DekTec's Participation in the DVB-T2 Working Group

1) Generation of Reference Streams for the DVB-T2 Verification and Validation Group

The objective of the DVB-T2 V&V group has been to verify that there is no bug nor ambiguity in the DVB-T2 standard and that everyone understands it the same way. The members of the group agreed to generate and publish their own reference streams at selected points of the modulator chain. Then it was possible to compare them and see where were the problems. Due to the large number of different modulation parameters, it was not possible to test every combination. So the members agreed on several sets of parameters to have a wide coverage of the standard.

DekTec was among the first to provide reference streams for all the sets of parameters, including the more difficult cases where several Physical Layer Pipes (PLP) are combined.

As the DekTec DVB-T2 modulator product is done in software, it was possible to use the same code for the DekTec DVB-T2 modulator products (DTC-378-T2) and to generate the test streams. It saved a lot of the time in coding and product validation. The V&V parameter sets (VV001, ...) are available as presets in the DVB-T2 modulators.

2) Participation in the DVB-T2 Simulation Group

This group evaluated the performance of the DVB-T2 standard over a range of selected terrestrial channels using simulations. Some of its results are found in the Chapter 14 of the DVB-T2 Implementation Guidelines.

DekTec had not originally planned to participate in this group because the goal was only to build a real time software DVB-T2 modulator. Simulations are more complicated because they additionally require a channel simulator and a demodulator.

While writing the software modulator at DekTec, a simplified demodulator was written to verify that no input bit is lost inside the modulator. The demodulator was simplified because it assumed that the input signal was perfect (no noise, no channel distortion and no synchronization problems). This simplified demodulator was of great help to test the modulator with complicated sets of parameters.

After some time DekTec realized it could be possible to improve the demodulator to do performance simulations. The important steps were to have a soft decision QAM demapper and a Low Density Parity Check (LDPC) code decoder. Everything was implemented in C using an in-house toolbox of signal processing routines. In order to keep the simulation time small, some time was spent optimizing each part of the demodulator. The most challenging one was the LDPC code decoder because it takes most of the decoding time at low Carrier to Noise Ratios, which are the most interesting ones in simulations.

The first simulations were done using the same parameters as the DVB-T2 Common Simulation Platform (CSP). The CSP contains contributions from many companies and is the reference to simulate the DVB-T2 standard. It was written using Matlab. The DekTec simulator was written in C and contains many code optimized for real time processing (in particular the modulator and the channel simulator), so it is much faster than the CSP. Instead of taking days to get one result, only less than one hour is needed using similar computing power. It was useful to explore many DVB-T2 parameter sets and channel model combinations. Although the channel models were slightly less accurate than the CSP, the Carrier to Noise Ratio needed to reach a given Bit Error Rate were comparable up to 0.1 dB.

In order to get more results, a distributed system was designed to automatically assign simulations to several PCs at the same time. With this system, it was possible to generate tens of results in a few hours.

3) LDPC Code Rate 2/3 Issue

When doing simulations with the CSP, strange results were noticed for a few points using the LDPC code rate of 2/3. The DekTec simulator helped confirm the problem and showed that it appeared for many channel types and QAM constellations. The result was that the DVB-T2 standard would have been less efficient than expected when using the LDPC code rate of 2/3. Unfortunately this code rate is critical in many broadcasting cases.

A special Simulation Task Force was set up to fix the problem. The DekTec simulator was used to evaluate all the proposed fixes with all the channel types using the same parameters as the high precision simulations done by the CSP. The results were confirmed by other companies at least for the Additive White Gaussian Noise (AWGN) channel.

Before the final decision for the fix, it was agreed that more extensive simulations were needed. The goal was to simulate the Quasi Error Free case after the LDPC code decoding, which means to have less than one error every 10^11 received bits. It corresponds to about one error per hour per TV channel. It was agreed that the transmission of at least 10^13 bits were needed to have a good enough confidence in the results. These simulations were much more demanding in terms of computing resources than the previous ones which involved 1000 times less transmitted bits per result.

DekTec used its software simulator to make the computations. Since it required 1000 times the computing power of the previous simulations and since only a few days were available to do the simulations, a set of dedicated PC servers provided by the Amazon Elastic Compute Cloud (Amazon EC2) were rented. The equivalent of 17000 hours of compute power of a single CPU core PC were used. As it was done in 3 days, about 30 eight-core PCs were rented. On a single CPU core, the simulator is about 6 times slower than a real time transmission. With the available computing power, the simulations were about 40 times faster than real time. A typical result required the equivalent of about 100 hours of real time DVB-T2 transmission to get the required 10^13 bits. With the simulation system, it took only 2.5 hour to get the result.

Other companies also confirmed the results using FPGA based hardware simulators. This simulation campaign confirmed to us that a pure software simulator can compete with such hardware solutions.

4) Real Time DVB-T2 Receiver

As the simulator was only 6 times slower than real time on a single CPU core, it showed that it was possible to do a real time DVB-T2 receiver using a single PC with a multi-core CPU. It was decided to improve the demodulation code so that it includes the missing parts to have a real time receiver. The missing parts were the channel estimation (the simulator uses an "ideal" channel estimator which cannot be used with real signals) and the time and frequency synchronization.

The real time PC based software DVB-T2 receiver was demonstrated at the DVB-T2 plug fest in March 2009 at the RAI Research Center in Turin.

Because the same code is used for simulation and for the real receiver, it is particularly easy to try new reception algorithms or to make various signal measurements.

At a later stage a user interface was added to the real-time DVB-T2 software, creating the commercial product T2Xpert.