
for different types of beamforming. Section III presents the E[xx, ] = Rx = Received Signal Spectrum
results and discussions where a comparison is made between One important thing about the Rx matrix is that, it is
the output plots of different beamforming techniques for dynamic in nature. As x varies in every sample snapshot
linear as well as nested arrays based on various governing (across all M elements of the array) so will Rx and hence
parameters of the arrays. Finally the conclusion of the paper the weights. Which is why, for a given approximately
is presented. stationary source and interference signal, a few number of
iterations should allow the estimated weights to converge
II. DESIGN OF LINEAR AND NESTED ARRAYS to a nearly optimum value.
The design methodology of non-adaptive and adaptive Algorithm for adaptive weights construction for a ULA:
beamforming for uniform linear array and nested arrays is 1. The same as Non-adaptive Beamforming for Steps 1 to
explained in the form of algorithms in this section. 10 apply here.
2. Set a value for K, the number of iterations.
A. Uniform Linear Array - Non Adaptive Beamforming 3. Repeat K number of times.
Algorithm Design 4. Received signal is varied every iteration x =
The Basic algorithm used for conventional non-adaptive INRvi (randn(1,K)+jrandn(1,K))
√
2
+ noise
beamforming on the ULA is given below: Where vi is the steering vector using the estimated AoA of
1. Choose M - number of array elements. the interference source(s)
2. Choose N number of samples from the signals. 5. Covariance Matrix of the received signal is calculated
3. Input the frequency of operation f; Calculate λ from f. Rx = k1 [xx/ ]
4. Calculate d from λ using the ideal spatial sampling 6. Inverse of Rx is found by iRx = Rx− 1
/
frequency condition. 7. Calculate the adaptive weights as cadaptiv e = iR
/
x vs
vs iRx vs
5. Acquire the DoA and the SNR of the SoI.
6. Compute us , the normalized spatial sampling frequency. Where vs is the steering vector using the DoA of the
SoI which can either be estimated or known before hand in and greater the signal gathering strength of the receiver as
certain applications. a whole. Therefore, nested arrays with the same number
8. Repeat Steps 4 to 7 K times. of elements N1 + N2 + N3 . . . + NK = M are capable
/
9. Output signal y = cadaptiv e ×x of gathering signals better than a regular M element ULA
10. Beampattern response B = cadaptiv e × vs since they are more spread out. However, such a performance
enhancement is usually quantified in terms of Degrees of
Freedom (DoF) of the array. DoF is merely a mathematical
C. Nested Arrays - spacing of the elements
concept which specifies the number of sources an array
Nested Arrays are a special case of linear arrays. Instead receiver can resolve. This means not only the number of SoI
of being Uniformly spaced (deducing the naming of the it can resolve as distinct but also the number of interference
ULA), the nested arrays have different spacing between sources it can suppress. And in that way, nested arrays
successive elements and find themselves in exciting places which use the concept of co-array sum and differences
to provide a very unique and valuable set of advantages over sport O(N2 ) DoF while ULA only resolve O(N) sources
their conventional counterparts. The spacing is what makes in comparison with the same number of elements. However,
them different from ULA and the conventional nested array while designing Nested Arrays, one must keep in mind few
contains a general formula employed for the construction of points: More than two levels of nesting requires a corre-
the physical array layout depending on the number of nested sponding order of statistical terms to estimate the adaptive
levels and the number of elements in each of the levels. weights from. This is why we seldom find more than 2
Assume there to be K number of nested levels. Let each levels of nesting. Nested beamformers show an incredibly
of the levels have elements given by Level 1: N1 elements, asymptotically optimal Loss in SINR performance. There is
Level 2: N2 elements, Level 3: N3 and so on until level K: always an optimal spacing condition for a given number
NK elements. of antenna elements as a nested array and it is usually
In this case, the first level consists of N1 elements, all not a two-level solution and hence cannot be implemented
separated by the regular d interelement spacing as shown in practice. Such an optimal spacing solution requires an
in the Figure 1. The second level consists of N2 elements M level array with M number of antennas arranged in an
exponentially ascending spacing given by the figure below:

Figure 1: Elements separated by uniform distance Figure 3: Optimally spaced multilevel nested array with
highest DoF
each with a spacing d2 = (N1 + 1)d1 from the reference
point, making the whole array look like: The maximum DOF is O(N2 ) = N2 (N1 + 1) – 1 for a
nonuniform nested linear array.
E. SMI Adaptive Beamforming
The sample correlation matrix is calculated as an
Figure 2: Elements separated by non-uniform distance estimate rather than using the known interference AoA
based steering vector in the non-adaptive techniques. Here,
Similarly, the next level will have a spacing of the estimated correlation matrix is,
d3 = (N2 + 1)d2
and the elements will be arranged in a similarly skewed Ri+n est = Ri+n + σ 2 I(1) (1)
manner until the K th level where the spacing will be
dK = (NK −1 + 1)dK −1 . Where σ 2 is an arbitrary constant. The effect of this constant
The algorithm for the nested array will remain the same is highlighted by the scaling it brings to the Identity matrix it
other than for the nested layout. Other modifications done is multiplied with. This product produces a diagonal matrix
to suitably alter the beam pattern response and other perfor- such that when added to the correlation matrix, it makes
mance parameters will be elaborated further. an important change to the mathematical property of this
second order term.
D. Nested Arrays - degrees of freedom In linear algebra, the second-order statistical quantity is
The main advantage that Nested arrays offer over the called the Sample Covariance matrix, or in this case called
conventional ULA is the increased virtual aperture area due the correlation matrix (since it is obtained as an autocorre-
to the nonuniform spacing between the elements of the lation of the received signal minus signal component with
different levels. Greater the aperture area, larger the gain, itself) is an indication of the variance of the quantities of the
array snap shot vector, and in this case literally meaning the
individual samples of the received signal across M elements
of the array. The taper sample number is optimised and
selected to be as many as possible to get closer to optimum
weights and an asymptotic beam pattern response and as less
as possible to have a system which is fast.

III. R ESULTS AND D ISCUSSIONS

A. Uniform Linear Array - Conventional Beamforming
Figure 4 shows the beam pattern response for the conven- Figure 5: Output signal of beamformer
tional condition when the steering angle is set to 0 degrees.
The pattern shows a steady, nearly inverse exponential drop
in the magnitude of the Side Lobe levels as it moves away Figure 6 shows that the side lobes are suppressed a great
from 0 towards 90 degrees on either side. The main lobe deal and made uniform in height (a side-effect of using an
is centred at the 0 degree steering angle as required, the equiripple filter for tapering) as the taper increases. That is,
peak at 0 dB and thus allowing any signals received at for 70dB taper, shown in the figure 6 as a red plot, the side
broadside to pass through unattenuated. The effect of other lobes are more compressed compared to that of 50dB taper,
steering angles can be observed and obtained by changing shown as the black plot.
the weights of the antenna elements such that the antenna
beam’s peak is at the desired steering angle - ϕs .

Figure 6: Comparison of Conventional and tapered beam-

forming output beam pattern
Figure 4: Output beampattern for different steering angles
C. Optimum Beamforming
B. Uniform Linear Array for tapered beamforming
Along with signal strength of 20dB for the SoI, three
The weights assigned to the antenna elements depend on interferences are assumed at three different angles, that is, at
the filter used to taper them, which is done to suppress 20, -30, 50 degrees with 35dB, 70dB, 50dB (INR) strength.
the side lobes while keeping the main lobe intact. The Here for the weights chosen, conventional beamformer does
filter used is a Chebyshev filter which displays equiripple not give the desired signal whereas optimum beamformer
behavior in the pass band resulting in a uniform sidelobe supresses all noise as well as interferers and the SoI is
level depending on the attenuation level specified. In tapered retained, showing as a distinct spike at sample number 50 in
beamforming, the magnitude of weight is maximum for the Figure 7. In the Figure 8 it is observed that at 20, -35 and 50
antenna element which is present at the centre and goes degrees there is a significant dip which shows the complete
on decreasing at either side. The filter command used was suppression of the interference sources known to be arriving
chebwin(number of taps, attenuation magnitude in dB). In from these angles. The side lobes show similar behaviour to
Figure 5 it is observed that for tapered weights the signals at that of the conventional beamformer response other than the
samples other than 50 (where it is expected our signal spike loss at the interference AoA. In the Figure 9 the comparison
to be present) are supressed as compared to the conventional between conventional and optimum beamforming is clearly
beamforming plot, the one in blue. The nature of the filter depicted. It is shown that in conventional beamforming
used reflects on the weights of the beamformer and in this suppression of interferer does not take place whereas in
case produces a taper proportionally steep to the attenuation optimum beamforming suppression of the interferer occurs
magnitude specified in the filter arguments. and the signal is retained at 0 degrees.
Figure 7: Comparison of conventional and optimum beam- Figure 11: With the loading value 0.1 the SMI and Diago-
former output in the presence of interference and noise nally loaded plots overlap as sigma is almost 0

found to be 100.

Figure 8: Output beampattern response for the Optimum

Condition

Figure 10: SMI adaptive beamformer beampattern

D. Adaptive Beamforming
Figure 10 shows that the adaptive beam pattern responses E. SMI Adaptive Beamforming with Diagonal Loading
with the interferer suppression at the angle of +20 degrees
Just as the SMI adaptive beamforming, the diagonal
where the interference source is modelled to have an AoA,
loading condition was also simulated using an iterative
main lobe at broadside complete with side lobe suppression.
technique to estimate the second-order interference plus
From the plot, it is observed that, as the number of samples
noise correlation matrix but with an extra term, a scaled
increases, the beampattern becomes finer. Depression at
identity matrix. However, a careful study of the values of the
20 degrees shows that the interferer is suppressed at that
co-efficient to be used for loading yielded interesting results.
angle, any signal arriving from that direction will be greatly
Figures 11 to 14 illustrate the progressive improvement in
attenuated. The number of samples-’K’ is optimised and
the Diagonal loaded results as the beam pattern response
converges towards the Optimum Pattern. And all of them ran
for the same number of iterations for weight computation.
The most suitable value for σ is found to be 5.

F. Nested Arrays with adaptive beamforming

Figure 15 shows exactly this for 5 cases of arrays with
different levels of nesting with respect to the same total
number of elements. 4 nested arrays and one 7 element
ULA with three interference sources each of 70dB strength
at angles -30, 20, and 50 degrees. The green plot in Figure
15 is the ULA plot for 7 elements and along with it are
4 cases of combination of 2-level nested array spacings: •
Figure 9: Comparison between conventional and optimum For N1 = 2 and N2 = 5 the position array has a total of 15
beamformers positions and 8 gaps
 111001001001001
Its weighted array was:
0 1 2 0 0 5 0 0 8 0 0 11 0 0 14
• For N1 = 3 and N2 = 4 the position array had a total 16
positions and 9 gaps
1111000100010001
Its weighted array was:
0 1 2 3 0 0 0 7 0 0 0 11 0 0 0 15
• For N1 = 4 and N2 = 3 the position array had a total of
15 positions and 8 gaps
111110000100001
Figure 12: With σ as 0.5, there is a difference between the The weighted array was:
two plots 0 1 2 3 4 0 0 0 0 9 0 0 0 0 14
• For N1 = 5 and N2 = 2 the position array had a total of
12 positions and 5 gaps
111111000001
Its weighted array was:
0 1 2 3 4 5 0 0 0 0 0 11

Figure 16 plots the loss in SINR plots for 10 sources

of interference for all 5 different arrays with 7 elements
each. The green plot with overall loss less than -100 dB for
the ULA shows its limitation with only 6 DoF at maximum
while each of the nested combinations performs better. Even
the N1 = 5 and N2 = 2 combination with only 11 DoF for
Figure 13: With loading coefficient 1.5, the black plot
this 10 + 1 = 11 source resolution performs much better,
corresponding to the Diagonal loading plot
albeit with quite a significant loss displayed at the center of
the plot, at around 90 degrees in the 0 to 180 angular sweep.

Figure 15: A plot of adaptive beam pattern responses for 5

cases Figure 16: loss in SINR plot for 10 interferences and 1 SoI

C ONCLUSION
Nested arrays promised an improved beamforming output
with lesser number of antennas forming a bigger virtual
array, thereby achieving a feat their uniform counterparts
could not achieve, they were found to be able to resolve a
greater number of source signals.

R EFERENCES
[1] W. Wang and C. Zhu, “Nested array receiver with time-
Figure 14: With σ as 5, Diagonal loaded plot takes shape of delayers for joint target range and angle estimation,” IET
Radar, Sonar and Navigation, vol. 10, Oct. 2016
optimum pattern
 [2] P. Pal and P. Vaidyanathan, “Nested arrays: A novel approach
to array processing with enhanced degrees of freedom,” IEEE
Transactions on Signal Processing, vol. 58, 8 Aug. 2010

[3] L. Yu, Y. Wei, and W. Liu, “Adaptive beamforming based on

non-uniform linear arrays with enhanced degrees of freedom,”
IEEETransactions on Signal Processing, Jun. 2015

[4] I. N. Psaromiligkos and S. N. Batalama, “Interference-plus-

noise covariance matrix estimation for adaptive space-time
processing of ds/cdma signals,” IEEE Fall Vehicular Tech-
nology Conference, Oct. 2014

[5] D. U. Nickel, “Fundamentals of signal processing for phased

array radar,” Advanced Radar Systems, Signal and Data
Processing

[6] D. G. Manolakis, V. K. Ingle, and S. M. Kogon, Statistical

and adaptive Signal Processing. Artech House Inc, 2005

[7] S. Cheng, Q. Zhang, M. Bian, and X. Hao, “An improved

adaptive received beamforming for nested frequency offset
and nested array fda-mimo radar,” MDPI, Sensors, Feb. 2018

[8] Z. Tian, K. L. Bell, and H. L. V. Trees, “A recursive

least squares implementation for lcmp beamforming under
quadratic constraint,” IEEE Transactions on Signal Process-
ing, vol. 49, Jun. 2001

[9] Z. Zheng and W. Q. Wang, “Robust and efficient adaptive

beamforming using nested subarray principles,” IEEETrans-
actions on Signal Processing, vol. 8, pp. 4076– 4085, Jan.
2020

[10] P. Pal and P. Vaidyanathan, “Beamforming using passive

nested arrays of sensors,” IEEE Transactions on Signal Pro-
cessing, pp. 2840–2843, 2010

[11] J. Yang, G. Liao, and J. Li, “Robust adaptive beamforming

in nested array,” Elsevier Signal Processing 114, Mar. 2015

[12] R. Abrahamsson, Y. Selen, and P. Stoica, “Enhanced covari-

ance matrix estimators in adaptive beamforming,” ICASSP
2007, pp. 969–972, Jun. 2007

[13] J. Liang, D. Jiang, Y. Han, Y. Zhang, and Z. Xie, “A robust

beamforming algorithm based on covariance matrix recon-
struction and steering vector estimation,” ICEMCE 2020,
2007

[14] P. Chen, Y. Zhao, and C. Liu, “Robust adaptive beamforming

using a low-complexity steering vector estimation and covari-
ance matrix reconstruction algorithm,” International Journal
ofAntennas and Propagation, 2016