This can only be done with the help of some technique that uniquely identifies the source of a pirated stream, which could be a set top box associated with a traditional pay TV service or, as increasingly the case, an IP stream within an OTT service. Indeed the growing availability of live sports via OTT services is itself a cause of rising piracy, while the advent of 4K Ultra HD content will provide a further substantial boost. The combination of high resolution 4K cameras and displays will allow content thieves to camcord digital copies of content with almost no loss in quality, fit for restreaming in at least full HD quality.

There is a firm consensus now that some form of forensic watermarking involving insertion of supposedly indelible marks in the content will be necessary to enable sufficiently rapid identification of pirated live streams being redistributed instantly. The watermarking movement gained a huge boost when MovieLabs representing the big studios in April 2014 mandated the technology as a key pillar for protecting 4K content, but the case is if anything even stronger for live sports.

Vendors of such watermarking systems were out in force at NAB not just peddling their wares but also pontificating over the scale that live stream piracy had already reached. One such vendor, Dutch based Civolution, intoned that Internet streaming piracy had scaled unprecedented heights and was disrupting the growth of pay TV and OTT services. Civolution highlighted the multiplicity of illegitimate sources on the Internet for premium sports content available to view without paying a subscription to the rights owner or an authorized distributor. These are fast becoming the predominant sources of illegitimate live sports content, taking over from longer established forms of piracy such as card cloning, or somewhat more recently control word sharing.

The exact extent of live sports piracy is hard to divine, but there is plenty of indirect data highlighting its growth and impact. Some sports distributors have privately admitted that up to half the people accessing their content had not paid for it, with a more common estimate being 25% to 30%, but often the total is unknown. Another content security firm specializing in watermarking, UK based Friend MTS, measured live streaming piracy of one specific event, the football match between Spain’s two top teams Real Madrid and Barcelona match on 25th October 2014. The firm used its digital fingerprinting and network forensics technology to identify 262 unique pirated streams spanning 402 unique servers, 57 hosting providers and 74 hosting sites. Of the unique streams identified, five were serviced by CDN (Content Distribution Network) provider Akamai with virtually unlimited viewer capacity. These five streams in turn were embedded on sites with a total estimated traffic of 75.7 million unique visitors per month, giving some idea of the piracy potential.

As Friend MTS has also noted, such piracy is not just depriving operators of subscription revenue but is also sucking up advertising income. There is relatively recent data on this, with the Digital Citizens Alliance, a non-profit organization working on Internet safety, estimating that almost $230 million was made in 2013 from advertising on pirate sites, with the 30 largest ones taking more than $4 million dollars a year each. Adverts featured many leading brands that did not deliberately seek out such sites but were drawn by the growing phenomenon of automated programmatic buying that matches products or services with online audiences wherever they are. This will be a growing potential source of income for pirates since automated buying is rapidly increasing, rising by 50% year on year to reach $21 billion in 2014, according to media buying unit Magna Global. By 2017 it will account for 87% of all online ads placed as this becomes the only practical way of dealing with the volumes involved.

Such factors are stimulating live sports piracy and Civolution argued at NAB that distributors urgently need the capability to take down pirate streams while the game is still on. Friend MTS goes further by arguing that this has to be done within minutes of the game starting to stem significant business damage.

The first part of the problem, detecting Internet streams of live events, is well on the way to being solved. Web monitoring companies are increasingly able to locate internet streams of live events, but the bigger challenge comes next, to identify quickly which ones are pirated and which are legitimate. Distributors do not want to throw out the baby with the bathwater and cut off paying subscribers, who would then desert to a rival operator. This is where watermarking comes in by combining with network forensic techniques to identify content in the process of redistribution and trace it back to the originating stream or source. The watermark will usually be inserted in the content at the origin server or headend so that it is linked to a given subscriber, so that all redistributed streams containing that mark can be associated with that account.

This raises several important points, the first one being that while watermarking is necessary, on its own it is not sufficient to combat piracy. It must be combined with network monitoring and forensic techniques to identify streams, along with digital fingerprinting to categorize and identify each item of content. Operators therefore need an integrated package to combat stream piracy combining all the techniques, and they want evidence that package works.

Secondly watermarking cannot even be guaranteed to perform its core task of identifying streams or unique instances of content effectively. Some techniques are better than others and watermarking systems themselves will come under attack from pirates in an attempt to prevent their offending streams being traced back to the source. So watermarking involves a trade-off between several conflicting requirements. On the one hand it must not discernibly affect picture quality, but it must also be robust against tampering and the marks must be efficient to insert and recover for identification.

Friend MTS argues that since pirates will attack watermarking systems it is essential that this can be detected almost immediately so that a response can be made. One known threat is the collusion attack, where several copies of the same video are synchronized and mixed together to scramble or weaken the watermark, aiming to prevent that stream from being uniquely recognized. The Friend MTS system can detect that and immediately change the codes inserted into the stream as it is being played out, which then enables the attackers to be flushed out. This adaptive response is possible because the content is live and so the attack has to take place in real time, which in turn enables the watermarking codes to be changed on the fly. Another approach adopted by some other systems is to avoid inserting watermarks into every video frame and just select some, then varying the sequence of marked frames with each user, to make it harder for pirates to identify where those marks are.

Another factor is that live video imposes tighter timing constrains on the watermark insertion and detection process, which has to be balanced against the impact on video quality. Typically the impact of the marks on quality is minimized by using a system called “non-blind” watermarking. This requires comparison of the video with the original non-watermarked source material in order to extract the marks. It allows the marks to be kept as small as possible so that they have no discernible impact on video quality, but the approach is not practicable for live video, because of the time scales and computational effort that would be involved. To get round this, so called “blind” systems were developed, which can extract the marks directly from the video without reference to source material. The disadvantage though is that this requires more information to be put into the marks, which makes them less transparent and more visible, particularly for higher resolution content such as 4K.

One compromise is the semi-blind method, which enables smaller marks to be used but still without the need for comparison with the original source material to extract them for identification of the stream. This involves comparison of video suspected of being pirated with a small snapshot of the original source video to extract the marks, which reduces the computational effort and ensures that it can operate in a live environment at large scale.

The wider point is that watermarking is not a commodity but a complex technology that requires careful assessment and competent deployment. It is not just a box that can be ticked by deploying any given product. It is though going to become a quintessential part of the pay TV ecosystem as we move into the 4K era and not just for OTT distribution.