Sine Approximation for Direct Digital

Frequency Synthesizers and Function Generators

Milan Stork
Applied Electronics and Telecommunications, Faculty of Electrical Engineering/RICE
University of West Bohemia, CZ
Univerzitni 8, 30614 Plzen
CZECH REPUBLIC
[email protected]


Abstract: - There are several compression methods for phase to amplitude converter for sine function. In this
paper, the linear and parabolic approximation of equi-section division utilizing the symmetry property and
amplitude approximation of a sinusoidal waveform for direct digital frequency synthesizer (DDS) and also for
other waveform applications, e.g. function generators and telecommunication devices is proposed. The
sinusoidal phase from 0 to /2 is divided into different numbers of equi-sections. The linear and parabolic
approximations are compared. The simulation results show spectral properties of booth approach. The
presented sine approximation is usable for FPGA implementation.


Key-Words: Direct digital synthesis, frequency spectrum, frequency synthesizers, linear approximation,
parabolic approximation, phase to amplitude converter, simulation.

1 Introduction
Direct digital frequency synthesis is increasingly
welcomed in modern communication systems and
precise electronic systems, due to their significant
advantages over phase-locked loop (PLL) based
synthesizers. Fast frequency switching, excelent
frequency resolution, continuous-phase switching
and low phase noise are important features easily
obtainable in direct digital synthesizers is a well-
known technique for the generation of a sinusoidal
waveform reconfigurably [1]. DDS techniques have
lots of advantages over analog approaches beyond
flexibility, including excellent frequency resolution,
fast response time, fast frequency switching, and
low phase noise.
The DDS architecture, is shown in Fig. 1. The
system has two inputs: a clock frequency fclk and a
frequency control word (FCW). The phase
accumulator accumulates the value of the FCW in
every clock cycle, periodically overflowing. Its
ramp output is considered as a phase in the interval
0, 2, while the phase-to-sinusoid amplitude
converter provides a digital amplitude value of the
sinusoidal waveform of that phase. The output
frequency of DDS is given by

2
o clk N
FCW
f f = [Hz] (1)
The sequences of digital amplitude values pass
through a digital-to-analog converter (DAC) and
usualy to low-pass filter (LPF) to generate a
sinusoidal signal [2].

F
R
e
g
P / D A
Clk
A
Phase accumulator

Fig. 1. The DFS architecture block diagram.
F- frequency control word, Clk- clock generator,
Reg register, P phase-to-sine amplitude
converter, D/A digital/analog converter.

There are a wide variety of methods for phase to
sine-amplitude conversion [3, 4, 5], from full-
memory to full-hardware approaches. The methods
can be placed into three categories.
a) Based on ROM lookup tables (LUT), containing
amplitude samples of the complete, half, or quarter
period of sinusoid, and addressed by the digital
phase, which is generated by the phase accumulator
[1, 2].
Latest Trends in Circuits, Automatic Control and Signal Processing
ISBN: 978-1-61804-131-9 127
b) Full computational methods, trying to compute
the sine amplitude samples from the digital phase,
which is generated by the phase accumulator. Sine
computation by using Taylor series [1] and
CORDIC algorithm [6] are two examples of this
category.
c) Initial gusess correction methods, in which an
initial guess for the sine function is generated by
digital hardware and then is corrected by small
ROM lookup table, containing the difference
between the initial guess and the accurate value for
the sine amplitude. If the initial guess is properly
provided, in each memory location just small
correcting data will be stored instead of the whole
sine amplitude. It is obvious that the closer the
approximation is to ideal sine function, the more
memory sawing there will be.
There are several ROM compression methods,
such as the quarter-wave symmetry method [7], the
sine amplitude approximation, eg.
Sine-Phase Difference Algori thm [2].
Modified Sunderland Archite cture [4].
Nicholas Architecture [2].
Taylor Series Approximation [5].
The another possibility is ROM-less DDS with
Sin-weighted DAC.
With the quarter-wave symmetry method, the
ROM only needs to store one-quarter (the phase
from 0 to /2) of the sine wave sequences [8 - 15].
The two most significant bits (MSBs) of the phase
value determine to which quadrant an angle belongs,
while the residual M2 bits are used to address a
one-quadrant sine LUT.
In this paper, the linear and parabolic
approximation of equi-section division is proposed
as a ROM compression method which utilizes the
property of the quarter-wave symmetry method as
well as sine amplitude approximation. The length of
period in all simulations is 4096 and sine amplitude
is 4095 (2
13
-1).
The MSE (mean square error) is calculated
according (2)
( )
2
1
N
i i
i
MSE y sin (.)
=
=

(2)


2 Linear and Parabolic
Approximation
The sinusoidal phase from 0 to /2 is divided into s
equi-sections, where s is the parameter by which to
decide the number of equi-sections. The coefficients
for linear and polynomial approximations are
derived by least squares sense method.
The example of linear approximation (for 4
sections) is shown in Fig. 2, and example of
parabolic approximation (also for 4 sections) is
displayed in Fig, 3. The error of approximation is
also shown. The sinusoidal phase from 0 to /2 is
divided into equi-sections.

Fig. 2. Example of linear approximation quarter of
sine (top) and approximation error (bottom).
Approximation solid line (red), sine dash line
(blue), shifted down.
(4 segments s1, s2, s3, s4 are used).

Fig. 3. Example of parabolic approximation quarter
of sine (top) and approximation error (bottom).
Approximation solid line (red), sine dash line
(blue), shifted down.
(4 segments s1, s2, s3, s4 are used).
Latest Trends in Circuits, Automatic Control and Signal Processing
ISBN: 978-1-61804-131-9 128
Equation (3) below gives the slopes and shift for
linear approximation (example for 4 segments) as

4095 2
4096
6.138 8; 0 256
5.205 261; 256 512
3.48 1155; 512 768
1.225 2892; 768 1024
x
sin y
x x
x x
x x
x x

| |
=
|
\ .
+ <

+ <

=

+ <

+ <

(3)
Result is shown in Fig. 2.
Equation (4) shows the equations for parabolic
approximation (example for 4 segments)
2
2
2
2
4095 2
4096
-0.0009 +6.37 1.98; 0 256
-0.0027 7.25 116; 256 512
-0.0040 8.59 456; 512 768
-0.0047 9.66 860; 768 1024
x
sin y
x x x
x x x
x x x
x x x

| |
=
|
\ .
<

+ <

=

+ <

+ <

(4)
Result is shown in Fig. 3.

Fig. 4. Example of error calculated for 1 period for
linear approximation and 4 segments.

3 Linear Approximation for
4, 8, 16, 32 and 64 segments
In this chapter the simulation results for segment
number of 4, 8, 16, 32 and 64 are shown. The MSE
(Mean square error) errors and spectral properties
were calculated and are presented in Tab. 1. The
example of error for 1 period is shown in Fig. 4,
spectral properties in Fig. 5 - 9. From Tab. 1 can be
seen, that almost optimal is 32 segments (quarter of
period, 0 to /2 is divided to 32 equidistant
segments).
Tab. 1. Linear approximations with different
number of segments.
No. of
segment
MSE dBc
4 1.13e+6 -47.9
8 7.10e+4 -60.1
16 4.76e+3 -72.1
32 612.2 -106.0
64 378.7 -104.9


Fig, 5. The output frequency spectrum, calculated
for 4 segments and linear approximation.

Fig, 6. The output frequency spectrum, calculated
for 8 segments and linear approximation.

Fig, 7. The output frequency spectrum, calculated
for 16 segments and linear approximation.
Latest Trends in Circuits, Automatic Control and Signal Processing
ISBN: 978-1-61804-131-9 129

Fig, 8. The output frequency spectrum, calculated
for 32 segments and linear approximation.

Fig, 9. The output frequency spectrum, calculated
for 32 segments and linear approximation.

Fig, 10. The output frequency spectrum, calculated
for 4 segments and parabolic approximation.


4 Parabolic Approximation for
4, 8 and 16 segments
In this chapter the simulation results for segment
number of 4, 8, and 16 are shown. The MSE (Mean
square error) errors and spectral properties were
calculated and are presented in Tab. 2. The spectral
properties are also displayed in Fig. 10 - 12. From
Tab. 2., can be seen, that 16 segments of parabolic
approximation is similar to 64 segments of linear
approximation.
Tab. 2. Parabolic approximations with different
number of segments.

No. of
segment
MSE dBc
4 1.5307e+003 -81.2
8 374.6 -95.8
16 356.9 -102.9


Fig, 11. The output frequency spectrum, calculated
for 8 segments and parabolic approximation.

Fig, 12. The output frequency spectrum, calculated
for 16 segments and parabolic approximation.


5 Conclusion
Ultrahigh speed direct digital synthesizers will play
important roles in next generation radar and
communication systems. Recent developments in
radar systems require frequency synthesis with low
power consumption, high output frequency, fine
frequency resolution, fast channel switching and
versatile modulation capabilities. Linear frequency
modulation or chirp modulation is widely used in
radars to achieve high range resolution, while pulsed
phase modulation can provide anti-jamming
capability. With fine frequency resolution, fast
channel switching and versatile modulation
capabilities, the DDS provides frequency synthesis
and direct modulation capabilities that cannot be
Latest Trends in Circuits, Automatic Control and Signal Processing
ISBN: 978-1-61804-131-9 130
easily implemented by other synthesizer tools such
as analog-based phase-locked loop synthesizers.
In this paper, the linear and parabolic
approximation of equi-section division utilizing the
quarter-wave symmetry method and sine amplitude
approximation was described, calculated and
simulated. The sinusoidal phase from 0 to /2 was
divided into different numbers of equi-sections, in
linear approximation from 4 to 64, in parabolic
approximation from 4 to 16 sections. The calculated
values were rounded to integer numbers and
frequency spectrum of simulated sine function was
presented. The proposed compression method can
be applied for DDS, software defined radio (SDR)
systems, spread spectrum frequency hopping
systems, and a variety of signaling applications.


Acknowledgment
This research was supported by the European
Regional Development Fund and Ministry of
Education, Youth and Sports of the Czech Republic
under project No. CZ.1.05/2.1.00/03.0094: Regional
Innovation Centre for Electrical Engineering (RICE)


References:
[1] J. Vankka and K. Halonen, Direct Digital
Synthesizers, Kluwer Academic Publishers,
ISBN 0-7923-7366-9, 2001
[2] H. T. Nicholas, H. Samueli, and B. Kim, The
Optimization of Direct Digital Frequency
Synthesizer in the Presence of Finite Word
Length Effects Performance, Proc. 42nd Annu.
Frequency Contr. Symposium, June 1988, pp.
357-363.
[3] P. W. Ruben, E. F. Heimbecher, II, and D. L.
Dilley, Reduced Size Phase-to-Amplitude
Converter in a Numerically Controlled
Oscillator, U. S. Patent 4855946, Aug. 8, 1989.
[4] D. A. Sunderland, R. A. Strauch, S. S.
Wharfield, H. T. Peterson, and C. R. Cole,
CMOS/SOS Frequency Synthesizer LSI Circuit
for Spread Spectrum Communications, IEEE J.
of Solid State Circuits, Vol. SC-19, 1984, pp.
497-505.
[5] L. A. Weaver, and R. J. Kerr, High Resolution
Phase To Sine Amplitude Conversion, U. S.
Patent 4905177, 1990.
[6] J. E. Volder, The CORDIC Trigonometric
Computing Technique, IRE Trans. On
Electron. Comput., EC-8:330-334, 1959.
[7] D. De Caro, N. Petra, and A. G. M. Strollo,
Reducing lookup-table size in direct digital
frequency synthesizers using optimized
multipartite table method, IEEE Trans. Circuits
Syst. I, Reg. Papers, vol. 55, no. 7, 2008, pp.
2116-2127.
[8] J. M. P. Langlois, and D. Al-Khalili, Phase to
sinusoid amplitude conversion techniques for
direct digital frequency synthesis, IEE Proc.-
Circuits Devices Syst., vol. 151, no. 6, 2004,
pp. 519-528.
[9] J.M.P. Langlois and D. Al-Khalili, Hardware
optimized direct digital frequency synthesizer
architecture with 60 dBc spectral purity, IEEE
International Symposium on Circuits and
Systems (ISCAS 2002), May 2002.
[10] D. De Caro, and A. G. M. Strollo, High-
performance direct digital frequency
synthesizers using piecewise-polynomial
approximation, IEEE Trans. Circuits Syst. I,
Reg. Papers, vol. 52, no. 2, 2005, pp. 324-337.
[11] L. S. J. Chimakurthy, M. Ghosh, F. F. Dai, and
R.C. Jaeger, A novel DDS using nonlinear
ROM addressing with improved compression
ratio and quantization noise, IEEE Trans.
Ultrason., Ferroelectr., Freq. Control, vol. 53,
no. 2, 2006, pp. 274-283.
[12] A. M. Sodagar, and G. Roientan Lahiji,
Mapping from phase to sine-amplitude in direct
digital frequency synthesizers using parabolic
approximation, IEEE Trans. Circuits Syst. II,
Analog Digit. Signal Process., vol. 47, no. 12,
2000, pp. 1452-1457.
[13] A. M. Sodagar, and G. Roientan Lahiji, A
pipelined ROM-less architecture for sine-
output direct digital frequency synthesizers
using the second-order parabolic
approximation, IEEE Trans. Circuits Syst. II,
Analog Digit. Signal Process., vol. 48, no. 9,
2001, pp. 850-857.
[14] S. Hermann, and R. Klette, A Comparative
Study on 2D Curvature Estimators,
International Conference on Computing:
Theory and Applications, 2007, pp. 584-589.
[15] K.I. Palomaki and J. Niitylahti, Direct digital
frequency synthesizer architecture based on
Chebyshev approximation, Proceedings of the
34th Asilomar Conference on Signals, Systems
and Computers, 2000, pp. 1639-1643.
[16] B. G. Goldberg, Digital Techniques in
Frequency Synthesis, New York: McGraw-Hill,
1996.
[17] V.F. Kroupa, V. Cizek, J. Stursa, and H.
Svandova, Spurious signals in direct digital
frequency synthesizers due to the phase
truncation, IEEE Transactions on Ultrasonics,
Ferroelectrics, and Frequency Control, vol. 47,
no. 5, September 2000, pp. 1166-1172.
Latest Trends in Circuits, Automatic Control and Signal Processing
ISBN: 978-1-61804-131-9 131
Appendix: Spurious modulation of
DDS
The DDS output waves at the digital to analog
convertors (DAC), with normalized output
frequencies close to the ratio of small integers,
exhibit either a "quasi-amplitude" modulation in
some instances (see Fig. 13) and a "quasi-
superposition" (see Fig. 14) of the desired high
frequency wave on a low frequency signal, as
mentioned in [16, 17], e.g. for expansion:
FCW/2
n
= 253/1024
1
2
3
253 1 1
1 1
1024
4
1 1
21
12
X
Y
b
b
b
= = =
+ +
+ +
(A1)


Fig. 13. Example of quasi-amplitude modulation for
DDS, FCW/2
n
=253/1024


Fig. 14. Example of quasi-superposition for DDS,
FCW/2
n
=207/1024

The continued fraction expansion is given by
0 1 1 2 2 3 1
1 1 1 ( 1)
...
n
n n
X
Y B B B B B B B B
+

= + (A2)
where

0
1 1
2 2 1 0
3 3 2 1
1 2
1
....
n n n n
B
B b
B b B B
B b B B
B b B B

=
=
= +
= +
= +
(A3)
In previous example (X/Y = 253/1024) b
1
=4, b
2
=21,
b
3
=12 (function "rat" in MATLAB returns the
continued fraction representation, b
1
, b
2
, b
3
) and
according (A3):
B
1
=4; B
2
=21*4+1=85; B
3
=12*85+4=1024
and therefore:
253
1024
1 1 1
0.2470703125
3 4*85 85*1024
X
Y
= =
= + =

For second example X/Y = 207/1024, the b
1
=54,
b
2
=-19, b
3
=5, b
4
=2 and expansion is:
207 1 1 1
1024 5 5*( 94) ( 94) *( 465)
1
0.2021484375
( 465) *( 1024)
X
Y
= = +

=


The output sine wave of DDS can be simplified to:
1 1 2
( ) 2
1 1
2
X
s m sin m
Y
sin m
B B B

| |
=
|
\ .
( | |

( |
\ .
(A4)
The trigonometric expansion gives:
1 1 2
1 1 2
1 2
1 1
( ) 2 2
1 1
2 2
( ) ( )
s m sin m cos m
B B B
cos m sin m
B B B
s m s m


| | | |
=
| |
\ . \ .
| | | |

| |
\ . \ .
=
(A5)
From expansion can be seen that low frequency
component cause "quasi-amplitude" modulation or
"quasi-superposition" for some FCW numbers.

Latest Trends in Circuits, Automatic Control and Signal Processing
ISBN: 978-1-61804-131-9 132