Universitat des Saarlandes
Naturwissenschaftlich-Technische Fakultat I
Fachrichtung Computer-und Kommunikationstechnik
Bachelorarbeit
Variance of DVB-T2 Performance Gains over
different channels
Vorgelegt von: Julian Metzger
am: 15.06.2009
Betreut von: Prof. Dr.-Ing.Thorsten Herfet
Erster Gutachter: Prof. Dr.-Ing.Thorsten Herfet
Zweiter Gutacher:
Eidesstattliche Erklärung Ich erkläre hiermit an Eides Statt, dass ich die vorliegende Arbeit selbstständig verfasst und keine anderen als die angegebenen Quellen und Hilfsmittel verwendet habe.
Statement under Oath I confirm under oath that I have written this thesis on my own and that I have not used any other media or materials than the ones referred to in this thesis.
Einverständniserklärung Ich bin damit einverstanden, dass meine (bestandene) Arbeit in beiden Versionen in die Bibliothek der Informatik aufgenommen und damit veröffentlicht wird.
Declaration of Consent I agree to make both versions of my thesis (with a passing grade) accessible to the public by having them added to the library of the Computer Science Department. Saarbrücken,…………………………….. …………………………………………. (Datum / Date) (Unterschrift / Signature)
Contents
Introduction 3
1 DVB-T2 Overview 5
1.1 FEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 QAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Rotated Constellations and cyclic Q-Delay . . . . . . . . . . . . . . 8
1.2.2 2-dimensional demapping of rotated constellations . . . . . . . . . . 10
1.3 OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.1 New OFDM FFT-sizes . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.2 Extended Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.3 Guard Intervall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.4 Pilots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.5 Frequency Interleaver . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Peak-to-Average-Power-Ratio reduction . . . . . . . . . . . . . . . . . . . . 18
1.4.1 Active Constellation Extension . . . . . . . . . . . . . . . . . . . . 21
1.4.2 Tone Reservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5 Bitrates and Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.5.1 Theoretical Data Rates . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6 Physical Layer Pipes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.7 MISO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2 Simulation suppositions 30
2.1 Simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2 Eb/N0 and SNR in OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Channel Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.1 AWGN channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.2 Rayleigh channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.3 Rician channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3 Simulation results 38
3.1 AWGN channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2 Fading Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Rician Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4 Rayleigh Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1
4 Conclusions 42
References 44
List of Figures 45
List of Tables 46
Appendix 47
2
Introduction
The standard for digital terrestrial television DVB-T was published in 1997. DVB-T was
developed to replace the analog television broadcasting and is on air in more than 35
countries. Today it is the most widespread standard for digital terrestrial television in
the world. Over 90 million receivers and end devices for DVB-T are sold today and mass
market and cheap production led to low device prices. Furthermore, not only stand-alone
receivers like set-top-boxes are used in increasing numbers, but also integration of built-in
receivers in end devices, primarily TV’s. Also mobile and portable devices, like laptops,
car entertainment systems and mobile phones is growing.
Today, 12 years later, more sophisticated techniques together with powerful hardware are
available and can provide more performance, robusteness and help to reduce costs.
The DVB project released the DVB-T2 specification as Blue Book in June 2008. The
standard was released from the ETSI in Octobre 2008.
Benefits from DVB-T2?
In some countries DVB-T content is broadcasted with an MPEG-4 video codec, which
allows HDTV already with DVB-T. What is the benefit from DVB-T2 if HDTV trans-
mission would be able without setting up a new system?
T2 is not backward compatible with DVB-T, so changing to T2 means new receivers at
the end users side and new transmitters at the broadcasters.
The DVB project states one main reason. DVB-T was developed for the transition from
analogue to digital television. In most countries the complete switch off of analogue
television is already terminated or even completed. The analogue switch off will release
frequencies and many different services will compete for these. DVB-T2 is intended to give
the broadcasters the possibility to use the available bandwidth as efficient as possible.1
DVB-T2 will not replace DVB-T instantly but both techniques will coexist for some time.
Besides there are some other reasons where DVB-T2 is expected to bring advantages
compared to DVB-T. These are subject to this document and will be inspected in the
following sections.
1DVB Fact-Sheet to DVB-T2: http://www.dvb.org/technology/fact sheets/DVB-T2-Fact-Sheet.0409.pdf
3
The DVB consortium states some commercial requirements for DVB-T2:
• the intended receivers are primarily portable or fixed
• DVB-T2 should achive at least 30% higher data rates than DVB-T following the
same constraints
• DVB-T2 has to be able to use the existing transmit and receive antennas, so MIMO
is not provided and MISO optional
• the SFN performance compared to DVB-T should be improved
• an option to provide service specific robustness within one channel to target different
types of end devices
• DVB-T2 should provide high flexibility in bandwidth and frequency
• a peak-to-average-power ratio reduction technique ought be defined to enable trans-
mission cost reduction
Additionally the DVB project followed the principles to reuse existing solutions, to pro-
vide a coherent family of standards and to develop the standards in a way that makes
translation between standards as easy as possible.
For the development of DVB-T2 that meant particularly, the adoption of the FEC and a
part of the system layer architecture of DVB-S2.
Outline
The next section explains the techniques that are new to DVB-T2. At the end of the
section the different parts of T2 are summarized in a calculation of theoretical bitrates
and a consideration how much performance gain is brought in by each part.
In section 2 a Matlab model of T2 and channel models are described, which are used to
prove the performance gain of some parts of the T2 system by simulations.
Section 3 discusses the results of the simulations, which are made for different channel
models.
At last in section 4 a resume will be drawn.
4
1 DVB-T2 OVERVIEW
1 Overview of the DVB-T2 system and new techniques
in T2
1.1 FEC
At the time, when DVB-T was developed the Reed-Solomon-Codes and convolutional
codes were widely used and offered the best error correcting performance.
Short time later even more powerful error correction schemes were developed such as the
Turbo-Codes and the LDPC codes. Though the idea of the LDPC code was already
invented in 1963 by R. Gallagher, at that time practical implementation was impossible
and it passed out of mind. In 1996 the LDPC code was rediscovered by D. MacKay, and
the LDPC proved to be very performant, operating near the shannon limit. Today the
LDPC code belongs to the most performant error correcting techniques that are known.
The DVB-S2[2] standard, which was released in 2005 includes a powerful forward error
correction using the LDPC code as an an inner code and the BCH as outer code. Following
the DVB design principles T2 uses the FEC from DVB-S2 except for the code rate 3/5,
where slight modifications were made.
The data are arrangend in FEC frames of length 64800 bit (normal FEC frame) or 16400
bit (short FEC frame). L1 signaling has to be sent in short FEC frames. Data frames can
be sent in short or normal FEC frames. Normal FEC frames have a lower overhead and
hence offer a higher performance. Short FEC frames induce less delay, as the interleaving
size is smaller.
BBFRAME BCHFEC LDPCFEC
NLDPC
KBCHN = KBCH LDPC
Figure 1: format of FEC frame
KBCH number of data bits for BCH block (outer code)
NBCH −KBCH number of code bits in coded BCH block
KLDPC number of data bits for uncoded LDPC block (inner code)
NLDPC length of coded FEC frame
T2 defines six code rates for the normal FEC-frame, which are given in table 1:
5
1.1 FEC 1 DVB-T2 OVERVIEW
LDPC code rate KBCH NBCH = KLDPC terrorBCH BCH code bits NLDPC
1/2 32208 32400 12 192 648003/5 38688 38880 12 192 648002/3 43040 43200 10 160 648003/4 48408 48600 12 192 648004/5 51648 51840 12 192 648005/6 53840 54000 10 160 64800
Table 1: code rates for the normal FEC frame
Additionally for short FEC-frames a code rate of 1/4 is defined, which is only used for
error protection of L1 pre-signalling. The L1 pre-signalling LDPC frames are punctured
at 11488 bits. For short FEC-frames the nominal code rates of the LDPC are different
to the effective code rates Reff . The code rates of the BCH for the short FEC-frame is
128/135.
LDPC code rate KBCH NBCH = KLDPC terrorBCH BCH code bits Reff NLDPC
1/4 3072 3240 12 168 1/5 162001/2 7032 7200 12 168 4/9 162003/5 9552 9720 12 168 3/5 162002/3 10632 10800 12 168 2/3 162003/4 11712 11880 12 168 11/15 162004/5 12432 12600 12 168 7/9 162005/6 13152 13320 12 168 37/45 16200
Table 2: code rates for the short FEC frame
The code rates are anounced to receivers in L1 post signaling. The code rates of the BCH
code depend on the LDPC code rate and are not specified by signaling. The real code
rate differs slightly from the LDPC code rate, when the BCH code rate is considerated in
the calculation.
The calculated code rates of the BCH are shown in table 3. CRdif denotes the difference
between the nominal code rate and the real code rate.
LDPC CodeRate 1/2 3/5 2/3 3/4 4/5 5/6BCH CodeRate 497/500 199/200 249/250 249/250 249/250 997/1000CRdif 0.6% 0.5% 0.4% 0.4% 0.37% 0.3%
Table 3: code rates for LDPC and BCH
6
1.2 QAM 1 DVB-T2 OVERVIEW
1.2 QAM
The higher data rates in T2 rely far on the new QAM of order 256. DVB-T offers
QPSK, 16QAM and 64QAM, these are enhanced for 256QAM in the new standard. The
256QAM carries log2(256) = 8 bit information, that is 33% more than the 64QAM. The
drawback is the higher susceptibility to noise. The minimal euclidean distance in 256QAM
is 2√170
= 0.15, compared to 2√42
= 0.30 in 64QAM.
For a symbol error rate of 10−2, there is a loss of 5dB between 64QAM and 256QAM,
so Es/N0 should be about 5dB higher for the 256QAM to achive the same symbol error
rate.
0 5 10 15 20 25
10−4
10−3
10−2
10−1
100
ES/N0
sym
bol e
rror
rat
e
symbol error rate
QPSK16 QAM64 QAM256 QAM
Figure 2: theoretical symbol error rates of QAM
Due to regulatory limitations it is not possible to increase Es/N0, so T2 must compensate
this loss somewhere else. That is done in a large share by the new forwared error coding
(as described above in section 1.1). Besides T2 introduces bigger FFT sizes for OFDM
(see section 1.3.1) and a technique, that is completely new to DVB, called Constellation
Rotation.
7
1.2 QAM 1 DVB-T2 OVERVIEW
1.2.1 Rotated Constellations and cyclic Q-Delay
DVB-T2 offers the option of Constellation Rotation. If this mode is used, each Constel-
lation Point of the QAM is rotated on the complex I-Q-plane. The angles of rotation are
defined by the T2 standard and depend one the QAM order.
Table 4 shows the values stated by the specification.
Modulation: QPSK 16QAM 64QAM 256QAMϕ in degree: 29,0 16,8 8,6 atan(1/16)
Table 4: rotation degrees defined by the T2 standard
The rotation projects the Constellation Points unique on both axes such that each axis
carries sufficient information to decode the modulated QAM points, even when the other
dimension is erased. The distance between the Constellation Points on each axis is non-
uniform as outcome of the angles defined in the specification. The rotation-angles are
chosen by the DVB-Organisation in a way that the provided robustness is maximized
over all channels. The Constellation Points are modified by the rotation phasor
RRQD = ej2πϕ360 .
To exploit the advantages of Constellation Rotation the In-Phase and the Quadrate part
of the QAM points are separated and the Quadrature part is cyclic shifted one QAM-
symbol in each FEC-frame. Thus redundancy is introduced without affecting code rate
or spectral efficiency such that even a lost virtual QAM-symbol can be recovered. The
cyclic delay produces a virtual constellation with M2 constellation points. Figures 3(a)
and 3(b) show the resulting signal after rotation respective rotation and cyclic Q-delay.
The constellation points from which the virtual constellation is derived are marked red in
figure 3(b).
8
1.2 QAM 1 DVB-T2 OVERVIEW
Qua
drat
ure
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
rotated 16QAM
In−Phase
(a) rotated constellation
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Qua
drat
ure
In−Phase
Scatterplot 16QAM virtual constellation
(b) virtual constellation after cyclic Q-delay
Figure 3: Constellation Rotation
9
1.2 QAM 1 DVB-T2 OVERVIEW
The distance of the constellation points on each axis is non-uniform. This distribution
offers more performance than a uniform distance. All constellation points with a small
distance are farmost separated on the other axis. This tradeoff offers the biggest combined
distance.
The rotation and the Q-delay can be expressed as
g0 = Re(RRQDf0) + jIm(RRQDfNcells−1)
gq = Re(RRQDfq) + jIm(RRQDfq−1), q = 1, 2, . . . , Ncells − 1
where fq are the QAM symbols in the FEC frame (section 1.1), Ncells is the number of
QAM symbols per FEC frame and gq the virtual QAM symbols.
After Constellation Rotation the QAM-symbols are fed through the Frequency Inter-
leaver. The Frequency Interleaver permutes all QAM cells in one OFDM symbol pseudo-
randomly. In-Phase and Quadrature part of one QAM-symbol are separated one sym-
bol after cyclic Q-Delay. The Frequency Interleaver spreads neighboured QAM symbols
further apart. Hence the In-Phase and the Quadrature part of one QAM-symbol are
travelling on different, distant frequencies on the channel. If a frequency selective chan-
nel destroys now one subcarrier and so the virtual QAM-symbol, the information is still
received on another frequency.
The receiver reverses the frequency interleaving and the cyclic Q-Delay and can then
combine the information of I- and Q-part. If one part of the QAM symbol is lost, the
decoding will be much more susceptible to noise, as the distance on each axis is smaller
than the Euclidean Distance of the QAM.
The process of decoding is described below.
1.2.2 2-dimensional demapping of rotated constellations
At the receiver the information of I- and Q-part has to be combined to demodulate the
bits. After frequency deinterleaving the Q-delay is removed, the corresponding I- and
Q-part of one QAM symbol is rejoined.
The probability for the received values, if bi is transmitted as 0 is
p(I,Q|bi = 0) =1
2mπσ2
∑x∈C0
i
exp−(I − ρIIx)2 + (Q− ρQQx)2
2σ2
where∑
x∈C0iIxQx is the set of all 2m−1 CP given that bi = 0. It is assumed that the other
2m−1 states are transmitted with equal probability. I and Q is the received constellation.
10
1.2 QAM 1 DVB-T2 OVERVIEW
ρI and ρQ are the amplitude-fading factors that I,Q were exposed to on the channel.
The calculation for bi = 1 is done analog.
−1.5 −1 −0.5 0 0.5 1 1.5−1.5
−1
−0.5
0
0.5
1
1.5Scatterplot 64QAM bit 0
In−phase
Qu
ad
ratu
re
(a) bit b0 in 64QAM
−1.5 −1 −0.5 0 0.5 1 1.5−1.5
−1
−0.5
0
0.5
1
1.5Scatterplot 64QAM bit 5
Qu
ad
ratu
re
In−phase
(b) bit b5 in 64QAM
Figure 4: Constellation Points in 64QAM
Figure 4 shows the set of CPs of the rotated 64QAM for bits b0, b5. bi = 0 is marked in
red and bi = 1 in blue.
The log-likelihood-ratio is then calculated from these probabilites:
LLR(bi) = ln
(P (bi = 1|I,Q)
P (bi = 0|I,Q)
)= ln
(P (I,Q|bi = 1)
P (I,Q|bi = 1)
)= ln
∑x∈C1ie− (I−ρIIx)2+(Q−ρQQx)2
2σ2∑x∈C0
ie−
(I−ρIIx)2+(Q−ρQQx)2
2σ2
A LLR > 0 implies that bit bi was transmitted as a 1 with higher probability than as 0.
LLR < 0 instead indicates that bi is more probable 0.
Simplified LLR
For hardware implementation the LLR calculation can be simplified by applying the max-
log approximation:
ln
(∑i e− ai
2σ2∑i e− bi
2σ2
)= ln
(−e
a02σ2 − e
a12σ2 − . . .− e
ak2σ2
−eb02σ2 − e
b12σ2 − . . .− e
bk2σ2
)=
11
1.2 QAM 1 DVB-T2 OVERVIEW
=1
2σ2
ln (−ea0 − ea1 − . . .− eak)︸ ︷︷ ︸−min(eai )
− ln(−eb0 − eb1 − . . .− ebk
)︸ ︷︷ ︸−min(ebi )
i = 0 . . . k, k = 2m−1−1
LLR(bi) =1
2σ2
(minx∈C0
i
((I − ρIIx)2 + (Q− ρQQx)
2)− min
x∈C1i
((I − ρIIx)2 + (Q− ρQQx)
2))
Iterative Demapping
An exhanced form of demapping is expected to have an improvement to the BER of the
decoded QAM.
Instead assuming that for bit bi all other 2m−1 states are equiprobable, the demapper is
fed with apriori information from the LDPC decoder. The demapping process is done in
iterations until a certain threshold. After each iteration of demapping the LDPC decoding
is done and from the error rate in the LDPC a probability can be calculated.
ETSI
Draft ETSI TR 102 831 V<0.8.0> (2008-09)152
LLR bi ln Pr(bi 1 | I,Q)Pr(bi 0 | I,Q)
ln
p I,Q | bi 1 p I,Q |bi 0
lne
I I I x 2 QQQx 2
2 2
xC i1
e
I I I x 2 QQQx 2
2 2
xC i0
9.5.3.1.2 Computation of simplified LLR (suitable for hardware implementation)
The LLR computation can be simplified by applying the Max-Log approximation:
ln ea1 eak max
i1kai
The LLR becomes:
LLR bi 12 2 min
xCi0
I I Ix 2 Q QQx 2 minxC i
1I I Ix 2 Q QQx 2
9.5.3.2 2D LLR demapper with iterative demapping and decoding
When Iterative Demapping (ID) is applied, the demapper has to be slightly modified in order to take extrinsic bit information coming from the LDPC decoder into account as shown in Figure 81:
QAM demapper -rotated
constellation
LLR(b0) LLR(b1)
LLR(bm-2)
LLR(bm-1)
I
Q
:
A priori information from the LDPC decoder
Figure 81: LLR demapper for iterative demapping
9.5.3.2.1 Computation of perfect LLR with ID
With iterative demapping, the metric for bit bi should be calculated in the light of a priori knowledge of the likely state of the m – 1 other bits, obtained from the LDPC decoder in the previous iteration. Because of this knowledge, we should no longer assume that all states x of the constellation are equiprobable. To reflect this, we need a more complicated expression for the conditional pdf of the received values I, Q , given that bi was transmitted as a 1:
p I,Q | bi 1 12m 2 e
I I I x 2 QQQx 2
2 2 Prapriori x | bi 1
xCi1
This expression sums up the contributions from each of the 2m–1 possible transmitted points x in the half-constellation Ci
1 that is distinguished by our choosing bi = 1. Each point has its own probability of having being transmitted. This probability can be expressed as a function of the probabilities that the (m – 1) bits other than bi take the value 0 or 1, as
Figure 5: Iterative LLR-Demapper [4, p.152]
The LLR with Iterative Demapping multiplies each bit probability Pbi with a probability
for each of the 2m−1 Constellation Points.
LLR(bi) = ln
∑x∈C1ie− (I−ρIIx)2+(Q−ρQQx)2
2σ2
∏k 6=i PLDPC(bk = xk)∑
x∈C0ie−
(I−ρIIx)2+(Q−ρQQx)2
2σ2 PLDPC(bk = xk)
12
1.3 OFDM 1 DVB-T2 OVERVIEW
1.3 OFDM
1.3.1 New OFDM FFT-sizes
DVB-T offers two choices of OFDM carriers: 2k and 8k.
The OFDM is realized by an IFFT. Today the progress in silicon devices allow larger FFT
sizes, than at the times when DVB-T was developed.
DVB-T2 enhances the OFDM to 1k, 2k, 4k, 8k, 16k and 32k carrier. The decision of
OFDM FFT size has to take various aspects in account and DVB-T2 offers huge flexibility
to optimize the transmission parameters to the needs
Aspects for the Choice of OFDM size
The choice of the OFDM size depends on the application and expected channel character-
istics, so T2 gives the content distributors a wide range of possible parameters to adapt
to the desired behaviour and expected channel suppositions.
To give 2 examples: The 32k mode will be very useful for fixed antennas, which experience
slow fading effects and multipath echoes but no fast fading, as it offers the longest symbol
duration in T2 and thus is less susceptible to time variations in the channel.
The 1k mode instead is less susceptible to Doppler shift and offers better performance for
mobile applications.
The maximum Doppler shift depends on the carrier frequency: fDmax = v·fcc
. So it is
possible that the limit of allowed Doppler shift is reached at different speeds in different
frequency bands using the same FFT size. Said in other words, the used frequency band
can force to use a smaller FFT size, if a certain Doppler performance is desired.
Hence the selection of the FFT size in mobile applications does also depend on the fre-
quency band. The performance regarding time and frequency variations is also improved
by the scattered pilots (see section 1.3.4).
1.3.2 Extended Mode
The sharp out of band falloff in 8k, 16k and 32k outperforms the requirements of the spec-
trum masks. DVB-T2 takes an advantage from this and introduces the extended mode.
The extended mode adds at both sides of the spectrum additional data subcarriers, such
that the spectral requirements are still satisfied, but the bandwith efficiency is increased.
13
1.3 OFDM 1 DVB-T2 OVERVIEW
Figure 6 shows the OFDM spectrum for different FFT sizes.
3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 4.3 4.4 4.5
x 106
−50
−40
−30
−20
−10
0
10
frequency relative to center frequency in Hz
PS
D in
dB
DVB−T2 OFDM spectrum
2k8k normal mode32k normal mode32k extended modePAL normal mask
Figure 6: spectrum of OFDM in normal mode and extended mode
8k extended mode
The subcarrier spacing in 8k mode is 1Tu
= 1116Hz. The spacing between the first used
carrier Kmin and the last used carrier Kmax in normal mode is 6816 ·1116[Hz] = 7.61MHz.
In extended mode the spacing between Kmin and Kmax is 6912 · 1116 = 7.71MHz. The
gain for 8k is 966817
= 0.014.
16k extended mode
The subcarrier spacing in 16k mode is 1Tu
= 558Hz. The spacing between Kmin and Kmax
in normal mode is 13632 · 558[Hz] = 7.61MHz.
In extended mode the spacing between Kmin and Kmax is 13920 · 558 = 7.77MHz. The
gain for 16k is 28813633
= 0.021.
32k extended mode
The subcarrier spacing in 32k mode is 1Tu
= 279Hz. The spacing between Kmin and Kmax
in normal mode is ·[Hz] = 7.61MHz.
In extended mode the spacing between Kmin and Kmax is 27841 · 279 = 7.77MHz. The
gain for 32k is 57627265
= 0.0211.
14
1.3 OFDM 1 DVB-T2 OVERVIEW
The values in the extended mode are summarized in table 5.2 The additional bandwidth
is the bandwidth that is used for additional active carriers and is calculated from the
difference in spacing between the first carrier Kmin and the last carrier Kmax in normal
and extended mode.
8k 16k 32ksubcarrier in normal mode 6817 13633 27265subcarrier in extended mode 6913 13921 27841added subcarrier at each side of the spectrum 48 144 288additional bandwidth 0.1MHz 0.16MHz 0.16MHzperformance gain 1.4% 2.1% 2.1%
Table 5: OFDM normal and extended mode
The 8k OFDM gains 1.4%, the 16k and 32k OFDM gain 2.1% capacity in the extended
mode.
Figure 6 shows the spectrum of normal and extended mode.
1.3.3 Guard Intervall
If the FFT-size is chosen,the OFDM symbol duration enlarges and the subcarrier spacing
gets smaller. If the same Guard Interval fraction is maintained while the FFT-size of the
OFDM symbol is increased, this results in a longer guard interval, and thus a longer echo
travel distance.
Longer echo travel distances are helpful to build larger Single Frequency Networks, as the
distance between two senders of that SFN can be larger.
The other way, the Guard Interval fraction can be decreased with increased FFT-size -
the guard-intervall duration will stay the same - to achieve the same maximal echo travel
distance for ISI-free reception. Reducing the Guard Interval fraction leads to a shorter
OFDM symbol duration without affecting the useful part of the symbol and the shorter
OFDM symbols result in a higher throughput. The overhead between Guard Interval
fraction to useful symbol part is reduced. In the figure below the 1/4 Guard Intervall is
reduced to a 1/16 Guard Intervall in the 32k mode. Hence the Guard Intervall duration
in time is maintained. The capacity gain is 18.75%
2assuming 8MHz channels, the values for other bandwidths differ
15
1.3 OFDM 1 DVB-T2 OVERVIEW
ETSI
Draft ETSI TR 102 831 V<0.8.0> (2008-09)23
Figure 11: System behaviour with constellation rotation (blue) and without (black).
4.5.2.8 FFT sizes 16K and 32K accompanied by guard-interval fractions down to 1/128
Increasing the FFT size results in a narrower sub-carrier spacing, but longer symbol duration. The first attribute leads to greater difficulties with inter-carrier interference, hence lower Doppler frequency that can be tolerated, so this is not a setting preferred for mobile reception. However, the second attribute, longer symbol duration, means that the guard-interval fraction is smaller for a given guard-interval duration in time (see Figure 12). This reductioin in overhead leads to a increase in throughput ranging from 2.3% to 17.6%.
Figure 12: Guard interval overhead reduction with larger FFT size
Other advantages consist of better robustness against impulsive noise, quasi-rectangular spectrum down to lower power-spectral-density levels and the option to interpolate in the frequency direction only. The memory requirements for interpolation in the receiver are in the same order for 32K as for 8K, but for 16K they are doubled. The FFT-calculation complexity is only slightly increased.
In summary, 32K together with a guard-interval fraction 1/128 might be a pragmatic setting for targeting stationary (and portable) devices.
4.5.2.9 LDPC/BCH error control coding
Whereas inner and outer error-control coding was realised with convolutional and Reed-Solomon codes in the DVB-T case, ten years of technological development mean that the higher complexity of LDPC decoding can now be handled in the receiver. DVB-T2 uses concatenated LDPC/BCH coding, as for DVB-S2. These codes assure a better protection, allowing more data to be transported in a given channel; they also show a steeper behaviour in the relation of BER to C/N, i.e. they move closer to the ideally desired 'brick wall' behaviour (see Figure 13).
Figure 13 shows the results before outer decoding (RS or BCH respectively), and the Reed-Solomon decoder has more powerful error-correcting capabilities than the BCH decoder. A BER of around 10-4 before Reed-Solomon is usually assumed to give "quasi-error-free" performance after Reed Solomon. The gain achieved here is therefore in the order of 5 dB.
Figure 7: GI overhead reduction [4, p.23]
The maximal echo travel distance for different FFT-sizes and Guard Interval fractions is
shown in table 6. For a SFN these values are the maximal spacing between two transmit-
ters. The yellow marked cells show the configurations available in DVB-T.
1/4 19/128 1/8 19/256 1/16 1/32 1/1281k 8.4km - 4.2km - 2.1km - -2k 16.8km - 8.4km - 4.2km 2.1km -4k 33.6km - 16.8km - 8.4km 4.2km -8k 67.2km 39.9km 33.6km 19.95km 16.8km 8.4km 2.1km
16k 134.4km 79.8km 67.2km 39.9km 33.6km 16.8km 4.2km32k - 159.6km 134.4km 79.8km 67.2km 33.6km 8.40km
Table 6: maximum echo travel distance
Each bigger FFT size allows to reduce the Guard Intervall about 1/2 while still keeping
the same GI duration in time.
1.3.4 Pilots
DVB-T defines a static Pilot Pattern, which is the same for all configurations of OFDM
size, carrier frequency, Guard Intervall, etc. This static Pilot Pattern offers the same
performance in channel estimation for all FFT-sizes. In many cases a more flexible choice
of Pilot Pattern is desired and now induced by T2.
T2 defines eight different Pilot Patterns for the Scattered Pilots, which differ in their
performance regarding frequency and time interpolation, but also have different overhead.
For example receivers with fixed roof-top antenna do not face doppler shift and hence
the pilots that detect Doppler shift can be reduced to a minimum. On the other hand a
service for mobile receivers can trade off a decreased bitrate (especially as the screensize
of the devices is small) against achieving a better BER in fast fading channels.
Not all combinations of Pilot Patterns and FFT-sizes are possible. If PP8 is used, the
channel estimation is done with the useful data instead of pilots.
16
1.3 OFDM 1 DVB-T2 OVERVIEW
PP1 PP2 PP3 PP4 PP5 PP6 PP7 PP8overhead 8.33% 8.33% 4.17% 4.17% 2.08% 2.08% 1.04% 1.04%
Table 7: Pilot Patterns in T2
The Continual Pilots in T2 are defined by FFT-size and the Scattered Pilot Pattern.
Most of the Continual Pilots are replaced by Scattered Pilots. The overhead, if only the
Continual Pilots are included in the caluclation, that are not replaced by Scattered Pilots,
ranges from 0.59% in the 1k mode to 0.12% in the 32k mode.
1.3.5 Frequency Interleaver
The Frequency Interleaver of T2 is adopted from DVB-T, but enhanced for the other FFT
sizes of T2. In T2 the Frequency Interleaver is essential for the Constellation Rotation, as
it separates the In-phase and Quadrature in frequency. Without frequency interleaving,
a smallband frequency selective noise, which cancels out more than one carrier, would
cancel out the benefits from the Constellation Rotation.
The Frequency Interleaver is implemented with a shift register and a wires permutation.
These generate a pseudo-random address between 0 and Nmax. Nmax is the maximal num-
ber of active data carriers in one OFDM symbol. The data from the OFDM Framebuilder
are then permuted accordingly to that address vector. The permuted data cells are sent
to the OFDM-Modulator.
17
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
1.4 Peak-to-Average-Power-Ratio reduction
Besides the many advantages of OFDM it also brings a drawback with it. An OFDM
symbol with FFT-size N is a superposition of N subcarriers with gaussian distributed
amplitude values, and thus can have a N times higher Peak-to-Average-Power-Ratio than
a single carrier symbol.
A high PAPR is unfavorable for some reasons.
• The transmit high power amplifiers have to be operated with a power back-off
between the average signal energy and the clip-off for high power peak amplitudes.
An amplifier transfer curve with a 6dB back-off is depicted in figure 8. The axes are
normalized to Eav. The nonlinear transfer curves are modelled by
g(A) =A(
1 +(AA0
)2pk)−2pk
.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
idealnonlinear pk=4nonlinear pk=8
back−off
clip−off
Eav
Figure 8: back-off between Eav and clip-off
To prevent distortions fo the signal the amplifiers need to operate in the linear region
and require the back-off to be large enough that the peak-values are not clipped.
If the back-off between Pavg and Ppeak is too small, high peak amplitudes will be
clipped in the amplifier and cause a degraded BER and increased out-of-band energy.
Figure 9 illustrates the resulting spectrum after clipping and the original spectrum.
The difference in out-of-band energy is clearly visible.
18
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
−4 −3 −2 −1 0 1 2 3 4
x 106
−50
−40
−30
−20
−10
0
10
frequency arround carrier frequency
PS
D in
dB
DVB−T2 OFDM spectrum
12dB backoff16dB backoff20dB backoff
Figure 9: spectrum off an 8k OFDM-symbol with clipping
On the other hand large back-offs diminish the efficiency of the high power amplifiers
because a large operation range is idle the most time but kept in reserve for some
peak amplitudes. Especially for the high power amplifiers used for broadcasting this
induces high costs.
• The maximum transmit power is restricted by regulations. The amplification of the
signal is thus limitated by the maximum allowed peak amplitude. As a consequence
the SNR of low power portions of a signal with high PAPR can not be sent with
higher SNR because other parts of the signal prohibit higher transmit power.
The cummulative distribution function of the PAPR of OFDM signals with random data
and 256QAM is shown in figure 10. The influence of Pilots is neglected.
19
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
15 20 25 30 35 400
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PAPR in dB
P(d
B)
PAPR cummulative distribution function
2k OFDM8k OFDM32k OFDM
Figure 10: PAPR CDF for different OFDM signals
DVB-T2 introduces two techniques to reduce the PAPR:
In figure 11 the effect of clipping an OFDM signal on the BER is depicted. It shows
clearly, that OFDM signals with more subcarriers need a higher backoff, as their peak-to-
averge-power is higher.
0 5 10 15
10−4
10−3
10−2
10−1
100
Eb/No
bit e
rror
rat
e
16QAM BER with OFDM and different clip−off values
theoretical8k OFDM no clipping8k OFDM with back−off 6dB8k OFDM with back−off 12dB32k OFDM with back−off 12dB32k OFDM with back−off 18dB
Figure 11: BER for clipped OFDM signals
DVB-T2 introduces two techniques to reduce the PAPR:
20
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
• Active Constellation Extension
• Tone Reservation
The use of PAPR is optional and the T2 signal can be generated with eiter one or both
PAPR techniques together. The use of ACE and Rotated Constellation is not possible
as it would destroy the virtual constellation. Hence ACE PAPR reduction and Rotated
Constellations are not allowed by the specification.
1.4.1 Active Constellation Extension
The ACE produces a signal xACE that replaces the original signal.
ACE extends the outer constellation points of the QAM symbols, such that they have
higher amplitude values. This results in higher Pavg, because some symbols have a higher
symbol energy ES, so that the ratio between Pavg and Ppeak is lower. The mimimum
distance between the constellation points stays unchanged.
A block diagram of the ACE algorithm is depicted in figure 12.
x′ is generated from X by a four times oversized IFFT. The four times oversized IFFT
effects as oversampling of factor four and a low pass filtering.
x′′ is then obtained by applying the clipping operator Vclip on x′. Vclip is given as parameter
to the ACE algorithm and can be chosen between 0dB and +12dB above the standard
deviation of the signal.
x′′k =
x′k if ||x′k||2 ≤ Vclip
Vclipx′k||x′k||2
if ||x′k||2 > Vclip
XC is calculated from x′′ by an four times oversized FFT. The oversized FFT works as a
downsampler and a low pass filter.
The difference between XC and X is added to X with gain G.
X ′C = X +G · (XC −X)
G is given as parameter to the ACE algorithm and can be chosen between 0dB and 30dB.
The extension of the Constellation Points is now limited by a saturation operator L.
L can be chosen between 0.7dB and 1.4dB and is added to the maximal value a outer
Constellation Point can have. For instance with L = 1dB and 64QAM, the extension will
21
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
be bound to ±7/√
(42) · 1.26 = ±1.36.
Saturation is done for In-Phase and Quadrature separately. If I- or Q-part exceed the
extension limit their value is attenuated to L with according sign.
At the end the ACE-algorithm constructs Xace from values of the original signal X and the
modified values from X ′C . The modified values are taken into Xace, if they are extendable,
the modified value is bigger than the original value and if both values have the same sign,
else the original value from X is kept. Only the outer Constellation Points are extendable,
as the extension of other Constellation Points would diminish the Euclidean Distance.
Xace replaces the original signal X.
Figure 12: block diagram of the ACE algorithm from [1, p.99]
The receiver of an OFDM signal with ACE has to consider the altered constellation
points in the QAM demapper. Usually the deviations from the QAM grid are caused by
the presence of noise. ACE however produces also modified constellation points and the
demapper has to consider this in the demodulation process.
The QAM constellation of a signal that was modified by ACE is shown in figure 13. The
original Constellation Points are marked red, the modified Constellation is marked in blue.
22
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
Figure 13: QAM constellation after ACE
1.4.2 Tone Reservation
The Tone Reservation PAPR reduction algorithm tries to find a impulse-like kernel which
cancels the signal peaks out. TR reserves a number of carriers which are not carrying
data but the peak reduction kernel. The obtained peak reduction signal is added to the
OFDM signal x′ = x+ c(i). The receiver needs no other information but the L1 signalling
that TR is used and discards the reserved carriers.
The use of TR inhibits a small loss of capacity as the number of data symbols per OFDM
symbol is reduced. The number of TR carriers depends on the FFT-size of the OFDM.
The capacity loss per OFDM symbol is around 1% and calculated exactly in table 8.
FFT-size: 1k 2k 4k 8k 8k ext 16k 16k ext 32k 32k extTR cells: 10 18 36 72 144 288TR overhead: 1.17% 1.05% 1.05% 1.05% 1.04% 1.06% 1.03% 1.06% 1.03%
Table 8: Tone Reservation carriers and overhead
TR provides two parameters to controll the peak reduction signal, the maximum number
of iterations imax and Vclip, the clipping magnitude level. Vclip is the same as for ACE
between 0dB and +12.7dB above the standard deviation of the signal. Choosing too
23
1.4 Peak-to-Average-Power-Ratio reduction 1 DVB-T2 OVERVIEW
high clipping values leaves the benefit of PAPR unexploited. If too low clipping values
are chosen, the algorithm will not be able to fulfill these and produces noise due to the
imperfectness of PAPR reduction.
The TR algorithm is described in the following:
A reference kernel is defined as
p =
√NFFT
NTR
IFFT (1TR)
where NTR is the number of tone reservation carriers, as defined in 8. 1TR is a vector with
ones at the position of the TR carriers and zeros else.
The peak reduction signal is initially set to all zeros. The following steps are done in
iterations until the maximal number of iterations is reached or the maximum addition of
the OFDM-symbol x and the peak reduction signal c(i) in the i-th iteration is below the
clipping threshold Vclip:
1. the maximum value of xn + cin and its index mi is searched; n = 1 . . . NFFT
if max(xn + cin) is below Vclip the iterations are aborted
2. update of the peak reduction signal c(i) = c(i−1) − αip(mi)
αi =xmi+c
(i−1)mi
yi(yi − Vclip)
p(mi) is the peak reduction kernel shifted cyclic to the right.
3. unless the maximum number is reached return to first step and increase i
4. the PAPR reduced signal is x′ = x+ c(i)
The PAPR techniques are particularly important for DVB-T2, as the new OFDM of 32k
and 16k will also implicate higher PAPRs than the 8k in DVB-T. The use of a 32k OFDM
in T2 should of course not be ruled out by unreasonable amplifier requirements.
24
1.5 Bitrates and Capacity 1 DVB-T2 OVERVIEW
1.5 Bitrates and Capacity
r =ld(M) · CRLDPC · CRBCH · SPov · CPovL1ov
TGS
CRLDPC and CRBCH are the code rates of the forward error correction. SPov, CPov and
L1ov is the overhead from Scattered Pilots, the Continual Pilots and the L1-signaling
overhead.
The L1-signalling overhead depends on the configuration of the T2 signal and splits up in
L1-presignalling and L1-postsignalling. L1-signalling is only done at the beginning of a
superframe and the configuration values communicated by the L1-signalling are valid for
the whole superframe. The L1-presignalling has a fixed length of 1840 BPSK symbols.
The length of L1-postsignalling depends on the configuration.
1.5.1 Theoretical Data Rates
QPSK 16QAM 64QAM 256QAM1/2 7.5 15.0 22.6 30.13/5 9.0 18.1 27.1 36.22/3 10.0 20.1 30.2 40.33/4 11.3 22.6 33.9 45.34/5 12.0 24.2 36.2 48.45/6 12.6 25.2 37.8 50.4
Table 9: theoretical bitrates in T2
Figure 14 illustrates the bitrates from table 9. For calculation a 32k OFDM with extended
mode and a Guard Intervall of 1/128 is chosen. This configuration achives the highest
data rate.
25
1.5 Bitrates and Capacity 1 DVB-T2 OVERVIEW
QPSK 16QAM 64QAM 256QAM5
10
15
20
25
30
35
40
45
50
55bit rates of T2
bitr
ate
in M
bit/s
Figure 14: bitrates in DVB-T2
Table 10 compares the bitrates of DVB-T and DVB-T2, if the same QAM and Code Rate
but biggest OFDM and smallest Guard Intervall is chosen. The performance gain of T2
is in this case only earned by the extended mode and the reduced overhead of the Guard
Intervall and the Pilots.
QPSK 16QAM 64QAMDVB-T DVB-T2 DVB-T DVB-T2 DVB-T DVB-T2
1/2 6.03 7.54 12.06 15.08 18.1 22.613/5 - 9.05 - 18.11 - 27.172/3 8.04 10.00 16.09 20.1 24.13 30.223/4 9.05 11.33 18.1 22.66 27.14 33.994/5 - 12.09 - 24.18 - 36.275/6 10.04 12.60 20.11 25.20 30.16 37.807/8 10.56 - 21.11 - 31.76 -
Table 10: comparison of bitrates between DVB-T and T2
The gain in the bitrates is about 20%.
Table 11 summarizes the enhancements of T2 to DVB-T:
26
1.5 Bitrates and Capacity 1 DVB-T2 OVERVIEW
DVB-T DVB-T2- concept of PLPs- optional MISO
OFDM subcarriers 2k, 8k 1k, 2k, 4k, 8k, 16k, 32k
extendend mode -8k (+1,4%), 16k(+2,1%), 32k (+2,1%)
Guard Intervall 1/4, 1/8, 1/16, 1/321/4, 19/256, 1/8,19/128, 1/16, 1/32,1/128
QAM QPSK 16QAM, 64QAM QPSK, 16QAM, 64QAM, 256QAM
FEC Code-RateRS + CC1/2, 2/3, 3/4, 5/6, 7/8
BCH + LDPC1/2, 3/5, 2/3, 3/4,4/5, 5/6
Scattered Pilots 8% overhead 1%, 2%, 4%,8% overhead
Table 11: comparison between DVB-T and DVB-T2
Table 12 summarizes the gain of each part of T2. The Code Rate is not considered in the
calculation.
QAM 256QAM: +33%extended mode 8k: +1.4% 16k: +2.1% 32k: +2.1%Guard Intervall 8k 1/4 16k ext, 1/8: +12.5% 32k ext, 1/16: + 18.75%Scattered Pilots PP3,PP4, +3.8% PP5,PP6 + 6% PP7,PP8 +7%
performance gain +60.85%
Table 12: gain of different parts in DVB-T2
27
1.6 Physical Layer Pipes 1 DVB-T2 OVERVIEW
1.6 Physical Layer Pipes
In DVB-T all services in one bouquet have to use the same configuration of modulation
and code rate.
DVB-T2 gives the content providers more flexibility with the Physical Layer Pipes. Each
PLP can have its own parameters of modulation order, code rate and time interleaving
depth. The parameters can be chosen by the service provider. Hence the concept of PLPs
allows to allocate a different bit rate and robustness to each service. That makes it possible
to adapt the configuration for a specific type of receiver, like a fixed roof-top-antenna or
a mobile device, without making compromises in the other services.
A second advantage is, that the PLP concept can be used to save energy in the receiver.
A receiver can turn on to receive the PLP, it is currently focused on. When the other
PLPs in the service group are transmitted, the receiver can turn off and thus save power.
As each group of services shares common information like the PSI/SI tables and this in-
formation should not be transmitted in duplicates, each group of PLPs contain a common
PLP, which contains the shared information. Therefor each receiver has to be able to
decode at least two PLPs at the same time.
DVB-T2 does not necessarily need more PLPs to be transmitted, but even if the parameter
configuration for all PLPs is the same, the use of more PLPs is advantageous in respect
of power saving and a longer time interleaving.
1.7 MISO
The decision in the commercial requirements, that existing DVB-T transmit and receive
antennas should further be usable in T2, ruled out that MIMO was included in DVB-T2.
To give broadcasters however the possibility to improve transmission in fading channels
by diversity, T2 includes optionally MISO.
MISO needs no extra antennas at the receiving end and hence is not visible for the user.
At the end device the different MISO signals are received as overlapping multipath signals.
These can be combined and contribute to the received energy so that the signal-to-noise
ratio is improved.
The MISO in DVB-T2 is based on the Alamouti technique, with a pair of transmitters.
Both tranmitters send a slightly modified version of the original signal. Preamble symbols
are not modified, but sent twice the same at each transmitter.
In fading channels the received signal strength can vanish due to deep fading. In conse-
quence, parts of the signal are erased.
28
1.7 MISO 1 DVB-T2 OVERVIEW
MISO can help to prevent this, as more than one signal is transmitted to the receiver and
the probability that all MISO signals are completly erased is low.
29
2 SIMULATION SUPPOSITIONS
2 Simulation suppositions
2.1 Simulation model
For simulations a part of the T2 system was modeled in Matlab.
The following Matlab function:
T2Modulation(bitstream,qamSize,constRotation,ofdmSize,GI,extendedMode)
takes a bitstream as input. A channel bandwidth of 8MHz and a normal FEC-frame with
64800bits is assumed. T2Modulation() calles a number of functions which generate a
modulated T2Signal.
The bitstream is processed in FEC-frames of 64800 bits, as the cyclic Q-Delay is done
on 64800/ld(M) cells. If neccessary, the bitstream is padded to required length of n ·64800, n = 1 : NFECFrames.
At the Bit to Cell Demultiplexer, the bitstream is segmented into substreams for the
QAM. The order of the bits is permuted. The number of substreams in the long FEC-
Frame is, except for the QPSK, 2 · ld(M). The cell words of size ld(M) are then generated
from the substreams. The first bits of the first ld(M) substreams form the first cell word,
the first bits of the second ld(M) substreams form the second cell word and so on.
The Constellation Mapper forms the complex amplitude values from the cell words. If
Constellation Rotation is set, the Constellation Mapper rotates the cells by multiplication
with the complex phasor and shifts the Quadrature part cyclic in each FEC-frame.
To normalize the QAM amplitudes, the Constellation Points are multiplied with the
normfactor. The normfactor is√
2 for QPSK,√
10 for 16QAM,√
42 for the 64QAM
and√
170 for the 256QAM.
The OFDM-framebuilder arranges the data in groups of Nmax symbols. Nmax is defined
by the OFDM FFT-size and extended mode and indicates the maximum number of data
QAM symbols, that can be transmitted per OFDM-Frames. If another pilot pattern than
PP7 is used, the number of active data cells per OFDM-Frame is less than Nmax. This
Matlab model neglects the influence of pilots, the pilot subcarriers are set to zero. Nmax
data carriers are used, that means PP7 is assumed.
The Frequency Interleaver permutes the Nmax QAM-symbols in a pseudo-random manner.
The Matlab model implements the algorithm given in [1].
After frequency interleaving an OFDM signal is generated by IFFT.
The Guard Intervall is inserted, the last GIfraction ·NFFT samples are inserted before the
useful OFDM symbol.
30
2.1 Simulation model 2 SIMULATION SUPPOSITIONS
Time- and Cell-Interleaving is not done in this model, the whole FEC is idle. Instead of
using Code Rate× 64800bits, the 64800 NLDPC bits are data bits.
The obtained T2Signal is fed through the different channel models which are described in
section 2.3.
The receiver transforms the signal back into a bitstream.
First the Guard Intervall is removed and a FFT applied to demodulate the OFDM. The
groups of Nmax data symbols are reappended to frames of 64800/ld(M) QAM symbols.
These frames are sent to the LLR-demapper. Here the simplified LLR calculation as
described in section ?? is implemented. The Demapper removes first the cyclic Q-Delay.
Now the corresponding I- and Q-part are rejoined in one QAM-symbol. For each of the
ld(M) bits the demapper calculates the probability, if this bit is more likely a one or
zero. For this calculation it compares the received Constellation Points of bit bi with all
2(m−1) Constellation Points, with bi set to zero and with bi set to one. The sum of the
squared distances between the In-Phase parts and the Quadrature parts is calculated.
The distances are weighted with the amplitude fading factors ρI and ρQ.
The decoding performance relies on exact values of ρI and ρQ.
In a real system these have to be gained by channel estimation through the pilots. In the
simulation model perfect channel estimation is assumed and the values, as the channel
characteristics are known, are fed directly into the Demapper.
The code for the demapper is appended on 47.
After demapping, the bits from the cell words are reassembled to a bitstream.
The received bitstream is compared with the original bitstream and the bit-error-rate
calculated from the difference.
31
2.1 Simulation model 2 SIMULATION SUPPOSITIONS
bit t
o ce
llde
mul
tiple
xer
cons
tella
tion
map
per
b 0b1..
.bFF-1
c c c0 1 N
cp0 1 N
cp cp
ofdm
framebuilder
0 0
cp
...
cp
0,kmin
m,kmin
cp
...
cp
0,kmax
m,kmax
frequency
interleaver
0 0
cp'
..
. cp'
0,kmin
m,kmin
cp'
..
. cp'
0,kmax
m,kmax
ifft
s .
.. s
0,0
0,NFFT
s .
.. s
m,0
m,NFFT
...guard
intervall
s'
... s
'0,0
0,G
s'
... s
'm,0
m,G
...
+no
ise
+ m
ultip
ath
+ fa
ding
channel
cell
wor
d to
bi
tm
ultip
lexe
r
2D L
LR
dem
appe
rb 0
b1..
.bFF-1
c c c0 1 N
cp0 1 N
cp cp
ofdm
fram
eto
QA
M c
ells
0 0
cp
...
cp
0,kmin
m,kmin
cp
...
cp
0,kmax
m,kmax
frequency
de-
interleaver
0 0
cp'
..
. cp'
0,kmin
m,kmin
cp'
..
. cp'
0,kmax
m,kmax
ffts
...
s0,0
0,NFFT
s .
.. s
m,0
m,NFFT
...guard
intervall
removal
~~
~
~ ~ ~
~ ~ ~
~~
~~
~~ ~
~~
~~
~s'
...
s'
0,0
0,G
s'
... s
'm,0
m,G
...~
~~
~
BER
Figure 15: Modulation
32
2.2 Eb/N0 and SNR in OFDM 2 SIMULATION SUPPOSITIONS
2.2 Relation between Eb/N0 and SNR in OFDM
The simulations scale the bit error rates to Eb/N0. The conversion from Eb/N0 to SNR is
SNR =EbN0
· ld(M)
or in logaritmic scale
SNRdB =EbN0 dB
+ 10 · log10(ld(M))
This holds for single carrier transmissions. For OFDM transmission the calculated SNR
differs for some reasons:
• the bandwidth for transmission is reduced by the guard band
• the symbol duration is prolonged by the guard intervall TGS = TU + TG
• not all subcarriers are modulated with the same order QAM, the pilots are encoded
in BPSK
In the following a formula for the SNR degradation through OFDM is derived. The
number of pilots depends on the Pilot Pattern. Here the PP7 for the Scattered Pilots is
used, as the number of data carriers is maximal for PP7. If other Pilot Patterns are used,
the SNR degradation is a bit worse.
Es/N0 = Eb/N0 ·(Ndata
CN
)·(
TUTU + TG
)+ ·(
(CT −Ndata) · ApCN
)
)
Ndata maximum number of data carriers
CT number of active OFDM subcarriers
CN size of OFDM FFT
TU useful part of the OFDM symbol in seconds
TG guard intervall fraction in seconds
Ap boosted amplitude for the pilot
Table 13 gives some example values for the Es/N0 degradation in OFDM for PP7:
33
2.2 Eb/N0 and SNR in OFDM 2 SIMULATION SUPPOSITIONS
8k 32kGI 1/8 1/32 1/32 1/128QPSK -0.52dB -0.39dB -0.39dB -0.36dB16QAM -1.07dB -0.81dB -0.81dB -0.73dB64QAM -1.62dB -1.22dB -1.22dB -1.11dB256QAM -2.16dB -1.63dB -1.63dB -1.48dB
Table 13: degradation in SNR for different sizes of QAM and OFDM FFT
If the degradation through OFDM is not considered in the simulations a difference in the
BER is visible between the theoretical curve for single carrier transmission and the curve
from the simulation.
34
2.3 Channel Models 2 SIMULATION SUPPOSITIONS
2.3 Channel Models
To estimate the system behaviour possibly near to real world conditions, channel models
are used for simulations in Matlab. There is a number of models which simulate different
aspects of the wireless transmission. The most known and used models are the
• AWGN channel model
• Rayleigh channel model
• Rician channel model.
These three are used in the simulations for this document and described in detail below.
2.3.1 AWGN channel
The AWGN channel model is a simple but basic concept for modelling channel effects on
electromagnetic signals in communication systems.
The AWGN channel adds white noise n(t) to the signal s(t):
y(t) = s(t) + n(t)
The noise has a constant spectral density and the amplitudes are normal distributed with
variance σ2 = N0/2. N0 is the single-sided noise spectral density.
White Noise is existent in all communication systems independent of their progagation
and induced by many sources like thermal noise in electronic circuits, terrestrial noise,
and cosmic noise.
Hence the AWGN channel model is essential but not sufficient to model terrestrial prop-
agation effects.
The terrestrial propagation faces further effects like multipath, slow and deep fading,
which can affect the channel severe. To consider these, other channel models have to be
used additionally. The Rayleigh and the Rician channel model are common reperesenta-
tives of these and described below.
2.3.2 Rayleigh channel
The Rayleigh channel model assumes that at the sink a number of signals with varying
amplitude and delay are received. The multipath components of the signal are reflected on
still obstacles i.e buildings, mountains, water surfaces and moving obstacles like vehicles
and aircrafts.
Moving objects change their positions and hence the received multipath components re-
35
2.3 Channel Models 2 SIMULATION SUPPOSITIONS
flected from these vary over time. This effect is called slow fading.
Additionally fast fading can complicate the mobile reception. Fast fading is induced by
the Doppler effect and is encountered on moving receivers. The amount of Doppler shift
depends on the velocity, the carrier frequency and the angle between moving direction
and direction of the sender. The maximum Doppler shift is
fDmax =v · fcc
.
The Doppler shift depending on the angle between sender and reciever is
fD = fDmax · cosα =v · fcc· cosα.
V
V
Figure 16: Multipath channel
The In-Phase and the Quadrature part of each QAM symbol face staticstically indepen-
dent normal distributed variance. The sum of these variances is the sum of two zero-mean
gaussian distributions and called Rayleigh distributed:
f(x, σ) =x
2σex2
2σ2
Figure 17 shows the fading in received signal strength after a signal is passed through a
Rayleigh channel with a maximum Doppler shift fDmax of 10Hz.
36
2.3 Channel Models 2 SIMULATION SUPPOSITIONS
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1−35
−30
−25
−20
−15
−10
−5
0
5
10fading in Rayleigh Channel
time in seconds
pow
er in
dB
Figure 17: received signal strength in fading channel
2.3.3 Rician channel
Contrary to the Rayleigh channel the Rician channel assumes an aditional direct path, the
line of sight, between transmitter and receiver. The ratio of signal energy from the direct
path and the multipath contributing to the energy of the received signal is expressed by
the K − factor
K :=|E0|2∑Nn=1 |En|2
.
For K → 0 the Rician Channel approaches the Rayleigh Channel. For K → ∞ the
channel has only the line-of sight path.
The Rician distribution is
f(x) =x
σ2exp
(x2 + s2
2σ2
)I0
(x · sσ2
).
I0 is the modified Bessel function, s2 = |E0|2, the energy of the line-of-sight path. For
s = 0 the Rician distribution is a Rayleigh distribution.
In [1] the Rayleigh channel is used to simulate mobile reception, while the Rician channel
is used for the simulation of fixed receivers.
37
3 SIMULATION RESULTS
3 Simulation results
3.1 AWGN channel
The advantages of the OFDM are idle in the AWGN channel and the new FFT-sizes do
not improve the bit error rates in T2 compared to DVB-T. There is also no benefit from
the rotated constellations and frequency interleaving. Hence the bit error rates without
error coding in DVB-T and DVB-T2 are the same.
The simulation results for the AWGN channel are shown below. Figure 18 shows a com-
parison between the DVB-T modulation and the T2 modulation with rotated constellation
points.
0 2 4 6 8 10 12 14 16 1810
−4
10−3
10−2
10−1
100
Eb/No
bit e
rror
rat
e
BER in AWGN channel (8k OFDM)
16QAM DVB−T16QAM rotated DVB−T264QAM DVB−T64QAM rotated DVB−T2
Figure 18: comparison BER of DVB-T and DVB-T2
Figure 19 proves that the constellation rotation in T2 has no improvement in the AWGN
channel.
38
3.2 Fading Channels 3 SIMULATION RESULTS
0 2 4 6 8 10 12 14 16 1810
−4
10−3
10−2
10−1
100
Eb/No
bit e
rror
rat
e
BER in AWGN channel (8k OFDM)
16QAM DVB−T16QAM rotated DVB−T264QAM DVB−T64QAM rotated DVB−T2
Figure 19: comparison BER with and without constellation rotation
3.2 Fading Channels
The gain in the BER with rotated Constellations becomes clearly evident in deep fading
and erasure channels. In fading channels with less severe fading, where all carriers are
transmitted despite beeing faded, with a channel estimation the fading effect can be coped
with, also with Non-rotated Constellations
The advantage of the Rotated Constellations turns out when some carriers are completely
vanished due to frequency selective fading. The combination of In-Phase and Quadrature
diversity, Q-Delay and frequency interleaving can then recover lost QAM symbols.
Theoretically up to Ndata/2 carriers can be lost and still all information can be recovered.
Ndata is the number of active data carriers in one OFDM symbol. In practice such a big
erasure will degrade the bit error rate for two reasons:
• The information that is carried on each axis has a much smaller distance, than the
minimum distance of a complete QAM symbol.
Hence, if a QAM symbol must be decoded from one part only it is much more
susceptible to noise
• The decoding process in the case of erasure depends on the position of the corre-
39
3.3 Rician Channel 3 SIMULATION RESULTS
sponding I- and Q-part after frequency interleaving.
In unfortunate cases it can happen that both parts of a symbol are erased on differ-
ent frequencies. For a large number of erasures the probability that both parts are
lost increases.
3.3 Rician Channel
The performance gain of Rotated Constellations in the Rician channel depends on the
ratio between direct path and the multipath components. For large K-factors the LOS
dominates and the advantage of the Rotated Constellations vanishes. For K → 0 there is
no line of sight and the results match the results in the Rayleigh channel.
3.4 Rayleigh Channel
Figure 20 shows the results in a frequency selective fading channel with erasures. The
erased QAM symbols can be recovered from the Rotated Constellations, while the infor-
mation in the transmission without rotated constellations is lost.
Perfect channel estimation with phase detection and AGC is assumed.
The values of ρI and rhoQ are not estimated but given the demapper in the simulation
as parameters.
40
3.4 Rayleigh Channel 3 SIMULATION RESULTS
0 5 10 15
10−4
10−3
10−2
10−1
100
BER in fast fading channel
Eb/No
bit e
rror
rat
e
Rotated 16 QAMNon−Rotated 16QAMtheoretical AWGN
Figure 20: comparison BER with and without constellation rotation in Rayleigh channel
41
4 CONCLUSIONS
4 Conclusions
These days the DVB-T standard is more than twelve years old.
Progress in research and development offer today more powerful techniques and can en-
hance the digital terrestrial television.
The 32k OFDM e.g. was not possible to realize in hardware in 1993, but can now be
implemented in silicon devices.
So consequentially, if the enhanced techniques are thrown together, like a bigger OFDM
FFT or a higher order QAM, a successor to DVB-T with higher performance is thinkable.
In some parts DVB-T2 is a further development of DVB-T, with the added OFDM sizes
of 16k and 32k and the QAM of order 256; but T2 adds some techniques that are beyond
an enhancement of techniques that exist (in lower ability) in DVB-T. So T2 is in all
directions a bit smarter than DVB-T. Of course, this adds also to the complexity of the
system.
The weighing, why DVB-T2 should be introduced, instead using DVB-T further on, might
be different, depending on the preconditions. In the UK, where only the 2k mode of DVB-
T is deployed, the desire for a more powerful system is rather given, as in other countries,
like France, where the video coding with MPEG-4 already makes HDTV with DVB-T
possible.
The DVB project gives the broadcasters T2 to the hand for a time after complete analogue
switch off (ASO), and not an instant replace of DVB-T.3 Both service will probably coexist
for some time.
T2 is more bandwidth efficient than DVB-T and analogue TV, so that after ASO, the
released frequencies of analogue television can be used more efficient or released to other
competing services.
The higher bitrates offered by T2 can not only be used to tranmit HDTV, but also to
carry more different services per channel.
DVB-T2 offers many advantages on one hand to the end user, who benefits from higher
data rates, thus better service quality respectively a bigger number of services and a more
noise robust signal.
A energy saving mode for handheld or portable devices can be implemented, utilizing the
concept of the Physical Layer Pipes.
On the other hand T2 is also profitable for the service providers.
3http://www.dvb.org/technology/dvbt2/index.xml
42
4 CONCLUSIONS
T2 comprises some parts where costs can be reduced.
This is firstly the PAPR reduction techniques that are included into T2. The PAPR
reduction allows the service providers to dimension their high power transmit amplifiers
more efficiently and thus saves energy.
Secondly the higher bit rates allow to multiplex more services into one RF channel, so for
a given number of services transmitted with T2 less RF channels are necessary than in
DVB-T.
The 32k OFDM allows, owing to the longer symbol duration, a larger distance between
transmitters and hence nation-wide SFNetworks can be thought of.
Emphasized should be the high flexibility of T2. Where DVB-T offers only static settings
or a small number of choices, T2 can be configured in a wide range to adapt the service
parameters to the demands.
The different Pilot Patterns, OFDM size, Guard Intervall, forward error coding and mod-
ulation give a wide range of possibilities to address different types of receivers, and can
be adjusted to any scenario.
Also two services with the same content but varied configuration for e.g. fixed receivers
in high quality or handhelds in lower qualitiy but error robust reception are thinkable.
Though mobile receivers were not in the focus of development, the parameters can be
optimized for mobile reception. A high performance in mobile scenarios is expectable.
43
References References
References
[1] DVB: Frame structure channel coding and modulation for a second generation digital
terrestrial television broadcasting system (DVB-T2). Rev. 5. 2008-06
[2] ETSI: Digital Video Broadcasting (DVB) User guidelines for the second generation
system for Broadcasting, Interactive Services, News Gathering and other broadband
satellite applications (DVB-S2). V1.1.1. 2005-02
[3] ETSI: Digital Video Broadcasting (DVB); Framing structure, channel coding and
modulation for digital terrestrial television. V 1.6.1. 2008-09
[4] ETSI: Digital Video Broadcasting (DVB); Frame structure channel coding and modu-
lation for a second generation digital terrestrial television broadcasting system (DVB-
T2). V1.1.1. 2008-10
[5] Kuhn, Manfred: Der digitale terrestrische Rundfunk: Grundlagen, Systeme und
Netze. Heidelberg : Huthig, 2008
[6] Proakis, John G.: Digital communications. 4. ed., internat. ed. Boston : McGraw-
Hill, 2001
[7] Reimers, Ulrich: DVB - Digitale Fernsehtechnik: Datenkompression und Ubertragung
; mit 38 Tabellen. 3. Auflage. Berlin, Heidelberg : Springer-Verlag Berlin Heidelberg,
2008
44
List of Figures List of Figures
List of Figures
1 format of FEC frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 theoretical symbol error rates of QAM . . . . . . . . . . . . . . . . . . . . 7
3 Constellation Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Constellation Points in 64QAM . . . . . . . . . . . . . . . . . . . . . . . . 11
5 Iterative LLR-Demapper [4, p.152] . . . . . . . . . . . . . . . . . . . . . . 12
6 spectrum of OFDM in normal mode and extended mode . . . . . . . . . . 14
7 GI overhead reduction [4, p.23] . . . . . . . . . . . . . . . . . . . . . . . . 16
8 back-off between Eav and clip-off . . . . . . . . . . . . . . . . . . . . . . . 18
9 spectrum off an 8k OFDM-symbol with clipping . . . . . . . . . . . . . . . 19
10 PAPR CDF for different OFDM signals . . . . . . . . . . . . . . . . . . . . 20
11 BER for clipped OFDM signals . . . . . . . . . . . . . . . . . . . . . . . . 20
12 block diagram of the ACE algorithm from [1, p.99] . . . . . . . . . . . . . 22
13 QAM constellation after ACE . . . . . . . . . . . . . . . . . . . . . . . . . 23
14 bitrates in DVB-T2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
15 Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
16 Multipath channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
17 received signal strength in fading channel . . . . . . . . . . . . . . . . . . . 37
18 comparison BER of DVB-T and DVB-T2 . . . . . . . . . . . . . . . . . . . 38
19 comparison BER with and without constellation rotation . . . . . . . . . . 39
20 comparison BER with and without constellation rotation in Rayleigh channel 41
21 Matlab code of the demapper . . . . . . . . . . . . . . . . . . . . . . . . . 47
45
List of Tables List of Tables
List of Tables
1 code rates for the normal FEC frame . . . . . . . . . . . . . . . . . . . . . 6
2 code rates for the short FEC frame . . . . . . . . . . . . . . . . . . . . . . 6
3 code rates for LDPC and BCH . . . . . . . . . . . . . . . . . . . . . . . . . 6
4 rotation degrees defined by the T2 standard . . . . . . . . . . . . . . . . . 8
5 OFDM normal and extended mode . . . . . . . . . . . . . . . . . . . . . . 15
6 maximum echo travel distance . . . . . . . . . . . . . . . . . . . . . . . . . 16
7 Pilot Patterns in T2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
8 Tone Reservation carriers and overhead . . . . . . . . . . . . . . . . . . . . 23
9 theoretical bitrates in T2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
10 comparison of bitrates between DVB-T and T2 . . . . . . . . . . . . . . . 26
11 comparison between DVB-T and DVB-T2 . . . . . . . . . . . . . . . . . . 27
12 gain of different parts in DVB-T2 . . . . . . . . . . . . . . . . . . . . . . . 27
13 degradation in SNR for different sizes of QAM and OFDM FFT . . . . . . 34
46
List of Tables List of Tables
Appendix
Matlab Code of the Demapper:13.06.09 05:56 D:\Uni\BA\Matlab\BA\demap.m 1 of 1
function result = demap(cell,M, Nldpc, rot, rI, rQ) if nargin < 5 rhoI = ones(length(cell),1) rhoQ = ones(length(cell),1)end if (rot) cell = removeQDelay(cell, M, Nldpc)end numBits = log2(M) numCP = 2^(numBits-1) result = zeros(1,numBits) % return all possible 2^(m-1) constellation points if bit_j is zero and if% bit_j = 1x0 = []x1 = []for j=1:numBits x0 = [x0 returnConstellationPoints(M,j-1,0,rot)] x1 = [x1 returnConstellationPoints(M,j-1,1,rot)] j = jend for l = 1:length(cell)% LLR calculation for each bitfor k = 1:numBits sum0 = [] sum1 = [] for i = 1:numCP sum0 = [sum0 ((real(cell(l,1)) - rhoI(l,1)*real(x0(k,i)))^2 + (imag(cell(l,1)) -rhoQ(l,1)*imag(x0(k,i)))^2)] sum1 = [sum1 ((real(cell(l,1)) - rhoI(l,1)*real(x1(k,i)))^2 + (imag(cell(l,1)) -rhoQ(l,1)*imag(x1(k,i)))^2)] i = i end % if LLR > 0 bit is a 1 else bit is a 0 llr = min(sum0) - min(sum1) if llr >= 0 result(l,k) = else result(l,k) = end k = kend l = lend
Figure 21: Matlab code of the demapper
47
Top Related