INTRODUCTION
The market for low power, shortrange wireless sensor and control
technology is expected to show explosive growth during the coming years.
The typical applications are home automation, industrial control and monitoring,
automatic meter reading and alarms. There are so many applications for
this low data rate shortrange wireless sensor such as industrial and
commercial, home automation, personal computer peripherals, consumer electronics,
personal health care and game that should be able to operate for five
years under battery energy supply.
Recently developed new standards are under consideration for wireless
sensor networking. Examples include Bluetooth (IEEE 802.15.1), UWB (IEEE
802.15.3) and Zigbee (IEEE 802.15.4). Each of these standards is accompanied
by limitations for sensor networks. For example, currently available Bluetooth
devices show excessive power consumption for sensor network applications
(Trung et al., 2006).
The Electronic Communications Committee (ECC) has established new strategic
plans for the future use of the 863870 MHz band for Short Range Devices
(SRD) (Electronic Communications Committee, 2002).
Therefore, the 863870 MHz band, which is available only in Europe, is
a good field to test new ideas and concepts toward the development of
a low power transceiver for short range and low data rate applications.

Fig. 1: 
Wireless sensor transceiver architecture 
The direct conversion transmitter has recently attracted widespread attention
for its simple architecture and easy integration with the base band circuit,
as well as for its low power consumption and low manufacturing costs (Ahmadreza
et al., 1998).
Thus, our objective consists in the design of an independent wireless
sensor in order to set up a wireless sensor network inside a building.
The transmitter is designed for BFSK modulation, with a data rate of 20
kb sec^{1}. This modulator is designed using the DDFS.
In this study, a lowpower technique using a small ROM and 4 pipelined
phase accumulator were used to implement DDFS.
Transmitter Architecture The transceiver architecture is presented
in Fig. 1. The transmitter part operates as follows:
The digital FSK modulator modulates the data and synthesizes in a quadrature
outputs a FHSS signals at base band.
A fixedfrequency oscillator up converts these outputs to the ISM band
in a singlesideband mixer. With an 866.5 MHz oscillator, in the center
of the ISM band either the upper or the lower sideband is selected to
place the instantaneous carrier frequency anywhere in the band. The modulator
is based on DDFS. The DDFS output frequency needs only to span from 0
to 3.5 MHz to cover the ISM range of 866.5Â±3.5 MHz.
The digital data is carried by symbol tones at an offset of either Â±20
kHz from the carrier frequency. This system can achieve a maximum data
rate of 20 kb sec^{1} by sending two symbols per hop.
The theoretical study for determining the data rate is presented in the
following section.
Data Rate Selection If at each transmission the node sends one full packet, then the
overall average power consumption Pd of the transmitter node can be approximated
by:
where, P_{tx} is the power radiated from the antenna, T_{tx}
is the time required for each transmission, T_{wu} is the wakeup
time of the transmitter (i.e., the time required to start up the circuitry),
P_{idle} is the power dissipated in the idle mode when most of
the transmitter functions are off, Pdiss is the nonradiated power dissipated
by the transmitter and T is the time interval between two consecutive
transmissions. From the ShannonHartley theorem, it is known that:
where, B is a fixed bandwidth and D is the maximum data rate achievable
on an additive white Gaussian noise channel. The term S/N is the signal
to noise ratio at the demodulator side (SNRin, dem) and can be related
to the signal to noise ratio at the receiver input (SNRin, rec) by:
where, NF is the noise factor of the receiver. Suppose that the received
signal arrives attenuated by a factor PL(path loss), then:
where, NAWGN is the power spectral density of an additive white Gaussian
noise channel. Substituting (2) in (4),
yields an expression for the transmitted power as a function of all the
other parameters:
Now the transmission time is defined as:
where, Lpacket is the packet length. Substituting (5)
and (6) in (1), an approximated formula
is obtained for the power dissipated by the transmitter as a function
of the data rate. If the receiver has a noise factor of 20 dB, then the
overall power consumption, as a function of D and Pdiss appears as shown
in Fig. 2 for a packet length of 100 bits. Here each
node transmits every 1 min, with a bandwidth of 80 kHz, the channel attenuation
is 102 dB, the wakeup time is 1 msec and the Pidle is 10 Î¼W. From
this Fig. 2, it is clear that the data rate can be lowered
to a certain value. Below some hundreds of bits per second the duty cycle
will increase to a point for which, the power dissipated in the transmitter
circuitry will dominate, increasing the overall power consumption.
In conclusion, target of 20 kb sec^{1} is selected with a balance
between the requirements of lower power, bandwidth efficiency favouring
closely spaced channels and a BFSK tone frequency that avoids the impact
of CMOS flicker noise and DC offset.
Frequency Allocation The transmitter is designed for BFSK modulation, with a data rate
of 20 kb sec^{1} (Trabelsi et al., 2006). The power spectrum
of a BFSK signal is represented in Fig. 3. The transmission
bandwidth B of the BFSK signal was calculated to be 80 KHz.
Fig. 2: 
Power consumption as a function of data rate and power
dissipated by transmitter circuitry 
Fig. 3: 
Power spectrum of a BFSK signal (bit rate = 20Kbps) 
A maximum of 58 channels has been chosen to fulfil FHSS requirement and
ETSI regulations (which demand a minimum separation of 25 kHz between
adjacent channels) (Electronic Communications Committee, 2002; Trabelsi
et al., 2006). Therefore, the separation between adjacent channels
has been chosen equal to 40 kHz. The allocation of the 58 channels is
shown in Fig. 4.
Modulator Design The modulator is implemented using scalable cells, which have been
sized down to the minimum required clocking speed of 43.4 MHz. Frequency
is selected with a 20 bit frequency control word to obtain a Fclk/2^{N}
frequency control resolution, where N is the number of frequency control
bits and Fclk is the sampling frequency of the DDFS/DAC. For Fclk = 43.4 MHz, the smallest
frequency resolution is 41.29 Hz. The frequency control word sets the
accumulation rate in the phase accumulator, which addresses ROMs containing
coefficients of the trigonometric sine and cosine waveforms to produce
digitaldomain quadrature outputs at the programmed frequency. The choice
of internal word length guarantees that in the worse case, imperfections
in the DDFS will result in spurious tones of at least 72.6 dBc relative
to the fundamental frequency (Yang et al., 2004).

Fig. 4: 
Channels allocation in the 863870 MHz band 

Fig. 5: 
The Modulator block diagram 
The architecture proposed of the modulator system is shown in Fig.
5. Essentially, a modulator consists of a DDFS, a PN code generator
and a mixer. The DDFS creates digital samples of a baseband sinusoidal
waveform by addressing a sine ROM at a frequency set by a 20bit control
word. The PN code is a random generator code, which corresponds to the
hopping patterns. The mixer selects one of the two codes generated by
both PN code generators.
The PN code generator and the DDFS used in the modulator are explained
in the following sections.
Pn Code Generator In its frequency hopping sequence, the signal hops to a different
channel during each hopping period. The discrete hop frequencies are determined
by a PN code sequence stored in the memory before the input control register
of the DDFS. The carrier frequency hops at 20 hops sec^{1} among
these 58 channels, according to a pseudo random sequence.
The PN code is a random generator code which has a 20 bits length and
works at the rising edge of a clock frequency which is lower than the
sampling frequency. At each clock edge the PN code will generate a binary
word N (frequency control word), which serves as incremental phase for
the DDFS, as shown in Fig. 6. The N binary words are
stored in a ROM memory that will be sent to the DDFS at each clock period.
The hardware description and design of the PN code generation has been done using the VHDL. Post simulation of the designed modules has
been done using the VHDL simulator to verify the description.

Fig. 6: 
Simulation results for PN code 
DDFS Description The overall architecture of a typical DDFS system is shown in Fig.
7. Essentially, a DDFS consists of a phase accumulator and a sine/cosine
generator. The phase accumulator input is a frequency control word (Fcw).
The phase accumulator is an overflowing N bit accumulator whose value
specifies the instantaneous phase. The rate at which this phase ramp overflows
gives the generated frequency which is proportional to Fcw.
The amplitude of a sinusoidal signal is digitally stored in a ROM and
is consecutively fetched by the output of an accumulator, which feeds
the address line of the ROM (Tierney et al., 1971). The output
sine wave produced has the same frequency as the phase ramp rate overflow,
which is proportional to Fcw. The output sine samples can be converted
to an analog waveform using a digitaltoanalog converter (DAC) and a
lowpass filter (LPF) (Lee and Park, 2003).
The accumulator input and its wordlength determine the output frequency
value and resolution, respectively.
The generated frequency Fout is related to the frequency control word
and the reference frequency Fclk by:
where, N is the accumulator wordlength (Akram and Swartzlander, 2003).
The two main parts, the phase accumulator and the ROM, of the DDFS are
explained in the following sections.
The Phase Accumulator The phase accumulator, shown in Fig. 8, is a 20
bit adder that repeatedly increments the phase angle by Fcw. Therefore,
its output increases by Fcw at each clock cycle. At the time n, the phase
accumulator output is given by:
and the sine/cosine generator must compute sin (2Ï€n Fwc/2N) and
cos (2Ï€n Fwc/2N).

Fig. 7: 
Typical DDFS system 

Fig. 8: 
The phase accumulator block diagram 

Fig. 9: 
A conventional pipelined accumulator 
A large phase accumulator is frequently used in DDFS for the fine frequency
resolution at high clock frequency. However, this large accumulator cannot
finish one addition in a small single clock period because of the delay
caused by the carry bits propagating during the addition. A typical solution
is to pipeline the phase accumulator as m stages of L bits each, such
that, mxL = N as shown in Fig. 9. Each adder generates
L+1 bits output: L sum bits and one carry output bit. The carry output is latched between successive adders. Every new frequency control word
is moved into the pipeline circuits consisting of Dflipflops (DFF) and
delay elements. The speed of the phase accumulator, based on this architecture,
can be increased up to m times. For the DDFS implementation, we used a
four level pipelined phase accumulator.

Fig. 10: 
Block diagram of proposed DDFS 
However, the pipeline circuit requires considerable area and power and
introduces more frequency switching latency. At the same time, increasing
the number of pipelined blocks would increase the loading of the clock
network.
The Read Only Memory A first step towards the reduction of sine/cosine generator complexity
consists in truncating the least significant bits (LSBs) from the phase
accumulator (Fig. 10). This introduces spurious noise
in the DDFS outputs (Strollo et al., 2007; Jafari et al.,
2005), which should be carefully taken into account in the design phase.
Another common approach, usually used to simplify the sine/cosine generator,
exploits the quadrant symmetry of trigonometric functions and identities.
For a quadrature DDFS, this reduces the task of the sine/cosine generator
to the calculation of sine and cosine functions for angles belonging to
the interval only. For a singlephase DDFS, sine calculation for phase
angles, belonging to the first quadrant, is required. Simple DDFS implementations
use a ROM lookup table to calculate trigonometric functions.
One effective method of reducing the ROM size is to exploit the quarterwave
symmetry properties of the sine curve. Thus the quadrantdecode technique
is employed in order to generate a full sine wave.
The second MSB is used to determine whether the phase accumulator output
has to be inverted and the first MSB determines whether the sine amplitude
output of the LUT has to be inverted. This method produces a fourfold
decrease in the ROM size.
MODULATOR SYSTEMLEVEL SIMULATION
Modulator Simulations The most demanding characteristics of modulator are SFDR and output
spectrum. The simulation purpose is to determine the design performance
of the modulator. The design in Fig. 5 is coded by VHDL,
which is then simulated by Modelsim. The VHDL code is then synthesized
by Quartus.
Figure 11 shows the modulator output for maximum code.
At each clock rising edge (clk1), the PN code will generate a binary word
N (frequency control word), which serves as incremental phase for the
DDFS. At each DDFS increment (clk2), the modulator will generate a sample
of sin and cos, as shown in Fig. 11. At each signal
edge (s), the mixer switches between both PN code generators. The output of the modulator for minimum code is presented in Fig.
12. Figure 13 shows the quadrature phase between
sine and cosine. We note that if the sine is zero, the cosine is in its
maximum and when the sine takes maximum values, the cosine begins to take
negative values. Figure 13 shows the transition of sine
and cosine signals corresponding to two different codes before and after
the inversion.

Fig. 11: 
Simulation results for maximum code 

Fig. 12: 
Simulation results for minimum code 
SpuriousFree Dynamic Range (SFDR) The SFDR is defined as the ratio of the desired frequency component
amplitude to that of the largest undesired frequency component at the
output of a modulator, as shown in Fig. 14. Its value
is usually expressed in decibels. An undesired (spurious) frequency component
is often called a spur. Modelsim of Mentor and MATLAB of Mathworks are
the software tools to perform systemlevel simulations. Meanwhile, the
decimal output data in a 10bit format are collected. The Fast Fourier
Transform (FFT) command of MATLAB is executed to obtain the spectrum as
shown in Fig. 14, which illustrates that the spurious
performance of the proposed method is as high as 88 dBc. It is far better
than any prior works.
The data shown in Table 1 compare the performances
of the developed DDFS with recently published stateoftheart circuits,
using CMOS technology, SFDR and clock frequency similar to the ones considered
in this paper. The proposed IC exhibits a power dissipation reduction
by a factor larger than 10 and a substantial decrease of the maximum clock
frequency, with respect to the solutions proposed in (Langlois and AlKhalili,
2003). The DDFS of (De Caro et al., 2004) dissipates 51% more power
and has higher output resolution and SFDR. However, it is faster than
our design.
Table 1: 
Performance and comparison of proposed DDFS 


Fig. 13: 
Modulator Block simulation results 

Fig. 14: 
Spurious performances of the proposed
modulator for the 24th code 

Fig. 15: 
The modulator chip design 

Fig. 16: 
The modulator chip layout 
Chip Implementation The design was implemented using the AMS 0.35
Î¼m CMOS standard cell library, triple metal technology with a linear
capacitor. To generate a hardware model from software algorithm some simple
logic and arithmetic blocks (such as: multipliers, adders and logic gates)
have been used. The DDFS can be built using VHDL description. First, the
design was coded in VHDL and verified using NClaunch. Second, the Cadence
Ambit tool was used to perform logical synthesis and optimization. Finally,
the placement and routing and the Design Rule Check (DRC) were done using,
respectively, Cadence Silicon Ensemble and Cadence virtuoso. The chip
design is broken down into thirteen major subsystems (Fig.
15).
Simulation results show that the average power dissipated is 47.7 Î¼W
at 43.4 MHz. It is interesting to conclude that the DDFS power dissipation
is close to 63% of the total power. The power consumption of the PNcode1
and PNcode2 is 17 and 18%, respectively. The power dissipation of the
accumulator is not negligible since it requires about 19% of the total
power. The power consumption of the lookup tables is as low as 22%.
The chip operates from a nominal 3V power supply. The modulator chip
layout, shown in Fig.16, contains 1205052 transistors
and has a die size of approximately (0.396x0.396) mm^{2}.
CONCLUSION
In this study, we have presented a novel FHSSFSK modulator implementation
design. First, we have presented a novel method exploiting the quadrant
symmetry of trigonometric functions and trigonometric identities for the
purpose of reducing DDFS spurious tones. A new DDFS architecture based
on this technique was presented. A DDFS has been presented which uses
a smaller lookup table for sine and cosine functions and four level pipelined
phase accumulator. Furthermore, we described the modulator simulation
results, which have been designed using the proposed DDFS. It enabled
us to generate BFSK signal with frequency hoping. The design was implemented
using a 0.35 Î¼m CMOS technology. It occupies a core area of 0.16
mm^{2}. The entire chip operates at a low supplyvoltage of 3
V. Simulation results show that the average power dissipated is 47.7 Î¼W
at 43.4 MHz with an 88dBc SFDR.