HARDWARE EFFICIENT SCALING FREE VECTORING AND ROTATIONAL CORDIC FOR DSP APPLICATIONS

International Journal of VLSI design & Communication Systems (VLSICS) Vol.4, No.3, June 2013
DOI : 10.5121/vlsic.2013.4302 15
HARDWARE EFFICIENT SCALING FREE VECTORING
AND ROTATIONAL CORDIC FOR DSP
APPLICATIONS
Anita Jain1
and Kavita Khare2
Department of Electronics and Communication Engineering, MANIT, Bhopal, India
1
anitajainone@gmail.com,
2
kavita_khare1@yahoo.co.in
ABSTRACT
The COordinate Rotation DIgital Computer CORDIC algorithm has proved its versatility in computing
various transcendental functions by only using the shift and adds operations. This paper presents a new
hardware efficient scaling free CORDIC algorithm to operate in vectoring and in rotation mode. The
micro rotation of the vector is always in one direction with no scale factor correction. The Range of
Convergence RoC is from 0 to 2π. No pre and post processing circuitry is required. 16 bit Scaling free
CORDIC Pipelined architecture based on the proposed algorithm is synthesized on FPGA Xilinx VirtexII P
device coded in Verilog. Synthesized results show totally scaling free performance with very small
dynamic power consumption of .06 mW and maximum delay of 4.123 ns and 9.925 ns in the rotational and
vectoring modes respectively. The minimum BEP of the proposed algorithm implementation is 12.
Proposed algorithm is faster and efficient in terms of area and accuracy as compared to conventional
CORDIC.
KEYWORDS
DSP, Scale Free CORDIC, Systolic array, rotation mode, vectoring mode, RoC, pipeline architecture.
1. INTRODUCTION
Many Digital Signal Processing applications are based on either trigonometric functions
computation or matrix computations. The trigonometric functions are used for calculation of
various transforms such as fast fourier transform (FFT), discrete sin/cosine transform
(DST/DCT), discrete hartley transform (DHT) and Hough transform (HT)[14] etc. Other signal
processing applications, such as the adaptive beamforming, MIMO decoders, the 3G wireless
communication, the software defined radio and cognitive radio uses matrix computations such as
the QR decomposition [12] etc. To factorize the matrix, systolic arrays processors are used which
comprise of CORDIC as processing element (PE). The number of processing element (PE)
increases with the order of matrix. Also the CORDIC used as PE needs to be operated in either of
rotation mode or vectoring mode depending on their position in systolic array. CORDIC
processors are also used to generate signals during modulation and to estimate phase and
frequency parameters during demodulation [13].
Efficient hardware implementation of CORDIC on Field Programmable Gate Arrays (FPGA),
become necessary for above applications. This paper proposes a hardware efficient scaling free
CORDIC algorithm to operate in both modes rotation and vectoring. Being Scaling free
CORDIC with no pre and post processing circuitry, the complexity and latency of the system
employing large number of CORDIC as PE is reduced significantly.

16
This paper is structured as follows: Section 2 explains the overview of CORDIC algorithm;
Section 3 details the steps of proposed scaling free CORDIC algorithm. FPGA Implementation
and results together with error analysis is furnished in Section 4; Section 5 summarizes the
conclusion drawn; references are listed in Section 6.
2. CORDIC ALGORITHM OVERVIEW
CORDIC Algorithm is based on rotation of vectors in two dimensional coordinate space using
simple shift and add operation. CORDIC Algorithm can operate in two modes namely: rotation
and vectoring. In rotation mode, the objective is to convert polar coordinate of a vector into
Cartesian where as vice versa in vectoring mode through a series of iterations. The rotation
trajectory can be linear, circular or hyperbolic depending upon the requirement.
2.1. Conventional CORDIC Algorithm
The conventional CORDIC algorithm [4] is derived from general equation of vector rotation. If a
vector V with components (Xi, Yi) is iteratively rotated through an angle αi, a new vector V’ with
components (Xi+1, Yi+1) is formed. In matrix form, the value of vector after this micro rotation
can be represented as:
The sign sequence di Є{1,-1} is so selected that:
Where, ‘w’ is the word-length in bits.
Note that the range of convergence of this algorithm is limited to [-99.99°, 99.99°], which can be
extended to entire coordinate space using the properties of sine and cosine functions, using an
extra iteration for full-range rotation.
The overall scaling-factor of above CORDIC iterations is given by (2)
2.2. Unified CORDIC Algorithm
Walther [5] has extended the scope of conventional CORDIC algorithm suggested by Volder to
include linear and hyperbolic trajectory along with circular trajectory. Due to this extension
exponential, logarithmic and various other functions also can be computed using CORDIC. Thus
the application and usefulness of it is broadened. A variable (m) for defining the trajectory was
introduced to modify the basic CORDIC rotation matrix and elementary angle ‘α’ as:

17
2.3. Scaling-Free CORDIC
Scaling-free CORDIC [8] was the first attempt to completely dispose of the scale-factor. Here,
the sine and cosine functions were approximated to:
However, the approximation imposes a restriction on the basic-shift i=⌊((w-2.585))⁄3⌋. For
16-bit data, the basic-shift = 4 results in extremely low range of convergence. However, modified
virtually adaptive scaling-free algorithm [9] extends the range of convergence over the entire
coordinate space and introduces an adaptive scale-factor.
2.4. Review of Existing CORDIC Architectures
Various architectures have been suggested to improve the efficiency and reduce the complexity of
the CORDIC algorithm. The path of development is given chronologically in the paper ’50 years
of CORDIC’ [3]. Vectoring CORDIC processor using conventional CORDIC was first developed
by [11] for an arbitrary target value. It has limitation of bulky scale factor compensation circuit
and higher hardware cost. Virtually scaling free CORDIC suggested by K. Maharatna eliminates
the scaling factor from the CORDIC equations, but the major drawback of this technique is very
small range of convergence due to which it can neither be applied for the rotation operation with
large angles nor for the vectoring mode of operation. A virtually scaling free vectoring CORDIC
using combination of conventional and modified CORDIC was suggested by Stapenhurst [6]. But
it has still fixed value of scaling factor and complex pre and post processing unit. Use of radix- 4
booth recoding for reduction of number of iteration is suggested in Enhanced scaling free
CORDIC [2] for rotation mode of CORDIC. In revised new CORDIC algorithm [1] an area and
time efficient scaling free CORDIC is designed using leading one bit detector. This has efficiently
implemented completely scale free rotational CORDIC.
3. PROPOSED ALGORITHM FOR SCALING FREE CORDIC
The proposed algorithm for scaling free CORDIC is based on third order approximation of Taylor
series.
The Taylor series expansion of sine and cosine of an angle is:
But, this approximation imposes a restriction on the allowed values of iterations i as:
For 16 bit word length, the initial value of ‘i’ required for maintaining the accuracy of the
calculations comes out to be 2. It divides the coordinate space into eight equal sectors, each of 45
degrees.

18
CORDIC equation for the above approximation reduces to:
With this division the range of convergence is (0, π/4). To extend the RoC to entire coordinate
space quadrant mapping is done as per table 1. for rotation mode and table 2. for vectoring mode.
3.1. Proposed Algorithm for Scaling Free Rotation CORDIC
In rotation mode, angle θ is given as input for which sin θ and cos θ is to be determined as per
the CORDIC equations. The Pseudo code for the proposed algorithm is:
Input: z
Output: X, Y
Begin
Step 1: Identify and map the sector and quadrant of the input angle
of the vector.
Step 2: Initialize i = 2; v = word length w;
Step 3: If z [v] = 0; skip the rotation;
Else go to step 4;
Step 4: Rotate the vector and compute its next iterative value.
Step 5: Increment i; Decrement v;
Go to step 3
Step 6: Repeat until (i <= w)
Step 7: Out X, Y
Step 8: Restart with new data in pipeline.
End
Table 1. Quadrant Mapping For Rotation Mode
Range of Target Angle X Y
Degrees Radians
[0°, 45°] [0, π/4] x y
(45°, 90°] (π/4,π/2] y x
(90°, 135°] (π/2,3π/4] -y x
(135°, 180°] (3π/4,π] -x y
(180°, 225°] (π,5π/4] -x -y
(225°, 270°] (5π/4,3π/2] -y -x
(270°, 315°] (3π/2,7π/4] y -x
(315°,360°] (7π/4,2π] x -y
3.2. Proposed Algorithm for Scaling Free Vectoring CORDIC
In vectoring mode, Cartesian to polar coordinate conversion is carried out as per the CORDIC
equation. The Pseudo code for the proposed algorithm is:
Input: X, Y
Output: z, Xout
Begin
Step 1: Identify and map the sector and quadrant of the input vector.
Step 2: Initialize i = 2;

19
Step 3: Rotate the vector and compute its next iterative value.
Step 4: If MSB of yi = 1; z = 0; i = i+1; go to step 3,
Else z = 1;
i = i+1;
Go to step 3
Step 5: Repeat until (i <= word length)
Step 6: Out Xout = mag and z = phase
Step 7: Restart with new data in pipeline.
End
Table 2. Quadrant Mapping For Vectoring Mode
X Y Sector Final Angle θ
x y I θ
x y II (90°-θ)
-x y I (180°-θ)
-x y II (90°+θ)
-x -y I (180+θ)
-x -y II (270°-θ)
x -y I (360°-θ)
x -y II (270°+θ)
4. FPGA IMPLEMENTATION OF PROPOSED ALGORITHMS
Functional simulation and Hardware implementation of the proposed algorithms for pipeline
architecture is carried out in Xilinx ISE9.2i on VirtexII pro device using Verilog. Test bench
waveforms are used for simulation purposes. Summary of hardware used is given in table 3.
Table 3. Hardware Summary
Parameter Rotation mode Vectoring mode
Flipflops 387 462
LUTs 1344 1482
IOB 86 66
BUFG 1 1
Maximum delay 4.123 ns. 9.925 ns.
Dynamic power dissipation at 20 MHz is found to be .06 mW.
4.1. Error analysis and Mathematical Verification of Simulated Values.
There are two types of error in the results produced by CORDIC algorithm namely quantization
error [16] and truncation error. The first arises due to rotation of vector by finite number of
iterations and second one is due to truncation of intermediate values because of fixed word
length.
The cumulative effect of both these error is analysed by verifying the simulated values
mathematically. The error is specified in terms of Bit Error Position BEP. According to Kota
[15], using 16 bit word length, the achievable upper bound of the error is approximately 10 b.
Mathematical verification is shown in Table 4. and 5. Results show that the proposed algorithm is
better in performance as its minimum BEP is 12 i.e 12 bits are error free out of 16.
This algorithm has used 16 bit fixed point binary representation with 1 represented as
0100_0000_0000_0000. X and Y outputs shown are hexadecimal values.

20
Table 4. Mathematical Verification And Bit Error Position (Rotation Mode)
Angle X output BEP Y output BEP
degree hexade
cimal
Theoretica
l value
Actual
Value
Theoretic
al value
Actual
Value
30° 2183 376D 376C 15 2000 2002 14
60° 4305 2000 2001 15 3763 3764 14
135° 96CC D2BF D2BA 13 2D41 2D56 12
180° C910 C000 C000 16 0000 0000 16
Table 5. Mathematical Verification And Bit Error Position (Vectoring Mode)
X Y BEP
Hexadecimal
Value
Hexadecimal
Value
Theoretical
value
Actual
Value
C89D DFEE EA92 EA94 14
DFEE C89D 10C15 10C18 13
2012 C89D 14F1A 14F1C 14
0000 C000 12D98 12D98 16
5. CONCLUSIONS
This paper has presented a novel algorithm for vectoring and rotation mode of CORDIC which is
free from any scaling factor correction and complex circuitry for pre and post processing of
vectors. It has also proved its efficiency in hardware implementation. Logic blocks used for its
implementation are less as compared to conventional CORDIC. It consumes less power and is
faster in convergence with RoC covering complete coordinate from 0 to 360 degrees. Its latency
is also low as it skips the operation not required as per the algorithm and also by rotating
unidirectionally. The accuracy of the proposed algorithm is better as it uses higher order
approximation of Taylor series for deriving its CORDIC equation.
Future scope of this algorithm can be extended to cover three dimensional coordinate space as
well as other DSP applications.
REFERENCES
[1] Supriya Aggarwal, Pramod K. Meher, and Kavita Khare, (2012) “ Area-time efficient scaling-free
cordic using generalized micro-rotation selection”, IEEE Transactions on VLSI Systems, 20(8):1542–
1546.
[2] F. J. Jaime, M. A. Sanchez, J. Hormigo, J. Villalba, and E. L.Zapata, (2010) “ Enhanced scaling-free
cordic. IEEE Transactions on Circuits and Systems”, 57(7):1654-1662.
[3] Pramod Kumar Meher, JavierWalls, Tso-Bing Juang, K. Sridharan, and Koushik Maharatna,
(2009)“50 years of cordic : Algorithms and architectures and applications”, IEEE Transactions on
Circuits and Systems I, 56(9):1893–1907.
[4] J. E. Volder, (1959) “The CORDIC Trigonometric Computing Technique, “ IRE Trans. Electronic
Computing, Vol. EC-8, pp. 330-334.

21
[5] J. S. Walther, (1971) “A unified algorithm for elementary functions”, Spring Joint Computer Conf.,
Proc., pp. 379-385.
[6] R. Stapenhurst, K. Maharatna, J. Mathew, J.L. Nunez-Yanez, and D.K. Pradhan, (2007) “A cordic
like processor for computation of arctangent and absolute magnitude of a vector”, In Proc. IEEE
International Symposium on Circuits and Systems ISCAS 2007, pages 3002–3005.
[7] J. Valls, T. Sansaloni, A. Perez-Pascual, V. Torres, and V. Almenar,(2006) “ The use of cordic in
software defined radios: A tutorial”, IEEE Communication Mag., 44(9).
[8] Maharatna, A. Troya, S. Banerjee, and E. Grass,(2004) “Virtually scaling free adaptive CORDIC
rotator,” IEE Proc.-Comp. Dig. Tech., vol. 151, no. 6, pp. 448–456.
[9] K. Maharatna, S. Banerjee, E. Grass, M. Krstic, and A. Troya, (2005) “ Modified virtually scaling-
free adaptive cordic rotator algorithm and architecture”, IEEE Trans. Circuits Syst. Video Technol.,
11(11):1463–1474.
[10] K. Maharatna, A. Troya, M. Krstic, E. Grass, and U. Jagdhold,(2004) “ A cordic like processor for
computation of arctangent and absolute magnitude of a vector”, In Proc. ISCAS, volume 2, pages
713–16.
[11] T. Lang and E. Antelo,(1998) “Cordic vectoring with arbitrary target value”, IEEE Transactions on
Computers, 47(7):736–749.
[12] Dongdong Chen and Mihai SIMA,(2011) “Fixed-Point CORDIC-Based QR Decomposition by
Givens Rotations on FPGA”, International Conference on Reconfigurable Computing and FPGAs.
[13] José Marcelo L. Duarte, Francisco das Chagas Mota,(2007)” Digital PM Demodulator for Brazilian
Data Collecting System”, Proceedings of the 20th annual conference on Integrated circuits and
systems design , Pages 142-146
[14] K. Maharatna, A. S. Dhar, and S. Banerjee, (2001)“A VLSI array architecture for realization of DFT,
DHT, DCT and DST,” Signal Process., vol. 81, pp. 1813–1822.
[15] K. Kota, J.R. Cavallaro, ( 1993 )“ Numerical Accuracy and Hardware Tradeoffs for CORDIC
Arithmetic for Special-Purpose Processors”, IEEE Transactions on Computers,vol. 42, no. 7, pp. 769-
779.
[16] Y. H. Hu, (1992) "The Quantization Effects of the CORDIC Algorithm," IEEE trans. Signal
Processing, Vol. 40, No. 4, pp. 834-844.

22
AUTHORS
Anita Jain received the B.Tech in 1994 and M. Tech degree in 2009 with
specialization in Electronics and Communication Engineering . Currently, she is
pursuing her Ph.D. as full time research scholar in Electronics and Communication
Engineering in MANIT, Bhopal, India. Her fields of interest are VLSI design. Her
research mainly includes front end designing of VLSI circuits. She has ten papers in
the proceedings of various Intern ational and National Conferences. Ms. Jain is a life
member of ISTE national society.
Kavita Khare received the B.Tech degree in electronics and communication engg in
1989, M.Tech. degree in digital communication systems in 1993, and the Ph.D.
degree in the field of VLSI design in 2004.Currently, she is working as Professor in
Electronics and Communication Engineering in MANIT, Bhopal, India. Her fields
of interest are VLSI design and communication systems. Her research mainly
includes design of arithmetic circuits and various commn algorithms related to
synchronization, estimation and routing. She has nearly 110 publications in various
international conferences and journals. Dr. Khare is a Fellow of IETE (India) and a
Life Member of ISTE.

HARDWARE EFFICIENT SCALING FREE VECTORING AND ROTATIONAL CORDIC FOR DSP APPLICATIONS

More Related Content

What's hot (19)

Viewers also liked (19)

Similar to HARDWARE EFFICIENT SCALING FREE VECTORING AND ROTATIONAL CORDIC FOR DSP APPLICATIONS (20)

Recently uploaded (20)

HARDWARE EFFICIENT SCALING FREE VECTORING AND ROTATIONAL CORDIC FOR DSP APPLICATIONS