Sei sulla pagina 1di 7

Options for creating a DAB receiver using Global

Silicon X-Core DSP and custom acceleration logic


By Morgan Colmer, July 2000, Global Silicon R&D Labs, Cambridge, UK

1. Introduction means that the US has developed a rival system


rather than adopt anything 'foreign' as standard.
1.1 Basic DAB requirements The obvious choice of name for such a, US-
For a number of years, the radio broadcasting only, standard is "WorldSpace". Eureka-147
community has been researching a digital and WorldSpace are technically as different as
broadcast method to overcome the problems can be imagined, removing any [commercially
that analogue systems suffer from, in particular, viable] possibility of dual system receivers.
multipath interference, fading, Doppler effects,
and poor spectrum utilisation. In addition to WorldSpace does have its advantages. The data
overcoming the technical failings of encoding is not quite so complex and so can be
conventional radio, the designers attempted to decoded by a less sophisticated (and cheaper)
afford a method of real-time, medium receiver. The power consumption of receivers
bandwidth data delivery to listeners. is also lower than it's Eureka-147 counterpart.
Being a newer specification, it was also able to
The solution they came up with, imaginatively take advantage of the later MPEG-2.5 audio
call "Digital Audio Broadcasting" or DAB is encoding which has a number of advantages
necessarily complex due to the conflicting over the MPEG-2 system adopted by Eureka-
requirements it addresses. The standard that 147.
was developed was named "Eureka-147" and is
the European DAB standard. WorldSpace probably wouldn't be such a threat
to Eureka-147 if it hadn't already launched 3
Eureka-147 is unique in that it is a single satellites covering Africa, Asia, the Caribbean,
frequency system, meaning that the entire Latin America, the Mediterranean basin and the
country would be on one single [centre] middle east, potentially reaching 4.6 billion
frequency. Because DAB can operate in the listeners. WorldSpace has also bought a
presence of fairly extreme multipath substantial stake in BayGen the company
interference, other transmitters on the same licences the clockwork radio (Trevor Baylis)
frequency will just be seen as an artificial source and are expected to start producing large
of multipath distortions. quantities with consumers from the developing
nations in mind. The WorldSpace system is
Because DAB is transmitted on only a few/one also gaining support from silicon vendors, with
[centre] frequencies, overall around the country, several decoder chip sets already available from
the utilisation of available spectrum is less, even supplies such as STMicroelectronics and
though the DAB channel is far wider than any Micronas Intermetall.
single (say) FM station.
The WorldSpace specification is lower quality
DAB is transmitted with convolution coding to than the Eureka-147 system and only partially
spread the spread the transmitted energy evenly overcomes the problems it set out to solve. But
across the spectrum and makes use of modern history is littered with examples of technically
digital techniques such as ECC (error correction superior systems being commercially overtaken
codes). This means that the decoded signal is by their lower quality rivals (VHS-Vs-Beta).
less prone to fading or other interference based
errors.
2. Brief Technical Overview of
DAB
1.2 WorldSpace - The Fly in the NOTE: For the purposes of this document, the
Ointment term DAB refers to the European Eureka-147
The Eureka-147 specification has been drawn
standard, any reference to US standard will be
up and approved by ETSI, the European
referred to as "WorldSpace".
standards agency. The ETSI specification has
also been adopted by Canada, Australia, Japan,
The Eureka-147 DAB system is a very
China and the majority of the Far Eastern
sophisticated digital communication system that
countries as the definitive standard for digital
has only in recent years become a viable
audio broadcasting. Such unanimity obviously
technology. The ready availability of powerful
DSP's has allowed the modulation method, The de-interleaved data is then fed to a Viterbi
COFDM (coded orthogonal frequency division decoder and error correction is performed.
multiplexing), to be economically realised. From this corrected data, the audio channels
COFDM consists of a number of orthogonal must be demultiplexed and then decompressed
carriers being differentially phase modulated from MPEG2 format.
with a quadrature constellation. Because the
carriers are QPSK modulated (have four states) Much of this information is taken from the only
they can code up to 2 bits/Hz/second. currently available commercial DAB chip. The
Hitachi SH7490 is a combination of a 60 MIPS
In DAB there are 1536 of these channels, each 1 DSP and some digital logic designed by Roke
kHz wide, giving a total bandwidth of 1536 x 1 Manor Research Ltd. Funded by the BBC.
kHz x 2 bits = 3.072 Mbps, or 384 Roke Manor have produced a commercial
Kbytes/second. Within this digital data must be design for a DAB receiver using the Hitachi
all national DAB stations, programmes and data. chip, called the "GoldCard" design which is an
This is the equivalent of a single 96 kHz stereo OEM module that supports dual band DAB.
16 bit audio channel, how does it all fit?

The simple answer is audio compression. Audio 3. Technical Difficulties


data is compressed using the MPEG2 (as is
DVD) perceptual encoding algorithm, but with a One thing that DAB does have a wealth of is
variable compression ratio. For data that cannot technical difficulties. A few of the more
be compressed with any loss, the compression obvious ones have been listed here.
engine is relaxed, then as the audio data blocks
start, the perceptual encoding is brought into 3.1 Complex RF circuitry
operation. I just can't wait to see the look on David's face
when he sees what's involved! The radio
The RF front end of a DAB receiver must be receiver will be using mobile phone type
capable of dual band operation, VHF-3 (174 - technologies and frequencies. Even the VHF
240 MHz) for terrestrial reception and L-band band is far above the FM that we have been
(1452 - 1492 MHz) for satellite reception. This "used to". This is not a trivial exercise in RF
makes for a complex RF front end with three design.
local oscillators and many filtering stages. First
the L-band signal is demodulated down to be 3.2 Local Oscillator Phase Noise
within the VHF band, at which point both signal The 1st local oscillator from L-band is used to
options follow a common path. The first stage directly convert down to VHF. The DAB
is to generate the 1st IF frequency of 38.912 carriers are only 1KHz apart, any phase noise in
MHz, then again down to 2.048 MHz. This 2nd excess of this will destroy information. The
IF signal is then synchronously sampled with an shear ratios of these two frequencies (1251.456
ADC at 8.192 MHz, then digitally I/Q MHz Vs 1 kHz) requires stability levels that will
demodulated then a complex FFT algorithm is get the average analogue chip designer a little
used to decode the phase of each of the worried. Add to this the mixed-signal approach
individual carriers. that GS and others are adopting and the problem
of generating this pure signal becomes even
Because of the requirement for such a low over- more extreme.
sampling ratio (to reduce the computational
bandwidth), the ADC must sample the spectrum 3.3 High Processing Bandwidth
synchronously, and this requires a PLL based The complex nature of the COFDM
digital clock source. The clock can be adjusted demodulation process (2048 point complex
under software control to achieve a FFT) means that to produce dedicated hardware
synchronously sampled data source. The data to perform this would be very expensive in
spectrum is only over-sampled by 4:1 but has to silicon. The alternative is to perform this
decode QPSK data within it, which has 4 states, computation in software. It is estimated that the
synchronous sampling is the only way this will FFT computation would take approximately 60
work with the bandwidth available from a MIPS of processing bandwidth even with the
typical DSP. inclusion of several dedicated hardware
'accelerators'.
Once the phase of each of the 1536 carriers is
known, the signal must be de-interleaved,
requiring a minimum of 256 Kbytes of memory.
DAB Feasibility Study
© Global Silicon Ltd, 2000 Page 2 of 7
3.4 Synchronous Sampling of the IF an implementation that already exists shows us
In order that the processing bandwidth be inside that it is possible. This part of the circuitry
of the operational envelope for an economic wound not be suitable for inclusion with Arrow,
DSP architecture, the IF signal must be sampled the requirements for low phase noise in the
synchronously. One approach to generating this oscillators, and GHz operation make this
clock is to use an analogue PLL under software something of a remote possibility.
control. This technique is used when the system
clock is below about 100 MHz, if the system 4.2 COFDM Demodulator
clock is above this figure, it might be possible to This is perhaps the most 'fearsome' aspect of the
use a digital PLL to generate the sample clock. design phase. This could be performed on a
revised Dart chip. Dart would not only require
3.5 MPEG-2 more memory, more speed, and more
The decoding of perceptually encoded and instructions (specialised for FFT processing).
compressed audio signals is also an area where There will need to be digital logic to I/Q
the processing requirements are not at all demodulate, decimate and bandpass filter the IF
simple. An estimate of 40 - 50 MIPS is signal, which is a fairly large chunk of silicon.
reasonable even with hardware acceleration The software would then (with the aid of
logic. hardware acceleration logic) perform the 2048
point FFT on the data samples. The FFT would
3.6 Viterbi Decoder probably occupy the total available bandwidth
Again another technically challenging area. The of the current Dart chip (DART-1) alone, so a
processing requirements of a Viterbi decoder 2x Dart is a basic requirement and not a luxury.
are well beyond the abilities of even an
advanced DSP today. The Viterbi error 4.3 De-interleave Logic
correction scheme would have to be This would again require some dedicated
implemented in dedicated digital logic hardware, but the largest demands this places
upon the system is in terms of memory. The de-
3.7 Convolution decoder interleave buffer would need to be at least 256
The signal is convolution encoded to spread the Kbytes and would probably be located in
data energy evenly across the RF band. This DRAM as the most economical form of memory
encoding needs to be removed (or decoded) that can be attached to Dart.
with a subtle mix of hardware and software in
the receiver. Again this is a fairly advanced 4.4 Viterbi and Convolution decoding
topic in signal processing theory. Again this requires so many MIPS that a
hardware solution is the only real possibility.
4. Possible Global Silicon Viterbi decoding is a specialist area of signal
processing and would occupy at least 20 K
Implementation ASIC gates in silicon. The convolution decoder
A practical implementation of a Global Silicon would be included in the Viterbi logic.
DAB receiver could possibly take a number of
pointers from the Roke Manor Research 4.5 MPEG-2 Decoding
implementation of the "GoldCard" receiver. This function would be performed mostly in
This device uses a Hitachi SH-DSP core to hardware with some software support to take
perform COFDM demodulation in software. care of the higher-level functions. MPEG-2
This is similar to the strategy employed by GS decoding is something of a specialised area,
to gain cost reductions from conventional HiFi. Global Silicon should consider using external
Using the "GoldCard" receiver as a model, GS support for this. The alternative is to use about
could attack the main sections as follows: - 40 MIPS of CPU bandwidth to do the decoding.
This would only leave about 35 MIPS after
4.1 RF Front End COFDM & MPEG-2 decoding, 30 of which is
Use the National Semiconductor chipset for required for the graphic equaliser. Global
generating the three synthesised sources Silicon is also considering the inclusion of DVD
(LMX2331 and LMX1551) required for to the system an this would require an external
demodulation. This is a fairly safe route as the MPEG decoder - DAB could 'hi-jack' this.
"GoldCard" receiver uses the same chipset and
successfully performs the RF down-conversion. 5. Summary
This is not meant to trivialise the problems Designing a DAB receiver is not an easy task,
associated with this type of design exercise but and one that Global Silicon should be very wary

DAB Feasibility Study


© Global Silicon Ltd, 2000 Page 3 of 7
of undertaking at this stage. The "Cambridge- http://www.bbc.co.uk/rd/pubs/reports/1996
rumour-mill" has it that Symbionics (now a _08.html
division of Cadence) recently attempted to • "why do we need digital radio"
design a DAB chipset. Symbionics are without RadioScape web site
doubt consummate experts in the field of mixed- http://www.radioscape.com/digital-
signal RF chip design, however, they appear to broadcasting/why-do-we-need.htm
have dropped the project for reasons of • "Brief Litreature Review on OFDM"
technical difficulty (again rumour based). http://www.drb.crc.ca/ottawa/cofdm.htm
• "What is COFDM ?"
The software and hardware requirements are http://www.cclinf.polito.it/~s83797/cofdm.
such that to implement a DAB receiver is going htm
to take a serious design effort in the order of • "DAB Explained by CCETT"
several man-years. Global Silicon should be http://www.cclinf.polito.it/~s83797/ccett.ht
prepared to allocate at least one man-year of m
effort for each of the areas of software, digital • Philips SAA3500 web page http://www-
hardware and RF system design, i.e. a minimum us.sv.philips.com/dab/Products/Chip_sets/S
of 3 man-years. AA3500/saa3500.html
• Philips "Fadic & Silvic" Web Page
Obviously the emergence of a rival standard in http://www-
to the digital radio arena also complicates the us.sv.philips.com/dab/Products/Chip_sets/F
situation, with semiconductor companies like adic___Sivic/fadic___sivic.html
Texas Instruments and OKI adopting a holding • EDN report "Will DAB finally take off?"
pattern. The "WorldSpace" Vs Eureka-147 http://www.ednmag.com/ednmag/reg/1998/
issue looks unlikely to be resolved in the next 090198/pdfs/18ecs2.pdf
12 months and this may apply a brake to the • Article "Digital Radio - The Eureka 147
emergence of consumer acceptance of either DAB Standard" A J Bower (BBC)
one. http://www.bbc.co.uk/rd/pubs/papers/paper
_21/paper_21.html
Put simply, for Global Silicon to implement a
• "A Technical Overview of Digital Radio"
DAB receiver based upon the Dart/Arrow
Stephen Baily (BBC)
chipset would require the following: -
• http://www.bbc.co.uk/rd/pubs/papers/pdffil
es/ps0199.pdf
• Designing a specialised RF dual band radio
• Press release from Roke Manor research
receiver as a separate item to Arrow
http://www.roke.co.uk/news/archive/new6.h
altogether.
tml
• Designing dedicated digital logic for Dart,
• The world DAB forum - Web page
which would effectively double the silicon
http://www.worlddab.org/
area of the chip.
• DAB Developments - German company
• Doubling the software overhead of the
specialising in DAB software products
system.
http://www.esslinger.de/dab/
• Including a DRAM memory chip.
• "The Effects of Phase Noise in COFDM" J
• To have already designed the DVD version
Stott (BBC)
of Dart (Dart-II).
http://www.bbc.co.uk/rd/pubs/papers/pdffil
es/jsebu276.pdf
6. Further Reading • COFDM Vs 8-VSB http://web-
star.com/hdtv/cofdmvs8vsbfcc2.html
• "An introduction to digital modulation and
OFDM techniques" M C D Maddocks BBC • Ensigma - DAB software specialists - web
research department report. BBC RD site
1993/10 http://www.ensigma.com/News/InTheNews
http://www.bbc.co.uk/rd/pubs/reports/1993 /IP99TunesIn.htm
_10.html • Data sheet for the Hitachi SH7490 DAB
• "The how and why of COFDM" J H Stott receiver
BBC research department report. http://semiconductor.hitachi.com/products/
http://www.bbc.co.uk/rd/pubs/papers/pdffil pdf/h1htd001d1.pdf
es/ptrev_278-stott.pdf • Evaluation board based upon Hitachi
• "The COFDM modulation system" P SH7490 http://www.super-
Shelswell BBC research department report. h.com/eng/virtual/panel/p34.html
BBC RD 1996/8
DAB Feasibility Study
© Global Silicon Ltd, 2000 Page 4 of 7
• EE Times Article 24S0030
http://www.eetimes.com/story/OEG199903
7. Appendices
7.1 Appendix1 - The OFDM Spectrum

The OFDM spectrum - 1536 x 1 KHz spaced channels

The OFDM spectrum - bad sketch of some of the


channels in close up

DAB Feasibility Study


© Global Silicon Ltd, 2000 Page 5 of 7
7.2 Appendix 2 - Simplified block diagram of a typical DAB Receiver RF Front-end

L-Band Input
(1452-1492 MHz)

2nd L-Band
Roofing Filter

Fo Fo

Broadband Filter AGC

LO
To digital
1251.456 MHz demodulator

VHF-Band 3 Input
(174-240 MHz) Software
ADC controlled AGC
8 MSPS

Fo Fo Fo Fo

Broadband Filter AGC VHF band select 1st IF Amp SAW IF Filter IF Amp 2nd IF Filter
38.912 MHz

LO LO

213-279 MHz 40.96 MHz


7.3 Appenxix 3 - Simplified block diagram of a typical DAB digital decoder

2 Mbits SRAM
OFDM Demodulator

IF Spectrum I/Q Down 2048 point De-interleaver Viterbi Convolutional MPEG-2


ADC converter FFT Decoder Decoder Decoder Audio

AGC RF Control
DAC Sync &
Tracking

Synthesiser
control

DAB Feasibility Study


© Global Silicon Ltd, 2000 Page 7 of 7

Potrebbero piacerti anche