Studios' watermarking agenda could benefit cable...
...But MSOs should get involved now
Hollywood studios may soon use digital watermarking as an additional layer of protection in the distribution of high-quality, high-value video content. Forensic steganography, as the technique is formally called, will be fundamental to an important new content opportunity for cable. It also poses new challenges.
Already widely used with applications in distribution monitoring, copyright notification and forensic tracking, movie studios and record labels currently use watermarking to keep track of content through the production process prior to release as protection against internal theft.
Watermarking is not a replacement for digital rights management (DRM), however. Where DRM techniques are designed to control the ability to copy a video file, watermarking is designed to make copies traceable to their sources.
The film industry is interested in watermarking because DRM alone is not solving the problem. The industry is losing $6.1 billion dollars a year to piracy.1
Engaging in watermarking is in the cable industry’s interest, because it will encourage content owners to release the most desirable content to cable outlets earlier, when the content is still most valuable.
Figure 1: Substitutive watermarking. Source: Thomson
Why consider watermarking now? Cable industry involvement is imperative at this stage to insure that whatever conditional access, forensic, and rights-management protocols content owners decide to employ, those methods do not compromise content quality, raise barriers to the legitimate subscriber, or favor other means of distribution to cable’s detriment.
Though the digital data carried by a watermark can include virtually any information, the primary application of robust (indelible) watermarking is event-specific, containing information about the content, its sources, and means of acquisition. In the cable environment, watermarking would ideally be integrated at source, intermediate, and terminal nodes – the terminal being the set-top box.
Therein are many of the challenges.
With ticket sales and DVD profit margins decreasing, studios are exploring ways to drive revenue by better-addressing a changing market. The studio initiative bodes well for cable because of the cable industry’s history as an early adopter of wholly-electronic content distribution and its excellent record of securing that content over a quarter century.
Now VOD plays center stage. To many studio executives, the upward trend in VOD buy rates suggests the time has come to leverage new content protection technology and build upon the appeal and economic advantages of electronic file-based feature film content distributed via cable.
A film’s greatest appeal is when it’s new. Historically, nearly every popular feature film is first exhibited in theaters. Some months later, that same title is released to home video. But distribution to both these markets is dependent upon transporting indelible physical media such as theatrical release prints and digital versatile disks (DVD) to their geographical regions of exhibition, and doing so prior to a feature’s release date in that market.
The costly distribution infrastructure required to manufacture, duplicate, and move such physical media around the country (and subsequently, the globe) has become increasingly prohibitive. Further, the wide distribution – generally consistent with a film’s anticipated popularity – exposes the physical media to illegal interception. As such, the expense of moving and protecting such media is directly proportional to market penetration.
Electronic distribution, however, particularly via C-band commercial satellites feeding closed-circuit cable plant and its entire physical distribution media, is extremely reusable. Despite such an advantage, VOD and PPV – two examples of wholly electronic distribution – do not see new release feature content until some 40 days after home video (DVD) release. But now, the release windows for DVD and VOD are converging, are in fact trending toward concurrence. There’s also a growing consensus that early release transactional content be distributed cross-platform. That means Internet and mobile as well as digital cable. Few distribution platforms can deliver on so broad a mandate as effectively as digital cable, and with superior quality.
Figure 2: Source to sink architecture.
Security and more security
While early release means more value, it also mandates higher standards of protection.
It’s been established that some 72 percent of all domestic feature film piracy occurs in the window starting just prior to theatrical release and ending shortly after DVD release.
VOD shouldn’t afford pirates yet an additional early interception opportunity.
Cable already protects and secures content, in large part through encryption. Just look at the performance history. A mere 5 percent of all domestic content interception is perpetrated during the VOD/PPV window, and much of that through the analog hole or residual DVD duplication. Virtually no interception occurs through defeat of the ciphers.
But should content be released on VOD/PPV closer to the DVD release date, that’s just incentive for pirates to redouble their efforts, and sooner or later, some interception is inevitable.
Here’s where watermarking takes center stage. Watermarking, for the first time, accords direct physical evidence to non-physical media. It eliminates the anonymity inherent to the theft of non-physical media such as MPEG files, thus making illegal interception and subsequent distribution a much riskier business.
The traceability provided by watermarks is the best means we know to track content from source to sink, identify the interception point, and lead to its elimination and the subsequent mitigation of loss.
The robust watermark must be invisible on display systems. It must survive compression, encryption, transcoding and error control with no reduction of video fidelity discernable by virtue of the mark. These conditions must persist from low resolution copies to 1920 x 1080 rendition. Further, it must survive the permutations of piracy such as digital to analog and scan conversion.
In an effort to test and evaluate watermarking systems to these extremes, TVN Entertainment began a watermark integration effort with its primary signal security supplier, Widevine Technologies.3
The program’s aim was to evaluate the suitability of several forensic systems to satellite and cable distribution, and anticipate implications to our own network, as well as those of our cable affiliates.
The tests were designed and carried out as follows:
Using 128 bit AES encryption, MPEG-2 compression levels between full 3.75 Mbps (CableLabs) and an extremely low (500 kbps) sampling rate, the embedded watermark was repeatedly recovered from any group of 2,000 concatenated frames (approximately 1 minute) of reproduced video to which it was applied. On full 50 Mbps contribution quality files, at full D-1 resolution, the watermark remained invisible to human subjects viewing 32-inch CRT monitors. In the most extreme test of content-permutation, watermarked video was MPEG-2 compressed at 3.75 Mbps, projected onto a wall, cam corded directly from that wall, and re-encoded to H.264, at a bit rate of 500 kbps.
The detector was able to recover the watermark from this minimally-coherent video in all instances within the first minute (1,800 frames) of the captured file.
These conditions and results were met by all of the proponent systems tested. Among these were technologies provided by Thomson, Dolby’s Cinea, and Philips incorporated into Widevine’s Mensor server at TVN’s Burbank N.O.C.
The tested systems all insert marks through a discrete cosine transform (DCT). By replacing less significant frequency coefficients (those representing samples indistinguishable from random noise in the transformed signal) with discrete data, and doing so over concatenated frames, a 64 bit watermark is formed.
Given the substitutive nature of this data, aggregation of marks does not add substantively to the recovered file’s size or effectively change its payload. In the Mensor system, the tested file sizes increased less than 1 percent by virtue of the insertion data, but zero by virtue of the watermark payload.
By pseudo-randomly replacing frequency coefficients within spatially specific pixel basis-functions across a spread frequency spectrum referenced to the DCT wave table and a secret key, the watermark is inserted securely and invisibly.
The technique – while allowing the inserter to add information that is virtually invisible in video renditions of the highest fidelity – affords recovery of that data from even the lowest-quality coherent copies.
When applied to protection of digital entertainment content for general market distribution, all logical intercept points across the distribution chain could embed code which identifies that location specifically.
In a cable environment, a watermark capable of delineating manufacturer, unique system and STB identifiers, and time stamps accurate to one minute in a four year epoch, is a reasonable requirement. Shared across these tasks, the 64 bit payload could still yield a potential one trillion unique permutations.
As with every potential intercept node from acquisition to exhibition, there are watermark integration issues unique to the cable plant. These are exacerbated by the highly asymmetric nature of the distribution chain.
For example, while it might be economically feasible to integrate a single watermark analysis and insertion system to a solitary one-million dollar SAN device at TVN’s operations center (the source), these same economics hardly accrue to placing individual embedders into a few million low-cost set-top boxes (the sinks) at the exhibition points.
Nonetheless, the ability to trace an illegal copy back to a specific subscriber remains the grail. Here is where the issue of integration becomes specific to the cable system physical plant and that most numerous and often least extensible network asset: the set-top box.
It is at the exhibition point – the STB – that session based, and thus precise, forensic traceability is most valuable. This implies (but does not mandate) an analysis and insertion capability at the set-top box or receiver itself. Further, such a mark would ideally include the terminal device’s unique identifier and timestamp. However, the challenges to watermark analysis and insertion of even such fundamental data, within the processing capabilities of today’s digital set-tops, are as much economic as they are technical. While advances are already apparent, traditional watermarking technologies have not been developed with the cable TV market as a first adopter.
Motion picture camera acquisition and digital cinema exhibition dominated early design philosophies. This high resolution, symmetrical production, distribution, exhibition paradigm, as evidenced by the Digital Cinema Initiative’s (DCI) specifications, is vastly different both technically and economically from that required and practicable for cable television use and security.2
While the fidelity and security of such techniques is paramount, analysis, processing, and insertion of even the 21 bit watermark payload specified by DCI has been shown to require 70 to 100 MIPS of computational power. Thus, while suitable for a terminal device such as a digital cinema projector, these traditional watermarking techniques place heavy ancillary processing demands upon both current and legacy set-top box architectures: necessarily among the most numerous, simplest and least extensible elements in a highly asymmetrical cable TV architecture. Further, when allocated across the tasks of manufacturer and set-top identification, and time stamp insertion, even 21 bits falls short of yielding the permutations required in a multiple-million STB universe characteristic of cable TV or even of a single large MSO.
Things are changing, though, and quickly. For despite these high hurdles, security vendors are now concentrating on the emerging cable TV and IPTV markets. An increasing number have already, or are planning to release watermarking systems suitable to STB applications. The Digital Watermarking Alliance (DWA) provides an excellent resource for timely information on systems, vendors, and status.
Among Digital Watermarking Alliance members, at least two in the TVN/Widevine early-adopter program have announced watermark systems that are both technically and economically suitable to integration with many of today’s digital cable set-tops while meeting expectations for both fidelity and security indicated by the content owners.
The thornier issue of session marking in systems using legacy STBs that have neither ancillary processing capabilities nor available memory is being addressed by application designers thinking outside the box – literally.
Here’s where the Widevine Mensor and TVN approach is unique and highly applicable to the asymmetric cable TV network architecture. By relying on the processing and computational power of the source embedder at TVN, this technique imposes no demands upon set-tops whatsoever. It is thus applicable to not only the most feature-rich STBs, but the simplest digital terminals as well, and everything in between.
Simply stated, the unicast nature of VOD allows for session-based marking without relying on a terminal device CPU or an incremental client application, however light. By inserting an indelible mark at the aggregator, in this case TVN’s operations center, and modifying that mark uniquely and securely as it moves from source, through the intermediate and terminal nodes, the need for a process or client application to do computationally-rigorous analysis at the headend or STB can be completely obviated.
Widevine Technologies has developed a bridge inserter that installs in a one rack unit housing at the headend and takes advantage of the secure watermark insertion and analysis done at the source node (TVN).
By including secure metadata in each unicast stream post the VOD server and prior to the STB, robust session marking can be accomplished which would mark each instance of content access requested by a specific STB. The STB mark would be a unique variant of the source watermark inserted at TVN, would develop through little more than a byte offset calculated and imposed at the bridge, and would include each unique STB identifier and a time stamp of content access.
Working with Widevine and TVN, at least one major MSO is already integrating this very promising and highly secure technique.
Some level of agreement within the cable industry in consultation with content and equipment suppliers as to what the identifier information generated by set-top or session-based watermarking should contain would increase effectiveness and simplify identification and recovery of the marks while minimizing the number of permutations a system need compute.
Though but a single approach to cable TV watermarking has been described here, there is no doubting the mandate and its economic potential. Neither is there any doubt about the resolve of feature film studios in their search for ways to securely expand the market for file-based content.
Watermarking is not the only security enhancement under consideration by content owners, of course, but with discussions underway between the studios and watermarking vendors as well as device manufacturers, and through deployment by networks such as TVN, the process has already found its way into the cable TV realm.
Timely action will equip our industry with the technical support systems that provide the studios and content owners the confidence they are seeking as we all look for ways to boost on-demand revenues, and do so to mutual advantage. As it has so often in the past, cable could very likely be the proximate means for successful realization of these very practical ambitions.
Given that precedent, TVN and its security partner, Widevine, have taken what we believe are critical first steps.4
1. Motion Picture Association of America, “U.S. Theatrical Market Statistics Report,” 2006.
2. DCI Specification V1.0, Digital Cinema Initiatives LLC, July 2005, http://www.dcimovies.com/DCI_Digital_Cinema_System_Spec_v1.pdf
4. “Scaling Watermarking For IPTV,” Reza Rassool, Chief Engineer, Widevine.