DYNAMIC BANDWIDTH ALLOCATION IN SCPC-BASED SATELLITE NETWORKS

Transcription

1 DYNAMIC BANDWIDTH ALLOCATION IN SCPC-BASED SATELLITE NETWORKS Mark Dale Comtech EF Data Tempe, AZ Abstract Dynamic Bandwidth Allocation is used in many current VSAT networks as a means of efficiently allocating satellite bandwidth resources. In many Time Division Multiple Access (TDMA) based networks, rate-based allocation mechanisms are used in order to provide the best user experience within the delay environment of satellite communications. Rate-based dynamic allocations are also possible in Single-Carrier-Per-Channel (SCPC) based network architectures. Key advantages and limitations of SCPCbased implementations are discussed. Quantitative comparisons between dynamic SCPC and the DoD s Joint IP Modem TDMA architecture are provided, and it is shown that under realistic conditions, the dynamic SCPC approach reduces the required return link bandwidth by greater than 40% relative to that required by a Joint IP Modem based system. Introduction Two-way satellite communications support a variety of applications. For constant bit rate, point-to-point connections, single-carrier-per-channel (SCPC) modems are typically used, and provide the optimal bandwidth and power efficiencies. However, for networks of terminals, other topologies are often used. A common topology connects many remote terminals to a common Hub, where all terminals share a pool of satellite bandwidth in each of the forward (Hub to Remote) and return (Remote to Hub) directions. These networks are commonly referred to as Hub- Spoke networks. In Hub-Spoke networks, the potential to share Hub and satellite bandwidth resources across all Remotes presents an opportunity to gain efficiencies through statistical multiplexing and dynamic bandwidth allocation in both the forward and return link directions. Basic mechanisms for statistical multiplexing and dynamic bandwidth allocation are well known. Very Small Aperture Terminal (VSAT) systems are typically based on Time Division Multiplexing (TDM) in the forward link, and Time Division Multiple Access (TDMA) in the return link. An alternate architecture that achieves the same objectives, uses TDM in the forward link but dynamic SCPC (dscpc) in the return link. Hub-Spoke TDM/TDMA Systems Figure 1 shows a high-level depiction of a standard Hub- Spoke architecture. This basic architecture applies equally to either TDM/TDMA or TDM/dSCPC systems. In the forward link of the TDM/TDMA system, traffic from all users share a common carrier. This enables dynamic allocation of capacity based on the instantaneous demand of the various terminals. For example, if all traffic is destined for single remote terminal at a given time, this terminal can be assigned all the capacity of the forward link carrier. Later, if a different set of terminals has different traffic demands, capacity can be reallocated to fairly share the available capacity. If the terminal traffic patterns is random and bursty, this mechanism provides for a much more efficient use of the forward link satellite capacity. In the return link, the TDMA carrier is subdivided in the time domain into multiple time slots. Terminals share a given carrier (channel) by taking turns making bursts in the defined time slots. Capacity is dynamically assigned in the return link by allocating time slots more or less often to a given terminal. This basic concept is illustrated in Figure 2. Exact details are specific to a given vendor s implementation. However, there are some common principles that all TDMA based return links share: 1. Burst overhead (usually in the form of preambles) is required to enable burst reception at the Hub. 2. A network timing synchronization mechanism is required, to maintain the precise burst time alignment between terminals, 3. Data packets have variable sizes, but nearly all TDMA systems use fixed burst sizes. This requires data packets to be fragmented and reassembled, which in turn requires overhead to support. TDMA systems sometimes suffer from even more significant packing inefficiencies in the event that implementation restrictions preclude efficient fragmentation and packing of data packets into return link bursts /08/$ IEEE Page 1 of 6

2 4. Burst processing has demodulation and forward error correction (FEC) disadvantages relative to continuous streams, both theoretically (due to small burst sizes) and in implementation practice. 5. In the TDMA return link, a single terminal is never allowed to have an average capacity assignment equal to the capacity of the entire return link carrier (if it did, it would effectively become an SCPC link). The maximum practical average level is 50% utilization or lower. Most often, many TDMA terminals share a given return channel, and average channel utilization of any given terminal is much lower than 50%. 6. TDMA users are required to close the link margin of the shared channel. For example, if multiple users are sharing a 1.0 Mbps link and a given terminal is assigned 100 kbps, the terminal must have the antenna size, RF power, etc. to close the 1.0 Mbps link, not the lower requirements associated with a dedicated 100 kbps link. Dynamic SCPC Systems Like TDM/TDMA systems, in TDM/dSCPC systems the forward link traffic from all terminals shares a common carrier, enabling the advantages of dynamic bandwidth allocation and associated statistical multiplexing. In the dscpc return link, each terminal is assigned a dedicated carrier. Dynamic allocation of return link capacity is achieved by varying modulation, FEC type and/or channel bandwidth of return carriers assigned to individual terminals in response to traffic demands. This basic concept is illustrated in Figure 3. As with TDMA system, exact details are specific to a given vendor s implementation. However, there are some common fundamental principles of dscpc based return links that are important to note: 1. The Hub must reacquire the return link carrier from a terminal when a parameter such as carrier frequency or symbol rate is changed in response to a capacity assignment. 2. Once made, the SCPC connections are continuous, and hence there is no burst overhead during transmission. 3. Because there is no need to control burst times, network timing synchronization is not required. 4. Data packets can be packed sequentially into the continuous SCPC stream very efficiently (less overhead than required for burst fragmentation and reassembly). 5. In the dscpc return link, a single terminal always uses the entire channel. 6. Modulation and FEC can be selected for maximum bandwidth efficiency for a given terminal, and return link uplink signals are often bandwidth rather than power limited on the satellite. For example, consider the case where a terminal can be assigned a maximum return link data rate of up to 1.0 Mbps, but has a current assignment of 100 kbps. If the terminal used the same RF power, modulation, and FEC when assigned 100 kbps as 1.0 Mbps, it would operate with 10 db excess link margin at 100 kbps. Instead of operating with excess margin, the modem has the option of using higher order modulation and higher rate FEC to improve bandwidth efficiency. Bandwidth Efficiency Comparisons In order to make quantitative comparisons between the dscpc and TDMA approaches, the Joint IP Modem () TDMA waveform will be compared with a dynamic SCPC system using the DVB-S2 waveform. The basic waveform has been publicly defined. The comparison is particularly relevant since has been stated to be an important future standard for IPbased satellite communication. DVB-S2 is an open standard used in both Government and commercial products, whose parameters and performance is well known. Forward Link: Both the and the dscpc system use a TDM forward link based on DVB-S2. Forward link control overhead (e.g. for sending system information (SI) tables in and dscpc control management messages) are implementation and configuration specific. However for simplicity, the overhead required will be assumed to be the same. Hence forward link bandwidth efficiency is identical for these two systems. Return Link: The overhead required for control messaging (e.g. to make bandwidth allocation requests, provide state-ofhealth information, etc.) will be assumed to be the same between the two approaches (essentially the same information is transmitted). In the system, burst timing network synchronization is maintained by periodically transmitting special synchronization (SYNC) bursts. This overhead is completely absent from the dscpc system. Although this overhead can be significant, very high stability components and/or intelligent Hub processing can minimize the need for this overhead component in the system, and its Page 2 of 6

3 Shared Forward Link TDM Return Link #1 (SCPC or TDMA) Return #2 (SCPC or TDMA) Return #N (SCPC or TDMA) Remote #1 N Remote #2 Hub Remote #N Figure 1: Generic Hub-Spoke Satellite Communications Architecture return link frame time L L Remote #1 burst time slots within frame Dynamic Bandwidth Allocation Remote #1 assigned more burst time slots (higher data rate) Figure 2: Example TDMA Burst Assignments L time L time return link frequency band Remote #1 Return Link frequency Dynamic Bandwidth Allocation Remote #1 assigned more bandwidth (higher data rate) frequency Figure 3: Example dscpc Bandwidth Assignments Page 3 of 6

4 contribution to reducing overall efficiency will be neglected. With these assumptions, the difference in return link bandwidth efficiency can be written as: BW eff (dscpc) = F OH F LM F AQ BW eff (TDMA) Where F OH, F LM, and F AQ are the ratios of dscpc bandwidth efficiency to TDMA bandwidth efficiency due to burst overhead, link margin, and SCPC carrier acquisition factors respectively. Burst Overhead Factor: The exact amount of overhead required for TDMA and dscpc systems depends on several factors. However, typical scenarios can be analyzed to give approximate comparisons. The TDMA system is based on DVB-RCS, which most commonly uses an ATM mechanism for fragmentation and reassembly of packets. ATM cells are composed of a 5 byte header, followed by a 48 byte payload field, hence the overhead factor due to ATM overhead is F FRAG = 48/53 = DVB-RCS also has an option to use a satellite access control (SAC) field, which adds 2-4 bytes of additional overhead. For analysis purposes, it will be assumed that the SAC field is not present. burst overhead depends on implementation, but can be estimated for a typical mode. For example, an ATM burst using QPSK ¾ modulation and coding requires ceiling[53*8*4/3] / 2 = 283 symbols. Typical preamble + guard time lengths for this type of burst are approximately 32 symbols or greater so the overhead factor due to burst overhead is at least 283/(283+32) = Hence the total fragmentation plus burst overhead for the system is at least OH = (0.906)(0.898) = 0.81 SCPC links do not have burst overhead. However, overhead is required to insert and remove data packets into/from the continuous data stream. SCPC modems typically use High-Level Data Link (HDLC) protocol flags in order to accomplish this task. This adds approximately 1 additional byte of overhead to each packet. IP packets range from 46 to 1500 bytes. Which means that the largest possible HDLC overhead is approximately OH SCPC(max) = 46/47 = Hence the ratio of the SCPC bandwidth efficiency to the TDMA bandwidth efficiency due to overhead factors (F OH ) is at least OHSCPC 0.98 FOH = = = 1.2 OH 0.81 Link Margin Factor: Use of larger block sizes coupled with the continuous nature of SCPC processing gives dscpc an inherent advantage over TDMA. The performance specification is compared to DVB-S2 in Table 1. Values for the DVB-S2 performance specification are calculated for all DVB-S2 modes in Table 3. Table 1: dscpc vs. Processing Advantage [db] Mode FEC BW Eff. Specification Threshold E s /N 0 (db) DVB-S2 Threshold E s /N 0 (db) Diff (db) QPSK 1/ QPSK 2/ QPSK 3/ For the example modes shown in Table 1, the smallest processing advantage of DVB-S2 based dscpc approach relative to is 1.2 db. An even more important link margin advantage is due to channel utilization. As noted previously, a TDMA based system must close the link budget for the entire shared channel, regardless of the percentage utilization it is granted in that channel. Conversely, dscpc based system need only close the link budget for the actual capacity assigned. Table 2 shows the dscpc advantage as a function of % utilization, and the total link margin advantage considering % utilization and the worst-case processing advantage from Table 1. Table 2: dscpc vs. - Total LM Advantage[dB] Individual Terminal Utilization [%] LM Util factor (db) Processing Factor (db) (QPSK 1/2) Total LM advantage for dscpc (db) 50% % % The total link margin advantage enables dscpc systems to use more efficient modulation and FEC combinations. For example, if a system is using a QPSK ½ based return link (3.2 db required E s /N o ), then for terminal channel utilizations of 50, 25, and 10%, a dscpc based could equivalently enjoy E s /N o values of = 6.2 db, = 9.2 db and = 13.2 db respectively. Table 3 can then be used to obtain the appropriate dscpc bandwidth efficiencies for a dscpc system using DVB-S2. Page 4 of 6

5 Table 3: Example DVB-S2 Performance Specification Mod Code Rate Spectral Efficiency Ideal Es/No (64k blocks) [db] Imp. Loss [db] Margin for spec Possible Es/No [spec] (QEF) Eb/No [db] (QEF) QPSK 1/ QPSK 1/ QPSK 2/ QPSK 1/ QPSK 3/ QPSK 2/ QPSK 3/ QPSK 4/ QPSK 5/ QPSK 8/ QPSK 9/ PSK 3/ PSK 2/ PSK 3/ PSK 5/ PSK 8/ PSK 9/ APSK 2/ APSK 3/ APSK 4/ APSK 5/ APSK 8/ APSK 9/ APSK 3/ APSK 4/ APSK 5/ APSK 8/ APSK 9/ Table 4 shows the results of this exercise, and calculates examples of the net bandwidth efficiency improvement (F LM ) for the example % utilizations, assuming that the return link is bandwidth limited on the satellite. Table 4: Link Margin Scale Factor F LM Individual Terminal Utilization [%] Mode BW Eff. Eq. dscpc DVB-S2 Mode DVB-S2 BW Eff. 50% QPSK 1/2 1.0 QPSK 5/ % QPSK 1/ PSK 3/ % QPSK 1/ APSK 5/ F LM dscpc Acquisition Factor: When a dscpc system reassigns capacity amongst the remotes, in general in addition to changing the modulation and code rate, the symbol rate and center frequency are also modified (see Figure 3). This results in the need for the Hub demodulators to reacquire the reconfigured return link signal. Data packets can be buffered to eliminate packet loss during the transition. Also, execution of configuration commands at the hub and remote can be timed so as to minimize outage time. However, some outage time is inherent in the dscpc approach (TDMA systems suffer no similar outage). In order to gauge the impact on bandwidth efficiency, the case where a command to reconfigure capacity is Page 5 of 6

6 sent to all terminals at a fixed frequency (or time interval) will be considered. If a command is sent every T update seconds and causes an outage on all terminals of T outage, the overhead factor associated with dscpc acquisition is simply: Toutage Tupdate - Toutage F AQ = 1- = Tupdate Tupdate Note that unlike the other factors F OH and F LM, F AQ is less than unity, indicating a disadvantage for the dscpc approach in this factor. The amount of outage time is implementation dependant, and depends on the data rate of the return link signal (the higher the data rate, the shorter the required acquisition and outage time). For return links of several Mbps down to 100 kbps, typical required outage times are in the range of 20 ms to around 100 ms. In practice, the frequency of the bandwidth reallocation commands would never be less than the round-trip satellite delay of roughly 500 ms, since sending commands more often than this would imply that changes in the allocated data rates were being made before having a chance to observe the impact of the previous change, or obtain any feedback from the remotes. The minimum practical rate of the command to update the data rate allocations is roughly once per second. In practical usage, update commands could be sent significantly less frequently. Using values from the typical ranges that gives the largest penalty for dscpc results in an estimated value for F AQ of F AQ = = Table 5 shows the approximate net scaling factor, F Total = F OH F LM F AQ and the percentage return link bandwidth reduction enabled by the dscpc approach vs.. Table 5 shows that using realistic and conservative assumptions, the same return link capacity can be provided by a dynamic SCPC system with 44% to 72% less satellite bandwidth compared to a Joint IP Modem based TDMA system Conclusions The proposed dynamic SCPC and TDMA systems use identical forward links based on DVB-S2, hence the bandwidth efficiency in the forward link is identical for each of these approaches. It has been shown that the return link bandwidth efficiency of the dscpc approach is dramatically improved relative to the TDMA system. Using realistic example parameters, the return link satellite bandwidth required by dscpc was reduced by 44% to 72% relative to a TDMA system while providing the same capacity, with equal or better link margins. The reasons for the improvement are fundamental, and result from inherent advantages in (a) lower overhead of the SCPC waveform relative to a burst system, (b) the performance advantages of continuous vs. burst the processing and (c) the requirement of TDMA systems to close the link budget for the shared channel, while on average using less than the full capacity of the channel. The reacquisition disadvantage for the dscpc approach was calculated to be far smaller that the combination of dscpc advantages. Hence the dynamic SCPC approach presents an opportunity to provide the same dynamic bandwidth allocation functionality of TDMA based systems, while dramatically reducing required return link satellite bandwidth. This capability has the potential to provide significant cost savings and/or capacity improvements to the DoD. Table 5: Net RL Improvement: dscpc vs. TDMA Individual Terminal Util [%] F OH F LM F AQ F Total Reduction dscpc BW [%] 50% % % Page 6 of 6