4 General description of the technical characteristics of the DVB-S2 system
6.1 IP Unicast Services
6.1.1 Single Generic Stream and ACM command
According to this system configuration, the DVB-S2 ACM modulator receives two input signals. The first is the data stream, continuous or packetized. The second is the ACM command, carrying the MODCOD information used by the modulator for encoding and mapping each specific portion of the input data stream. This approach has the important implication that the scheduling function, which performs the selection and aggregation of the information to be transmitted in each frame, is located outside the DVB-S2 modulator subsystem. This strategy leads to a complete transparency of the modulator to layer 2 functions and therefore has the important advantage of an absolute flexibility in
It is worth noting that in ACM systems the choice of the physical layer mode to be used in each frame is necessarily linked to the scheduling process, as all the UPs included in one frame are transmitted with the same physical layer parameters. From a different perspective, the data for filling a specific frame need to be selected taking into account the physical layer mode requested by the STs to whom they are addressed. This is the reason why the MODCOD
information is generated at the same time of the user data selection and sent as an input to the DVB-S2 ACM modulator.
Figure 20 shows the block diagram of a possible architecture for buffering and processing the data prior to conveying them to the ACM modulator. The input data stream, composed of a sequence of User Packets, is routed according to the addressed user and his QoS requirements. Therefore, if L is the number of active users and N the possible QoS levels in the network, L*N is the maximum number of buffers needed to appropriately discriminate UPs. As active users we mean those STs, which have an open session. When no traffic is sent to an individual ST for a certain period of time, a time-out is exceeded and the user is not considered active any more. As a result, the associated buffers are de-allocated. In general, a user can simultaneously support applications requiring different QoS levels. However, as most of the users will probably support less than N simultaneous different applications with different QoS requirements, the number of allocated buffers is consequently reduced. In the simplified case where the service level agreement defines only one QoS level for each user, the number of buffers equals the number L of active users.
Figure 20: Block diagram of a possible system architecture with single generic stream input to the ACM DVB-S2 modulator
For each frame the merger selects from the input queues a number of packets, and combines them for building a set of information bits. Frame by frame successive data sets composed in this way are sent to the ACM modulator, together with the associated transmission parameters. When the number of bits in one set is not sufficient to completely fill the BBFRAME, the modulator will provide padding by automatically choosing the most suitable type of FECFRAME, with short or normal length.
The merger selection is driven by an ACM Routing Manager, which is responsible for packet scheduling. The scheduling policy is application dependent and needs to be designed for maximizing system efficiency while meeting QoS requirements. In order to achieve these goals, the ACM Routing Manager can take advantage of the channel status information reported by the STs, of the different priority levels and QoS requirements of the input queues, and finally of the information concerning the buffer occupation. In fact, the first type of information is needed in order to combine in one frame packets with the same transmission parameters; the second allows for meeting QoS requirements (maximum delay, minimum rate, etc.); the third can be used e.g. for satisfying QoS requirements without sacrificing efficiency (see clause 6.1.3) in presence of scarce traffic associated to a certain physical layer mode. In this case indeed, a smart scheduler policy could decide of merging these few packets with others more numerous requiring a lower efficient physical layer mode.
It is worth to mention that the high level of discrimination of the input flow (packets are routed according both to their QoS level and to the addressed user) increases the degree of choice in the packet selection process, allowing for a
Furthermore, in the analysed system architecture the loop delay is minimized, thanks to the fact that the choice of the transmission parameters is made immediately before the encoding and mapping functions. The waiting time in the buffers before the scheduling process is not contributing to the loop delay, as a channel variation occurring when UPs are waiting in the queue does not lead to a wrong physical layer mode selection. On the other hand, no queues are present within the DVB-S2 modulator and the time interval comprised after the decision on the physical layer mode and before signal transmission is minimized. The complete separation between scheduler and modulator allows for a robust system, whose dynamicity is not affected by data buffering delay within the GW.
The system good performances in terms of flexibility, channel tracking capabilities and QoS satisfaction need nevertheless to be traded off against system architecture complexity. The quite large number of buffers, though of limited memory size requirements, and their dynamic allocation dependent on user traffic variations make
implementation a challenging task. However, a functionally equivalent solution can be implemented with only N input buffers, in which the user packets are separated per QoS levels. As for each frame the ACM Routing Manager shall aggregate packets with the same requirements in terms of transmission parameters, there is a need for accessing all the buffers memory locations, instead of the first one only. Thus, the L*N FIFO buffers of figure 20 can be in principle replaced by N buffers of larger dimension, where all the packets in the buffers are made available to the merger for building the data field. This prevents the need for dynamic buffer allocation, thus simplifying implementation