Hardware Considerations for Digital Audio Broadcasting System

Nariman Moezzi-Madani, Ehsan Rohani and S. Mehdi Fakhraie {n.moezi, e.rohani})@ece.ut.ac.ir, [email protected] Silicon Intelligence and VLSI Signal Processing Laboratory ECE Departrnent, University of Tehran, Tehran, IRAN. Abstract-In recent years, several implementations have been reported for Digital Audio Broadcasting (DAB) systems. Normally, implementation parameters of these systems are extracted from extensive system level simulations to adjust various parameters while maintaining the required performance. In this paper, the bit-true model of a DAB system is extracted and an accurate model simulation for the system is performed to find the word lengths of various parameters to approach the best trade off between performance and hardware cost. Here, the decimation-in-time algorithm for FFT/IFFT and adaptive LMS algorithm for time equalizer is adopted.

(ISI). The generated OFDM signal is amplified and sent to the RF block. The transmission frame consists of a sequence of three groups of OFDM symbols: synchronization channel symbols, Fast Information Channel (FIC) symbols and Main Service Channel (MSC) symbols. The synchronization channel symbols are comprised of the null symbol and the phase reference symbol [1]. The DAB system operates in 4 modes, and each mode has its specific characteristics, as shown in table I.

Index Terms-Digital Audio Broadcasting (DAB), Bit-true, Word-length, OFDM, Equalizer. 1. INTRODUCTION The Digital Audio Broadcasting (DAB) system, described in the European Eureka-147 standard [1], offers high-quality audio services, supports multimedia data to mobile reception and is likely to replace the traditional radio systems. Fig. I shows the block diagram of a conceptual DAB signal transmitter. The source encoder for the DAB system is the MPEG Audio Layer II encoder with restrictions on some parameters and some additional protection against transmission errors. The MPEG 11 audio signal and the other data are the input services to DAB transmitter. Each service signal is coded individually at source level in the transmitter, error protected, and then time interleaved in the channel coder. Each service is independently error protected with a coding overhead ranging from about 25% to 300%, the amount of which depends on the requirements of the broadcasters (transmitter coverage and reception quality).Then the services are multiplexed in the Main Service Channel (MSC), according to a predetermined and adjustable, multiplex configuration. Finally, Orthogonal Frequency Division Multiplexing (OFDM) is applied to the signal to shape the DAB signal which consists of a large number of sub-carriers [1]. The OFDM signal generation involves the processes of Quadrature Phase Shift Keying (QPSK) mapping of arrival signals, frequency interleaving, and differentially modulating. When the signal is generated, it is defined in the frequency domain, and is then transformed into its time domain representation by Inverse Fast Fourier Transformer (IFFT). The last complex output values of the IFFT are copied to the front of the symbol to add the guard interval (cyclic prefix). The guard interval is used to avoid inter symbol interference

mH

b--

mm

MU

mim
IV 48ms 648us 123us 48km 1.5GHz 500us 623us 768

Fig. 1: Conceptual block diagram of DAB transmitter [I]. TABLE I Different operation modes for DAB system [1]. II III I System Parameter 24ms 24ms 96 ms Frame Duration 168us 324us 1297us Null Symbol Duration

Guard Interval Duration Transmitter Separation Frequency Range Useful Sym. Duration Total Symbol Duration
No. of Sub-Carriers

246us 96km 375MHz Ims 1246us 1536

312us 384

62us 24km 1.5GHz 250us

3lus 12km 3GHz 125us 156us

As shown in Table I, DAB system works in three frequency modes: 375 MHz, 1.5 and 3 GHz. For the nominal frequency ranges, the transmission modes have been designed to suffer neither from Doppler spread nor from delay spread, while both of these exist in mobile reception with multipath propagation. In our simulations, we have considered mode I that system works in 375 MHz frequency band. This mode is most suitable for a terrestrial Single-Frequency Network in the VHF range, because it allows the greatest transmitter separation (96 Km) [1]. The spectrum of the signal is approximately rectangular, Gaussian noise-like, and occupies a bandwidth of 1.536 MHz.

0-7803-9262-0/051$20.00

92005 IEEE

209

Fig. 2: DAB transmitter, receiver and channel system-level diagram.

There are papers that include simulation of DAB system [2], [3], however there is no reported accurate model that considers the effects of limited word-length of the implemented hardware on the performance. In the proposed work, we have modeled and simulated all parts of the DAB system, but MPEG 2 audio coder and encoder. The bit-true model of this system was then extracted, and the simulations were repeated with different word-lengths of the system parameters. In some blocks of the receiver, the word-lengths of the blocks have a great effect on the overall power and area of the module, so finding the best word-lengths for modules help us to design a more efficient and compact system.
11. CHANNEL MODELING

the effects of Doppler shift. In Table II, we have shown the parameters of multipath channel used in the model [6]. The model was simulated for three vehicle speeds: 60, 120 and 180 km/h. The results are shown in Figures 4-6.
III. EQUALIZER AND FFT ALGORITHMS

In an OFDM system data is transmitted at a low symbol rate using many parallel narrow-band carriers rather than at a high rate using a single wide-band carrier. The bit rate for each carrier is inversely proportional to the OFDM symbol duration. A lower bit rate means that the received data suffers less from Inter-Symbol Interference (ISI) in the presence of multipath propagation. Consequently, OFDM is less sensitive to this type of propagation than a wide-band single carrier system [4]. By adding a guard interval between successive symbols, as it is done in DAB, the effect of ISI can be completely eliminated, as long as the delay spread of the received multi-path signal does not exceed the duration of the guard interval [5]. So, in the channel model, the delay time of the received signals should not exceed the guard interval time. In a wireless system, a signal transmitted into the channel interacts with the environment in a very complex way. There are reflections from large objects, diffraction of the electromagnetic waves around objects, and signal scattering. The result of these interactions is the presence of many signal components, or multipath signals at the receiver. Another property of wireless channels is the existence of doppler shift, which is caused by the motion of the receiver, the transmitter, or any other objects in the channel [4]. The DAB signal is affected by multipath and Doppler shift in addition to AWGN (Additive White Gaussian Noise). The implemented channel model includes multipath fading with a LOS (Line of Sight) component as well as the additive Gaussian noise considering

A. Equalizer In the model for simulation we selected an adaptive LMS time equalizer. The structure of this equalizer is shown in Fig. 3. The only parameters that we have to set are number of taps and step-size of the equalizer. To reduce the power and complexity of the circuit in the design, step-size was selected negative power of two, so we can have a simple shift instead of one multiplication. By using this equalizer, the required BER defined by standard can be obtained. The required BER for this system is about 0.0001 after forward error correction.

Fig. 3: Adaptive LMS equalizer TABLE 11 Channel Characteristics

Path
2
3
I

T(us)
0

Gain(dB)
0

30
80

-3
-3

210

B. FFTAlgorithm The algorithm used in the IFFT block of the transmitter and FFT block of the receiver is decimation-in-time radix-2 FFT. Of the several important FFT algorithms which have been developed for efficient computation of the discrete Fourier transform (DFT), the most widely used is the Radix-2 FFT, where the transform length (N) may be any positive integer power of 2 [7]. We have used decimation-in-time radix-2 FFT algorithm because this algorithm has better finite word length properties rather than the decimation in frequency algorithm [8]. As the simulation mode I is selected among other modes for this investigation, 2048 point FFT and IFFT are required.
IV. B1T-TRUE MODEL SIMULATION

mode 1; where the average code rate of approximately 0.34.

the system is

V. CONCLUSION In this paper, an accurate model simulation of the DAB system was performned, using Simulink. In this model, fixedpoint effects of the system components were taken into account. The simulations show that how the word-length of the components can affect the performance of the system. Using appropriate word-length for the system, there is an efficient tradeoff between performance and the cost of the implementation of the system.

To make a circuit or system ready for implementation, all the words, accumulators and coefficients must be defined with a specific word length. Basically, two strategies are employed to implement the DAB receiver: the DSP-based architectures [9], [10] and the ASIC-based implementation [11], [12]. The forner has the characteristics of maximum flexibility, ease of use, and simple programming, but it can provide only limited processing capability. On the contrary, the ASIC-based implementation has the potential of supporting real-time symbol decoding and low-cost implementation [13]. In both cases, designers use either fixed point or floating point representations. Floating point representation has the disadvantage of higher complexity and so more power consumption. To lower hardware costs, most implementations of digital systems rely on binary fixed-point (FP) number system using either two's complement or unsigned-magnitude with round-off and truncation quantization [14]. Therefore before starting hardware implementation, the effects of quantization error and the limited bit-width error must be taken into account. In our system, the more sensitive blocks to artifacts of fixed point design are equalizer, FFT, IFFT and QPSK mapper. We have used two's complement representation for arithmetic operations, saturation for overflow mode, and truncation for rounding mode. The results of the BER simulation of the system are shown in Figs. 3 and 4 versus SNR of the channel and with different word-lengths of the system variables. In system design we want to specify word lengths that provide a good trade off between BER performance and hardware complexity. Considering all of the states of the different wordlengths needs a huge space and enormous simulation time. Therefore the word-length of the Equalizer and FFT are considered equal, and the QPSK mapper output is quantized with 16 bits word-length. In the attached figures, the difference between system BER of different word-lengths of the equalizer and FFT block and the floating-point representation are shown. As seen, SNR variation does not have strong effect on BER of the floating point model, but is effective on BER of the fixed-point mode. The diagrams of the coded data are obtained taking into account the effect of forward error correction and the un-coded diagrams are before the operation of error correction. The simulations were performed for the protection level one of the standard and

ACKNOWLEDGMENT The authors wish to express their thanks to Iran Telecommunication Research Center (ITRC) for the financial support during the course of this research.
REFERENCES

[1l ETS 300 401. Radio Br-oadcasting System: Digital Auidio

Broadcasting (DAB) to Mobile, Portable and Fixed Receivers. ETSI. Td edition. May 1997. [2J R. D. Gaudenzi and F. Gianneti "Analysis of an advanced satellite digital audio broadcasting system and complementary terrestrial gap-filler single frequency network," IEEE Trcansactions on Vehiclulcar Technology, vol. 43, no. 2, May 1994. [3] M. Chrysochoos, Y. Guo and J. Kim "Performance of an in-band adjacent-channel DAB systemn for frequency-selective rayleigh and ricean slow fading channels under analog FM interference," IEEE

Transactions on Broadcasting, vol. 44, no. 2, June 1998. [4] Theodore S. Rappaport, etal. "Overview of spatial channel models
[5]

[6]
[7]
[8]

[9]

[101

[HI]
[12]

[13]
[141

for antenna array communication systems," IEEE Personal Communications, Feb. 1998. ETSI TR 101 496-3, Digital Auclio Broadcasting {D.4B): Guidelines and Rules for lIn?olementation and Operation: Part 3: Broadcast network. ETSI. ITd edition May 2001. G. Zimmermann, M. Rosenberger and S. Dostert "Theoretical bit error rate for uncoded and coded data transmission in digital audio broadcasting," IEEE Im1t. Con. Communication 1996 (ICC 96), vol. 1, pp. 297-301, June 1996. 1. R. Mactaggart and M. A. Jack "A single chip radix-2 FFT butterfly architecture using parallel data distributed arithmetic," IEEE Journal of Solid State Circuits, vol. 19, no. 3, June 1984, RB. Perlow and T.C. Denk "Finite wordlength design for VLSI FFT processors," IEEE Con?f Signals, Systems and Computers 2(001, vol. 2, pp. 1227-1231, Nov. 2001. J. A. Husiken, F V. Laar. A. Delarueele. and N. .J. L. Philips, "Specification, partitioning and design of a DAB channel decoder," VLSI Signal Processing, vol. 6, pp. 21-29, Oct. 1993. M. Bolle. D. Clawin. K. Gieske, F. Hofmann, T. Mlasko. M. J. Ruft G. Spreitz "The receiver engine chip-set for digital audio broadcasting." in Proc. 1mit. Synp. Signals, Systems and Electronics, pp. 338-342, 1998. A. Delaruelle, J. Iluisken, J. V. Loon and F. Welten, "A chip set for a digital audio broadcasting channel decoder," in Proc. IEEE Custom Integrated Circtit ConJference, pp. 13.4.1-13.4.4. 1995. Delaruelle. J. Huisken. J. van Loon. F. Welten, "A channel demodulator IC for digital audio broadcasting." in Proc. IEEE Custom Integrated Circuits Conference, pp. 47-50. 1994. C.M. Wu et al. "Implementation of channel demodulator for DAB system," IEEE Int. Symp. C'ircuits and Svstems 2003, vol. 2, pp. 137-140, May 2003. M. A. Cantin, Y. Savaria, and P. Lavoie, "A comparison of automatic word length optimization procedures," IEEE Int. Symp. Circuits and Systems, 2002, vol. 2, pp. 612 -615.

211

.-I

BER for floating piont repesentation

A :~ --

IOJ i

I
lo'

EEE-E--_EE-EEEE-E-EE-----

--___

_____

E!EEE-!E

EEEEEEEEEEDE

I0
a

.=.=

T=_=====_==,======10

11

SNR

12

13

14

15 1R

Fig. 4: Uncoded and coded BER vs. SNR, at v = 30 km/h for floating point model.
______---_- _

BEP forUocoded data

______
14

lop
It! IC'

BER for coded data

--------------------W- 10
12

_________

______ ________~~v

10
10

EEEESSSEEESEE
!mEmeemmm

o10'

-------___
-______o

u. 10

---------

10
109

_-__-_ -_ -__-__-_-_
2t-,
0

10

11 12 SNR

13

14

15

-A ,I,. 10- 6 8 7 9

t 011

SNR

12

13

14

15

16

Fig. 5: Uncoded and coded BER vs. SNR, at v = 30 km/h. (The numbers in the legend are word-length ofthe equalizer and FFT/IFFT blocks.)
10

BER for Urmoded data

BER for coded data

-1.
-13--- 14

--10~'

Ej
15
16

leI
tO.

EE-

H\~~~~

--

-101-

lo'

le0*

=\---

- -

--

__

---_-_-------->_sE-------9 10
11

-___

10

11

12 SNR

13

14

10

12 SNR

13

14

1s

16

Fig. 6: Uncoded and coded data BER vs. SNR, v = 60 km/h.

Io10

BER for coded data

- B-

'BBB2

--

- 1

p-------------------1+a1

It.

<

=====================-+- 140
-====================q--41 12

---------------------

it, 1

Crelmss;E-- --l,

--------8 9 10

F______

--

------D

le

it-a

---------------------_
9

10

11

12

13

14

SNR

15

16

11

SNR

12

13

14

15

16

Fig. 7: Uncoded and coded data BER vs. SNR, v = 90 km/h.

212