International Journal of Computer Applications (0975 – 8887)

Volume 67– No.20, April 2013

Matlab Simulation of Cordic based Adaptive Filtering for

Noise Reduction using Sensors Array
Anamika Gupta Rajesh Mehra
ME. Student Associate Professor
Department of ECE Department of ECE
NITTTR,Chandigarh NITTTR, Chandigarh

ABSTRACT important. There are several algorithms to achieve the

The process of suppressing noise and interference in any type calculation of coefficients in a given system, which vary in
of signal for transmission has created challenges in the area of complexity. Among the most simple is the Least Mean Square
digital signal processing. Apart from mobile communication algorithm (LMS). This algorithm is widely used because of its
devices, there are a huge number of applications, in which it is ease of implementation and low utilization of computer
difficult to have a good acoustic interface for accurate voice resources. When the medium is highly dynamic, requires
control or smooth audio communication. So for this it is very algorithms that adapt quickly to changes, for these cases the
essential to enhance the signal by removing its noise.For this a LMS algorithm do not provide a good Performance [3].
cordic based adaptive algorithm has been designed and If LMS algorithms represent the simplest and most easily
simulated for digital signal processing. When the input signals applied adaptive algorithms, the recursive least squares (RLS)
statistics to the filter are time varying or unknown and the algorithms represents increased complexity, computational
design requirements for fixed filters cannot easily be specified cost, and fidelity. In performance, RLS approaches the Kalman
then adaptive filter is useful. Examples of such applications filter in adaptive filtering applications, at somewhat reduced
are system identification, channel, equalization /identification, required throughput in the signal processor. Compared to the
noise reduction in digital signal processing. These algorithms LMS algorithm, the RLS approach offers faster convergence
are also used in case of beamforming. Array of sensors and smaller error with respect to the unknown system, at the
(microphones) play an important role in noise reduction and expense of requiring more computations. In contrast to the
speech enhancement. In this paper a cordic based qrd_rls least mean squares algorithm, from which it can be derived,
adaptive (CQR_RLS) algorithm is developed and simulated the RLS adaptive algorithm minimizes the total square error
using MATLAB. Instead of taking signal from one sensor here between the desired signal and the output from the unknown
array of sensor is used for signal as well as interference.Matlab system.
simulation results are calculated and compared for different
iteration. Suppose in a room an array of sensor is places In the case of
speech signal, a person who is near to first sensor of the sensor
General Terms array take voice signal as a useful signal while all other sensor
Adaptive algorithm, Matlab
takes this signal as interference signal, because that signal is
Keywords not in the frequency range of the sensors. If person moves
Qrd_rls, cqr_rls from one place to another, then the sensor which is near to the
person takes person’s voice as a useful signal. So for this
1. INTRODUCTION reason an array of sensor is placed in place of one sensor.The
In the last many years the day by day advances in digital paper is organized as follows: Section 2 explains the basic
circuit design have been the main key technological concepts adaptive algorithms in general. Section 3 shows the
development that increased a growing interest in the field of architecture and design platform. This paper shows the
digital signal processing. The main attraction of digital signal implementation of RLS algorithm which is based on cordic
processing systems is their accuracy, less cost, reliability, processing. The proposed algorithm is developed in MATLAB
small physical sizes, and flexibility. The most efficient and after that the simulated results for different iterations are
example of DSP is filter. The main aim of filtering is filter out compared.
the noise and interference contained in the signal. A filter that 2. ADAPTIVE ALGORITHM
processes discrete-time signals represented in digital format is An adaptive system can be modeled as shown in Figure 1. As
called digital filter. In some cases when using digital filters, can be seen there is a plant with defined characteristics, the
signals or systems may undergo some changes with time or output is input to the adaptive mechanism after being
some noise and interference occurs but the exact nature of subtracted of a desired signal, whereby adaptive mechanism
change is not predictable then in such cases it is highly can calculate the new coefficients needed to adapt the response
required to design a filter that can learn from the process itself, of the plant to the desired. The operation model equations
way that can be adapted to handle the situation. To resolve showed an adaptive system [4].
many of these problems, it is proposed to use adaptive filters
[1].Today adaptive systems have found their way into many
applications where learning capacity of the system is a factor
y(n)=x(n)*w(n) (1)
important. An adaptive filter is a filter that self-adjusts e(n)=d(n) - y(n) (2)
its transfer function according to an optimization algorithm e(n)=d(n) – [x(n)*w(n)] (3)
driven by an error signal [2].
Today adaptive systems have found their way into many Where x(n) is the input signal, y (n) is the filter output, d (n) is
applications where learning capacity of the system is a factor the desired output signal and e(n) is the error between d(n) and

29
 International Journal of Computer Applications (0975 – 8887)
Volume 67– No.20, April 2013

y(n). In this case, the signal input x(n) moves into the plant There are four major types of adaptive filtering configurations,
that contains the coefficients w (n) (FIR filter) and returns a adaptive system identification, adaptive noise cancellation,
signal y(n) whose result is shown adaptive linear prediction, and adaptive inverse system. All of
the above systems are similar in the implementation of the
algorithm, but different in system configuration [5]. Figure 3
shows array of sensor based adaptive algorithm system in
which 3 sensors are used for interference signal in the place of
one.
Y[n] d[n]
Filter DESIRE
x(n) SIGNAL Sensor 1
Σ
-
W[n] e[n
Adaptive
algorithm
NOISE
SIGNAL Sensor 2 ADAPTIVE
FILTER 1
Fig 1: Diagram of an adaptive system

in equation 1. Then an error signal e(n) is computed by the

NOISE Sensor 3
subtraction of result y(n) from a signal d(n) ,whose result is SIGNAL ADAPTIVE
shown in equation (2), which is the parameter that tells the FILTER 2 Σ
adaptive mechanism how "far" is the mechanism to have a
response similar to the desired signal d(n). With the help of
this error signal and the input signal new coefficients w (n) are
calculated for the Filter using an Adaptive algorithm. NOISE
SIGNAL Sensor 4 ADAPTIVE
In the noisy environment speech signal is affected by the FILTER 3
presence of noise signal (acoustic noise).To solve this problem
one possible solution of obtaining a better recording of desired
signal is simple sensor array system with adaptive filter as
shown in figure 2.
Fig 3:sensor array based adaptive noise cancelation
system

signal sensor 1 desire signal 2.1 LMS Algorithm

source Σ The traditional LMS filtering algorithm is an approximation to
using gradient descent to find the optimal filter coefficients by
finding the minimum mean square error (MMSE) between the
filter output and some desired output. It is an iterative
F
- Error procedure where the coefficients can be updated according to
signal the gradient of the MSE .Least Mean Square (LMS) is the
sal most common and popular algorithm The LMS algorithm is
noise Adaptive very popular and has been widely used due to its extreme
source filter simplicity [5].

2.2 RLS Algorithm

sensor 2 On the other hand RLS (Recursive Least Square), is generally
preferred for its fast convergence. The direct calculation of the
new vector of coefficients involves matrix inversion, which is
sal usually unwanted in implementations of hardware due to the
sal high consumption of resources. The based matrix
Fig 2: adaptive noise cancelation system decomposition schemes such as least squares, SVD (Singular
Value Decomposition) and explicitly avoiding QR matrix
inversion are more robust and more affordable implementation
In the above figure 2 the path at which signal coming from the in hardware [6].Table 1 shows the comparison between LMS
noise source to the sensor 1,which is primary sensor is as and RLS.
unknown FIR channel F.If an adaptive filtering is applied to
the noise source at the sensor 2,then it is possible to employ an
adaptive algorithm to train the adaptive filter.

30
 International Journal of Computer Applications (0975 – 8887)
Volume 67– No.20, April 2013

Table 1 : Comparison Table of LMS and RLS

E (n)  Q(n)1/2 (n)e(n) 2 (9)
E (n)  Q(n)1/2 (n)d (n)  Q(n)1/2 (n) A(n) w(n) 2

Mult. Add. Sub. (10)

Algorithm Computational Convergence
complexity speed Where λ is a forgetting factor slightly less than 1, and

LMS O(N) LOW N(2M+1) N(2M) 2N  1/2

 diag ( (n  1),  (n  2).... (0)) .The
minimization problem defined by the cost function, the unit
RLS O(N2). FAST N(6M) N(3M) N(2M)
matrix Q (n) is chosen to triangular matrix data exponentially
weighted such that

2.3 QRD_RLS Algorithm  R ( n) 

Q(n)1/2 (n)d (n)    (11)
QR-RLS algorithm is used to solve linear least square
problems. The decomposition is the basis for the QR
 0 
algorithm. As all methods are iterative, their development and Where R (n) is an upper triangular matrix dimension k x k and
constant improvement aim for reduction of computational 0 is a zero matrix of dimension (n-K) x K. The desired signal
complexity, increased speed of convergence, and robustness vector, after being converted, is defined by:
against round-off errors. QR matrix decomposition (QRD) is  p ( n) 
the decomposition of a matrix (A) into an orthogonal matrix Q(n)1/2 (n)d (n)    (12)
(Q) and an upper triangular matrix (R). QRD is useful for  v ( n) 
solving least squares’ problems and simultaneous equations. Where p (n) is a vector of elements K X 1, v (n) is a vector (n-
K) x 1 element, then we can rewrite the cost function as
3. QR DECOMPOSITION follows:
The QR algorithm, which is based on the QR decomposition of  p ( n)   R ( n) 
A, is still considered one of the most important methods E ( n)     w(n)
2
(13)
developed so far. the QR algorithm decomposes A into a  v ( n)   0 
unitary matrix Q and a (different!) upper triangular matrix R,
instead of the lower and upper triangular matrices from
elimination. The QR algorithm uses successive unitary  p(n) R(0) w(n) 
transformations, which render the method superior with E ( n)   
2
(14)
respect to numerical stability and computational requirements  v ( n0 
A  QR (4) Obviously, the least squares estimation for the weight vector
where must satisfy that:
Q   q1 q2 qN  (5)
And w '(n)  R1 (n) p(n) (15)
The unitary matrix Q (n), the upper triangular matrix
 r11 r12 r1N  R (n), and the vector p (n) can be calculated recursively using
R   0 r 22 r 2 N  (6)
 0 rNN 
 R ( n) p ( n) 
There are three methods for factorization of matrix.The first 0(n  K  1) xK 0(n  K  1) 
classical and the modified versions of the Gram–Schmidt

  ( N ) 
orthogonalization method based on projections. Next is
Householder orthogonalization method based on reflections 01xK
and the last one is Givens orthogonalization method based on
rotations. Among all three Gives rotation is the most suitable
and economical [7]. The QR decomposition method starts
from the data matrix using unitary transformation. An error  1/2 R(n  1) 1/2 p(n  1) 
vector is defined as  
 Q '(n) 0(n  K  1) xK 0( N  K  1) x1
e(n)  d (n)  A(n)w(n) (7)
 u T ( n) d ( n) 
Now the cost function can be define as  
(n)  t 1 |e(i ) |2
n
(8)
Q(n  1) 0
Q(n)  Q '(n) 
1 
(16)
For a given matrix Q(n), the cost function may be expressed  0
as[8]:
Therefore, the optimal vector of coefficients can be
obtained. But in some applications such as noise reduction and
linear prediction e (n) is the signal output. Developing the

31
 International Journal of Computer Applications (0975 – 8887)
Volume 67– No.20, April 2013

previous equation can obtain e (n) directly without removing d(n)+N(n)

the weight vector explicitly. This result can be summarized
following equation: noise signal
 R(n) p(n) R (n)u (n)  1 from sensors
array N(n)’
y(n)
  Filter
(17)
Σ
 01xk  (n) ( (n))* 
1/2

1/2 R(n  1) 1/2 p(n  1) 0kx1  e(n)

Q ''(n)  T 
 u ( n) d ( n) 1  (18) CQR_RLS
ALGORITH
4. CQRD_RLS ALGORITHM
Efficient generation of trigonometric as well as exponential
functions without much increase in hardware complexity has Fig 4:sensor array based cqr_rls adaptive system
always been a challenge, owing mainly to their importance and
widespread use in Digital Signal Processing applications 5. SIMULATION RESULT
besides other areas.
In this paper we have designed and simulated proposed
algorithm using MATLAB. Then the results are compared for
One such algorithm which is very much effective for the
different adaptive iteration. Figure 5 shows the input signal
calculation of trigonometric, hyperbolic, exponential, linear
which is useful signal and the interference signal used in
and logarithmic functions is the CORDIC algorithm .CORDIC
cqr_rls adaptive system.
(Coordinate Rotation Digital Computer) is an iterative
algorithm for the calculation of the rotation of a two-
dimensional vector, in linear, circular and hyperbolic
coordinate systems, using only add and shift operations. A
CORDIC describes a method to perform a number of
functions, including trigonometric, hyperbolic and,
multiplication with the help of addition and shifting only. The
algorithm is very much hardware efficient because it omits the
dependence on multipliers and is rather a combination of shift-
add operations [9]. The CORDIC is a shift-and-add technique
for computing a large class of mathematical functions in
hardware. The algorithm can be derived from the rotation
transform.
Fig 5:Input signal and interference signal
x '  x cos   y sin  (19)
y '  y.cos   x.sin  (20)

On rearrangement of the terms, this can be given as

x '  cos [ x  y.tan  ] (21)

y '  cos [ y  x.tan  ] (22)

The implementation of these equations is still complex due to

the presence of the trigonometric functions. If the rotation
angles are restricted to values such that tan ϕ = ± 2-i , the
multiplication by the tangent can be greatly simplified as it can Fig 6:Input signal contaminated with inteference signal
be implemented using simple shift and addition operations.
Normally the calculation of an FFT consists complex Figure 6 show the input signal contaminated with interference
multiplications. These requirements can be minimized by signal and its frequency domain. Now we will discuss the
replacing the complex multipliers by applying CORDIC desire signal and output signal for different iterations.
algorithm. For implementing QR decomposing RLS algorithm
there are three methods for factorization of matrix. Among Figure 7, 8, 9 and 10 shows the result of the proposed
these the last one is Givens orthogonalization method based on algorithm for 60, 80,150 and 250 iterations, which shows that
rotations is the most suitable and economical. But the after 60 to 80 iterations we are able to get the output signal
hardware consumption is somewhat high. So in the proposed very close to the desire one but still have some noise signal.
algorithm the Givens rotation is implemented by CORDIC After 150 iterations output is very similar to the desire one
algorithm and shown in figure 4. almost negligible noise as shown in fig. And when this
algorithm is computed for 250 iteration then the output is
almost a replica of the desire signal as shown in figure 10.So
we can see as the number of iteration increases the output
moves toward the desire signal.

32
 International Journal of Computer Applications (0975 – 8887)
Volume 67– No.20, April 2013

Fig 7 Result of CQR_RLS algorithm for 60 iteration

Figure 8 Result of CQR_RLS algorithm for 80 iteration

Figure 9 Result of CQR_RLS algorithm for 150 iteration

33
 International Journal of Computer Applications (0975 – 8887)
Volume 67– No.20, April 2013

Figure 10 Result of CQR_RLS algorithm for 250 iteration

6. CONCLUSION Microphone Array Algorithms for Speech

Enhancement”, IEEE Conference on Application
The CQR-RLS algorithm is an excellent way to filter out any Specific Systems, Architecture and processors, pp.203-
signal using a signal reference d(n) as a model,. The proposed 208, 2008.
algorithm in which decomposition is based on Cordic
algorithm is more efficient in terms of resource utilization. [5] Jing Dai and Yanmei Wang,”NLMS Adaptive
This adaptive algorithm is also used in beamforming for Implement Based on FPGA”, IEEE Conference on
signal enhancement Array of sensors (microphones) play an Intelligent Networks and Intelligent System, pp-422-425,
important role in noise reduction and speech October 2010.
enhancement.Instead of taking signal from one sensor here [6] B. Yang and JF Bohme, "Rotation-based RLS
array of sensor is used for signal as well as algorithms: Unified derivations, numerical properties,
interference.Matlab simulation results are calculated and and Parallel Implementations, "IEEE Transactio
compared for different iteration. on Signal Processing, vol. 40, pp. 1151-1167, May 1992.
7. REFERENCES [7] Ka Fai Ceric and Yiu,Chun Hok Ho,”Reconfigurable
Acceleration of Microphone Array Algorithms for
[1] S. Haykin, "Adaptive Filter Theory", 3rd ed. Upper
speech Enhancement”,IEEE Confrence,2008
Saddle River, NJ: Prentice-Hall, 1994.
[2] Emanuël A. P. Habets, Jacob Benesty, and Patrick A. [8] M. E. I. Martínez, “ Implementation of QRD-RLS
Naylor.”A speech distortion and interference algorithm on FPGA. Application to Noise Canceller
rejectionconstraint beamformer”, IEEE transactions on System “, IEEE latin america transactions, vol. 9, no. 4,
Audio, Speech and Language Processing, vol 20, no. 3, pp.458-462, july 2011.
pp. 854-867, March 2012.
[9] Leena Vachhani, K. Sridharan and Pramod Kumar
[3] Saeed V. Vaseghi, Advanced Digital Signal Processing
Meher,” Efficient FPGA Realization of CORDIC With
and Noise Reduction, Third Edition, pp.45-46, 2006.
Application to Robotic Exploration” IEEE Transactions
[4] Yiu,K.F.C.;HoKHOChun,Grrric,Nedelko;Lu;Yao;Shi,Xi on Industrial Electronics, vol. 56, no.12,pp.4915-4919,
aoxiang;Luk,Wayne S., “Reconfigurable Acceleration of December 2009.

34