SWITCHED INFINITY: SUPPORTING AN INFINITE HD LINEUP WITH SDV First Presented at the SCTE Cable-Tec Expo 2010 John Civiletto, Executive Director of Platform Architecture. Cox Communications Ludovic Milin, Director of Product Management SDV, BigBand Networks Introduction Switched Infinity is a state where new content can simply be added to the Switched Digital Video (SDV) lineup with no impact to bandwidth utilization. This paper will present empirical data and learnings based on a long term SDV field trial conducted by Cox Communications to validate that cable networks can support a nearly infinite number of channels while mitigating pressure on bandwidth utilization. Simulations based on the collected viewership data during the trial showed that by appropriately reducing the number of tuners per Service Group (SG), virtually unlimited bandwidth becomes available. In other words, the lineup size, and more specifically the number of HD channels, could be further increased without requiring additional spectrum. The field trial encompassed 23 SDV Quadrature Amplitude Modulations (QAMs) deployed on four service groups in a select Cox market. SDV provided enough bandwidth to support the equivalent of 43 broadcast QAMs and a lineup of 268 standard definition (SD) and 75 high definition (HD) channels without sacrificing video quality across a total of 34 QAMs. In this scenario, SG size was significantly larger than the current SDV standards at 700-1100 tuners per SG. An advanced SDV monitoring tool was used during the trial to report on the system utilization, performance and end user experience. Migration to HD MPEG-4 and Switched Unicast delivery are also simulated from the existing viewership data.
SWITCHED INFINITY: SUPPORTING AN INFINITE HD LINEUP WITH SDV SDV Overview and Benefits Unlike legacy broadcast systems that broadcast all programs to all subscribers all the time, in an SDV network, video programming is only transmitted to subscribers within a network node, or SG that specifically request such programming. In an SDV network, cable spectrum is not viewed as a static collection of channels that transmits all channels at all times, but rather as a dynamic pool of digital transmission resources that, at any moment, could be serving any broadcast programming from any source. If a subscriber wants to watch a program that is currently being delivered to other subscribers within the same node, the new viewer simply joins the existing switched session. As a result, no additional capacity is consumed by the incremental subscriber. SDV architecture has proven to drive savings in bandwidth consumption and improved efficiencies, allowing cable operators to increase revenues via expanded program offerings and expanded content as well as the delivery of personalized, high-margin services encompassing narrowcast VOD content, interactive television, or other digital services developed and introduced in the future. Current State of Typical SDV Deployments SDV deployments have evolved significantly over the last four years from large SGs with 4 to 8 QAMs per SG carrying only long tail and niche SD channels to 16 to 24 QAMs per 500 to 700 tuners SG and a mix of SD and HD channels representing more than half of the total lineup available to subscribers. A key driver for this phenomenon is the need to reclaim more bandwidth to reach the 100 HD channels mark. These systems typically reach 3:1 or better over-subscription. Over-subscription in this case is the ratio of the amount of bandwidth required in broadcast to that used in SDV. Getting to Infinite Bandwidth/Unlimited Lineup The goal now is to reach at least 200 HD channels, as HD tuner penetration has reached more than 50 percent, service providers are offering many of their channels in HD and continue to expand this offering as more content becomes available. In parallel, time shifted television, HD video on demand (VOD), DOCSIS 3.0 high speed data (HSD) service and recently 3D HD, are putting additional pressure on the available spectrum. The goal for operators with SDV is to be able to add new channels at will, either on the SDV lineup assuming none of the new channels introduced are very popular channels for which there is no over-subscription gain with SDV, or moving channels between broadcast and SDV, and reach a state where new content can simply be added to the lineup with no impact to bandwidth utilization. As shown in Figure 1, this state is called switched infinity where, for a given SG size and number of SDV QAMs combination, the over-subscription rate is controlled by the demand side rather than the supply side, e.g. no matter what the size of the lineup offered, there can only be so many channels watched simultaneously in any given SG. In practice, operators are not necessarily looking to achieve this theoretical infinite capacity target, they are essentially looking to support their channel lineup on the available spectrum while meeting their internal quality of service targets, namely maximum allowed rate of blocking events. Field Trial Description and Goals In 2010, BigBand Networks and Cox conducted a field trial in four SGs where the goal was to switch as many channels as possible and determine the viability of getting the network to a switched infinity state. This paper will show that the observed over-subscription was actually higher because the SDV QAMs were never fully utilized. A main requirement of the trial was that all possible channels had to Figure 1: Switched Infinity Concept Idealized Curve be switched from day one while ensuring that there were no blocking events or noticeable impact to the subscriber viewing experience. The actual SG sizes of 700 tuners (identified as SG 1710 and 1833) and 1100 tuners (identified as SG 1650 and 1694) were larger than typically found in SDV deployments. The reasoning behind this was to consider the most difficult scenario where large node sizes prevent easy and economical SG split. The channel lineup was split as follows between broadcast and SDV: 15 Broadcast QAM channels 76 SD and 15 HD 23 SDV QAM channels 192 SD at 3.8 Mbps Total 60 HD at 15.6 Mbps, higher bitrate than typical to enhance overall Video Quality 268 SD and 75 HD in 38 QAM channels (228 MHz) An equivalent lineup would require at least 55 QAMs (330 MHz) if carried all in broadcast mode. This represented only a modest global over-subscription ratio of 1.75. The selected broadcast channels included very popular SD and HD content programming as well as those required to be broadcast due to the local system configuration. What is interesting to note is for this latter category, additional bandwidth savings would have been realized had these programs been carried over the SDV tier. Field Trial Results The field trial was monitored on a near-real time basis using a Switched Video Analysis (SVA) tool to track SDV QAM utilization, overall SDV performance and subscriber experience, and broadcast and SDV viewership data. Figure 2 provides overall channel popularity for the Broadcast channels and Figure 3 highlights the overall channel popularity for SDV channels in the field trial that was conducted. Only two HD channels appear in the top 25 most popular broadcast channels, and only 6 show up in the top 50. This is typical of most SDV systems analyzed to date, despite HD tuner penetration being about 50 percent. The top 60 broadcast channels have at least one viewer in all SGs over 80 percent of the time and the top 37 channels have at least one viewer in all SGs monitored during the trial at least 90 percent of the time. Additionally, since these channels are always on at peak time (7 p.m.-11 p.m.), there would be no bandwidth gain if they were moved to the SDV tier.
The 25 least popular broadcast channels are watched less than 50 percent of the time and should be moved to SDV reducing the overall broadcast QAM requirements. Figure 2: Broadcast Channels On Time Ranking As further shown in Figure 3, SDV channels popularity was as expected with only a handful of channels being watched more than 70 percent of the time and the others displaying very typical long tail behavior. The four channels above 70 percent ON time, which include one HD channel, may be moved back to broadcast, which combined with moving the 25 least popular broadcast channels to SDV would lead to an additional overall spectrum savings of one QAM. Figure 5: Historical View Maximum Number of Viewed SDV Channels per SG The maximum occupied bitrate concentration ratio or percentage of SDV QAM bandwidth occupied at peak time per SG, shown in Figure 6 tracked the maximum number of watched SDV channels noted in Figure 5 above. Figure 3: SDV Channels On Time Ranking The maximum number of concurrent tuners per SG tuned to SDV channels is shown on Figure 4. One can easily observe the typical daily variations peaking between 7 and 11 p.m. as well as the weekly spikes on Saturdays and Sundays. The maximum number of active tuners correlated directly with the total number of tuners in the SG. Figure 6: Historical View of the SDV QAM Utilization per SG The number of channel change requests handled by the SDV system is presented in Figure 7. It varied by over an order of magnitude during the course of the day, peaking at about 1.5 channel changes per second per SG. All SGs peak at the same time compounding the load on the SDV session manager. Figure 4: Historical View of the Maximum Number of Concurrent SDV Viewers per SG The total number of concurrently watched SDV channels in each SG is presented in Figure 5. When viewed in conjunction with the per SG bandwidth utilization detailed in Figure 6, one can see that maximum number of channels viewed at any given time does not exceed bandwidth allocated to that SG. Figure 7: Historical View of Number of Channel Change Request Per Second Per SG
Summary statistics in Table 1 point to the following results in the field trial: Less than 70 percent of the tuners in a SG were concurrently active; Less than 18 percent of the tuners per SG were concurrently watching SDV channels; Only 30 to 40 percent of the SDV channels were watched simultaneously, larger SGs lead to slightly more channels watched but the difference tapered out for large SG sizes and were likely influenced by local demographics; Almost all broadcast channels were watched at peak time, and overall 45 to 55 percent of the entire lineup was watched simultaneously; and As the number of tuners per SG increased, so did the average number of tuners per watched channel, showing a strong benefit of the multicast approach for large SG sizes and the bandwidth challenge of migrating to unicast in that scenario. Conversely, smaller SGs are more aligned with a unicast delivery scenario without any modification required to the SDV algorithms. Max number of concurrent SDV Tuners Max number of concurrent Broadcast + SDV Tuners Max number of simultaneously viewed SDV channels 132 18% 500 70% 87 35% 127 18% 476 68% 81 32% 193 17% 769 69% 105 42% 187 17% 731 66% 99 39% SDV Multicast gain 1.5:1 1.6:1 1.8:1 1.9:1 Max number of simultaneously viewed Broadcast + SDV channels 159 46% 152 44% 185 54% 174 51% Broadcast + SDV Multicast gain 3.1:1 3.1:1 4.1:1 4.2:1 Table 1: Summary Metrics, Maximum Number of Concurrent Tuners and Channels Watched With all other parameters being equal, the number of tuners per SG is the variable with the most effect on the efficiency and bandwidth gains of the SDV system. As shown in Table 2, going from 700 to 1100 tuners per SGs required up to 4 more QAMs (23% more spectrum) to support the same SDV lineup. Along the same lines, the measured over-subscription dropped from about 2.7:1 at 700 tuners per SG to less than 2:1 at 1100 tuners per SG. Maximum bandwidth utilization 23 SDV QAMs 78% 78% 95% 91% Measured over-subscription ratio 2.7:1 2.7:1 2.0:1 2.1:1 Number of SDV QAMs required to support SDV lineup 18 18 22 21 Table 2: Measured Over-Subscription Ratios and Minimum Number of QAMs Required Table 3 summarizes the total number of channel change requests per SG during the 7.5 weeks of the trial. It also gives an idea of the scalability required for the SDV session manager, edge device, and the data warehousing and analysis package in order to handle all these events without any latency or errors. Broadcast channels only make up about one third of the total number of channels in the lineup but typically generate over half of the channel change requests. Total number of SDV channel change requests Total number of Broadcast + SDV channel change requests 762,011 531,481 920,666 934,320 1,414,826 1,117,166 1,837,256 1,852,451 Table 3: Total Number of Channel Change Requests Recorded in the Field Trial Taking the Next Step: Simulations Based on Collected Viewership Data Looking at Ways to Further Improve Bandwidth Reclamation The software tool used in this trial provided a simulation engine that allowed Cox to test the impact of changing some of the SDV system configurations. These configurations included the number of QAMs per SG, SG splits, channel bitrate, the SDV lineup by adding or removing channels, and replays of the collected viewership data to estimate the QAM utilization ratio per SG as well as the number of blocking events if the system is over-utilized. Using this simulation engine, a few use case scenarios to illustrate how additional bandwidth could have been reclaimed are highlighted in Table 4. Reducing SG size through a SG split produces a 5 to 6 SDV QAM per SG saving. This is a quick and easy change in cases where node de-combining is all that is required but would be more involved and expensive if node splits had to be considered. Initial number of tuners 716 704 1114 1108 Number of tuners after split 358 352 557 554 Initial Number of SDV QAMs 18 18 22 21 Number of SDV QAMs 13 13 16 16 Over-subscription 3.3:1 3.3:1 2.7:1 2.7:1 Table 4: Simulation Results for Number of SDV QAMs Required after SG Split Migrating half of the HD channels in the SDV lineup to MPEG-4 at 8Mbps is another alternative, assuming the deployed set-top boxes (STBs) already support MPEG-4. In the field trial conducted, it could have resulted in a saving of 1 to 2 QAMs per SG as shown in Table 5. Number of tuners after split 18 18 22 21 Number of SDV QAMs 17 17 20 19 New over-subscription 2.2:1 2.2:1 1.8:1 1.8:1 Table 5: Simulation Results for Number of SDV QAMs Required after Migrating Half of the Long Tail Content to MPEG-4
Migrating all of the HD channels on SDV to MPEG-4 would have provided another 4 to 6 QAMs saving as shown in Table 6, albeit at a significant re-encoding cost. Cox utilizes a centralized architecture which allows these costs to be leveraged by multiple systems making this less of an issue. As MPEG-4 is being deployed and until all legacy HD STBs that support only MPEG-2 are cycled out, content would have to be sent in both MPEG-2 and MPEG-4 format, which would limit bandwidth savings, particularly for larger SGs where the probability of having two viewers on the same session increases significantly and may require sending both MPEG-2 and MPEG-4 streams simultaneously to cater different generation STBs. This limitation, however, is easy to address with a smart SDV session manager that can recognize STB capabilities and adapt the source being streamed accordingly. Also as the SG sizes decrease, the likelihood of more than one tuner per active SDV session decreases significantly as illustrated by the data in Table 1 above. Number of tuners after split 18 18 22 21 Number of SDV QAMs 12 12 25 15 New over-subscription 2.6:1 2.6:1 2:1 2:1 Acronym List High Definition (HD) High Speed Data (HSD) Quadrature Amplitude Modulation (QAM) Service Group (SG) Standard Definition (SD) Switched Digital Video (SDV) Set-Top Boxes (STBs) Video on Demand (VOD) Table 6: Simulation Results for Number of SDV QAMs Required after Migrating All of the Long Tail Content to MPEG-4 As noted earlier when presenting the average number of tuners per active SDV session, switched unicast using these large SGs where SDV channels have an average of 1.5 to 2 viewers at peak time would require in excess of 45 to 50 SDV QAMs per SG, typically more than is available on the spectrum. This can be effectively addressed by decreasing the average SG size by a factor of 3 or 4 assuming node de-combining is an available option. Conclusion This field trial demonstrated that using SDV with a large number of QAMs per SG and all but the most popular channels on the SDV line-up, is in fact a valid solution for significant bandwidth reclamation. Even for large SG sizes and large and complex SD and HD lineup, over-subscription ratios of 2:1, even approaching 3:1 for smaller SGs, can still be attained with no impact to the overall subscriber experience. Future trials will target smaller SG sizes and additional HD content as it becomes available. Simulations based on actual viewership data demonstrated that for example, splitting the current SGs or migrating HD channels to MPEG-4, the number of SDV QAMs required per SG would significantly decrease and therefore provide spectrum for additional channels. It is expected that this would be further assisted by the decreasing and competing popularity of any new channel addition when the linear lineup reaches over 350-400 channels. With the saved bandwidth achieved by a switched infinity implementation, operators will be able to quickly respond to competitive pressures by introducing new and revenue generating services. With virtually unlimited bandwidth, service providers have a means to rapidly introduce additional SD and HD programming, personalized services and IP video delivery to multiple devices and associated advertising all while leveraging existing SDV and broadcast infrastructure and without having to add bandwidth capacity. United States Corporate Headquarters 475 Broadway Street Redwood City, CA 94063 United States phone: 1.650.995.5000 fax: 1.650.995.0060 bigbandnet.com 2010 BigBand Networks, Inc. BigBand Networks brand and product names are service marks, trademarks or registered trademarks of BigBand Networks, Inc. in the United States and other countries. All other marks are the property of their respective owners. 1417-1010