Professional Documents
Culture Documents
00025
Department
1. Introduction
The advantages of OFDM (Orthogonal Frequency Division Multiplexing) have been
well known; they are exploited to combat impairments in wireless and optical communication systems. The principal mechanism of OFDM is to generate parallel
369
370
00025
L. K. Dang et al.
00025
371
(1)
where fk is the corresponding frequency with respect to the kth subcarrier within
the OFDM symbol.
The OFDM signals of N subcarriers can thus be represented as:
N 1
1
am,k k (t),
Sm (t) =
N k=0
di , N
am, k
am, N
Baseband
signals
1
Baseband
signals
Sm (t )
Sm (t )
DAC
Baseband modulation
S-P-Conversion
{dl }
di,k
(2)
a m ,1
d i ,1
0 < t < N T,
Baseband
signals
N
Fig. 1. Principles of construction of OFDM signals. DAC: digital to analog conversion. S-P: serial
to parallel.
372
00025
L. K. Dang et al.
where am,k is the kth complex symbol, NT is the length of OFDM symbol containing
N subcarriers, and T is the sampling interval. The subcarriers are equally spaced
by f = N1T , fk is estimated by
fk =
k
.
NT
(3)
These fk values can then be mixed with the passband carrier, e.g. microwave or
lightwave, to generate a set of orthogonal passband subcarriers. Each signal Sm (t)
is equivalent to a point in the Euclidean N -dimensional space which is called the
signal space. Each point is thus represented by a set of values (am,1 , am,2 , . . . , am,N ).
In the case of continuous transmission, m is an integer and dependent on the
length of the input data. All multiplied signals are then added and the nal signal is the passband time-dependent modulated signal waves which can then be
transmitted over the transmission medium. Due to the fact that fk is modulated
at the carrier frequency fk = kf Hz, the OFDM technique is commonly known
to be composed of N subcarriers, each carrying much lower speed data symbols
ROFDM = RS /N with Rs as the original sampling rate or the bit rate of the input
digital sequence. Thus for OFDM the bit rate of each subcarrier channel is the
transmission bit rate of the OFDM frame.
2.2. Implementation of OFDM system using IFFT/FFT
The multiplexing of subcarriers is not a major issue but the ltering of each subcarrier channel is the principal task. Originally they were ltered by several bandpass
lters and thus their design and performance characteristics, especially the very
sharp roll-o, cannot be easily satised. This deters the uses of OFDM in the initial development phase till the proposal of using inverse discrete Fourier transform
(IDFT)6 of the sequence ak , written as
N 1
1
nk
Sm (nT ) =
am,k ej2 N
N k=0
0 n N 1.
(4)
Naturally IDFT is commonly available in almost every digital signal processor. Likewise the demultiplexing of the subcarrier channel at the receiver can be
easily performed using DFT. This has facilitated the simplication for practical
implementation of OFDM. Figure 2 depicts the implementation of OFDM signals
using inverse Fourier transform IFFT and FFT which further simplify the generation of orthogonal channels. Therefore, a FPGA-based system can assist with this
implementation.
2.3. Cyclic prex
Communication systems employing OFDM would face two main problems. Firstly,
the channel separation between subcarrier channels is narrow and can easily
lead to intercarrier interference (ICI). Secondly, OFDM symbols are continually
00025
Fig. 2.
am, N
Sm (t )
DAC
Sm (t )
di, N
Baseband modulation
{dl }
am , k
IFFT
a m ,1
d i ,1
d i ,k
373
transmitted, so if the delay or distortion eects occur repeatedly, intersymbol interference (ISI) would happen. So if the guard band consists of all 0s, the processing
can minimize the ISI but not the ICI. Thus the prex is suggested by Peled and
Ruiz6 in 1980 by copying parts of the message signal and inserting it to the beginning of the message signal. Therefore the problems of ISI and ICI can be resolved.
The notation ts is dened as the symbol period, and Ts is the symbol period plus
protection interval G such that G t < 0. Thus the OFDM signals including
the prex can be expressed as:
N 1
1
am,k k (t),
Sm (t) =
N k=0
G < t < N T.
(5)
Usually the prex is selected such that it is longer than the transmission delay
time of the transmission medium and can be tuned to achieve the maximum transmission quality. Thus the protection interval plays an important part in the mitigation of the ISI and ICI. However by using the cyclic prex, some energy must be
paid to this extra part of the signals to be transmitted. Naturally the total energy
to be consumed by cyclic prex OFDM signals would be now evaluated for each
subcarrier channel as:
NT
.
(6)
|k (t)|2 dt =
N T G
Thus the SNR penalty per subcarrier channel at the receiver is given by
NT
.
N T + G
Hence the total SNR penalty at the receiver is given by:
G
SNR loss = 10 log10 1
.
NT
Eloss =
(7)
(8)
Therefore it is naturally expected that the longer the cyclic prex interval, the
higher the SNR penalty.
In summary, the structure of the OFDM symbol is arranged as shown in Fig. 3.
374
00025
L. K. Dang et al.
Fig. 3.
d i ,1
S m (t )
ADC
S m (t )
1
am, k
demodulation
am, N
demodulation
Fig. 4.
am ,1
demodulation
N
Principles of demodulation of OFDM signals.
di , k
di , N
{dl }
00025
Fig. 5.
am, k
am, N
d i ,1
di , k
di , N
{dl }
FFT
S m (t )
S m (t )
A/D Converter
am ,1
375
with fDmax = v
fc
,
c
(10)
where v is the relative velocity of the MS to the BS. fc is the carrier frequency, c is
the velocity of light in vacuum, is the angle of the direction of movement of the
MS with respect to the BS.
376
00025
L. K. Dang et al.
x (t )
0 (t )
1 (t )
k (t )
L (t)
y (t )
Fig. 6. Delay model for transmission of purely delay eects such as wireless transmission and
polarization mode dispersion in single mode optical bers.
00025
377
impulse response is h(t) and output is r(t). In the case of AWGN n(t), we have
r(t) = h(t) s(t) + n(t).
(13)
(14)
where S(f ) is the frequency domain representation of the OFDM signal. The recovery of S(f ) is thus mainly ltering of the noises contributed by N (f ).
Thus the equalization can be implemented after the FFT subsystem, and only
the multiplication or division operations are needed and not integration or dierentiation. This oers signicant advantages in signal processing in real time.
The eectiveness of OFDM technique depends on the shifting of the frequency
and phase of the subcarriers. If Doppler eects exist in the transmission medium,
then the phase lock loop at the receiver front may have to be used. If errors in
frequency and phase happen, then the rotation of the constellation of the received
signals would happen and the detection must be corrected. The third problem is
that the clock signals of the ADC and DAC at the front end of the transmitter and
receiver can be dierent, thus creating some extension of the OFDM symbol which
may be dierent from symbols to symbols. Thus there must be a synchronization
of the symbols. These problems aect the recovery of the transmitted signals and
must be dealt with.
The use of the cyclic prex can solve the problem of synchronization of the
symbols. Furthermore, if the delay time varies and is less than the cyclic prex time
interval, then one can use this interval to assist the synchronization of the OFDM
symbols. Then the FFT with N points from this location would signicantly reduce
the mismatch in the subcarrier phase. Once the synchronization is complete, the
eects of shift in the frequency and phase due to the Gaussian noises can be resolved
by increasing the spectral distance between the subcarriers. However, the technique
employed in this work is to insert pilot carriers at the known subcarrier location in
OFDM symbols. This technique is described in the next subsection.
3.3. Pilot signals
At the receiver end, the received and original values of the pilot allow the estimation of the eects of the transmission at the frequency location of these pilots
and thus one can deduce the eects of the transmission medium on all subcarrier
channels of the OFDM symbol. The original OFDM symbols can then be recovered
without much diculty. The pilots can be inserted into OFDM symbols as shown
in Fig. 7. The frequency spacing between pilots must follow the sampling rules in
the frequency and time domain.
As mentioned above, the change of frequency of the transmission medium is
dependent on the maximum delay time of the medium. Let rf be the sampling
378
00025
L. K. Dang et al.
Fig. 7.
ratio in the frequency domain and f be the spacing between the subcarriers, then
the spacing between the pilots must satisfy:
rf =
1
1.
Df f max
(15)
Thus the minimum sampling ratio must be rf = 1. When rf < 1, the transmission
channel cannot be fully recovered via the pilots.
Similarly in the frequency domain, the spacing between the pilots must satisfy
rt =
1
1,
2fDmax Dt (TS + G )
(16)
(17)
00025
379
For the sake of simplicity, we can ignore the noises due to AWGN, then
Rpilot (f ) = Spilot (f )Hpilot (f ).
(18)
Hence the frequency response of the transmission medium evaluated at the locations
of the pilots is given as
Hpilot (f ) =
Rpilot (f )
.
Spilot (f )
(19)
From Hpilot (f ), the overall frequency response H(f ) of the transmission channel
can be derived by using available techniques such as WienerHop ltering or other
complex ltering methods.
There are several possibilities of inserting pilot carriers into the spectrum of
OFDM signals as proposed in Refs. 7 and 8. Each scheme oers its strong point
depending on the transmission characteristics of the transmission channel. Thus
there is possibility for further research in the allocation of the pilots, especially
when single mode optical bers and online optical ampliers are employed in the
systems in which the nonlinear phase noises and quadratic distortion characteristics
of the bers are combined with the random noises (ASE = amplied stimulated
emission) of in-line cascaded optical ampliers.
4. OFDM Systems Design
4.1. Structures of OFDM symbols
OFDM symbol is dened as a set of subcarrier channels whose number determines
the number of the FFT and IFFT to be used. First, the data carrier is employed
for transmission. Second, the pilot carriers would be used for synchronization and
estimation of the eects of the transmission medium. Finally, the null subcarrier for
protection band and DC carrier can be added. The usefulness of the protection band
ensures the sharp roll-o of the brick-wall-like passband of the OFDM symbol.
Depending on practical systems, the number of subcarriers of each portion of the
message symbol may vary. For example, the symbol structure of Standard 802.16
can be {28 zero, 100 data, zero, 100 data, 27 zero} where the data are imbedded with
pilot at special locations of 88, 63, 38, 13, 13, 38, 63, 88. Each complex data
sequence, when imbedded into the symbol, would be indexed from 128 to +127.
4.2. Estimation of the design parameters
In order to design OFDM symbols, we need to specify the following parameters:
BW: spectral width of OFDM symbol; Nused : number of subcarriers to be used in
an OFDM symbol; n: the sampling factor, together with BW and Nused the spectral width of the carrier can be specied based on the symbol time as dened in
Eq. (3); G: ratio between CP and the symbol time; NFFT : number of FFT points
which are selected to be in the order of 2N and greater with N being an integer.
NFFT is greater than Nused ; Fs : sampling frequency, Fs = f loor(nBW/8000)8000;
380
00025
L. K. Dang et al.
= BW.
(20)
NT N
T
Thirdly, OFDM is very sensitive to ICI so the subcarrier spacing N1T should be
considered as the best when it is greater than the frequency shift due to Doppler
eect fD so that the subcarrier orthogonality would remain satised.
1
1
BW
fD N
=
.
(21)
NT
T fD
fD
Thus,
BW N
BW
.
fD
(22)
00025
Fig. 8.
381
Hard- and software experimental platform of the OFDM digital transmission systems.
signals at the baseband level; thus the signal spectrum is evaluated on the I- and
Q-components.
4.5. Functionalities of OFDM system blocks
The principal function of a communication system is to transport information with
an assurance of the bit error rate (BER) as pre-determined. Therefore the OFDM
system consists of the following blocks:
The randomizer blocks. At the randomizer the input data initially would be split
into groups of 1s and 0s with a distribution by using XOR and a random bit
pattern generation. The bit pattern generation block would be associated with the
pseudo-random bit sequence (PRBS) generator. The eects of the randomizer are
to avoid identical bit patterns so as to avoid the diculty in recovery of signals to
its original bit sequence. The derandomizer at the receiver performs reverse process
of the randomizer.
Channel encoder. This is the most important block and plays the major part in
the coding and assists in the recovery of the decoder block with redundancy bits. In
this work the convolutional coder employed is associated with the Viterbi algorithm
to ensure perfect corrections of error. Turbo algorithm can also be employed to
further improve the performance of the decoder.
IQ mapper block. This block converts the bits into a set of bit patterns corresponding to the states of the constellation of modulation scheme. In OFDM this is
called the mapper which can be formed to further increase the bit rates and hence
the spectral eciency by higher order or level of the constellation. QPSK or M-ary
QAM can be used. In the reverse process the demapper is used to transform the
constellation points back to the bit pattern.
OFDM symbol generator block. After conversion of the bit sequence into symbol, the serial bit sequence would be converted to parallel blocks and assembled
382
00025
L. K. Dang et al.
with pilots or DC symbol and protection symbol; these symbols form the OFDM
symbol. Each symbol represents a frequency spectrum so as to superimpose on the
subcarriers.
OFDM signal block. The OFDM symbol would then be IFFT-transformed to
generate OFDM signals. Thus OFDM signal is a combination of all the spectra
of the OFDM symbols. Afterwards the IFFT cyclic prex would be inserted at
the beginning to form the OFDM signal for transmission over the channel. These
signals are converted to the analog form via the DAC and then converted to the
wireless or optical domain depending on the transmission medium.
The transmission channel is simulated using delay and summation to represent
the delay paths of signals over a wireless medium. If optical ber is used, then the
Schr
odinger equation of the complex amplitude would be used in association with
the split step process to propagate the OFDM signals. The noises are modeled using
AWGN method.
Other blocks in the receiving ends would be formed with their operations in a
reverse order and in complement to those described for the transmitting end blocks.
00025
383
the Stratix EP1S25 with the main device 25660 logic elements at the fastest speed,
packaged with FineLine BGA and a total memory of 1,944,576 bit RAM.
The development kit consists of (i) two ADC-12-bit, maximum sampling rate 125
Msamples/s, 2-complement data converter, frequency of input signals greater than
1 MHz, peak-to-peak input amplitude 2 V; (ii) two DAC-14-bit, operating speed
165 MHz, conversion to analog signals from digital form, peak-to-peak amplitude
of 1 V; 2 Mb SRAM with 7.5-ns synchronization, each SRAM with 18 address lines,
36 data bus and 1 MB 32 Mbit ash memory.
In general the library of the DSP Builder consists of a number of blocks
whose functionalities include4 : Library AltLab, Block SignalCompiler, Block
SignalTap II Analysis, Library Arithmetic, Library Board, Library Complex Type, Gate and Control, Library IO and Bus, Rate Change, Library
Storage, and Library MegaCore.
5. OFDM System Platform
This section gives the principal function of the design and associated blocks of
the OFDM system including I/O ports, randomizer, convolutional code block, and
structures of OFDM symbol. Employment of the cyclic prex is based on 802.16.
5.1. Random generation of data signals
5.1.1. Principles of generation
The logic diagram of the randomizer would consist of a shift register and two exclusive OR gates. The principles of operations of this block are briey described above.
5.1.2. Design of the pseudo-random generator
The MATLAB Simulink model of the generator is given in Fig. 9, it is integrated in
the DSP Builder. The pseudo-random binary sequence generator can be constructed
in this block using the sequence 1 + X14 + X15 . The pseudo-random sequence
Fig. 9.
384
00025
L. K. Dang et al.
can be generated from the initial sequence set 100101010000000. The input is
XORed with the output of the pseudo-random binary sequence generator. The
signals ena and rst enable the reset and rest of the pseudo-random generator. The
derandomization at the receiver can be done by passing the signals again through
the randomnizer. In this work the input is in the form of binary sequence, thus
there is no need for the conversion from the decimal form to the binary form.
5.2. Channel encoding
The encoder is shown in Fig. 10 for generation of the convolutional code with
the determined speed of 12 . The length of the encoder is seven symbols and the
polynomial used is G1 = 171OCT and G2 = 133OCT . The encoder is formed by
using six tap delays and two XOR gates with ve inputs. In CC is the input of the
encoder and out CC1 and out CC2 are the two outputs corresponding to G1 and
G2 . The signals ena and rst are for enabling and resetting the encoder.
5.3. Decoding using Viterbi algorithm
The decoder is designed using Viterbi algorithm. The algorithm consists of four
main blocks: branch metric, add compare and select, survivor path metric, traceback
and output decoding block.
The branch metric is used to estimate the Hamming distance of the bits, e.g. 2
bit if the speed of codes is R = 12 with branches. The summation block conducts
selective addition and is responsible for estimating the total Hamming distance of
the branches in the current state and it keeps only the shortest distance branch.
This block selects and determines the optimum sequence. The traceback and output
decoding block is responsible for selection, rechecking the sequence at the optimum
state and determine the output sequence which is the decoding sequence of the
algorithm.
Figure 11 shows the decoding techniques model using MegaCore so as to construct the decoder using Viterbi algorithm. The Viterbi algorithm is based on the
best suitable sequence. Therefore the determination of the length of the sequence
Fig. 10.
00025
Fig. 11.
385
inuences the eectiveness of the decoder. If the length is too long, then there would
be more numerical operations, increasing the memory storage and the delay time.
On the other hand, if the length is too short, then it limits the error correction
potential of the algorithm.
5.4. Signal constellation
5.4.1. Constellation
Constellation is the modulation technique to transform the sequence of m bit into
a complex phasor in the form of a + jb. The number of bits m depends on the
number of states of the constellation. For example, a 16-QAM has four bits per
symbol, which has 16 points on the constellation. The mapper uses the method of
modulation but the modulation function rests at the IFFT and the DAC at the
output. There are a number of methods for constructing the mapper by using the
look-up table based on data stored in ROM memory. In this work, the mapper uses
the structure of look-up tables which are classied for the I- and Q-components.
The data used for forming the constellation are stored in the look-up table. The
output values vary from 1 to +1. The output of the DAC is represented by a 14-bit
word. Thus, this work normalizes the sequence into xed point arithmetic values.
Figure 12 shows the MATLAB Simulink model of the constellation generator.
5.4.2. Reconstruction from constellation
At the receiving end, the constellation points representing the states of the OFDM
symbols at the output of the transmission medium, are accumulated. Thus the constellation demodulator must set the decision levels so as to determine the constellation points of the receiver. The demapper is constructed using Verilog language.
The decision point is based on the shortest Euclidean distance to the received signals. When QPSK modulation is used, the demapper can simply determine by
386
00025
L. K. Dang et al.
Fig. 12.
Fig. 13.
evaluating the most signicant bit of the bit sequence received which is the sign
bit of the received sequence. For the 16-QAM multilevel modulation scheme, the
demapper is designed based on the rule of IF-THEN. The output data are fed into
the FIFO and then into the decoder. In the OFDM symbol there are 192 data values to be transmitted. Thus the FIFO needs 192 memory pair shift registers so as
to store the I- and Q-component values. Figure 13 illustrates the Simulink model
of the demapper for the 16-QAM modulation.
5.5. OFDM symbols
An OFDM symbol consists of the data payload, the pilot, and the 0 patching up
values and the inserted symbol pilots. These pilots are used to estimate the eects
of the transmission medium at the receiving end.
5.5.1. Generation of pilot signals
In order to generate the pilot, we need to use the pseudo-random generator/
randomizer. The polynomial for the PRBS is (1 + X 9 + X 11 ) as shown in Fig. 15.
The coecients of the polynomial are based on 10101010101B. The pilot signals for
00025
Fig. 14.
Fig. 15.
387
Demapper block.
the kth OFDM symbol can be derived from the value wk . The interleaving values
can be determined as c88 = c38 = c13 = c38 = c63 = c88 = 1 2wk and c63 =
c13 = 1 2(!wk ).
The Simulink model for generating the pilots is given in Fig. 15 which consists
of a LFSR block to generate random sequence, a block for generating the location
of !wk and nally the arithmetic operator to generate (1 2wk ) or (1 2(!wk )). If
we index the symbol from 0 to 255 then the values 63 and 13 would become 65
and 115 which are the time indices for determining the location of the bit in the
OFDM symbol. The signal ena and rst are used for enabling and resetting the
circuitry. The pilot signal is the output of the block.
5.5.2. Assembly and separation of symbols
The assembly and separation of the constituents of OFDM symbols are implemented
as shown in Figs. 16 and 17, respectively.
The arrangement of the data sequence of OFDM symbols is based on Standard
IEEE 802.16 and dened as follows: {28 0s, 100 data, zero, 100 data, 27 0s} in
which the locations of the pilots are at c88 , c63 , c38 , c13 , c13 , c38 , c63 and c88 .
Verilog language is used to generate this arrangement. The inputs of assembler are
the index of the symbol, the data to be transmitted, pilot and zeros for patching and
388
00025
L. K. Dang et al.
Fig. 16.
Fig. 17.
data for the I- and Q-components as shown in Fig. 16. The output of this assembler
is IFFT-transformed to generate OFDM signals with control timing signals as shown
in Figs. 17 and 18. The disassembling of the OFDM signals at the receiving end
can be implemented with the input to the FFT block. The data information is then
fed into the equalizer as shown in the separator of Fig. 19.
Fig. 18.
OFDM generator.
00025
Fig. 19.
389
Fig. 20.
390
00025
L. K. Dang et al.
Fig. 21.
Fig. 22.
00025
Fig. 23.
Fig. 24.
391
beginning and at the end of the symbol are patched up with zeroes. Thus, with eight
pilots used, the total number of bits for data payload would be 192. Convolution
code is used with a factor of 12 . Figure 24 shows the structure of the block for control
signals generation.
5.7. Hardware implementation
The schematic of an OFDM system developed in this work is shown in Fig. 25.
Similar shaded blocks are used to indicate the block and its counterparts at
the transmitting and receiving ends respectively. The functions of each block are
described in the previous section. The functions of the blocks of this diagram are
listed in Table 1.
6. Simulation and Experimental Platform
The OFDM signals are monitored by the software platform SignalTap which is
integrated in the hardware system, in particular the FPGA-based section. The
results obtained are then displayed on a desktop computer. SignalTap also updates
signals in real time. The development board is also updated via software platform
board using JTAG standard. In order to study the functions and performances
of each block of the system as described above, we monitored and accumulated
392
00025
L. K. Dang et al.
Fig. 25.
data at the inputs and outputs of each block.11,12 Delay adjustments are made to
compensate for the data accumulation. This section thus presents the experimental
results obtained including the monitoring of the spectra of OFDK system via an
external spectrum analyzer.
6.1. Randomizer
The data passing by the randomizer twice would give the same sequence at the
input. Once randomized, the data are indexed with 1 or 0 which appear continuously. Figure 26 depicts the waveforms of the data sequence at the input and
output of the randomizer after one and two passing. This conrms the working
principles of the randomizer as described above.
6.2. Encoder
The input randomized data sequence would be encoded using convolutional code
principles with a coding speed of 12 . After coding, noises are superimposed by a
bit-complementary transmission channel. At this stage the probability of error is
1
4 with an error sequence of 01000100. After this the Viterbi algorithm is used
to decode the encoded data sequence; the decoded sequence must be the same as
the original data sequence. Figure 27 shows the performance of the encoder with
the input data sequence, its output after the encoder, and then the output of the
decoder which is identical with its original image. Thus this conrms the working
of the encoder developed using FPGA-based processor. The coding speed of 12
means that for each bit entering the encoder, there would be two bits following the
pattern g1 and g2 which are represented as a pair of bits (g1 g2 ) where g1 is the most
00025
Functions
Data lines
Names
Sink
393
index out
rd data
Functions
Ena
data out
Indexing of symbol
Reading data after constellation at the
FIFO
Enabling data output
Output data
Randomizer
Randomization of
data sequence
ena
in ran
rst
Enabling processing
Data to be randominzed input
Reset
Channel
Encoder
Encoding
ena
in cc
rst
out cc1
out cc2
Enabling processing
Data to be encoded input
Reset
Output g1 = 171oct
Output g2 = 133oct
Mapper
Mapping to
constellation
points
rd data
wr data
in map
out mapQ
sclr
out mapI
Assembling
Symbol
OFDM
Generating OFDM
symbol by
assembling data
and inserting pilot
and zero patching
index
in dataI
in dataQ
out I
out Q
Symbol index
Inphase input
Quadrature phase input
Inphase output
Quadrature phase output
Signal
OFDM
Generating OFDM
signals by IFFT
index in
in real
in imag
out real
out imag
out sop
out rst
Symbol index
Real input
Imaginary input
Real output
Imaginary output
Start of fame at output
Reset output for new symbol
Cyclic Prex
in real
in imag
sclr
out real
out imag
Real input
Imaginary input
Erase index of data at output
Output real
Output imaginary
Remove
Cyclic
Prex
Removing cyclic
prex
in real
in imag
out real
out imag
sop rx
Input real
Input imaginary
Output real
Output imaginary
Start frame at receiving end after
removing CP
End of frame at receiving end
eop rx
394
00025
L. K. Dang et al.
Table 1.
Blocks
(Continued )
Functions
Data lines
Names
Functions
Signal
OFDM
Decomposition of
OFDM signals by
FFT
in realrx
in imagrx
in sop
In eop
out realrx
out imagrx
sop rx
Input real
Input imaginary
Start of receiving frame
End of receiving frame
Output real
Output imaginary
Start of frame output
Disassembling
Symbol
OFDM
Decomposition of
OFDM signals to
collect data, pilot
in realrx
in imagrx
sop rx
data I
data Q
pilot I
pilot Q
wr data
index
Inphase input
Quadrature phase input
Start of output fram
Inphase output
QUadrature phase output
Inphase pilot output
Quadrature phase pilot output
Writing into FIFO memory
Index of received data
Equalizer
Estimation of
transmission channel
characteristics
data I
data Q
pilot I
pilot Q
rx I
rx Q
Demapper
Conversion of
constellation points
to data sequence
rx I
rx Q
wr data
rd data
data out
Channel
Decoder
Decoding using
Viterbi algorithm
in dec
sink val
out dec
out ena
Derandomizer
Decoding by
de-randomization
in ran
ena
out ran
signicant bit and g2 is the least signicant bit. The clock period of the decoder
is four times longer than the recovery temporal length. Thus for a recovery length
of 42, the length of the decoder is 7, and a delay time of the decoder is 168 clock
intervals. This would allow us to set the delay time for comparison between the
input encoded data sequence and that of the output of the decoder. The sequence
waveforms in Fig. 27 can then be conrmed to demonstrate the working of the
encoder and decoder. Note that for every four bits transmitted, there would be one
error bit and this error bit is corrected by the Viterbi algorithm.
00025
(a)
395
(b)
(c)
Fig. 26. Experimental observation of the randomizer. (a) Original data sequence. (b) Randomized
data sequence. (c) De-randomized data sequence.
(a)
(b)
(c)
(d)
Fig. 27. Experimental signals obtained at the encoder. (a) Original data sequence. (b) Encoded
data sequence. (c) Encoded data with added noises. (d) Decoded data sequence.
396
00025
L. K. Dang et al.
(a)
(b)
(c)
Fig. 28. Data input to the mapper and after the mapper for mapping to the constellation for
QPSK modulation. (a) Input data sequence to the mapper. (b) Inphase component sequence at
the output of the mapper. (c) Quadrature phase component sequence at the output of the mapper.
(a)
(b)
(c)
(d)
Fig. 29. Structure of an OFDM symbol. (a) Rearranged data in the I-component to bypass pilot
location. (b) Q-component. (c) Final OFDM symbol in I-axis. (d) Final OFDM symbol in Q-axis.
00025
397
with a structure of {28 0s, zero, 100 data bits, zero, 100 data bits, 27 0s}. In
this structure, the data-bit sequence consists of eight pilots at the locations of c88 ,
c63 , c38 , c13 , c13 , c38 , c63 and c88 . These pilots are not standardized. Thus,
they would have values greater than the data bits. The input data sequence would
bypass the locations of these pilots and the inserted pilot. The output of the OFDM
symbol block is consistent with its symbol structure.
6.3.4. Generation of OFDM signals
After forming the OFDM symbol, the IFFT would be used to form OFDM signals
(Fig. 30). Both I- and Q-data sequences are fed into the real and imaginary inputs
of the IFFT block. Thus the output of the IFFT is a complex sequence and we may
observe the PAPR eects of the waveform at this output as described in the section
above.
6.3.5. Insertion of cyclic prefix
Figure 31 shows the waveform after insertion of cyclic prex. The length of the
cyclic is 14 . The cyclic is inserted at the beginning of OFDM signals. The data at
(a)
Fig. 30.
(b)
OFDM signals. (a) I-component. (b) Q-component.
(a)
(b)
(c)
(d)
Fig. 31. Transmitted signals before and after adding cyclic prex after 192 clock period delay.
(a) I-component without cyclic prex. (b) Q-component without cyclic prex. (c) I-component
with cyclic prex. (d) Q-component with cyclic prex.
398
00025
L. K. Dang et al.
(a)
Fig. 32.
(b)
(a)
(b)
(c)
Fig. 33. Transmitted signals after AWGN channel. (a) Random noises generated by the randomizer. Superimposed signals and noises of (b) the I-component, and (c) the Q-component.
the end would be copied to the beginning so as to form a signal with 320 intervals
in the time domain. These signals are then passed through the DAC to give the
baseband OFDM signals. The output of the DAC is unipolar 14 bit thus the IFFT
would be set up so that there is unipolar waveform at the output of the IFFT. The
DC component would be added with an appropriate value of 8192. Figure 32 shows
the waveform after this superposition.
6.3.6. Transmitted signals
The signals at the transmitter would be added with noises which have a natural probability density distribution to mimic the AWGN channel noises that are
normally presented in wireless transmission medium or polarization dispersion and
optically amplied ber cascaded spans. Figure 33 shows the transmitted signals
after superimposing noises to give a SNR of 20 dB. Naturally when the signals pass
through the medium, errors would occur. However, the decoder can perform error
correction.
6.4. Processing of signals at the receiving ends
The rst block at the receiving end of the OFDM system has the responsibility
of removing the cyclic prex. In this work the removal is done by controlling the
00025
399
start of sop and nish of the frame eop. These two signal indicators indicate
the FFT to receive only 256 values of the OFDM signal frame. The output of the
FFT would then be fed to the blocks responsible for disassembling the OFDM
symbols, demapping the constellation points, decoding and error correction and
derandomization. At the same time the output of the receiving end block would be
compared with the data sequence at the transmitting end and error counting would
be performed to obtain the BER of the system.
6.4.1. FFT transform to recover OFDM symbol
OFDM signals are passed through the FFT so as to receive the data sequence as
shown in Fig. 34. Due to superimposition of the noises on the signal, correction is
implemented at the decoder using the Viterbi algorithm.
6.4.2. Disassembling OFDM symbol block
The OFDM signals are then disassembled into data, pilot and eliminated zero patching sections. The pilot parts would be used to estimate the distortion eects of the
transmission channel and then used for the equalizer of the receiving end section.
Figure 35 shows the waveforms obtained at the transmitting and receiving ends at
the assembling block and disassembling block.
6.4.3. Constellation and encoder
The output of the disassembling block would be fed into the constellation mapper.
The output would be a pair of bits for QPSK modulation states on the constellation.
With the errors accumulated, the Viterbi algorithm associated with the decoder can
correct without much diculty, as proven in our experimental system.
(a)
(b)
(c)
(d)
Fig. 34. Data waveforms after FFT. (a) I-channel waveform of OFDM symbol at the transmitting
end. (b) Q-channel waveform at the transmitting end. (c) I-component of OFDM signals at the
receiving end. (d) Q-component of the OFDM signals at the receiving end.
400
00025
L. K. Dang et al.
(a)
(b)
(c)
(d)
Fig. 35. Comparison of OFDM symbol at the transmitting and receiving ends. (a) I-component
data sequence of one symbol at the transmitting end. (b) Q-component data sequence of one
symbol at the transmitting end. (c) I-component data sequence of one symbol at the receiving
end. (d) Q-component data sequence of one symbol at the receiving end.
Table 2.
STT
Resources
Use
1
2
3
4
5
6
7
8
17,389/25,660 (68%)
14,900/29,168 (51%)
2020/2566 (79%)
54/598 (9%)
19/224 (8%)
136/138 (99%)
460,520/1,944,576 (24%)
637,632/1,944,576 (33%)
00025
401
(a)
(b)
Fig. 36. Spectra of noiseless OFDM signals as obtained on MATLAB Simulink. (a) I-component.
(b) Q-component. Normalized frequency.
402
00025
L. K. Dang et al.
Fig. 37. Nonnormalized spectrum of OFDM signals superimposed with noises as observed on
MATLAB Simulink platform. Note that no cyclic prex is included at this observation.
(a)
(b)
Fig. 38. Spectra of received signals as observed at the output of the DAC on a spectrum analyzer.
(a) I-channel at 8.7 dB/div. (b) Q-channel at 8.7 dB/div.
low-pass ltering of the DAC subsystem at the transmitting side. This could be
corrected without much diculty.
7. Concluding Remarks and Further Research
In this paper, we have demonstrated an OFDM communication platform using
both MATLAB Simulink and software development facilities of an FPGA-based
development hardware to prove the principles and performances of systematic
blocks of the system for generation of OFDM symbols, encoding, and mapping
00025
403
to QPSK symbols with the possibility of upgrading to 16-QAM models of transmission medium. The use of pilots has enabled the estimation of the transmission
medium characteristics and hence their employment in the equalizer at the receiver.
The waveforms are monitored at the transmitting end blocks and at the receiving
end so as to conrm the working principles of the design of various blocks of the
OFDM communication systems.
The models presented in this paper will be modied and integrated into a
MATLAB development platform for simulation of OFDM signals transmission
through optically amplied multi-span optical ber communication systems.13 This
will be used to study various OFDM signaling techniques to combat impairment
and associated mitigation technqiues for ultra-long transmission without using dispersion compensating modules and associated optical ampliers. These works will
be reported in the future.
Furthermore, the developed system reported here would be expanded to include
all kinds of wireless transmission media and channel estimation for broadband
wireless access for the 4G wireless communication networks with the bit rate reaching 1 Gb/s and above.
For long-haul optical amplifed ber transmission systems, we identify the following issues/problems for further research:
Techniques for the reduction of the peak to average values of OFDM signals in
order not to drive the optical modulator into the nonlinear region, especially when
an optical interferometric modulator is used. Optical phase modulator may be
used in association with clamping of the peak amplitude. The error contribution
to the signal recovery due to this clamping will be studied and investigated by
simulation.
The model developed in this paper will be integrated with a MATLAB Simulink
platform for multi-span optically amplied ber transmission to study these
eects and the mitigation techniques to reduce the impairment.
Under long-haul multi-span transmission, OFDM will suer the eects of linear chromatic dispersion and polarization mode dispersion, and nonlinear eects
such as self-phase modulation, cross-phase modulation and four-wave mixing.
However, we believe that the self-phase modulation eects will aect the I- and
Q-components of the passband OFDM signals and the cross-phase modulation
eects will create interchannel interference eects in the frequency domain of
adjacent optical channels (i.e. at dierent wavelengths of systems employing dense
wavelength division multiplexing).
Mitigation techniques will be employed to overcome the distortion eects identied above such as equalization of OFDM, especially the equalization in the
frequency domain.
A parallel bank of these FPGA-based processors will be structured to form
extremely wideband signals for modulating an IQ optical modulator for
long-haul optical transmission, metropolitan or access optical Internet operating with multi-Gb/s bit rates.
404
00025
L. K. Dang et al.
References
1. Garca J., Cumplido R., On the design of an FPGA-Based OFDM modulator for
IEEE 802.16-2004, Proc. Int. Conf. Reconfigurable Comp. FPGAs, ReConFig 2005,
pp. 14, 2005.
2. Chang K.-C., Sobelman G. E., FPGA-based design of a pulsed-OFDM system, Proc.
APCCAS 2006, IEEE Asia Pacific Conf. Circ. Syst, 2006.
3. Dang L. K., Nguyen T. A., Bui H. P., Nguyen H. P., Hardware implementation of
OFDM systems, J. Dev. Res. Sc. Tech., Public University of Natural Science, HCM
City, 2009.
4. Altera Corporation, Stratix EP1S25 DSP Development Board, 2005. http://www.
altera.com/technology/dsp/dsp-index.jsp.
5. Weinstein S. P., Ebert P. M., Data transmission by frequency-division multiplexing
using the discrete Fourier transform, IEEE Trans Comm. Tech. 19(5): 628634, 1971.
6. Peled A., Ruiz A., Frequency domain data transmission using reduced computational complexity algorithms, Proc. IEEE Int. Conf. Acoustics Speech Sig. Proc.
(ICASSP80), pp. 964967, 1980.
7. Hara S., Prasad R., Multicarrier Techniques for 4G Mobile Communications, Artech
House, 2003.
8. Barkat M., Signal Detection and Estimation, Artech House, Norwood, MA, 2005.
9. Binh L. N., Digital Optical Communications, CRC Press, Florida, USA, Chapter 11,
2009.
10. MathWorks, access date March 2010. http://www.mathworks.com/products/
simulink/.
11. Altera Corporation, Viterbi Compiler User Guide. http://www.altera.com/technology
/dsp/dsp-index.jsp.
12. Altera Corporation, FFT MegaCore Function User Guide. http://www.altera.com/
technology/dsp/dsp-index.jsp.
13. Binh L. N., MATLAB Simulink modeling of Raman hybrid amplification for long
distance hut-skipped undersea optical fiber transmission systems, Opt. Eng. 48(10):
105005-1105005-12, 2009.