J D Mitchell (BBC)
and
P Sadot (LSI Logic, France)
BBC Research and Development and LSI Logic are jointly developing a front end for digital terrestrial television transmitted according to the DVB-T specification.
The front end consists of two separate components. First, an analogue down-converter that converts the input signal from UHF to a low IF. Second, an integrated circuit that accepts the analogue signal from the down-converter and performs the required DSP operations, which include synchronisation and demodulation, to form a stream of soft decisions suitable for presentation to an FEC decoder.
The development process began by agreeing a set of requirements to which the two components must conform. This paper begins by outlining these requirements. During the development of the components, many issues have been considered and resolved. A selection of the key issues and the decisions that were reached is given and, finally, a discussion of the architecture that results from these decisions is presented.
BBC Research and Development and LSI Logic are working together on the development of a digital terrestrial front end which is capable of decoding transmissions compliant with the European DVB-T specification (1). This development unites the BBC's system expertise in COFDM, see Nokes et al. (2), Stott (3) and (4), and LSI Logic's long-established chip design capability.
As part of the front end development, LSI Logic and the BBC are working together on two distinct, but related, development projects: a single CMOS chip implementing a complete OFDM demodulator, and a new terrestrial down-converter designed for use with the digital chip. The aim of the overall system design is to optimise the cost effectiveness of the complete front end. This paper specifically addresses three technical issues:
The target of LSI Logic and the BBC was to implement a fully DVB-T compliant COFDM demodulator. This section describes the initial set of requirements which were the basis of the architecture of the digital demodulator chip, called the L64780 in LSI Logic's DVB product range.
Full compliance to the DVB-T specification means that the chip must be capable of decoding signals transmitted in the following modes:
The front end architecture must provide the best possible performance under actual operating conditions. There are several key types of channel impairments that the front end must be adept at dealing with. Some of the key channel impairments are explained below:
The down-converter architecture must cope with the specific requirements of COFDM whilst operating in the channel conditions described above. This means:
The time and frequency synchronisation algorithms were defined with two objectives in mind:
The output interfaces of the chip must allow for easy integration with existing chip-sets.
Fig. 1 shows the overall structure of a DTT receiver. The key components to the system are the front-end (which for terrestrial applications consists of a down-converter, the OFDM demodulator chip and an FEC decoder), the transport stream demultiplexer, the audio and video decoders and the overall system microcontroller.
Fig. 1 - Top level structure of a DTT receiver.
The OFDM demodulator is the first chip in LSI Logic's digital terrestrial television product range. It will permit first generation DTTV set-top boxes based on a down-converter, the L64780 itself and an FEC chip such as the L64705 or the FEC part of the L64724.
In a fully integrated set-top box, front-end components for satellite and cable may also be provided.
The microprocessor interface of the OFDM demodulator chip can operate in two modes, either parallel or serial (I2C like).
During the development of the front-end many architectural trade-offs had to be considered. This section provides a discussion of some of the key trade-offs that were made.
The most significant problem that was encountered during the design of the integrated circuit was the amount of RAM that the chip required, even taking into account state-of-the-art developments in RAM technology. In planning the architecture, we wanted to make the best possible use of the RAM that we could fit onto the chip. Some of the blocks of memory, such as the FFT and symbol deinterleaver, require fixed amounts of RAM and it is not possible to reduce them (except by reducing the word widths and so degrading the performance). Other blocks, for example the timing synchronisation, required some algorithmic alterations for the sole purpose of reducing the amount of memory but without degrading the performance. The final technique that was employed to make best use of the available memory was to "reuse" some of the memories. For example, the data delay required to implement common-phase-error correction doubles as the first data delay in the channel equaliser. This means that only two additional data delays were required to implement full linear temporal equalisation.
Table 1 shows the final allocations of RAM that were made in the chip. As this table shows, the highest memory usage is in the FFT circuitry and the smallest is in the timing synchronisation circuitry. We have found that this memory allocation provides the best compromise between performance and cost.
Table 1: Proportion of RAM used by different elements of the DSP.
One of the processes that is required in the synchronisation of the demodulator is to obtain frequency synchronisation. In our front-end architecture we always make the measurement of the AFC error digitally, but we had the choice as to whether to apply the required frequency shift as an analogue correction in the down-converter, or as a digital frequency shift in the chip.
If we choose to implement the frequency correction by adjusting the frequency of the reference crystal in the down-converter, then we have to provide a control signal from the output of the integrated circuit back to the down-converter. This method has the advantage that the SAW filter inside the down-converter can be made as narrow as possible. The disadvantages are twofold. First, the integrated circuit must pass a control signal back to the down-converter. Second, the architecture of the down-converter is made more complicated since the control signal must adjust the reference crystal within the search range of the AFC.
If we choose to implement the frequency correction in the integrated circuit, then the architecture of the down-converter is made much simpler since there is no longer any need to have a control signal from the chip, and the loop in the down-converter that drives the reference crystal is no longer required. The disadvantage of this method is that the bandwidth of the SAW filter must be increased by the AFC search range. This causes a significant penalty in terms of the adjacent channel protection ratio when the receiver is used in an environment where the existing analogue services are operated in adjacent channels to the new digital services.
After the launch of the new digital services it is expected that the analogue services will continue for many years. It is therefore clear that the adjacent channel performance of the front-end is an extremely important parameter when deciding on an architecture for the chip, or assessing its suitability for operation in an integrated receiver. Therefore the architecture we have selected will permit both forms of frequency correction.
The next component in the architecture we will discuss is the channel equaliser. The signal coming from the output of the FFT will be affected by all the impairments caused by the channel; for example in the presence of a single echo, the FFT output will suffer from frequency selective fading. The purpose of the channel equaliser is to rotate and scale the constellation so that the constellations on all the carriers are all of a known size (but not necessarily of the same reliability).
The process is performed by using the scattered pilot information contained in the COFDM signal. The scattered pilots provide a reference signal of known amplitude and phase on every third OFDM carrier. Since this scattered pilot information is subject to the same channel impairments as the data carriers, the scattered pilots are noisy.
One possible solution to this noise problem is to average a large number of these pilots over time. This provides a much less noisy estimation of the channel frequency response. This will provide an improvement in a Gaussian channel of approximately 2dB. The penalty that is paid for this improvement is that if the channel response changes (i.e. it is time varying) then the channel equaliser may not be able to track this and the decoder will fail.
The simplest solution to cater for time-varying channels is to use a sample-and-hold mechanism. In this case, the most recently received scattered pilot is used as the reference for all the data cells until the next scattered pilot is received, at which time the new value becomes the reference. This solution provides the possibility of tracking time-varying channels but totally sacrifices the advantages of averaging the continual pilots.
The solution that we have adopted for this chip, is to perform linear interpolation between two received scattered pilots and to use these interpolated values as the reference to equalise the data. Since scattered pilots at the same time duration are spaced 4 OFDM symbols apart, a compensating data delay of 3 OFDM symbols must be provided to permit this option. This solution provides the ability to track channels that are changing more rapidly than the sample-and-hold technique could follow and also provides a degree of averaging, since at least two scattered pilots are used to equalise every data carrier.
In summary, we have chosen the final option, known as linear temporal equalisation, because it is vital to be able to track time-varying channels, and because it is more important to protect against "difficult" channels than against "easy" channels.
The down-converter performance has a different set of requirements from those demanded by down- converters suitable for analogue television. For example, in a down-converter designed for analogue television, particular attention must be given to the group-delay characteristics. However, COFDM has been specially designed to be robust to this type of distortion, and so the group-delay performance is much less important.
Another difference between the two requirements is in the local oscillator phase-noise performance. It can be shown, Stott (3), that the addition of local oscillator phase-noise to an OFDM signal has two principal effects:
It is possible to remove the common phase-error component caused by phase noise added in the down-converter by digital processing in the chip. This processing is performed by the common-phase-error correction block in the architecture.
The common-phase-error correction block is able to remove the common phase error because all carriers within a given symbol suffer the same common phase error. By measuring the continual pilots, whose intended phase is the same from symbol to symbol, the common phase error is determined and then subtracted from the phase of all the data cells in the same symbol. There are sufficient continual pilots (which in any case are transmitted with a power approx. 2.5 dB greater than data cells) that the effect of thermal noise on this measurement can be rendered negligible by averaging.
There are essentially three components required to implement common-phase-error correction in the chip. These are:
These three factors, which together form the "cost" of implementing the feature on the chip, must be balanced against the cost of improving the performance of the down-converter so that the phase-noise it introduces is negligible. In our architecture we decided that the cost of including a common phase error correction circuit was substantially less than the cost of eliminating phase-noise in the down-converter, and so the chip includes circuitry to perform common-phase-error correction.
Fig. 2 shows an overview of the architecture that has been selected for the DTTV integrated circuit. The input to the device is an analogue signal with a centre frequency of 4.57 MHz (exactly 32/7 MHz), referred to as the low IF frequency. The signal would come directly from the down-converter having been shifted in frequency from UHF or VHF as appropriate.
Fig. 2 - Architecture of the DTTV integrated circuit.
The output from the chip is a stream, either in serial or parallel form, of soft-decision metrics suitable for presentation to a Viterbi decoder.
There are many options available that control the functionality and performance of the chip, and these are controlled by an external microcontroller, such as the microcontroller that will control the overall operation of a set-top box.
Once the chip receives the analogue signal from the down-converter, the first operation performed is to sample the signal at 18.29 MHz. This operation can be performed by the ADC internal to the chip, or for down-converters which already have an ADC included, the chip will also accept a parallel digital input.
The digital samples are then processed by a block called the real-to-complex converter. This block takes the "real" input signal centred on 4.57 MHz and outputs a complex signal centred on zero frequency by digital signal processing. These signals are then fed in parallel to the timing-synchronisation block and the FFT.
The timing-synchronisation block uses the complex input signal to derive a value that can be used (via a DAC) as the control voltage input to a voltage controlled oscillator (VCXO). This block, therefore, provides all that is necessary to synchronise the timing of the demodulator.
The FFT has four modes of operation. First it is capable of performing either a 2048 point or an 8192 point transform. Second it is capable of performing the transform in either direction. The inverse FFT functionality is provided so that the integrated circuit could be used in applications requiring OFDM modulation.
Immediately after the FFT, in a side chain, frequency synchronisation is performed. This block looks at the post-FFT (frequency domain) signal and provides an output signal to be used in the down-converter to adjust the frequency of the reference crystal.
The output data from the FFT is then passed to the common-phase-error correction block. This block calculates the common phase error of the signal and applies the necessary correction.
The output of the common-phase-error correction circuitry is then passed to the channel equaliser. This block first performs linear temporal equalisation followed by frequency equalisation using a high order interpolating filter. This block then outputs an equalised constellation (which can also be viewed off-chip for monitoring purposes) to the channel state block.
The channel state information block generates three- or four-bit soft decisions which are suitable for presentation to a Viterbi decoder. It is also possible to view the channel state response for monitoring purposes. For a detailed explanation of channel-state information, see Stott (4).
The final process that is performed by the chip is to deinterleave the soft decisions. This consists of (i) performing the symbol deinterleaving and (ii) performing the bit deinterleaving. Once this process is complete, the signals are passed off-chip for input to an FEC chip.
Obviously the next step is to integrate all the digital functions into a single chip, from the analogue-to-digital converter down to the MPEG transport-stream interface. This second-generation chip will be available by the end of 1998.
LSI Logic is closely monitoring the progress of direct-conversion tuner technology which allows the implementation of single-bipolar-chip tuners. When this technology is sufficiently mature for DTTV applications, LSI Logic will modify its single-chip COFDM receiver to permit seamless integration with these new tuners. This will enable a two component, extremely low cost front end for the high volume deployment of Digital Terrestrial TV in Europe and other countries, which we expect at the start of the next millennium.
BBC Research and Development and LSI Logic are working together to create a front end which consists of an integrated circuit that is capable of decoding signals compliant to the DVB-T specification in a wide variety of receiving environments and an analogue down-converter which is suitable for direct connection to the integrated circuit.
The L64780 is the first chip in LSI Logic's digital terrestrial television product range. It will permit a first generation of DTTV set-top boxes based on a down-converter, the L64780 itself and an FEC chip such as the L64705 or the FEC part of the L64724. The down-converter and the L64780 will be available by the first quarter of 1998.
The authors wish to thank the many colleagues, both within the BBC and LSI Logic, who have made the development of this chip possible.