Chapter 2 OFDM and DVB-T/H Technology
2.2 DVB-T/H Technology
2.2.5 DVB-H Particular
For portable devices, time slicing technology is employed to reduce power consumption. For the purpose to improve the system performance in mobile environment, forward error correction for multiprotocol encapsulated data (MPE-FEC) is adopted with powerful channeling and time interleaving. Since 8K mode has better performance in large single frequency network (SFN) but worse in against Doppler Effect and 2K mode has better performance in against Doppler Effect but not suitable for large SFN, a comprised 4k mode is proposed. Overall, the specification of DVB-T/H is listed in Table 2-3.
Table 2-3 Specification of DVB-T/H Transmission mode 2K, 4K, 8K
Number of useful sub-carriers 1705, 3409, 6817 Number of continual pilots 45, 89, 177 Number of scattered pilots 141, 282, 564 Number of TPS pilots 17, 34, 68 Radio frequency (MHz) 45~860
Guard interval 1/4, 1/8, 1/16, 1/32 Bandwidth (MHz) 5, 6, 7, 8
Elementary period (us) 7/40, 7/48, 7/56, 7/64
Channel model Rayleigh, Ricean
Forward error correct Convolution code with puncturing Reed Solomon Code (204,188)
Constellation QPSK, 16QAM, 64QAM, non-uniform
16QAM, non-uniform 16QAM Required BER 2 X 10-4 after Viterbi decoder
Quasi Error Free after Reed Solomon
Chapter 3
Symbol Synchronization Algorithms
Fig. 3.1 illustrates the block diagram of DVB-T baseband inner receiver and all synchronization processes in digital domain. The block diagram contains mode/GI &
symbol boundary detection, carrier synchronization loop, sampling synchronization loop, frequency domain channel estimation/compensation and hard demapper. This thesis will design the architecture based on this block diagram and focuses on the highlighted blocks.
Fig. 3.1 Block diagram of DVB-T baseband inner receiver
Timing synchronization plays an important role in digital communication systems. Without accurate timing synchronization process, the systems will fail to work in the beginning or get unreliable outcome. Timing synchronization includes mode/GI detection, coarse symbol synchronization (CSS), scattered pilot synchronization (SPS), carrier frequency offset (CFO) and sampling clock offset (SCO) issues. The last two topics have been discussed in [14]. This chapter focuses on mode/GI detection and coarse symbol synchronization problems. These two jobs should be down first in receiver. The FFT window has to decide the window length and locations with the correct transmission mode, guard interval length and symbol
boundary information. Otherwise, the feedback loops will have no idea to recover CFO and SCO and so do those parts behind inner receiver.
The goal of mode/GI detection is to get precise mode/GI parameters of transmitted symbols. By using the detected transmission mode, the system is able to set a FFT window, which length equals to transmission mode. Transmission mode and guard interval parameters make the system being able to compute boundaries behind the first boundary detected by coarse symbol synchronization. After mode/GI detection, coarse symbol synchronization process will adopt the parameters from mode/GI detection to do more accurate symbol boundary detection for the purpose to reduce timing offset effects and avoid ISI effect. Moreover, the results from mode/GI detection and coarse symbol synchronization make the system have enough information to determine the FFT window locations.
3.1 Mode/GI Detection
It seems that the system can get the information of transmission mode and guard interval length from TPS pilots discussed in 2.2.4. But without these messages, how can the systems set a correct FFT window length and correct FFT window locations?
That means FFT has no idea to start to work and leads to no TPS information. This phenomenon will become a vicious cycle and system will never start to work. In order to make the FFT start to work, it’s necessary to do blind mode/GI detection before other processes of inner receiver.
3.1.1 Introduction to Mode/GI Detection
Mode/GI detection algorithms are usually similar to coarse symbol synchronization theorems. There are many methods to detect the mode/GI. For example, [16] proposed a blind transmission mode detection process, [17] modified a
mode/GI jointed detection process based on [16] and a two-stage mode/GI detection is discussed in [18].
a) Mode Detection
The basic idea of mode detection is to use the characteristic of the inserted guard intervals, a copy of tail in OFDM symbols. After surviving from fading channel, the guard interval part will still have a high correlation with symbol’s tail. Fortunately, other parts will have low or even no correlation with guard interval. Thus, for a 2K transmission mode, the mode detector shall find the correlation of r(n) and r(n-2K), where r(n) is the nth received signal. For the purpose to ensure the correlation result, to accumulate the correlation results between r(n)×r(n-2K) and r(n+64)×r(n-2K+64) is necessary. Therefore, 2K+2×64, 4K+2×64 and 8K+2×64 long moving windows are required to detect the 2K/4K/8K transmission modes for DVB-T/H mode detection.
Fig. 3.2 illustrates a simple diagram of the correlation results of 2K/4K/8K mode detection windows under 8K transmitted symbols. As the figure shows, only the 8K mode window is possibly located at the region of guard interval and symbol’s tail at the same time and will have prominent correlation results. Eqn. (3.1) presents the computation required in Fig. 3.2 mathematically.
guard
interval symbol (n) tail
2K correlation 4K correlation 8K correlation
guard interval
symbol (n-1) symbol (n+1)
8192 4096
2048
× 8K moving window
4K moving window 2K moving window
×
×
t
Fig. 3.2 2K/4K/8K correlation under 8K mode
∑
=where N is the delay-line length which value will be 2K, 4K or 8K according to different transmission mode, r(n) is the received nth signal. Eqn. (3.2) is a modified version of Eqn. (3.1). By using an integration length equals to the minimum guard interval length of each tested transmission mode, the detector is able to have a reliable correlation result.
∑
−=
−
−
×
−
=
32 1
0
*( ) ( )
) (
N
i
N i n r i n r n
x (3.2)
Fig. 3.3 (a) and (b) show the different correlation results under different transmission modes with 1/4 guard interval, 12dB SNR, CFO=23.33 and surviving from Rayleigh channel. In Fig. 3.3 (a) only 2K correlation results have apparent plateaus and Fig. 3.3 (b) has the same situation for 8K correlation results.
0 1000 2000 3000 4000 5000
0 2 4 6 8 10 12
Amplitude
Sample Index
2K Correlation 4K Correlation 8K Correlation 2K Correlation
4K/8K Correlation
(a)
0 0.5 1 1.5 2
8K Correlation 8K Correlation
2K/4K Correlation
(b)
Fig. 3.3 2K/4K/8K correlation results under (a) 2K (b) 8K transmission mode Even though Eqn. (3.1) or Eqn. (3.2) have the ability to distinguish the transmission mode, there is still some aliasing peaks occurred due to channel noise and may possibly lead to a wrong detected mode. For example, in Fig. 3.3 (a), there is an aliasing peak near sample index 3000 of 2K correlation results which is almost as high as the lowest plateau value near to sample index 2000 for 2K. Thus it is a problem to decide the peak or plateau threshold since there is no flat plateau and unitary plateau value.
To eliminate the effect from channel noises and ease the decision of the threshold, [17] used a normalized method to detect the transmission mode shown in Eqn. (3.3).
∑
The correlation result will be normalized to 1 theoretically by dividing the power
term. Fig. 3.4 (a) and (b) are the results of Eqn. (3.3), using the same pattern with Fig.
3.3. The plateau is ideally close to “1” and no other aliasing correlation results are higher than “0.707”. The characteristic of the normalized flat plateau can be also used to calculate the guard interval length and it will be discussed later.
0 1000 2000 3000 4000 5000
0 0.2 0.4 0.6 0.8 1
Amplitude
Sample Index 2K Correlation
4K Correlation 8K Correlation
2K Correlation
4K/8K Correlation
(a)
0 0.5 1 1.5 2
x 104 0
0.2 0.4 0.6 0.8 1
Amplitude
Sample Index 2K Correlation
4K Correlation 8K Correlation
8K Correlation
2K/4K Correlation
(b)
Fig. 3.4 2K/4K/8K normalized correlation results under (a) 2K (b) 8K mode
b) Guard Interval Length Detection
The guard interval length will be detected after transmission mode. With an incorrect guard interval length, FFT won’t get correct patterns and sub-carriers after FFT won’t be the same with transmitted. Fig. 3.5 illustrates the situations of incorrect guard interval length. In Fig. 3.5 (a), a smaller guard interval length is detected and the second FFT window has a probability to get signals form the ISI destroyed region.
The third FFT window gets some signals from the correct symbol, some from ISI destroyed region and others from previous symbol, this will lose the orthogonality of OFDM symbols and FFT outputs are absolutely incorrect. FFT windows in Fig. 3.5 (b) also get incorrect signals either.
(a)
(b)
Fig. 3.5 Detected guard interval length is (a) smaller (b) larger than transmitted [18] used the minimum guard interval length, which is 1/32 of transmission mode, and accumulate the results using different delay-line length windows, 1/32, 1/16, 1/8 and 1/4. Only the correct guard interval mode will have a maximum peak after different length accumulation. [17] adopted an simple and less delay-line method to calculate the guard interval length. This method just calculates the length of plateau period and that will approximate to “guard interval length subtracts mode/32”.
3.1.2 Proposed Mode/GI Scheme
The normalized method seems very easy to realize and compute the guard
interval length, but a divider wastes large power and area. Two-stage mode/GI detection proposed in [18] needs extra delay-lines which size will be from 2K to 8K.
This is another penalty. As a result, a modified method based on normalized mode/GI detection is proposed in this thesis. Since a threshold value is determined to define the plateau region, it implies that all x(n) which are bigger than the pre-defined threshold belongs to the plateau. Therefore, two key observations below can be found:
1) If the plateaus exist, the transmission mode will be the same with the tested mode.
2) The period of the plateau represents the guard interval length.
Using the two key observations above, if x(n) is bigger than the threshold means x(n) belongs to the plateau region. Now the threshold is defined as 0.707 and the derivation of using a subtractor to replace the divider is shown in Eqn. (3.4).
0
First, move the denominator of the left term to the right term. Thus, a divider is replaced by a multiplier. Second, as the result of calculating the absolute value for complex numbers is too complicated to implement, squaring both sides of the equation is used to replace the absolute value calculating. In fact as Fig. 3.6 shows, modified from Fig. 3.4, square operation eliminates the noises. Then move the right term to left, this action makes the comparator replaced by a subtractor. Finally, the square of threshold becomes “0.5” that means only a bit shift instead of a multiplier is able to accomplish this job. Overall, the transmission mode can be tested by observing whether the “result >=0” and guard interval length can be detected by computing the period of the “result >=0”.
0 1000 2000 3000 4000 5000
0 0.2 0.4 0.6 0.8 1
Amplitude
Sample Index 2K Correlation
4K Correlation 8K Correlation
2K Correlation
4K/8K Correlation
(a)
0 0.5 1 1.5 2 x 104 0
0.2 0.4 0.6 0.8 1
Amplitude
Sample Index 2K Correlation
4K Correlation
8K Correlation 8K Correlation
2K/4K Correlation
(b)
Fig. 3.6 2K/4K/8K squared normalized correlation under (a) 2K (b) 8K mode
3.1.3 Performance Simulation
Fig. 3.7 illustrates the error rate under different threshold of the proposed mode/GI detection. The simulation environment is 1000 2K transmission mode symbols with 1/4 guard interval, 12dB SNR, 23.33 sub-carriers CFO and surviving from Rayleigh channel. As the simulation shows, the error rate of the proposed mode/GI detection method under the threshold 0.5 to 0.8 is zero. That means the pre-defined threshold value 0.707 locates at the reliable region. Because of noise and channel effect, the normalized correlation results are closing to 0.9 instead of 1.
Therefore, the reliability decreases while the threshold is defined close to 0.9. For threshold smaller then 0.5 cases, the noise will influence the detection result.
0.4 0.5 0.6 0.7 0.8 0.9 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Error Rate (%)
Threshold
Fig. 3.7 Error rate under different threshold of proposed mode/GI detection
3.2 Coarse Symbol Synchronization
Coarse symbol synchronization (CSS) is also named symbol boundary detection.
The goal of symbol boundary detection is to find out boundaries of transmitted symbols. Coarse symbol synchronization starts to work after finishing the mode/GI detection. With the information of transmission mode and guard interval length, symbol boundary detection has enough information to detect boundaries. After the first symbol boundary is detected, the system will use transmission mode, guard interval length and boundary location to derive the successive symbol boundaries.
In order to make the FFT windows locate on correct locations, symbol boundary detection needs to solve some problems. Since DVB-T signals are transmitted in SFN (Single Frequency Network), for a receiver there will have many signals from different transmitters with discordant delay time as Fig. 3.8 illustrates. The multipath effect leads to the same symbol with different delays overlaps and hard to detect the correct boundary of main path. Channel noises also reduce the reliability of coarse
symbol synchronization algorithms. The solution will be discussed in section.
≈≈≈≈
Fig. 3.8 Effect of multipath fading
3.2.1 Effect of Symbol Timing Offset
Before starting to introduce to coarse symbol synchronization algorithms, the effect of symbol timing offset must been derived. Symbol timing offset means the detected boundary does not locate on the true symbol boundary. This phenomenon is due to multipath effect, channel noise and aliasing. Two possible cases will occur, later or earlier than the true symbol boundary. Thanks to cyclic prefix, an earlier boundary location only causes a phase rotation which is able to be compensated by frequency domain equalization. The effect of the offset is derived in Eqn. (3.5).
kN
where X(k) is the respected result of FFT output andεis the number of offset samples.
The case of earlier offset only leads to a phase rotation. But the derivation above only works whenεis not too large to make FFT window locates on ISI region. Otherwise, just like the later case, FFT window will get signals from previous symbol which has no orthogonality with the current symbol. Fig. 3.9 represents the sub-carriers after FFT by early and late cases in constellation map. The early case leads to a phase rotation in constellation map while the late case leads to mix-up in constellation map.
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
IM(z)
Re(z)
(a)
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
IM(z)
Re(z)
(b)
Fig. 3.9 Sub-carriers of (a) early case and (b) late case for timing offset
3.2.2 Coarse Symbol Synchronization Algorithms
Coarse symbol synchronization aims at finding a rough symbol boundary. As a result of DVB-T system uses broadcasting technique to transmit signals in air, maximum-likelihood algorithms are not suit for DVB-T. Thanks to cyclic prefix again, like mode/GI detection, many coarse symbol synchronization algorithms are based on using the correlations between guard interval and symbol’s tail. There are three common algorithms, which are maximum correlation (MC) [19], normalized maximum correlation (NMC) [17] and minimum mean square error (MMSE) in [20].
In the following will compare their performance.
a) Maximum Correlation (MC)
)
Similar to mode/GI detection, maximum correlation algorithm uses guard interval length detected by mode/GI detection as Ng to detect the maximum peak value. Unlike mode/GI detection, there will only be a maximum peak exist in stead of a plateau. Ideally, when the moving sum window goes into guard interval region, correlation will start to grow up. When the moving sum window exactly fits the whole guard interval region, a maximum peak occurs. Correlation value will decrease when the moving sum window start to leave the guard interval region.
b) Normalized Maximum Correlation (NMC)
)
Normalized maximum correlation algorithm is similar to maximum correlation algorithm. By dividing its own power term, the peak is normalized to “1” ideally. A normalized peak makes the threshold easy to define and that’s why the modified normalized maximum correlation is adopted as mode/GI detection algorithm. But a penalty of an extra moving sum delay-line for power term and a divider are the disadvantages.
c) Minimum Mean Square Error (MMSE)
( )
⎟⎟⎠
⎞
⎜⎜
⎝
⎛ − × − − − − + − −
=
∑ ∑
−=
−
=
1 0
2 1 2
0
* ( ) ( )
2 ) 1 (
) ( max
arg g g
N
i N
n i
est r n i r n i N r n i r n i N
K (3.8)
This algorithm is similar to ML algorithm in [21], but the received signals replace the look-up table. It also likes a change of normalized maximum correlation algorithm. Since the maximum value of NMC is close to “1”, that means the numerator substrates the denominator will close to “0”. The characteristic of plateau, which is close to “0”, also can be used in mode/GI detection. But without a normalized threshold, MMSE is hard to tell from the plateau of 8K mode from bottom of other modes. As shown in Fig. 3.10 (a) and (b), the plateau value of 8K mode is closed to the bottom of 2K mode. Fig. 3.11 (a) and (b) show the boundary detection results of MMSE algorithm and the peak characteristic is similar to MC and NMC.
The patterns of Fig. 3.10 and Fig. 3.11 are the same with Fig. 3.3.
0 1000 2000 3000 4000 5000
8K MMSE 2K MMSE
8K MMSE
8K MMSE 8K MMSE
2K MMSE
(a) (b) Fig. 3.10 MMSE mode/GI detection under (a) 2K (b) 8K mode
0 1000 2000 3000 4000 5000
-60
8K MMSE 2K MMSE
8K MMSE
8K MMSE 8K MMSE
2K MMSE
(a) (b) Fig. 3.11 MMSE boundary detection under (a) 2K (b) 8K mode
As a result, NMC is adopted as mode/GI detection algorithm in this thesis because of the normalized plateau characteristic.
3.2.3 Performance Simulation and Comparisons
a) Accuracy Simulation
The symbol boundary location accuracy of MC, NMC and MMSE is simulated to compare the performance. The transmitted 1000 symbols will survive from two kinds of channels, Rayleigh and Ricean, with CFO equals to 23.33 sub-carriers, 2K transmission mode, 1/4 guard interval length, 50Hz Doppler spread and 12dB SNR.
As Fig. 3.12, Fig. 3.13 and Table 3-1 shows, MMSE owns the best accuracy. The accuracy of MC and NMC are closely.
-150 -10 -5 0 5 10 15 20
Boundary Offset (Samples)
-150 -10 -5 0 5 10 15 20
Boundary Offset (Samples)
(a) (b)
Boundary Offset (Samples)
(c)
Fig. 3.12 (a) MC, (b) NMC and (c) MMSE boundary offset distribution @ Ricean
-10 0 10 20 30 40
Boundary Offset (Samples)
-10 0 10 20 30 40
Boundary Offset (Samples)
(a) (b)
-10 0 10 20 30 40 0
1 2 3 4 5 6 7 8 9
Percentage (%)
Boundary Offset (Samples)
(c)
Fig. 3.13 (a) MC, (b) NMC and (c) MMSE boundary offset distribution @ Rayleigh Table 3-1 MC, NMC and MMSE boundary average/peak offset @ Ricean/Rayleigh
Name MC
(Average/Peak Offsets)
NMC
(Average/Peak Offsets)
MMSE
(Average/Peak Offsets)
Ricean 0.8051/18 0.7768/11 0.6020/7
Rayleigh 11.7687/36 11.4111/41 10.8273/29
b) Hardware Complexity
For 2K/4K/8K applications in DVB-T/H, all of the algorithms above must have at least an 8K long correlation delay-line, which is used to store the r(n-8K) signal and also can be used to store r(n-2K) and r(n-4K) signals, and a 2K (a quarter of 8K) long moving sum delay-line, which is used to integrate the 2K moving sum correlation values for 8K mode with 1/4 GI length, as Fig. 3.14 illustrates. Due to the correlation expression, a complex multiplier is necessary. Then two pairs of adders and subtractors, one for real part and another for image part, are used to do moving sum calculation. Finally, an additional complex multiplier is required to do the square operation. For NMC, an extra 2K moving sum delay-line, complex multiplier, adder and subtractor are needed for its power term denominator. The square operation of power term only needs a multiplier as a result of the power term is integer. Further more, a divider is needed for normalization operation. MMSE needs an extra complex
multiplier and adder comparing to the NMC denominator components and the divider is replaced by a subtractor. In summary, Table 3-2 lists the components required.
(a)
(b)
(c)
Fig. 3.14 Block diagram of (a) MC (b) NMC and (c) MMSE Table 3-2 Components required for MC, NMC and MMSE Name Correlation/Moving Sum
Delay-Line
Complex Multiplier
Adder/
Subtractor
Multiplier/
Divider
MC 8K/2K 2 2/2 0/0
NMC 8K/2K×2 3 3/3 1/1
MMSE 8K/2K×2 4 4/4 1/0
The reason why MC is adopted as symbol boundary detection algorithm in this thesis is because of the performance is not differ too much to NMC and MMSE, but it needs the fewest components and no divider.
3.2.4 Proposed Mode/GI and Symbol Boundary Detection Scheme
By observing the mode/GI and boundary detection expression, it is easy to find out their structures are very similar. The mode/GI detection block diagram is illustrated in Fig. 3.15. The architecture is modified from NMC architecture by using the subtraction to replace the division as Eqn. (3.4) derived. As a result, mode/GI and boundary detection is able to share the same hardware to detect mode/GI and symbol boundary by controlling the correlation and moving sum delay-line lengths. After mode/GI detection, controller changes the correlation and moving sum delay-line lengths for symbol boundary detection.
Fig. 3.15 Block diagram of mode/GI detection
Fig. 3.16 illustrates the proposed mode/GI and boundary detection hybrid architecture. In the correlation part, the functions of 2K, 4K and 8K delays for different transmission modes are realized by a 2K/4K/8K triple modes reconfigurable delay-line and defined as correlation delay-line in this architecture. For four guard interval lengths of three transmission modes, there are totally six possible moving sum lengths and realized by two 2K/1K/512/256/128/64 reconfigurable delay-line, which is also named the moving sum (MS) delay-line.
Correlator
[ ]*
FFT
Gate
Gate
D
| |2 r(n)
Moving Sum 2K/4K/8K
delay
2K/1K/512/256 /128/64 delay
MC2
Threshold
to FCFO
×
× D
CSS Controller FFT Out
| |2 2K/1K/512/256
/128/64 delay
Fig. 3.16 Architecture of mode/GI and boundary detection scheme
The proposed mode/GI and CSS scheme can be divided into three stages, the
The proposed mode/GI and CSS scheme can be divided into three stages, the