Technological Forecasting & Social Change 79 (2012) 1192–1216

Contents lists available at SciVerse ScienceDirect

Technological Forecasting & Social Change

Knowledge positions in high-tech markets: Trajectories, standards, strategies and true innovators Rudi Bekkers a, b, Arianna Martinelli c,⁎ a b c

School of Innovation Sciences, Eindhoven University of Technology, Eindhoven, The Netherlands Dialogic Innovatie & Interactie, Utrecht, The Netherlands LEM, Scuola Superiore Sant'Anna, Piazza Martiri della Libertá 33, Pisa, Italy

a r t i c l e

i n f o

Article history: Received 1 July 2011 Received in revised form 17 January 2012 Accepted 20 January 2012 Available online 15 February 2012 Keywords: Standardisation Essential patents Patent citation network Main path analysis

a b s t r a c t Essential patents refer to patents that are indispensable in order to make any product that complies with a technological standard. Portfolios of essential patents have often been used to indicate the strategic value of a firm's knowledge. We propose a range of alternative indicators based on a firm's position in patent citation network. Using a historical narrative and the actual licencing payments for the 3G W-CDMA standard in mobile telecommunications as a reference, we find that our alternative indicators provide better indicators for firms' knowledge positions and their long-term impact on technological change. Our proposed indicators can also be applied in markets that are not based on standards, and may not only be valuable to scholars but also to practitioners. Our findings also raise some concern over technology inclusion processes at standards bodies, and we recommend policy makers to consider our proposed method in order to critically look at these processes. © 2012 Elsevier Inc. All rights reserved.

1. Introduction Economic literature attributes considerable importance to the knowledge base – or knowledge stock – of firms [1,2]. The strategic value of this knowledge, however, may depend on how relevant that knowledge is in the context of an actual technology market. This paper focuses on the strategic knowledge positions of firms, which we believe to be a critical variable that provides a competitive advantage to the firm and can explain businesses success [3]. If on the one hand the concept is rather intuitive, its empirical operationalisation is largely overlooked. A notable exception is the work by Ref. [4], which explores firms' patent portfolios to characterise its level of specialisation and diversification. While these authors pick out just one or two firms in a given industry, one may also consider all relevant firms in order to understand their relative positions. Still, such approaches do not necessarily learn us about the value of this knowledge in the context of an actual technology market. Trying to address this, our paper explores several empirical approaches to quantify knowledge positions, and does this by focusing on markets where technology standards play a central role. In the last decades, there has been an increasing importance of what Ref. [5] has called “complex product industries”. In such markets, technology and knowledge have a systemic nature, relying on the integration of many different, interrelated and interdependent contributions. In the same industries, standards are becoming increasingly important, as they are needed to ensure interoperability between complex products and services at various points in the value chain. While such interoperability standards were initially found in the consumer electronics and telecommunications sector, now such standards start to become indispensable in other areas including service sectors (e.g. banking), IT systems, public transport, logistics and intelligent transport systems, biometrics and agricultural systems. Standardisation is an important yet underrated economic alignment mechanism, where the rate and direction of technological change is being negotiated between stakeholders [6]. Standards can strongly influence technical direction, activities and search heuristics, and thus influence technological change, whilst at the same time being the result of technological change. In many ⁎ Corresponding author. Tel.: +39 050 883591; fax: +39 050 883344. E-mail addresses: [email protected] (R. Bekkers), [email protected] (A. Martinelli). 0040-1625/$ – see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.techfore.2012.01.009

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1193

complex product industries fields, standardisation is the primary method of achieving alignment between actors. For any scholar interested in empirical research on knowledge positions, markets that are dominated by standards provide an interesting testing ground, as they offer several opportunities to compare the knowledge base with an actual technology market. Like in other markets, a strong knowledge position in standards markets offers competitive advantages that increase chances for market entry, sustainable participation, and market success. For instance, Ref. [7] shows how one single company, occupying a strong knowledge position, was able to fully dictate market entry into the emerging GSM market. A strong knowledge position may also contribute to bargaining power and, if secured in patents, be the source of substantial licencing revenues. In that sense, knowledge positions can have a positive effect on profitability. Without wanting to overemphasise this aspect, we observe that such licencing revenues can be substantial. For instance, holders of patents relevant for DVD players charge a total of approx. US$ 9 or more per player (depending on the implemented features); for mobile phones, firms pay approx. 8% (GSM) to 12% (GSM+ 3G) of the wholesale value of their products as running royalties; for the American digital TV standard ATSC, patent owners charge US$ 5.00 per receiver. 1 The CDMA patents of US firm Qualcomm, who are essential to various mobile telecommunication standards, earned this company billions of dollars in annual licencing revenue during the mid 2000s. Parties that own relevant patents themselves may enter into cross-licences, reducing the fees to be paid (which again confirms the monetary value of knowledge positions and patents). If knowledge positions are of such strategic importance, the question arises how one can measure these. For high-tech, standards-dominated markets, a common way to do this is to analyse the distribution of “essential patents”, i.e. patents that are considered to be indispensable in order to make any product that complies with the standard, because there is no alternative way to do so. This method relies on information that is generated as part of the Intellectual Property Right (IPR) procedures and processes that are implemented in most standards bodies, which we will now briefly outline. Standards bodies face the challenge of ending up in situations where patent owners would not be willing to licence their essential patents to other parties that want to adopt the standards. To this end, most formal standards bodies have adopted a FRAND (Fair, Reasonable and Non-Discriminatory) policy. Under such a policy, members are obliged to notify any essential patent they hold, and are requested to issue a public statement that they are willing to licence these under the FRAND conditions (which almost every member eventually does 2). Over time, the number of patents notified under FRAND policies has strongly grown. The largest numbers are found along mobile telephony standards, where for some standards a total over 1600 patent families is claimed by more than 60 owners [8]. This may lead to considerable transaction costs and delays, as well as to high cumulative licencing costs (“royalty stacking”), though the latter point is a subject of discussion — see Refs. [9,10] for proponents and opponents of this view, respectively. A number of recent papers have studied essential patents and essential patent portfolios. These include the work by Refs. [7,11–17]. Each of these studies has a somewhat different focus, but they all rely on public lists of patents declared by firms to be essential to a specific standard in order to comply with the IPR policies of standards bodies. While such lists are surely the most tangible expression of patents in relation to standardised technologies, such lists have some inherent limitations. First, patents greatly differ in actual value, and the field of high-tech, standards-based technologies is no exception to this rule. Just counting essential patents in order to estimate knowledge positions may therefore introduce a strong bias. A standard way to mitigate this problem is by weighting patent counts with citations. However, citations are far from a perfect indicator of economic value [18]. Second, given the strategic value that an essential patent offers to its owner, there is a concern that claims of essentiality are the result of strategic behaviour of the patent's owner instead (or in addition) of the actual technical relevance. A strategically operating patent owner might opt to get deeply involved in the drafting of the standard and use opportunities to suggest technologies that it owns patents on. If other participants have a similar agenda and incentives for such practices, it will result an increase of their own portfolio of essential patents. In a recent study [19] it was shown that strategic involvement was a better determinant of claimed essentiality than the actual technical merit of the patent in question. Third, the design of the IPR procedures creates some degree of uncertainty about using the lists of essential patents as indicator for knowledge position. In particular, there are at least four aspects to consider: (a) Companies are allowed to submit “blanket claims”, stating that they will licence essential patents on FRAND conditions. However, such blanked claims do not reveal individual patents. Companies that submit such claims may possess large portfolios of essential patents, but it is also possible that they do not own any essential patent at all. (b) There is some degree of strategic “over-claiming”, where firms declaring patents to be essential while in fact they are not. Such strategies are likely to differ between firms. (c) Standard bodies encourage early declarations, submitted before the patent is granted and/or before the standard is finalised. However, a granted patent may not be as broad as the original application and thus might not be essential anymore. Also, the final standard might be different from earlier draft versions, and disclosures that were appropriate for a certain draft version might not be essential for the final version of the standard. Since many standards bodies do not require parties to update or withdraw earlier disclosures, such declarations remain in the IPR database. (d) Patents owned by non-members may be missing. These parties are not obliged to disclose essential patents, although they may voluntary do so.

1 DVD fees estimates are based on fees for the Philips/Sony joint licensing programme (Philips, “Royalty rates for selected DVD and BD products”, retrieved on 2 February 2010 from https://www.ip.philips.com/services/?module=IpsLicenseProgram&command=View&id=27&part=8) (since replaced by the One-Blue patent pool) and the fees of the DVD6C Licensing Group (DVD6C, “Offer letter to Existing Licensees, 1 September 2010”, retrieved on 2 February 2010 from http://www.dvd6cla.com), as well as fees of the DVA Discovision Associates, (DVA “Licences”, retrieved on 2 February 2010 from http://www.discovision.com/ DVA/Licensing). Further licensing fees might be due to Thomson, the DVD Copy Control Association, and Microvision. ATSC and FireWire estimates are based on the licensing programmes published by the MPEG Licensing Administration (http://www.mpegla.com). Mobile telecommunications fees are based on [4]. 2 If a patent owner refuses to do so, the standards body eventually has to find an alternative definition for the standard, not drawing upon that patented technology, or has to abandon the work on the standard altogether.

1194

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Some studies have attempted to address the limitations of essential patent counting by analysing whether the disclosed patents are actually essential [11,20,16]. However, the reliability of these attempts has been heavily criticised by [17,21]. Indeed, the determination of the actual essentiality of even a single patent is a very specialist and resource-demanding activity (requiring the development of claim charts), and doing so for hundreds or even thousands of patents in a reliable way can be seen as virtually impossible. A more promising avenue is to complement the essential patent analysis with a time structure analysis. This provides valuable additional indications into actual knowledge positions [13], yet these insights are hard to quantify. Given the above concerns for using essential patent analysis in the context of understanding knowledge positions, it may be useful to look into better alternatives. Network-based analyses seem promising here, allowing not only for investigation of patent networks [22] but also the aggregation of these into networks of firms. This paper explores and tests a range of such measurements, including various centrality measures and core/periphery structure. We also investigate the development of the network over time, including density, fragmentation, and GINI coefficients for indegree and outdegree centrality. An even better understanding of knowledge positions may be offered by more advanced, network-based techniques such as the “main path analysis” pioneered by Hummond and Doreian in 1989 [23]. In the recent past, number of papers employed this approach for mapping technological trajectories (see, for instance, Refs. [23–28]). While originally devised for networks of citations between academic publications, these recent papers use this methodology for patent citations networks. Such networks link patents through the citations mapping the knowledge flows occurring between them. Specific algorithms can be used to identify the “main flow of knowledge” within the patent citation network. This main flow of knowledge is a set of connected patents and citations (i.e. a path) linking the largest number of patents of the network and therefore cumulating the largest amount of knowledge flowing through citations. This path represents therefore a local and cumulative chain of innovations consistent with the definition of technological trajectory put forward by Ref. [29]. Given the success of this approach in understanding the main flow and the development of patented knowledge, it might be promising for providing insight into the knowledge position of the firms that own those patents. At the same time, the granularity of this method might restrict it usability in this context: even if the full network comprises thousands or even ten thousands of patents, the identified main path of knowledge often comprises few dozen of patents or even less. To conclude, this paper has the aim to test, in standards-based high-tech markets, whether network-based methodologies do a better job in predicting actual knowledge positions than approaches based on essential patent analysis. We also aim to further extend such network-based methods, where possible. In addition to this methodological contribution we aim to make, we also expect that the empirical analysis itself can lead to valuable insights that can be relevant for businesses, standard-setting organisations, and policy makers. For our empirical data, we turned to the field of mobile telecommunications systems. Not only they do represent a market that is very sizable and of strategic value to its players, it is also one for which there is good availability of data, both for historical accounts and for patenting position. In order to fulfil its aims, Section 2 of this paper starts with an extensive technical narrative of the case study, which we will later use to test the outcomes of the various approaches. We believe that this narrative needs to go into a considerable degree of detail, not only to do justice to the quite complex development path of such technologies, but also to be able to judge upon the actual knowledge positions of actual firms. Knowledge positions are assessed upon the (a) actual contribution of firms to key technical advances and (b) the licencing payments between firms, which we believe reflects the bargaining position on the basis of knowledge position. Section 3 of this paper discusses the various methodologies we are using to test knowledge positions, and Section 4 presents the empirical findings. Where appropriate, we validate the knowledge position findings by comparing them with the outcomes of the technical/historical narrative and the current licencing position. We also provide a direct comparison of the outcomes of the two approaches. Section 5 draws conclusions and offers a discussion.

2. A technical narrative of 2G and 3G mobile telecommunications This section aims to introduce the main technological developments in the field of mobile telecommunications, the involvement of specific actors, and the associated standardisation efforts. Readers that already have knowledge of this sector, or that are only interested in the resulting outcomes in knowledge position, may skip this part and continue at Section 2.3. In this field, it is common to distinguish between four main technological generations, dubbed 1G to 4G. Each generation has its own, distinct standards. Table 1 provides an overview of the various aspects of the four distinct generations. Here we will specifically focus on the third generation (3G), which is the main focus of our empirical analysis. 3 When discussing technology and standardisation, we will pay specific attention the engineering challenges that came with the various new developments. While we aimed to keep this a brief narrative, we feel it is necessary to go into some degree of detail in order to be able to use this narrative as a reference point of the knowledge position of firms. Unfortunately, as with other treats on standards, the extensive use of acronyms is unavoidable. For the convenience of the reader, we do not spell each of them out in the text but instead provide an overview in Annex A. While this text offers some sources, we refer to Refs. [30–32] for a more complete listing of sources. The reminder of this section is organised as follows. Section 2.1 introduces the origin and development of the CDMA, the technology that eventually became the basis for 3G networks worldwide. Section 2.2 describes the development of standards for 3G networks and 3 For two different reasons, the other generations are not very suitable for our empirical analysis. At the time of the first generation, firms did not patent many inventions. The fourth generation yet has to crystalize; there is no good insight in the relevant or essential patents yet, and many patens will be relatively new and therefore have few forward citations, if any.

Table 1 Summary of main technological generations/standards.

Sub-standards /improvements Design requirements

2G

3G

4G

AMPS/TACS (1970s) NMT (1970s)

GSM (1986) IS-95 cdmaOne (1993)

W-CDMA/UMTS (1998)

1983 (US), NMT (1981)

1992 (GSM) 1995 (IS-95 cdmaOne) 2.5G: GPRS (2000): packet data services EDGE (2003) – High-capacity voice capacity at lower system price – Cost-efficient coverage in both urban and rural areas

2002

“3.9G”: LTE (frozen December 2008) 4G: LTE-A 2009 (small scale)

Various

– Low to medium capacity mobile telephony

Candidate technologies (*: dominant technology)

*FDMA (analogue)

*TDMA CDMA

Main technological challenges

Various, including mobility management, handover, and handsets

– Synchronisation and timing within a cell – Multipath fading (solved by the channel equaliser (Viterbi coder) and frequency hopping) – Efficient speech compression – Handover processes – Energy consumption

3.5G: HSPDA (2006): Improved data rates – Support wide diversity of services including internet access; substantial improvement in data speed – Low costs for terminals and networks (minimising required number of cell cites / antenna towers). – Low power consumption at terminals – Operation up to 300 km/h – Cost-efficient coverage in both urban and rural areas – Handoff to 2G systems Advanced TDMAb TDMA/CDMA hybridc *W-CDMAd MC-CDMA OFDM/ODMS – Power control within a cell – PN code sets – Timing/synchronisation between adjacent cells – Signalling / pilot channel – Integration with 2G (inc. handoff)

– Substantial improvement in data speed – Lowering infrastructure costs per capacity unit – All-IP core network integration – Flexible spectrum use

W-CDMA *OFDM

Increasing spectral efficiency

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Most successful standard(s), main decision Commercial servicesa

1G

a It is often hard to determine when the actual introduction of commercial services takes place, as technology demonstrators and trials gradually become commercial services. This row aims to indicate the date when which the first real commercial services with a substantial geographical coverage were offered. b Also known as A-TDMA or the “FMA-1 without spreading” proposal or the Gamma (χ) proposal. c Also known as TD/CDMA or the “FMA-1 with spreading” proposal or the Delta (δ) proposal. d Also know as DS-CDMA or the “FMA-2” proposal, or the Alpha (α) proposal.

1195

1196

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

discusses pre-competitive research, standardisation, and adoption. Finally, Section 2.3 presents an overview of actual knowledge positions on the basis of the presented narrative and on actual licencing fees. 2.1. The origin and development of the CDMA technology Mobile telecommunications network employ many different technologies, including speech and video compression techniques, call and data routing techniques, and much more. The single, most important technology, however, is the so-called air interface. This is the exact way in which signals get transferred on the radio path between the user terminal and the network. The radio interface defines how the radio capacity is shared between the different users, and how the signals of different users are distinguished from each other, and is therefore also known by the more technical term ‘Multiple Access’ (which include TDMA and FDMA technologies, among others). More than anything else, these multiple access technologies impact the system capacity, performance, and cost-effectiveness of a network. In the mid-1990s, when work on 3G technologies started, the most widely used air interface technology was TDMA. It had proven extremely successful as the technological basis of GSM, the dominant, second generation (2G) mobile telephony standard by that time. This technology distinguishes signals from different users by assigning unique time slots to them, which rotate up to hundreds of times in a second. Together with other valuable technologies (such as the Viterbi coder and ‘slow hopping’) this resulted in a robust and well-performing, cost-effective technology for telephony applications, and creating a mass market of billions of users worldwide — in fact, the large majority of current phones still include a GSM mode. While the GSM standard has a fascinating history with both winners and losers among the industry players (which is documented in [30–33]), the standard clearly had its champions and the supply market was rather concentrated. By 1996, five years after the first commercial network went live, Sweden's Ericsson had a 48% market share of GSM infrastructure, and Nokia, Siemens, and Alcatel shared another 45% [34]. The terminal market was similarly concentrated, with a particularly high share of Nokia from Finland. However, rather parallel to the roll-out and success of GSM, US company Qualcomm departed from the mainstream path and started working on an alternative technology called spread spectrum (or CDMA). In this technology, the transmissions of different users are identified by very fast, unique codes. The birth of CDMA can be traced back to the period of the Second World War, to an unprecedented story [35]. Trying to develop a radio link that was immune for jamming, multi-talented Hollywood movie star Hedy Lamarr and piano player George Antheil invented a method of radio communications that continuously jumped from one transmission frequency to the other, in a quasi-random matter. Both transmitter and receiver needed to know this secret, semi-random pattern [36]. In their patent, there are 88 frequencies – similar to the number of keys of a piano – and the pattern was coded in mechanical roll similar to the one in a pianola. Being resistant to jamming, they considered this system to be particularly useful for guiding torpedoes. In fact, their patent No 2,292,387 shows a remarkably detailed application. Lamarr and Antheil patented their invention and offered it to the US army at no charge, hoping to help the allied forces. The military showed no interest, whatsoever. Only in the 1960s, after the patent's expiration, that its value was recognised. This invention not only could withstand active jamming, but also offered excellent security against interception of sensitive communications (eves-dropping), and even dismissed the enemies' ability to locate military units through their radio transmission. The technology became standard in confidential military communications, but its knowledge and main patents remained suppressed until the late 1970s ([37] page 341). By the 1980s, some creative engineers realised that CDMA could potentially be a powerful and economical basis for large-scale mobile telephony networks. 4 Its broadband nature would – at least in theory – make it immune to many problems that limited the capacity of traditional systems, such as multipath fading. In contrast to military applications, a mobile telephony network has to with many different communications at the same time. Whereas almost all radio systems at that time were designed to minimise interference, CDMA went fully against that logic and has many different users transmitting on the same frequency and at the same time. A handbook on digital telephony technologies of the late 1980s comments: “viewed from [the] orthodox perspective, the vision of spread-spectrum transmission seems so contrary, even perverse, that it might almost be taken for a jest upon the inflamed sensitivities of the interference-bedevilled radio community” ([37] page 340). In order to use spread spectrum as the basis for a mobile telephony networks, some great hurdles needed to be overcome. One of them is known as the near-far problem. As explained above, multiple users would be transmitting on the same frequency and at the same time. To distinguish the signals of these users by their code, it is necessary that the received power of each phone at the base station would be almost identical. In a real life situation, where the actual received power constantly changes because of distance, obstacles and reflections, this deemed impossible by many an engineer. In fact, many initially regarded CDMA with great scepticism and claimed that it would never work in practice. Such beliefs are obvious from the following quote: “From the beginning, critics warned that the compelling theoretical potential of CDMA would never prove out in the field; dynamic power control in rapidly fading environments would be its Achilles heel; interference would vastly limit capacity; systems under heavy load would be unstable; and power balancing would make infrastructure engineering a nightmare”. 5 The sceptics proved to be wrong. Power control, the single biggest engineering challenge for a functioning CDMA system, could indeed be mastered. It was done by so-called open and closed loop power control methods that were conceived, developed and patented by Qualcomm. Soon after, Qualcomm developed a full mobile standard on its own, which was standardised as IS-95 CDMA in the US (later known as cdmaOne). As pointed out by Ref. [35], Qualcomm's IS95 system successfully addressed all the major and minor problems that were generally perceived 4 The earliest CDMA systems were based on a principle called Frequency Hopping (FH-CDMA). For mobile telephony, a somewhat different principle is used, known as Direct Sequence (CD-CDMA). 5 Source: Bill Frezza, Wireless Computing Associate, “Succumbing to Techno-Seduction,” Network Computing, April 1, 1995.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1197

Fig. 1. Overview of research and standardisation activities for W-CDMA.

to prevent the use of CDMA in a large scale mobile telecommunications system. While Qualcomm had developed the technology, this company did not have the means or the ambition to be a large-scale network infrastructure supplier (for which it is virtually necessary to have an existing switching platform). Instead, Lucent and Motorola adopted this technology relatively early and developed infrastructure products. By the end of the 1990s, 114 out of 431 US wireless service providers had chosen IS-95 CDMA as their technology, and South Korea and some other countries saw wide IS-95 CDMA adoption. Nevertheless, despite these successes, CDMA never dethroned GSM as the dominant 2G technology, with eventually over 80% of the 2G world market. 2.2. The development of standards for third generation mobile networks (3G) Although the various 2G technologies were later upgraded to support data transmission, their data speeds and other features made them quite unsuitable for the demanding data applications that were becoming popular in fixed networks, such as multimedia and internet access. It was perceived that a new, third generation of technologies would be necessary, capable of supporting a wide range of new services, including high-speed data transmission. At the same time, 3G systems were supposed to meet many other – sometimes conflicting – design requirements, as shown in Table 1. Perhaps most importantly, it was understood that subscribers wanted much higher data volumes but would not be willing to pay much more than they currently did. As a consequence, the new technology had to reduce the cost price per unit of data considerably.6 The actual development and use of 3G standards included the phases: pre-competitive research, standardisation, and adoption. These three phases are discussed below. Pre-competitive research. The success and extensive geographical coverage of GSM created high expectations from the public, raising the bar for 3G networks. R&D funding from the European Union aided the earliest investigations. In particular, the 2nd Research and Development in Advanced Communications Technologies for Europe (RACE) programme from 1992 to 1995 included specific grants for mobile phone technologies. Research efforts increased with follow-up research programmes funded by the European Commission, known as RACE-2, ACTS/FRAMES, and COST. With a budget of 100 million ECU for FRAMES alone, 6 As an illustration: per 2005, the network infrastructure costs for a subscriber that was generating 300 Mb/user/month accounted approximately 45 Euro for the older GSM/GPRS standard and approximately 7.5 Euro for the W-CDMA HSPA standard. Nowadays, with newer versions of HSDPA, the costs reduced further. Source: Source: GSA. (2005). Opinion paper: Radio Access Evolution. Available from http://www.gsacom.com.

1198

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

these projects were considerable in size. Contracts were awarded to several firms, including Ericsson, Nokia, Siemens, France Telecom, and CSEM/Pro, with participation from several European universities too. However, in the industry, opinions differed when it came to the most suitable technology to satisfy all the needs. Fig. 1 provides an overview of the research frameworks, as well as the competing technical proposal and standardisation efforts as described below. Within these frameworks, one group of firms worked on what essentially can be seen as extending the TDMA technology of GSM (dubbed A-TDMA, later known as FMA-1). While such extensions did allow for more capacity, it was increasingly understood that technology would be insufficient to really meet the design requirement for third generation systems. As the advantages of CDMA became clearer over time, the group added some CDMA elements to its design. Companies that were particularly active were Siemens and Nokia — although firms were not exclusively tied to one single group. Another group of firms was focussing on CDMA technology instead, as pioneered in the US for 2G systems. Their design was initially known as CoDIT and later as FMA-2. Particularly for 3G systems, CDMA would have additional benefits, being able to deal well with many different traffic patterns at the same time (e.g. telephony, video, internet traffic, telemetric). In terms of system capacity, these “Wideband CDMA” (W-CDMA) designs went quite some steps further than the existing 2G IS-95 CDMA system by Qualcomm. Nevertheless, they heavily drew upon the latter. In research reports, it can be seen that many studies evaluated system performances “based on a IS-95 like system”, and a number of tests were actually using IS-95 chipsets, because they are “readily available providing a very flexible solution” [38]. In the W-CDMA group, Ericsson was the primary contributor. This company also developed its own “test bed” in order to test features of the technology. Eventually, both groups pushed forwards their design as the basis for the European 3G standard. Standardisation. While research progressed rapidly, European standardisation efforts were simmering. The 3G developments were largely ignored by GSM operators – the principle customers – who were focusing on increasing subscribers' numbers of their existing 2G systems ([31] page 478). In Japan, where the domestic industry had very limited success on the global market for 2G, plans were made for a rapid standardisation. The alignment with European manufacturers was a key element of that plan, hoping to set a world standard. Before Europe decided on its 3G standard, NTT DoCoMo of Japan, at that time the largest mobile telephone operator of the world, decided to procure an experimental W-CDMA system. Orders were not only placed with domestic companies but also engaged foreign firms, including Ericsson, Nokia, Motorola, and Lucent. By involving foreign suppliers, NTT DoCoMo tried to increase its chances of having the W-CDMA technology adopted in other world regions. With NTT DoCoMo being so dominant on the national market, the Japanese standards body was placed at a fait accompli and eventually set W-CDMA as the formal standard. The actual design was in fact very close to the 3G system that Ericsson had been designing in the European research programmes. At about the time the Japanese contract was granted, Nokia – quite understandably – shifted most of its research efforts towards W-CDMA [39]. Under increased pressure from the events in Japan, Europe's standards body ETSI prepared itself to define the European standard. Fierce technical discussions took place, both within and outside ETSI. Some two dozen of proposals were categorised into five “concept groups”. Two strong, opposing camps formed. One camp, now including Siemens, Alcatel, Nortel, and Italtel, proposed what was called the Delta (δ) concept group. This was basically identical to the Advanced-TDMA/“FMA-1 with spreading”, the standard on which several of these firms already had been working on in the abovementioned research programmes.7 This proposal was notably different from the one adopted by the Japanese. The other camp supported the so-called Alpha (α) concept group, which is essentially the CoDIT/FMA-2 research work. Because this was also similar to what was already being built in Japan, it will come as no surprise that the chosen suppliers of the Japanese operator – Ericsson, Nokia, Motorola, Lucent and several Japanese firms – were the main backers. Would they be successful if they would have a significant head start, having already developed products for this technology? With stakes that high, a record number of ETSI members and representatives from many organisations, including the European Commission, gathered in Madrid on 15 to 17 December 1997 at ETSI SMG#24 meeting, on which the decision for the European standard UMTS was to be made. Not only the strongly divided opinions, but also potential patent issues made it impossible to reach a definitive decision at the meeting, which was then postponed for a month. These patent issues take up more than half of the length of the minutes of the meeting and already reflect that the ETSI members were quite aware that some contenders were heavily drawing upon technology that was developed elsewhere, and that the owner of the related intellectual property would require compensation. When it was proposed that the participations would pool their patents, one of them commented: “[…] in the case of a certain company outside the IPR pools asking for 6% licence fees, a pool licence agreement of other companies in order to keep the licence fees at 1%, would result in a 7% licence fee of total.”.8 There cannot be much doubt this “certain company” was Qualcomm, the company that pioneered CDMA for mobile telephone and owner of a large stock of related patents. Industry experts warned that “If forced to pay stiff royalties, Ericsson and Nokia may be unable to afford the cost of developing and manufacturing third-generation W-CDMA systems.”.9 The fact that many firms (including Nokia, Siemens, and Alcatel) were already Qualcomm licencees because they were supplying CDMA products long before their work on 3G (using the IS-95 2G technology) means that these firms must have had a very good understanding of the exact scope of this patent portfolio. In the month following the meeting, both camps undertook intensive lobby activities, trying to win the support of voting ETSI members (operators, administrations, other manufacturers). Eventually, the Alpha group seemed to have gained more

7 The Delta proposal is known as TD/CDMA. It was principally based on TDMA, like GSM, but over time, some CDMA elements had been added to improve its performance. 8 Source: ETSI. (1998). Consensus Decision on the UTRA concept to be refined by ETSI SMG2 (ETSI/SMG (98) 1 Annex 6 (DRAFT dated 17.2.98). Interestingly, this statement was removed in the final version of the minutes that were published several days later, possibly because of legal concerns. 9 Mobile rivals prepare for Paris take-off. (19 January 1998). CommunicationsWeek International.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1199

momentum than its competitor. The chairman of the responsible ETSI committee later described how a hotel room meeting of the high executives took place just before the next official meeting, and that these executives agreed on a compromise that was mostly based on Alpha but also had some elements of Delta (including so-called TDD-mode operations, which eventually never got implemented in commercial products) [32]. When the official meeting took place, on 28 and 29 January 1998 in Paris, the compromise proposal was put on the table and then easily won the required 71% of all votes. Understandably, the Japanese stakeholders warmly welcomed this decision and immediately announced they would ensure that their own standard would see some modifications so it would be fully identical with the European one. When also the Korean standards body TTA showed interest in cooperating, a new body called Third Generation Partnership Project (3GPP) was established. Its partners were standards bodies around the world (including ETSI from Europe, ARIB from Japan, and TTA from Korea). Finally, the work could commence on drafting the final specifications. It did not take long before a major conflict emerged. Qualcomm, being the owner of many CDMA patents but not having been involved in the European standardisation activities, felt that some implementation decisions were taken with the prime goal of creating incompatibility between the European standard and its own CDMA-based standardisation efforts. More specifically, the European proposal specified a so-called chip rate of 4.096 chips/s, whereas Qualcomm had settled on 3.6864 chips/s. An essential point here is that the latter was chosen in order to maintain upwards compatibility with an earlier standard (IS-95, with a chip rate that is exactly one third of that value), whereas the first is not designed to be compatible with any earlier radio standard and hence the designers were free to chose any value. Others argued that lowering the chip rate would result in a significant decrease of system capacity. Having a harmonised chip rate among different standards would facilitate multi-mode devices, lower market entry for chip makers, and drive down prices. At some point Qualcomm made the availability of its patents conditional to its request to harmonise the chip rate. When the (mostly European) companies refused to do so, the parties came into a clash. Eventually, operators around the globe, who feared a lack of interoperability and unnecessary high price levels, exerted pressure on their suppliers and ultimately commanded them to lower the chip rate to the value Qualcomm had been suggesting. The fact that operators forced this harmonisation of chip rate strongly suggests that there was not a significant downgrade of performance, as they would never have accepted that. After that final hurdle, the more detailed work of drafting the standard could be continued and the first “frozen” version of the standard, dubbed “Release 99”, was published early 2000. The standard is generally known as W-CDMA, but also referred to as UMTS (in Europe), 3GPP (worldwide), and FoMa (in Japan). From that point on, the standard has seen several new releases that improved stability and offered additional functionalities, and supported greater capacity and data speeds. The most significant addition was a higher data speed mode called High-Speed Downlink Packet Access (HSDPA) that increased the maximum data rate by a factor of 100 and further.10 Adoption. The commercial up-take of the W-CDMA standard get delayed for a variety of reasons, including steep spectrum licencing fees, the blow of the internet bubble, and slow initial adoption by end users. More recently, uptake started to grow with the popularity of smartphones such as the Apple iPhone 3G (launched in 2008). Some European operators noted almost a tenfold increase of data traffic as the result of smartphones. 11 W-CDMA – often combined with GSM – is currently becoming the dominant mobile telecommunications standard in the world. Elsewhere, competing 3G standards were developed, including Qualcomm-backed cdma2000 (in a body appropriately called 3GPP2), and TD-SCDMA, a technology proposed by the Chinese. None of these other standards, however, comes close to W-CDMA in terms of market success.

2.3. Market share and knowledge position for 3G technology suppliers In terms of equipment market share in W-CDMA, Ericsson and Nokia are currently (late 2010) the world largest suppliers of infrastructure and mobile phones, respectively. Asian firms have captured a substantial share of the market: in November 2009, Huawei from China became the world's second largest supplier of infrastructure, bigger than the merged Nokia Siemens Networks (NSN) company and the merged AlcatelLucent company. 12 Early 2010, South Korean companies like LG and Samsung were occupying the second and third position in mobile terminals, after Nokia. 13 Siemens left the handset market, and Motorola is in the course of doing the same thing, after having sold its wireless infrastructure division to Nokia Siemens networks in July 2010. Only judging on the integrated end-products might overlook the role of component suppliers, especially if these components embed a significant amount of the relevant functionality or knowledge (compare the role of Intel as a component supplier for the PC market). For the 3G market, the chip sets that provide the communications functionality (known as base-band chips) are the single most important component. In the market for these chips, the shares of the largest players are as follows: Qualcomm (38% revenue share), MediaTEK (18%), Texas Instruments (15%), ST-Ericsson (10%) and Infinion (7%). 14 This reveals that Qualcomm is not a Non-Producing Entity (NPE) even though it does not sell infrastructure products or end user products like mobile phones (anymore).

10 The original release supported a maximal data rate of 384 kbps in a wide area setting. HSDPA and its successors have driven that up 42 Mbps (release 8) and future releases even plan higher speeds. 11 See http://www.emerce.nl/nieuws.jsp?id=2965541&utm_campaign=rss&utm_source=rss&utm_medium=rss. 12 Source: ReThink Wireless, Huawei narrowly overtakes NSN in mobile infrastructure sales, 17 November 2009. 13 Source: IDC Quarterly Mobile Phone Tracker. See http://www.electronista.com/articles/10/01/29/idc.says.korean.and.us.phone.makers.benefit/. 14 The Free Library (March 15, 2010). MediaTek unseats TI as No.2 cellular baseband chip vendor.

1200

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Table 2 Knowledge positions on the basis of the historical narratives and the current licencing generation capabilities. Firm Qualcomm

Assessment

The technical narrative leaves little doubt as to Qualcomm has the strongest knowledge position in the CDMA technology field. Having pioneered the CDMA for 2G networks in the early 1990s and having found solutions to problems that others considered to be conceived unsolvable, they created the most valuable patent portfolio of any company. This is reflected by the royalty agreements and royalty fee levels: Qualcomm has over 140 licences and reportedly is able to almost invariably negotiate a licencing fee of approx. 5%, while all other 50 patent holders are sharing the remaining approx. 7%. A complaint at the EC against Qualcomm that alleged unreasonably high licencing fees was dropped. Interdigital The knowledge position of this company is very hard to determine. While it claims to own a large patent portfolio, parties have been questioning whether these patents are actually valid or actually infringed by standardised products. From Interdigital's annual report it can be concluded that many large firms have refused to sign licencing agreements. Recently, Nokia successfully challenged that its 3G products infringed Interdigital's patents. Overall we find it hard to estimate Interdigital's real knowledge position. Lucent Although not very actively part of the discussions for the W-CDMA standard, Lucent was already an early entrant to the IS-95 CDMA market, allowing it to build a good knowledge position in this technology. It later supported both W-CDMA and cdma2000. [Lucent and Alcatel merged in late 2006.] Motorola Motorola's position is fairly similar to that of Lucent and also was able to build a good knowledge position. Ericsson Being the first European manufacturers that fully concentrated on W-CDMA for 3G, its knowledge situation lags that of the US parties that were already active in CDMA markets but is stronger than that of other European companies such as Nokia. NTT Made a relatively early choice for W-CDMA, after having studied the features of this technology extensively around 1996. The only operator in this list. Nokia Having switched its research from ATDMA to W-CDMA at a relatively late stage (at the time it received a large contract from NTT and less than a year before W-CDMA was formally selected in Europe) its knowledge position is relatively weak. Its strong involvement in work on the standard and additions to it after the basis was already laid party made up for this. Having to pay 2% licencing fee to Qualcomm (and having had to transfer a number patents to get that deal) we can conclude its bargaining position in the knowledge space is considerable weaker than that of Qualcomm. Philips While Philips was quite active in research, it left the mobile infrastructure market already in the early 1990s and left and (re)entered the mobile terminal market repeatedly, making it hard to leverage their knowledge position in licencing agreements. Alcatel Supporting the losing TD/CDMA proposal to the last moment, resulting in a weak knowledge position in W-CDMA. [Lucent and Alcatel merged in late 2006.] Siemens Supporting the losing TD/CDMA proposal to the last moment, resulting in a weak knowledge position in W-CDMA. [Siemens sold its mobile terminal business to BenQ in 2005 and merged its infrastructure business with Nokia in 2007]

Estimated knowledge position Strong

Hard to estimate

Medium

Medium Medium

Medium Medium to Weak

Weak

Weak Weak

Patent licencing payments are one aspect that can indicate knowledge positions. In this industry, with more than 50 parties claiming to own essential patents, many firms cross-licence. As mentioned in the introduction it is estimated that the current aggregate licencing fee for a 3G-W-CDMA/GSM phone is around 12% for manufacturers that do not own relevant IPR themselves and therefore cannot cross-licence [8]. It is generally understood that Qualcomm charges approx. 5%. The other patent holding firms share the remaining part. In 2005 a number of companies filed a complaint at the European Commission, arguing that Qualcomm's rates were excessive. The companies apparently failed to persuade the Commission and the complaint was ultimately withdrawn. 15 In 2008, Nokia and Qualcomm entered into a cross licencing agreement. Industry experts commented that Nokia agreed to pay a 2% licencing fee to Qualcomm, 16 and some commented that, as part of this agreement, Nokia also transferred the ownership of number of its essential W-CDMA patents to Qualcomm. 17 This agreement can be seen as a reflection of the relative knowledge position of the two firms: while Nokia does own valuable patents (hence is not paying the 5% most other companies are paying), its knowledge position is considerable weaker than Qualcomm — otherwise it would be a cross licence with no monetary compensation, or net payments. Table 2 summarises our estimates of the actual knowledge positions of the most relevant firms in this space, on the basis of our historical/technical narrative and on the basis of the current licencing-generating capabilities. 18

15 Source: Bloomberg. Qualcomm Antitrust Probe Is Dropped by EU Regulators (Update2) November 24, and Daily News and Analysis. European Union studying anti-trust complaint against Qualcomm. Jun 17, 2010. 16 Source: The New York Times, July 24, 2008. In Settlement, Nokia Will Pay Royalties to Qualcomm. 17 The Chilli: Nokia and Qualcomm exchange patent currency. Retrieved on 23 February 2011 from http://www.thechilli.com/index.php? option=com_content&task=view&id=178&Itemid=335. This patent transfer is confirmed by the ETSI IPR database, where it can be observed that some patents that were originally applied for by Nokia are now claimed by Qualcomm. 18 We note that these firms might have improved their knowledge position by contributing to the field after the basic technologies were developed and/or commercialised, or by contributing technologies to future enhancements of the standard. Still, we believe that the knowledge that was held prior to the technology selection moment is the most relevant one.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1201

Table 3 Overview of empirical analyses. Analysis

Dimensions

1. Essential patents

1a. Count 1b. Time structure 2. Analysis of the network of 2a. Structure of network firms 2b. Firm's network position; core/periphery structure 2c. Firm's network centrality 3. Main path analysis

Principle unit of Data analysis

Empirical results presented in

Firms Firms Firms

Section 4.1

Firms Firms

3a. Conventional main path analysis

Patents

3b. Extended main path analysis

Patents–Firms

Claimed essential patents Claimed essential patents All patents in selected technology area All patents in selected technology area All patents in selected technology area All patents in selected technology area All patents in selected technology area

Section 4.2

Section 4.3 Section 4.4

3. Methodology and data Aiming to test whether network-based approaches are better at estimating knowledge positions than essential patent based analyses, this section starts by introducing the various methodologies we used for both type of analysis. It continues by describing the two data sets we developed. The actual empirical results of the analysis we performed will be discussed in Section 4. 3.1. Empirical methods In total, our study includes three sets of empirical analyses: the first one is based the essential patents, and the other two analyses use network-based methodologies. Our analyses are summarised in Table 3. 3.1.1. First set of analyses: essential patents As briefly discussed above, in both academic literature and industry reports, essential patent analysis has been used in order to understand knowledge positions. The rationale is that the more essential patents a firm owns the stronger is its knowledge position. Examples of such studies are Refs. [7,11–16,20], and our approach for this analysis builds on these earlier papers. One of the most critical elements here is how to count patent in order not to double-count. A single invention may be patented in multiple countries, and its owner may have disclosed all these individual national patents to the standards body. Another company might just have reported one family member. Simply counting all patents might result in considerable bias. It is also possible that a single invention is protected by multiple patents within a single country (e.g. continuation patents, divisionals, and divisionals-in-part; for a discussion on these see Ref. [40]). Also this can result in bias, especially if we consider that some patents in our data set have many dozens of such offspring. We adopted an approach in which we counted patent families instead of individual patents, thus solving issues of double counting. (More details are given below when we introduce our data sets). The insights that are created by essential patent analyses may be improved if also the time structure of the essential patent portfolios is considered [12,13]. Arguably, inventions that predate the work on the standard might be more ‘basic’ to the technology (and hence more valuable) than the later inventions (made more or less in parallel to the standardisation process), which might be more trivial. In this paper, we provide both a count analysis and a time structure analysis of essential patents. The results are shown in Section 4.1. 3.1.2. Second set of analyses: the network of firms As already claimed, the aim of this paper is to test whether network-based methodologies are more suitable for determining firm's knowledge position than essential patent analysis. In the context of our study, the starting point is a network of technological inventions, where nodes correspond to patents and the links are citations between these patents. If patent B cites patent A we can Table 4 Summary of key network measures. Variable Density Average distance

Definition

The density for a valued network is defined as the sum of all the values divided by the number of possible links. The average of geodesic distances between nodes in the network. The distance is the length of a geodesic between them, which is measured as the shortest path. Fragmentation Proportion of nodes that cannot reach each other. Reciprocity Percentage of reciprocated links on total links in the network. GINI coefficient for firms' Distribution of forward citations linkages measured by the GINI coefficient applied to outdegree centrality. Outdegree is the outdegree centrality number of forward citations (i.e. received) by a firm's patent portfolio. GINI coefficient for firms' Distribution of backwards citations linkages measured by the GINI coefficient applied to indegree centrality. Indegree is the Indegree centrality number of backward citations (i.e. made) by a firm's patent portfolio.

1202

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

imagine a directed knowledge flow linking patent A to B. In such a case we say that patent A received a forward citation, and patent B placed a backward citation. By collecting all the patents and all the in-between citations related to a specific technology one can easily build the relevant patent citation network. By definition, such a network is directed and acyclic. Note that while we may consider these links to represent knowledge flows, this is not a necessary assumption for the analyses we carry out (see [41] for an analysis to what degree patent citations indeed represent knowledge flows). As patents are assigned to firms, there is a straightforward way to aggregate the network of patents at firm level. The nodes of the resulting network represent the patent portfolios of each firm, and the links represent the accumulated citations between them, in either direction. We acknowledge that a patent citation network just a subset of the possible knowledge-based links between actors (joints ventures or research collaborations are other ways in which firms exchange knowledge), and we also note that the usual caveats about the use of patents and citations hold. Despite these limitations, we believe that this type of network can still offer valuable insights about firm's relative position in the knowledge space. Furthermore, by distinguishing different time periods (see below) one can also study changes over time. The literature on network analysis defines a number of indicators that describe the structure of the network. The ones we used for this study are presented in Table 4. Looking at their evolution over time can shed some light on the dynamics of knowledge flows [42]. Using graphic plots, one can consider the position of individual nodes in this network. In the context of our network, such a graph reveals all knowledge flow relations of any given firm: it does not only show with which firms there is a knowledge flow link, but also whether the net (cumulative) flow is inwards or outwards. The core/periphery structure approach suggests that network can contain of two types of firms, the first type belonging to a cohesive and dense core, and the second type to a sparse periphery [43]. The two groups of firms can be obtained by maximising the correlation between the data matrix and an ideal core/periphery structure matrix. 19 In the context of our network of firms, the core/periphery model may help to understand who are the key players. In network analysis, it is also common to look at the centrality measures for individual nodes in the network, which may be indicative of their importance. Betweenness centrality measures the number of time a node lies on the shortest path between any combinations of other nodes. In our context, it indicates that a company with high betweenness centrality score has good accesses the knowledge of other companies. An alternative centrality measure is known as degree centrality and considers the number of links a node has. Since our network is directed, we distinguish between indegree centrality and outdegree centrality. In the context of our network of firms, the first indicates the number of forward citations (i.e. ‘received citations’) and the latter the number of backward citations (i.e. ‘placed citations’). The net citation count – forward citations minus backward citations – separates the “net producers” of knowledge from the “net consumers”. Using the above methodologies, we have performed three sets of analysis on this network of firms: (a) we investigated the structural aspects of the network, (b) we investigated the position of specific firms in the network, and (c) we computed various centrality indicators for all individual firms. For each of these analyses we also looked at the dynamics by breaking up our data in five distinct time periods (we provide more details below when we discuss our data). The empirical results of the network of firms are presented in Section 4.2.

3.1.3. Third set of analyses: main path approach As we briefly described in the introduction, the main path analysis was pioneered in the late 1980s by Hummond and Doreian [23]. In essence, this is also a network-based technique, but being a connectivity approach it differs substantially from the network analysis that were discussed in the previous section. Hummond and Doreian developed and tested a set of algorithms on a citation network of scientific literature, where papers were considered as nodes, and citations between papers as the links. Their algorithms aimed to identify the main trajectory of knowledge within a body of literature. In the 1990s, several economists realised that the same method could be used on patent citation networks and thus could identify the main technological trajectory in a given field. Even in technology areas represented by thousands of patents, this method can pinpoint those that cumulate the largest amount of knowledge flow. Successful applications of this method have been done for fuel cells [24], medical knowledge [25], and Ethernet LAN communications technologies [27]. We will now provide a brief, simplified explanation of the methodology and the algorithm used; for more details, see [24,44] in particular. First, taking all the selected patents in a given technological area, it is investigated which of these are connected (“linked”) via citations. As a next step, each of the links is given a value (or “weight”). Ref. [23] proposes two alternatives to do this: SPNP and SPLC. The choice between these two algorithms is somewhat arbitrary, as they tend to give similar results in large networks.20 In this paper we use SPLC, which weights each edge proportionally to how often a given link is present all the paths that can be constructed form any startpoint (i.e. patents that do not cite any other patents) to any endpoints (i.e. patents that does not receive any citations). After assigning a weight to each individual link, the algorithm calculates trajectories from each startpoint. Each time a junction is found, the algorithm follows the link that has the highest SPLC value attributed to it; if there is a tie, both directions are followed until an endpoint is reached. In the end, for each startpoint, one or more trajectories will 19 There are several ways to define this “ideal” matrix. In this paper it consists of ‘ones’ in the core block interactions (all the firms in the core are connected) and ‘zeros’ in the peripheral block interactions. See Ref. [43] for details. 20 We calculated the trajectories using both SPLC and SPNP algorithm for the end years 2001, 2002, 2003, 2004 and 2005. The first ten patents of the top main path remained virtually identical for all these variants. As to be expected, the trajectories for the end years varied somehow, and was also dependent on algorithm used. Most notably, the Sharp and Matsushita patents that are in the top path presented in this paper disappeared in some other variants. Here, we note that our extended main path analysis places both companies fairly low, and because of its features is probably more robust in this respect.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1203

Fig. 2. Stylized example of the extended main path analysis.

be found. The trajectory that has the highest cumulative value of all the weights of its links (SPLC values) is called the “top main path” [26]. This path is considered to cumulate the most important knowledge flows in the network and represent the chain of most important inventions, consistent with the idea of technological trajectories put forward by Ref [29]. While the above method is widely found to result in a valid representation of the main path of technological development, and one may of course consider the ownership of the patents that form this path, such an approach is likely to have serious limitations. A typical trajectory includes only a dozen to two dozen patents, even if the knowledge field is as large as 10,000 patents or more. The question is whether such “over selective” trajectories lack the necessary degree of granularity. Some companies might have contributed important knowledge that became part of the trajectory but their patents are not part of the trajectory themselves, just because other patents happened to have a longer “trial” in history. Some studies relaxed the constraints of the top main path analysis by allowing it to produce multiple, divergent main paths. The problem here is that the newly founds paths usually overlap the existing path to a large degree (reference [26] shows some examples). And should new, independent paths emerge, they are likely to relate to different technological trajectories, thus not helping to provide less granular insights into the knowledge position of firms on the dominant trajectory. Recognising these restrictions, this paper proposes an extension to the main path analysis that makes it more apt to evaluate knowledge position. Considering the overall knowledge flows, our extension does not only consider patents that are actually on the main path, but also patents that feed knowledge into this main path, thus contributing to the trajectory. In contrast, we do not consider patents that just “use” knowledge of the main path but do not contribute to the path themselves. Fig. 2 illustrates the resulting three categories of patents. Subsequently, for each firm, we propose to calculate the relative share of patents in the free categories. The new “knowledge position indicator” results from the combination of the share of patents on the trajectory and the share of patents feeding into the trajectory. Using the above methodologies, we have performed both a regular main path analysis for our case, and the proposed, extended analysis that aims to provide a better indication of knowledge positions. The empirical results of these are presented in Section 4.3. 3.2. Data In order to test whether the above methodologies result in good indicators of knowledge positions, we need appropriate data for our selected case. The essential patent analyses, obviously, require a database of essential patents, whereas both networkbased analyses require a data set that contains all patents al all citation relations between these patents. Note that the latter data set is not restricted to essential patents only and is in fact considerably wider. Below, we will briefly describe both data sets. The first dataset, containing essential patents, builds on essential patent declarations as submitted to standards bodies. For the mobile technologies we are interested in, most declarations can be found at ETSI, the body that was involved in setting both the dominant 2G and 3G standards. However, since the dominant 3G standard was actually the result of a partnership between different standard setting bodies (3GPP, see Section 2), some firms might have disclosed their patents to other partnering bodies instead, such as the Japanese ARIB, the other standards body that was involved in early stage 3G developments. To prevent us from overlooking such patents, we complemented the list of ETSI declarations (as retrieved on February 14, 2010) by the list of patent families in the 3G licencing pool (all W-CDMA product categories, as last revised on September 8th, 2010). In total, we found approximately 23,500 ETSI declarations and another 345 patent families in the 3G licencing pool (which is structured differently). The first filtering step consisted of selecting only the declarations referring to 2G GSM or 3G W-CDMA technologies.21 The subsequent cleaning step aimed to identify the patents in the claims. Unfortunately, the information as provided by the patent owners, is far from consistent. Some owners offer patent numbers, other serial numbers of applications, other again publication numbers of applications, 21 More specifically, for GSM we selected the following project identifiers: “GSM”; “GPRS”; “GSM — Release 7”; “GSM/AMR-NB”; “DCS 1800”. For W-CDMA we selected the following project identifiers: “UMTS”; “3GPP”; “3GPP/AMR-WB+”; “3GPP/AMR-WB”; “UMTS/CDMA”; “UMTS Release 8”; “UMTS Release 7”; “3GPP Release 7”; “W-CDMA”; and “UMTS FDD”. The pool declarations by definition refer to W-CDMA.

1204

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Table 5 Patent ownership in the networks at the different time periods.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Firm

Data set

Ericsson Motorola Lucent Qualcomm Nokia NEC Interdigital Samsung Northern Matsushita Phillips Sony Fujitsu NTT Siemens Alcatel Toshiba Mitsubishi LG Hitachi Other Total

877 869 805 762 712 672 444 394 326 312 231 228 223 193 191 161 156 136 124 121 3215 12,289

1976–1985 network

1976–1990 network

4 18

12 49 38 6

28

61

2 5 6 3 2

27 1 11 5 3 2 1

66 150

4 185 452

1976–1995 network

1976–2000 network

Full time period network

140 214 113 61 47 143 33 5 15 31 55 18 33 31 19 35 30 16 2 14 657 1876

663 576 570 414 454 475 156 230 240 197 154 147 129 118 129 123 83 74 70 69 1860 7522

790 744 699 685 633 576 413 335 294 273 189 191 178 177 158 141 131 120 103 97 2530 10,363

Note: all the networks shown in this overview exclude isolates; i.e. patents that have no citing relations at all within the data set. The “Data set” column, however, does include isolates, as well of smaller networks (e.g. networks of just a single patent pair). Dates refer to the application (filing) dates.

and also within these categories there is a wide divergence of formatting. Quite often, the offered identities are incomplete or erroneous. For ETSI, we analysed all records that referred to patents at either the USPTO or the EPO. For the 3GL data, we took patents at the USPTO, EPO, and the Japan Patent Office. Serial numbers of applications were translated into publication numbers (preferably patent numbers, otherwise publications numbers of applications) by using the correspondence tables published by the USPTO.22 Unfortunately, these tables help to identify most patents applications numbers, but not all.23In fact, some companies claimed serial numbers of recent patents of which the application is not yet published (or possible withdrawn). We obviously could not identify such patents. In a next step, in order to validate the patent identity and to have access to patent metadata such as filing date, priority date, and patent family information) we linked all identified patents or applications to the EPO/OECD PATSTAT database. Approximately 88% of the ETSI records and 89% of the 3G licencing pool records could be successfully matched. This resulted in 2987 distinct patent numbers. This is a quite satisfying score, especially if we take into account that not all USPTO serial numbers can be successfully translated (see above) and that recent applications may not be available in PATSTAT because they are not yet published by the patent office. The next cleaning step involved the harmonisation of the firm's names. Duplicate names (over 400 in our database) were handled on caseby-case bases. If we were aware of a transfer of rights, or when such a transfer was registered in the legal registers (as reflected by the INPADOC database) we took the most recent owner. In case of multiple assignees of a patent we selected the “economically most active” owner.24 Finally, we used the concept of patent families to remove duplicate patent claims. In fact, for a single invention, some firms would submit up to dozens of declarations (for each different legislation in which a patent was applied for, multiple patents, re-issued patents, continuations and continuations-in-part), while other firms argued that one declaration sufficed — typically the corresponding USPTO or EPO patent. By using the INPADOC patent family definition, we were able to identify multiple entries for one single invention. This resulted in a significant reduction of overlap: the 2987 patents that we could match with PATSTAT were reduced to 1729 families. We note that the degree of reduction differs greatly by patent and by firm. For one particular invention, we found no less than 73 USPTO or EPO patents, all member of the same INPADOC-defined patent family. The dataset for the diverse network analysis was constructed using the Derwent Innovation Index (DII). One advantage of this database is that patent families are classified in a sensible way (see Ref. [45], among other papers, for a discussion on the different ways in which patent families can be constructed) and the so-called manual code and re-phrased abstracts help to adequately assess the scope of patents. On the basis of keywords and technological classes, aiming at a focussed set yet having a high recall, we identified 17,402 patent families that contained at least one US patent. Some of these families contained more than one US patent, such as patent continuations or divisional patents. (For a discussion on these see Ref. [40]). In such cases, we selected

22

USPTO (2010); “Filing Years and Patent Application Serial Numbers Since 1882”, available from http://www.uspto.gov/patents/process/search/filingyr.jsp. “In general, patent application serial numbers are assigned chronologically to patent applications filed at the U.S. Patent and Trademark Office. For this reason, application serial numbers and filing dates will generally correspond. Please note, however, that there are some applications for which the serial number and filing date may not fall within the time periods indicated in the above table.” http://www.uspto.gov/patents/process/search/filingyr.jsp. 24 For instance, for patents both (independently) claimed by “France Telecom” and by “L'Etat Français”, we selected the former. Similarly, if a one and the same patent was both claimed by a university and by a firm, we chose the latter. 23

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1205

Table 6 Firms claiming 50 or more patent families essential to W-CDMA. Firm

Essentially claimed patent families at ETSI and/or 3GL pool

Nokia Qualcomm Ericsson InterDigital Siemens LG Electronics Motorola NTT DoCoMo Samsung Electronics Other 43 firms Total

356 241 189 164 82 71 58 54 53 461 1729

all family members, resulting a database of 19,196 US patents related to our selected technological field. For constructing the citation relationships between the patents, we utilised the NBER patent database [46]. We used the update of this data set through 2006 that was compiled by Bronwyn H. Hall and made available in March 2009. Note that this data set does not include the most recent patents we retrieved, resulting in a final effective data set of 12,289 patents, with granting dates up to 2006. As the main technology decision for 2G/UMTS was taken in January 1998, and the first release of the standard was published in January 2000, we believe this time frame to be sufficient to analyse the technological field up to and including 3G. This dataset is also used for constructing the firms' network so the identification of the assignee was done via the DII database. 25 In cases where patents were assigned both to individual persons and to companies, we attributed the patent to the company in question. Table 5 shows all firms owning more than 120 patents, and their presence in the various networks (i.e. all patents that are not isolates). Note the relatively long tail; there are another 946 patent owners in the data set, of which 805 own 5 patents or less. To support dynamic analyses, we assigned patent to five different periods according to their application dates, as we believe this data comes closest to the actual invention. For each period, the largest network of connected patents is constructed. The “full time period network” is the network of all patents in our database, regardless their application dates. 4. Empirical analysis Now we present the finding of our empirical analysis, performed on the basis of the methodologies discussed above. It starts with two analyses of claimed essential patents, continues with the three analyses of the network of firms, and finally presents the results of the two main path analyses. For each analysis, where appropriate, we validate the knowledge position findings by comparing them with the outcomes of the technical/historical narrative and the current licencing position, as presented in Section 2.3. We finish in Section 4.4 by providing a more specific comparison between the outcomes of the essential patent analysis and that of the main path analysis. 4.1. Empirical analysis of claimed essential patents In this section, we estimate the knowledge position of firms using claimed essential patents. Table 6 presents the findings of our analysis of claimed essential patents, based on the database that was described in Section 3.2. We counted the number of unique patent families, thus preventing to double-count overlap between identical patents in different jurisdictions, or between continuation patents, divisional, or divisional-in-part patents concerning one and the same invention. By any means, Nokia claims the largest number of different patent families (356 families), almost 1.5 times larger than the second-largest owner, Qualcomm (with 241 families). This is even without the 32 patents claimed by Nokia Siemens Networks, in which Nokia has a 50% share. Ericsson comes just after Qualcomm, claiming 189 families, followed by Interdigital with 164 patents. At some distance, the second league of W-CDMA patent owners follows, each claiming considerable less than 100 patents. This second league includes Siemens, Motorola, and AlcatelLucent, among others. This outcome, and particularly the suggestion that Nokia occupies the strongest knowledge position in this technology, does not match very well with the historical account presented in Section 2. According to this account Nokia was originally on another technology path and “switched” its R&D to W-CDMA only after it was selected in Japan and just before it also was selected in Europe. We would therefore expect a weaker (or definitely not a leading) knowledge position for this firm. At the same time, the claimed essential patent families do not reflect the position that Qualcomm occupies in W-CDMA knowledge. All in all, we conclude that counts of claimed essential patents do not well predict actual knowledge positions. Some better insights into knowledge position are found when looking at the temporal evolution of the claimed intellectual properties. Fig. 3 shows the priority date of the patent families that are (eventually) claimed to be essential. Interestingly, Qualcomm's early work in the CDMA technology area can be easily recognised here. Between 1990 and 1995 it filed more patents in 25 In the DII database, owners are categorised into standardised names using a “who-owns-who”-type of approach, where all subsidiary owners for 50% or more are attributed to a mother firm. Some firms using different legal entities were merged manually.

1206

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Fig. 3. Time patterns of patent families claimed essential at ETSI and/or part of the 3G Licencing pool (five largest owners).

this area than all other companies together. It can also be seen that, of the other companies, Ericsson was the most prolific patentee in the early period, a finding that is in line with our historical narrative. Despite being the largest owner of claimed essential patent families, most of Nokia's patents date from the period just preceding or after the ETSI technology decision is taken. Also this matches our historical narrative, where we found that Nokia switched its research from another technology towards CDMA at a quite late stage. While this temporal analysis is insufficient to provide quantitative estimates of knowledge positions, it certainly provides signals that the value of essential patents is far from homogeneous and counting essential patents is not enough to understand knowledge positions. 4.2. Empirical analyses of the network of firms We now turn to out analyses of the network of firms. As briefly discussed before, we have created such networks for several time periods. In this part of the paper we focus on three most recent sub-networks, as we aim to focus on the 3G technologies, which coincide with these later periods. Table 7 provides a summary of the network size informing about the number of nodes (companies), the number of citations (links), the number of self-citations (loops), and the maximum value of the links (i.e. the maximum number of citations between any two specific firms). Detailed data about the citation behaviour of each of the firms in the whole set can be found in the Annex B, which also reports on the incidence of the self-citing. We observe a steep increase in network size along all the dimensions considered. Table 8 presents the evolution of the key structural indicators, again for different periods. We observe that the measures for density and cohesiveness (i.e. average distance, fragmentation, and reciprocity) grow over time. Given the cumulative nature of the network (in each period nodes and links are added to existing ones) these results are hardly surprising. However, we do observe some interesting features. For instance, despite the significant increase in the percentage of mutual links, reciprocity stabilises over time. This means some firms are persistently only using knowledge, or only producing knowledge. 26Table 8 also presents the development of degree centrality over time. Degree centrality indicates the number of links a node has. Here, it is important to note that citation relations are usually not symmetric: firm A may cite more patents of firm B than the other way around. Therefore, from a network perspective we have to distinguish between indegree centrality (i.e. forward citations) and outdegree centrality (i.e. backward citations). The last two rows in Table 8 show the GINI coefficient of these two measures, which capture the concentration of the knowledge linkages. When we subtract backward citations from forward citations the direction of the resulting net link indicates the direction of the knowledge flow. The increase of the GINI coefficient for outdegree reflects an increasing inequality in the distribution of forward citations by firms. This means that over time, fewer companies' patents are highly cited and therefore technological leadership emerges. On the contrary, the decreasing trend of the GINI coefficient for indegree shows that the extent to which companies “use” external knowledge is stabilising. Figs. 4–6 visualise the resulting network of firms, where the shown links represent cumulative citations, i.e. the sum of citations in both directions. Only links are shown that are larger than a defined cut-off point (reported in the bottom of each graph), and only those firms are shown that have any connections using that cut-off point. The size of the nodes in the figures is proportional to the number of self-citations. The three figures also visualise the outcome of the core/periphery structure approach: companies marked with dark circles belong to the core, and the companies marked with white squares are in the periphery. 1 First of all, we observe that the companies with a significant knowledge position are invariably part of the core network, not the periphery. 26 In this context “use of knowledge” refers to back-citing, whereas “production of knowledge” refers to forward-cited. If on the one hand, this jargon is rather simple, on the other hand, it clearly distinguishes between companies that only cites others and companies that only receive citations.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1207

Table 7 Summary of network size.

1975–1995 Network 1975–2000 Network Full time period network

Number of firms

Citations (links)

Self-citations (loops)

Max value of links

38 47 47

435 1151 1293

29 43 46

149 1075 1622

Secondly, we see that all these companies also have a large number of self citations (shown by the size of the node in the figures; see also Annex B). This could be interpreted as that important companies do also rely considerably on in-house knowledge. One could also argue, however, that firms with large patent portfolios are likely to display higher rates of self-citations, because they themselves represent a larger share of the prior art. The algorithm used for visualising the networks tends to place “similar” nodes closer together, where two nodes are “similar” to the extent that they have similar shortest paths to all other nodes. In the context of this knowledge network, close companies that are shown close to each other have similar citations patterns and therefore tend to exchange knowledge with the same companies. In this respect, we can see how Qualcomm, over time, becomes a central player with other companies “gravitating” it. Among these companies we find Ericsson, Lucent, and – to a lesser extent – Motorola. Here, it is hard to interpret the position of Interdigital. This company does belong to the core network, yet it is rather distant from the other manufacturers. Note that in our historical narrative we pointed out that this particular company indeed seems to occupy a peculiar position in the industry and its actual position in the knowledge network was hard to assess. Finally, we observe that the position if Nokia in these networks is in line with the narrative presented in Section 2. As a latecomer, it starts in the periphery and only later on reaches the core. In order to assess the knowledge positions of individual firms in a more systematic way, we will now consider various centrality measures for the full time period firm network as it was shown in Fig. 6. Table 9 reports the results for betweenness centrality and degree centrality. In our case, betweennes centrality is calculated on the binary transformation of the network, not the weighted network. The table also reports on net citations count. It is calculated by taking the number of forward citations minus the number of backward citations. This way, we can disfurnish the “net producers” and “net consumers” of knowledge. While many of the core firms are net producers of knowledge, Interdigital is again a remarkable outlier, being the only firm that is very large consumer of knowledge. When considering the historical narrative presented in Section 2, we conclude that betweenness centrality is a poor predictor of the actual knowledge position. Outdegree centrality does a somewhat better job, but still does not provide a perfect match with the outcomes of our historical narrative. The firm network analysis discussed above does offer a number of insights in knowledge positions. For instance, it shows that not all the companies contribute to the same extent to the evolution of the technology. In particular, it emerges a small group of active companies “inspire” the research of other companies. For instance, both Qualcomm and Interdigital belong to the core of the network but their role is very different: the citations net count shows a positive contribution of Qualcomm and a negative one of Interdigital. It follows that the latter seems to be able to plug (i.e. to cite) the relevant knowledge, without significantly contributing to it. Looking at betweenness centrality scores also supports this different behaviour. Interdigital displays higher betweenness centrality than Qualcomm highlighting the possibility to access a wider spectrum of knowledge. We can therefore conclude that network indicators allow to distinguish between different innovative behaviours undertaken by companies and to distinguish between truly innovators and companies waiting for the opportunities. While such insights are useful, our overall conclusion is that a firm-level network analysis does not yet allow for an assessment of the knowledge position of individual firms that well matches our historical narrative. 4.3. Empirical analysis using the main path approach The results of the main path analysis for our own case are shown in Fig. 7, which depicts the top main paths that represent the main flows of knowledge in the network over time. Patents are positioned from left to right on the basis of their time of granting. 27 This gives also the feeling of the overlapping part of the trajectories. The patents are labelled after their owner and a unique sequence number. The table in Annex C provides the full owner name and the full patent number. The top main path of the earliest network (1976–1985) includes seven patents, starting with a patent by NEC (labelled NEC1, patent US 4,028,496). Several other patents, owned by Lucent, IBM, ANT Nachrichtentechnik, and several by NEC itself, follow it. From Annex C, which summarises the main focus of each of the patents in Fig. 7, it can be seen that all patents in this earliest network are related to FDMA or TDMA systems, used in 1G or 2G systems. 28 Indeed, we do see the various engineering challenges that were presented in Section 2 above, such as time offset / advance timing and burst synchronisation / formatting. Channel equalisation techniques do not show up in the top main path. Also speech compression techniques are absent, but can be attributed to the fact that our data set focused on radio interface technologies, which is a distinctly different field. Extending the period up to 1990 “bends” the trajectory to

27 Whereas we based our time period distinction on the application dates of patents, this was not a suitable choice for the patent network, as the citation relation is acylcical on the basis of grant date. 28 Note that, because we have chosen an encompassing approach when selecting air interface technologies, we can also see other technologies that just CDMA surfacing, if these technologies have more valuable trajectories at that point in time.

1208

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Table 8 Changes in the knowledge network: descriptive comparative data.

Density with self citations Density excluding self citations Average distance (among reachable pairs) Fragmentation Reciprocity (mutual linkages on total linkages) GINI coefficient for firms' outdegree centrality GINI coefficient for firms' indegree centrality

1975–1995 network

1975–2000 network

Full time period network

2.486 2.110 1.699 0.224 0.571 0.650 0.667

9.3617 7.577 1.468 0.043 0.731 0.694 0.620

13.182 10.582 1.426 0.000 0.743 0.699 0.614

include some other patents, but the technology fields do not change much.29 Interestingly, we do not see any of the companies here that are considered to be the champions of GSM, like Ericsson (after whose demonstrator/“prototype” GSM was modelled), or Nokia.30 The explanation might lie in the history of the 2G GSM standard: because none of the involved companies believed that any of them would use patents as a strategic asset, they simply refrained from patenting their inventions before approx. 1988, leaving many key technologies unpatented. A fascinating example here is the Viterbi coder, which was integrated in GSM in order to deal with one of the main technical challenges, which was multipath fading. As its name suggests, this coder was developed – and remained unpatented! – by the Italian-American inventor Andrew Viterbi, one of the later founders of Qualcomm.31 From the authorised biography of the firm Ericsson [39] it can be understood how vital the Viterbi coder would prove to be for the performance of GSM. If we extend the time period up to the year 1995, something interesting happens: the new trajectory does not include the earlier TDMA patents, but instead now covers CDMA-related patents. The trajectory “breaks”. This is a feature that, to our knowledge, has not yet been observed in papers using this methodology in a technological field. There has been concern that the Hummon and Doreian methodology would have a (too) strong bias towards incremental, continuous technological paths (see [47] for a discussion). Our finding, however, refutes such concerns and shows that if a newer, robust trajectory is emerging, which is solidly linked to other sets of early patents, the methodology is able to abandon the original path instead of trying to stick to it. This third trajectory in Fig. 7, starting at the left bottom, coincides with the early development of the third generation CDMA systems. It starts with a patent from Harris, an American company that produces military equipment. Even though the CDMA technology originates from the military field, this particular patent is not really CDMA related and should be seen as an arbitrary starting point. That is not true for the two following patents, both invented by W. Schmidt of Philips Kommunikation Industrie (PKI) in Nürnberg, Germany, part of the Philips Company. These two patents are the earliest ones in our network actually using the words “Code Division Multiple Access”. Interestingly, the first patent concerns asymmetric multiplex technologies for the up- and downlink — the radio signal from mobile terminal to network and vice versa, respectively. Ultimately, this idea that was used for 3G but it would eventually become a cornerstone technical choice for 4G. If we look at the engineering challenges (see Section 2), we observe that CDMA came with its own, unique set of engineering challenges, often completely different from those relating to 2G/TDMA technologies. The major challenge, power control, is firmly embedded in the trajectory, including US patent No. 5,056,109, invented by K. Gilhousen 32 and assigned to Qualcomm. With no less than 632 forward citations in the Derwent Innovation Index (DII) database, this is one of the top cited patents in the history of the USPTO. The trajectory continues with a number of patents more diverse in nature but all relevant for CDMA. The fourth and fifth trajectory keep the same starting leg as the third one, but bend towards other patent sets, something that is often observed in main path analysis. Power control technologies (including open and closed loop ones) are becoming more and more prominent. Considering the ultimate trajectory, we observe that Qualcomm is the leading patentee. Not only is this the company that owns the largest number of patents in the trajectory (6 in total; see also Table 10, column 2); its patents are also spread over time and in the trajectory. Other prominent companies are Interdigital (4 patents in the ultimate trajectory, of which 3 are endpoints33), Lucent (2 patents) and Philips (2 patents). Finally, we performed an analysis using the new, extended main path approach, which we proposed in the methodology section. Table 10 presents the outcomes, presented at the firm level. While the main trajectory itself – as presented above – encompasses just 20 patents, the group of patents that contributes to the trajectory already includes 660 patents (of a total network of 8057 patents). Obviously, granularity is much better. The knowledge position indicator, shown in the right hand column, represents the estimated knowledge position of all firms. As explained in the methodology section, this proposed indicator is computed by combining the share of patents on the trajectory and the share of patents feeding into the trajectory. We conclude that the

29 Note that two of the three patents encompassed in this new trajectory are end points, and it is known that in the Hummon and Doreian methodology, the resulting start and end points of top main paths may be relatively arbitrary. 30 The patent labelled ERIC1 is assigned to Ericsson GE, a joint venture, but actually is a divisional concerning a patent that General Electric applied for before the joint venture came into existence. 31 David Morton. Andrew Viterbi, Electrical Engineer, an oral history. San Diego, California, United States. 1999-10-29. Retrieved from http://www.ieeeghn.org/ wiki/index.php/Oral-History:Andrew_Viterbi#Linkabit_and_M.2FA-COM.3B_development_and_consumers. 32 This inventor is one of the co-founder of Qualcomm and is listed as inventor in over 47 US patents, often together with another Qualcomm co-founder, I.M. Jacobs (who long served as chief executive officer of this firm). They both feature on two top citing patents, collecting a total of 1160 and 782 citations in DII respectively. Both men worked together on aeronautical research in the 1970s for NASA. 33 Note that endpoints are endogenous selected by the sampling method and the greedy algorithm used. Therefore, their interpretation can be cumbersome.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1209

Alcatel Toshiba

Matsushita Nokia

NEC Motorola

OKI

Philips IBM

Qualcomm

Ericsson Lucent

GE Omnipoint

NTT

Interdigital StanfordTelec

Hughes ITT Northern Fig. 4. Firms' network 1963–1995 (cut-off point 10).

scores on this indicator are indeed very consistent with the historical narrative and actual knowledge position, as summarised in Table 2 of Section 2. In fact, our new indicator provides a better match than any of the earlier analyses. 4.4. Investigating the presence of essential patents on the main path Assuming the main path we found is an accurate description of the most important contributions to the field, one might expect that most of the patents on this main path are indeed claimed to be essential to the standard (but not necessarily all, because the standard might not have employed all the top inventions in the field). Table 11 explores this question. Given the existence of some large patent families (and the observation than not all firms notify each patent family member as essential), we also identified whether any family member was claimed to be essential. As can be seen, approx. 33% of all patents lying on the trajectory is claimed essential (Annex C indicates which patents these are). That share rises to exactly 50% if we do not consider the start points and end points, which might be argued to be somewhat arbitrary. This makes good sense if one considers the interplay between standardisation and technological trajectories. The findings confirm that many important inventions on the top main path are indeed claimed essential. The opposite, of course, is not true: of all claimed essential patents, only very few are on the top path. Furthermore we can see in Table 10 that approx. 20% of all patents contributing to the trajectory is claimed essential. Of those patents that did not contribute to the trajectory, only approx. 9% is claimed essential. So we conclude that while technologically important patents have a higher chance of being claimed essential, the bulk of the patent claimed essential (898 of 1078, so 83%) is technologically not important (i.e. not contributing to the trajectory).

Hyundai OKI Samsung Siemens Nokia NTT

Tantivy

Northern Lucent

Ericsson

Philips Qualcomm

IBM Interdigital

Motorola Alcatel

Matsushita

GE

Omnipoint

NEC

ITT Sony

LG Hitachi Hughes Fujitsu Fig. 5. Firms' network 1963–2000 (cut-off point 40).

1210

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Fujitsu

Alcatel Hyundai

Tantivy OKI

LG

NTT NEC

Qualcomm

Hitachi

Samsung Lucent Northern

IBM

Matsushita

Ericsson

Interdigital

Nokia

GE

Motorola Philips

Omnipoint

Siemens

Hughes

ITT Sony Mitsubishi Fig. 6. Firms' network, full time period (cut-off point 50).

Overall, we conclude that there is only a weak link between the position of a patent in the network and whether the patent is claimed by its owner to be essential. 5. Conclusion and discussion This paper focused on the empirical measurement of the knowledge position of firms in high-tech, standards-based markets. We believe that being able to assess knowledge positions is important because they are assumed to increase chances for sustainable market participation, bargaining power, and licencing revenues. Our case study is on the W-CDMA, the most successful thirdgeneration standard for mobile telecommunications. As a reference point, we determined the actual knowledge position of the relevant firms on the basis of a historical/technical narrative of the development of this standard, as well as an examination of the licencing flows. Our main conclusions are: – The common way to measure knowledge positions in standards-based markets, using on an analysis of essential patents, does not match very well with the actual knowledge positions of firms. The priority date timing patterns of essential patents reveals interesting hints but does not yet allow the determination of knowledge position; – A network analysis of the firm position on the basis of patent citations reveals more insights and allows for the identification of specific outliers, yet does not allow for the systematic determination of knowledge position of individual firms either; – A Hummon and Doreian-type main path analysis does identify the most important technological advances and breakthroughs in the development of this technology, yet is too granular and selective to fully assess knowledge positions of firms; Table 9 Centrality indicators for the 16 firms with the highest outdegree centrality score (full time period network). Company

Betweennes centrality

Outdegree centrality (forward citations)

Indegree centrality (backward citations)

Net citations count

Qualcomm Motorola Ericsson Lucent Nokia NEC Interdigital NTT Philips Northern Toshiba IBM Hughes GE ITT BT

36 82 83 78 58 58 52 27 43 30 66 13 15 1 3 2

3989 2790 2770 2119 1510 1288 897 788 720 640 383 374 322 209 192 105

1682 1402 2126 1712 1558 1151 3012 454 335 658 190 100 160 76 36 29

2307 1388 644 407 − 48 137 − 2115 334 385 − 18 193 274 162 133 156 76

Note: self-citations are not considered.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

ERIC1

NEC3 NEC4 LUC1

1975-2000 Full network

1975-1995

GE2

IBM2

IBM1 NEC2

1975-1985 1975-1990

1211

GE3 QUA2

MOT3

GE1

ANT1

NEC1

MOT2

TER2

QUA7

MOT1

PHI1 HAR1

INTER2

ALC1

SAM4 INTER4 CYL1 LUC3 MAT4 SHA1 QUA6 NTT1 INTER3 SKY1 LUC4 INT1 NOR1 NOK1 ERIC3 QUA5 QUA1 MAT1 SAM1 SAM3 QUA4 NTT3 MAT3 QUA3 NTT2 SAM4 PHI3 SAM2 ERIC2

PHI2

TER3 AIR1 TER1 LUC2

MAT2

1977

1985

1977

2000

2005

Fig. 7. Technological trajectories and the patent's assignees, for five time periods.

– Our suggested extension of the main path analysis, as proposed in this paper, does result in a measurement of knowledge position that by and large matches the outcomes of the historical/technical narrative and an analysis of licencing flows. We believe that our paper offers two main contributions. Firstly, while our study does not counter claims that standard bodies are able to identify and endorse important technologies (compare with [14]), we show that the examination of the technologies included in the standard does not provide good insight into the actual knowledge positions of firms. Our finding that there is only a weak link between the position of a patent in the network (which we interpret as its technical merit) and whether the patent is claimed by its owner to be essential is indeed in line with the earlier findings in [19], which came to a similar conclusion using an entirely different methodology that uses a control group of patents. Second, we believe that our extension to the network-based trajectory analysis, as originally proposed by Hummond and Doreian [23], allows for a wider use of this methodology, beyond identifying breakthroughs and key contributions. A particular strength of the extension we propose is that it is much less granular than the regular analysis, making it well suited for the determination of individual firm's knowledge positions. This may be particularly useful in markets that are not based on technical standards, or markets where such databases of essential patents are available. Our proposed methodology to measure knowledge positions offers a wide range of applications. Because it does not rely on data of essential patents, it can just as well be applied to markets where standards do not play a role, or in markets where standards do play a role but the lists of disclosed essential patents are incomplete or lacking. It is important to realise, though, that by building on citation data networks, the method is unsuitable in areas where inventions are not commonly protected by patents, and is not suitable to investigate very recent technologies, as the network will be too incomplete. (Below, we will discuss limitations in some more detail.) Table 10 Patents on the trajectory and contributing to the trajectory (full network). Firm

Patents on trajectory

Patents contributing to trajectory

Patents non-contributing

Knowledge position indicator (see text)

Qualcomm Interdigital Lucent Motorola Philips Ericsson NEC Nokia NTT Alcatel Matsushita Samsung Harris

6 (30%) 4 (20%) 2 (10%) 0 (0%) 2 (10%) 0 (0%) 0 (0%) 1 (5%) 1 (5%) 1 (5%) 1 (5%) 1 (5%) 1 (5%)

63 (10%) 34 (5%) 56 (8%) 111 (17%) 22 (3%) 76 (12%) 66 (10%) 30 (5%) 28 (4%) 14 (2%) 7 (1%) 6 (1%) 0 (0%)

616 416 641 633 165 714 510 602 148 126 265 328 28

39.5% 25.2% 18.5% 16.8% 13.3% 11.5% 10.0% 9.5% 9.2% 7.1% 6.1% 5.9% 5.0%

1212

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Table 11 Comparison of essentiality claims and position with regard to network trajectory (full network). Patent or any of its family members

Patents on trajectory Patents contributing to trajectory Patents not contributing to trajectory

Claimed essential

Not claimed essential

All patents in network

7 (33.3%) 173 (20.1%) 898 (9.5%)

14 (66.6%) 688 (79.9%) 8581 (90.5%)

21 (100%) 861 (100%) 9479 (100%)

On the policy perspective, this study raises concerns over the technology inclusion processes in standards bodies. Although this study was not design to study the degree to which companies drive patents with little technical merit into standards, we do observe only a weak relation between technically important patents and patents that are claimed to be essential to the standard (Section 4.4). Including patents with little technical merit into a standard does not only result in a needless increase in rents, but can also unnecessarily complicate the standard from a technical point of view, a concern that was recently expressed by the chairman of ETSI's IPR Special Committee when he observed an increase in what he called “marginal patents”. 34 The methodology we propose can be used for critical investigations of the technology inclusions processes that are prevalent at standards bodies. The presented study and proposed methodology has a number of limitations. First, its use is limited to those areas where the large majority of technological contributions are indeed patented. In our particular case this condition is satisfied, but in other areas, trade secrets or other modes of protection might prevail. Second, the choice of measuring firms' position in terms of patents converging to the trajectory assumes that such convergence is positive and desirable. In our example of standardisation, this is indeed the case. But the method might not work when there are two competing, yet quite different dominant designs. Also in areas where knowledge is diverging (e.g. where a paradigmatic shift is about to take place) the method might fail to deliver sensible results. Finally, as with any study based on patents, the outcomes depend strongly on the researchers' ability of to select a patent data set that is well representative for the technological area in question. Recall (i.e. the degree to which all intended patents are present in the sample) as well as precision (i.e. the degree to which the sample includes non-intended patents) should be in balance. This may require specific knowledge of the area in question and might be harder for some technology areas than others. Another limitation that we would like to mention is that current knowledge positions are certainly not the only factor that determines the bargaining positions of firms. From interviews with portfolio licencing managers, we learned that also anticipated knowledge positions matter. A firm such as Huawei (see Section 2.3) may not have a strong position yet, but its investments in R&D for the coming years are so substantial that existing players do offer them attractive cross-licences, securing their access to the future patents of this firm. Finally, our method cannot capture the value of complementary patents that are outside the technological area of the trajectory, yet represent substantial value for other reasons. An example is Apple: just after it settled a court case with Nokia, Apple was granted a broad patent on the user interface gestures that have now become common among many smartphones.35 It is likely that such patents can leverage the positions of their owners, but will not show up in our analysis. List of acronyms. 3G 3GL 3GPP A-TDMA ACTS AMPS ARIB ATSC CDMA cdmaOne CoDIT COST DII DVD EDGE EPO ETSI FDMA FMA FOMA

Third Generation 3G Licencing Programme 3rd Generation Partnership Project Advanced TDMA Advanced Communications Technologies & Services Advanced Mobile Phone System Association of Radio Industries and Businesses Advanced Television Systems Committee Code Division Multiple Access Commercial name of a 3G CDMA system, a.k.a. IS-95 Code DIvision multiple Testbed European Co-operation in the field of Scientific and Technical research Derwent Innovation Indicators Digital Versatile Disc Enhanced Data rates for GSM Evolution European Patent Office European Telecommunications Standards Institute Frequency Division Multiple Access Frames Multiple Access Freedom of Mobile Multimedia Access

34 “[There is a] risk of complicating the solutions just for getting patented technology into the standard rather than to improve the standard.” Presentation of Dirk Weiler at the EC/EPO Workshop on “Tensions between intellectual property rights and standardisation: reasons and remedies”, Brussels, 22 November 2010. Available from http://ec.europa.eu/enterprise/sectors/ict/files/ict-policies/5_weiler_en.pdf. 35 FinancialBin (June 22, 2011). Apple Wins Broad Patent on iPhone Multi-Touch User Interface. Retrieved on June 27 2011 from http://financialbin.com/2011/ 06/22/apple-wins-broad-patent-on-iphone-multi-touch-user-interface/.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1213

Annex A (continued) FRAMES FRAND GPRS GSM HSPDA INPADOC IPR IS-95 LTE LTE-A MC-CDMA NBER NMT OECD OFDM PATSTAT PN codes RACE SPLC SPNP TACS TD-SCDMA TDD TDMA TTA UMTS USPTO W-CDMA

Future Radio Wideband Multiple Access System Fair, Reasonable and Non-discriminatory General Packet Radio Service Global System for Mobile Communications High-Speed Downlink Packet Access International Patent Documentation Center Intellectual Property Right Interim Standard 95 Long Term Evolution Long Term Evolution-Advanced Multi-Carrier Code Division Multiple Access National Bureau of Economic Research Nordic Mobile Telephone Organisation for Economic Co-operation and Development Orthogonal Frequency Division Multiplexing EPO Worldwide Patent Statistical Database Pseudonoise code Research and Development in Advanced Communications Technologies for Europe Search Path Node Pair Search Path Link Count Total Access Communication System Time Division Synchronous Code Division Multiple Access Time Division Multiplexing Time Division Multiple Access Telecommunications Technology Association Universal Mobile Telecommunications System US Patent and Trademark Office Wideband Code Division Multiple Access

Citation matrix for the full network. Cited → Citing ↓

Luc

Fuj

NEC

Phil

NTT

Sie

Mot

Nor

Int

Mit

Tos

Eri

Alc

Qua

Hit

Mat

Son

Nok

Sam

Total

% Self

Lucent Fujitsu NEC Philips NTT Siemens Motorola Northern Interdigital Mitsubishi Toshiba Ericsson Alcatel Qualcomm Hitachi Matsushita Sony Nokia Samsung Total

532 13 80 43 35 16 317 79 58 16 22 293 26 305 9 35 13 122 42 2056

23 32 67 5 29 4 45 12 11 7 16 28 1 37 6 13 8 28 4 376

88 50 299 26 98 14 112 23 39 19 39 134 8 164 30 77 23 72 25 1340

27 4 29 32 7 4 36 11 10 2 6 61 5 35 6 4 6 28 5 318

35 9 44 6 84 5 58 7 7 4 3 54 7 74 9 18 3 52 3 482

25 1 18 16 2 28 41 6 6 4 6 69 17 22 6 12 5 51 2 337

148 9 86 42 18 10 544 35 52 14 23 206 16 405 10 12 9 96 31 1766

70 6 27 6 12 7 88 69 15 2 8 92 10 166 6 3 1 66 24 678

382 11 145 106 81 56 186 40 762 19 21 453 22 772 38 24 29 133 26 3306

21 18 27 10 38 16 58 4 22 40 5 47 6 30 0 21 4 29 18 414

11 3 28 5 9 1 33 1 4 3 64 14 6 16 2 16 4 11 2 233

215 23 120 70 68 12 383 77 90 20 29 844 56 371 14 31 23 244 28 2718

21 5 33 14 13 3 47 3 15 0 9 75 29 35 3 14 4 32 8 363

276 19 78 83 52 36 330 76 72 9 14 195 19 1622 19 16 21 129 41 3107

25 5 35 6 38 6 30 19 20 3 8 31 5 84 42 21 13 38 6 435

35 13 83 10 102 2 74 9 27 6 29 100 3 122 17 91 17 52 5 797

20 2 28 11 15 2 64 7 7 8 15 33 0 44 5 11 65 31 4 372

114 13 100 55 46 18 213 47 49 22 41 360 44 206 11 31 16 569 18 1973

91 19 42 6 36 7 114 51 46 5 20 92 6 242 25 21 11 96 95 1025

2159 255 1369 552 783 247 2773 576 1312 203 378 3181 286 4752 258 471 275 1879 387

25% 13% 22% 6% 11% 11% 20% 12% 58% 20% 17% 27% 10% 34% 16% 19% 24% 30% 25%

Note: Self-citations are shown in bold. Trajectory evolution: analysis of the patents in the top main paths for each time period. Patent #

Assignee

Label

US US US US US US US

NEC NEC IBM NEC NEC Lucent ANT Nachrichtentechnik

NEC1 NEC2 IBM1 NEC3 NEC4 LUC1 ANT1

4,02,8497 4,10,7608 4,34,6470 4,71,5033 4,797,678 4,574,379 4,644,534

1985 1990 1995 2000 Full Priority Main challenge year addressed 1 1 1 1 1 1 1

1 1 1

1977 1979 1981 1985 1985 1986 1986

Claimed essential?

Handling frequency variations Burst synchronisation Burst synchronisation Burst formatting Time offset/advance timing Other Time offset/advance timing

(continued on next page)

1214

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Annex C (continued) Patent #

Assignee

Label

1985 1990 1995 2000 Full Priority Main challenge year addressed

US US US US US US US

4,418,425 4,835,731; 4,905,302; 5,020,132 5,131,007 4,528,656 4,697,260

IBM General Electric/ Ericsson GEa

IBM2 GE1 and GE2 and ERIC1

1 1

General Electric Harris Philips

GE3 HAR1 PHI1

1

US US US US US US US US US US US

4,765,753 5,056,109 5,164,958 5,295,153 5,363,404 5,530,716 5,642,348 5,629,934 5,768,269; 5,966,376 5,950,124

Philips Qualcomm Cylink Ericsson Motorola Motorola Lucent Motorola Terayon

1983 1988

Burst synchronisation Other

1991 1985 1986

Other Frequency allocation Asymmetric multiplexing for upand downlink Handover Power control (loop) Handover Frequency block allocation Other Identification of coded signal Other Power control (loop) Other

1 1

1 1

1 1

PHI2 QUA1 CYL1 ERIC2 MOT1 MOT2 LUC2 MOT3 TER1 and TER2

1 1 1 1 1 1 1 1 1

1 1

1 1

Aironet

AIR1

1

1997

US 6,137,840 US 5,805,583 US 5,267,262 US 5,383,219 US 5,461,639 US 5,570,353 US 5,694,388 US 6,034,952 US 6,385,184; US 6,487,188; US 6,526,032; US 6,590,883; US 6,490,263 US 6,512,931 US 6,654,358 US 6,831,910 US 6,747,969 US 6,868,279 US 6,999,427 US 6,311,070 US 6,795,712 US 6,055,231 US 6,208,632 US 6,490,263c

Qualcomm Terayon Qualcomm Qualcomm Qualcomm Nokia NTT NTT Matsushita

QUA2 TER3 QUA3 QUA4 QUA5 NOK1 NTT1 NTT2 MAT1, MAT2, MAT3, and MAT4

1 1

1997 1998 1993 1995 1995 1995 1996 1997 2000

Samsung Samsung Samsung Philips Ericsson NTT Northern Telecom Skyworks Interdigital Sharp Matsushita

SAM1 SAM2 SAM3 PHI3 ERIC3 NTT3 NOR1 SKY1 INT1 SHA1 MAT4

US US US US

6,564,067 6,748,234 7,106,700 6,934,526

Alcatel Qualcomm Lucent Samsung

US US US US

7,136,666 Lucent 6,907,010; Interdigital 7,126,922 6,985,473 Qualcomm

US 7,009,955

Interdigital

1 1 1 1 1 1 1

1 1 1 1 1

1 1 1 1 1 1 1 1

1987 1991 1992 1993 1994 1996 1996 1997 1997

1 1 1

2000 2000 2000 2001 2001 2001 2002 2004 1998 1999 2000

ALC1 QUA6 LUC3 SAM4

1 1 1 1

2001 2002 2002 2003

LUC4 INTER2, and INTER3

1 1

2003 2004

QUA7

1

2005

INTER4

1

2005

Dynamic parameters (e.g. PN codes) Power control (loop) Modulation/demodulation Power control (loop) Power control (loop) Power control (loop) Power control (loop) Modulation/demodulation SIR Pilot channel & power control combination

Power control (loop) Power control (loop) Signalling Signalling (power) Power control (loop) Power control (loop) Power control (loop) Modulation/demodulation Pilot channel Pilot channel & power control combination Power control (loop) Power control (loop) Dynamic parameters Dynamic parameters / system mode changes Power control (loop) Dynamic parameters

Claimed essential?

x x

x x x x

Xb

x x

Dynamic parameters / system mode changes Power control (loop)

Notes: patents that are members of the same family and present in the same trajectory are shown in one column. The years indicate the periods in question, being 1976–1985; 1976–1990; 1976–1995; 1976–2000; and full network. a The latest of these patents is assigned to Ericsson GE Mobile Communications Inc., a joint venture between Ericsson and General Electric (GE) formed in 1989. It is divisional application of a patent originally applied for by GE in 1987. b The main patent here, patent US 6,385,184, is not claimed essential as such but we found an INPADOC family member that was. c Part of the same patent family as US 6,385184.

Acknowledgments We would like to thank Roberto Fontana, Allesandro Nuvolari, Koen Frenken, Shane Greenstein as well as the participants of DRUID Summer Conference 2010, the Schumpeter Conference 2010 and the 7th European Meeting on Applied Evolutionary Economics (EMAEE 7) for valuable comments.

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

1215

References [1] B.J. Hall, A. Jaffe, M. Trajtenberg, Market value and patent citations, RAND J. of Econ. 36 (1) (2005) 16–38. [2] L. Nesta, P.P. Saviotti, Coherence of the knowledge base and the firm's innovative performance: evidence from the U.S. pharmaceutical industry, The j. of Ind. Econ. 53 (1) (2005) 123 142. [3] M.J. Zack, Developing a knowledge strategy, in: C.J. Choo, N. Bontis (Eds.), The Strategic Management of Intellectual Capital and Organizational Knowledge, Oxford University Press, New York, NY, 2002. [4] O, Granstrand, P. Patel, K. Pavitt, Multi-technology corporations: why they have “distributed” rather than “distinctive” core competencies, Calif. Manag. Rev. 39(4) (1997) 8 25. [5] J. M. Cohen, R. R. Nelson, J. P. Walsh, Protecting their intellectual assets: Appropriability conditions and why U.S. manufacturing firms patent (or not), NBER Working Paper No. 7552, 2000. [6] S.K. Schmidt, R. Werle, Coordinating Technology: Studies in the International Standardization of Telecommunications, The MIT Press, Cambridge, MA, 1998. [7] R. Bekkers, G. Duysters, B. Verspagen, Intellectual property rights, strategic technology agreements and market structure — the case of GSM, Res. Pol. 31 (7) (2002) 1141–1161. [8] K. Blind, R. Bekkers, J. Dietrich, E. Iversen, F. Köhler, B. Müller, Study on the Interplay between Standards and Intellectual Property Rights (IPRs). Final Report, Study commissioned by the European Commission Tender No ENTR/09/015 (OJEU S136 of 18/07/2009), European Commission, Brussels, 2011. Retrieved from http://ec.europa.eu/enterprise/policies/european-standards/standardisation-policy/policy-activities/intellectual-property-rights/index_en. [9] M.A. Lemley, C. Shapiro, Patent holdup and royalty stacking, Tex. Law Rev. 85 (2006) (1991) 2049. [10] D. Geradin, A. Layne-Farrar, A. J. Padilla, The complements problem within standard setting: assessing the evidence on royalty stacking, Boston Univ. J. of Sci. & Tech. 14(2) (2008). [11] D. J. Goodman, R. A. Myers, R. A., 3G Cellular standards and patents, Proceedings from IEEE WirelessCom2005, 2005. [12] A. Layne-Farrar, An empirical assessment of patenting within standard setting. Working paper, Available at SSRN: http://ssrn.com/abstract=1275968, 2008. [13] R. Bekkers, J. West, The limits to IPR standardization policies as evidenced by strategic patenting in UMTS, Telecommun. Pol. 33 (1–2) (2009) 80–97. [14] M. Rysman, T. Simcoe, The performance of standard setting organizations: using patent data for evaluation, J. of IT Stand. & Stand. Res. 5 (1) (2007) 25–40. [15] M. Rysman, T. Simcoe, Patents and the performance of voluntary standard-setting organizations, Manag. Sci. 54 (11) (2008) 1920–1934. [16] Jefferies & Company, Inc., Proprietary Essential LTE Patent Analysis. Retrieved from http://ipcloseup.files.wordpress.com/2011/10/rimm.pdf, 2011. [17] D. L. Martin, & C. De Meyer, Patent counting, a misleading index of patent value: A critique of Goodman & Myers and its uses, 2006, Available at SSRN: http:// ssrn.com/abstract=949439. [18] A. Gambardella, D. Harhoff, B. Verspagen, The value of European patents, Eur. Manag. Rev. 5 (2) (2008) 69–84. [19] R. Bekkers, R. Bongard, A. Nuvolari, An empirical study on the determinants of essential patent claims in compatibility standards, Res. Pol. 40 (7) (2011) 1001–1015. [20] Fairfield Resources International, Review of patents declared as essential to LTE and SAE (4G Wireless Standards) Through June 30, 2009. Darien, CT: Fairfield Resources International, 2010. Retrieved from http://www.frlicense.com/LTE%20Final%20Report.pdf. [21] K. Mallinson, Valuing IP in smartphones and LTE, WiseHarbor, 2011. Retrieved from https://sites.google.com/site/ipfinanceweblog/wiseharbor/ WiseHarborvaluingIP7Nov11pdf.pdf?attredirects=0&d=1. [22] S. Chang, K. Lai, S. Chang, Exploring technology diffusion and classification of business methods: using the patent citation network, Technol. Forcast. Soc. Change 76 (2009) 107 117. [23] N. Hummon, P. Doreian, Connectivity in a citation network: the development of DNA theory, Soc. Netw. 11 (1989) 39 63. [24] B. Verspagen, Mapping technological trajectories as patent citation networks: A study on the history of fuel cell research, Adv. in Complex Syst. 10(1), (2007) 93 115. [25] A. Mina, R. Ramlogan, G. Tampubolon, J.S. Metcalfe, Mapping evolutionary trajectories: applications to the growth and transformation of medical knowledge, Res. Pol. 36 (5) (2007) 789–806. [26] A. Martinelli, An emerging paradigm or just another trajectoryp Understanding the nature of technological changes using engineering heuristics in the telecommunications switching industry, Res. Pol. 4 (2) (2012) 414–429. [27] R. Fontana, A. Nuvolari, B. Verspagen, Mapping technological trajectories as patent citation networks. An application to data communication standards, Econ. of Innov. and New Techn. 18(4) (2009) 311 33. [28] D. Barbera-Tomas, F. Jimenez-Saez, I. Castelló-Molina, Mapping the importance of the real world: the validity of connectivity analysis of patent citations networks, Res. Pol. 40 (3) (2010) 473–486. [29] G. Dosi, Technological paradigms and technological trajectories, Res. Pol. 11 (3) (1982) 147–162. [30] R.N.A. Bekkers, Mobile Telecommunications Standards: GSM, UMTS, TETRA and ERMES, Artech House, Boston, MA, 2001. [31] G.A. Garrard, Cellular Communications: World-wide Market Development, Artech House, Norwood, MA, 1998. [32] F. Hillebrand (Ed.), GSM and UMTS: The Creation of Global Mobile Communication, Wiley, Chichester, UK, 2002. [33] A.T. Manninen, Elaboration of NMT and GSM Standards: From Idea to Market, Dissertation, University of Jyväskylä, 2002. [34] R. Bekkers, I. Liotard, The tense relation between mobile telecommunications standards and IPR, Europ. Intellect. Prop. Rev. 21 (3) (1999) 110–126. [35] R. Barton, Hedy Lamarr: The Most Beautiful Woman in Film, University Press of Kentucky, 2010. [36] A. Couey (1997). About Spread Spectrum. Retrieved from http://people.seas.harvard.edu/ jones/cscie129/nu_lectures/lecture7/hedy/lemarr.htm. [37] G. Calhoun, Digital Cellular Radio, Artech House, Norwood, MA, 1988. [38] European Commission, COST Action 231: Digital mobile radio towards future generations systems, Final report. (EUR 18957), 1999. [39] S. Karlsson, A. Lugn, Changing the World — The Story of Lars Magnus Ericsson and His Successors, Sellin & Partner, 2009. [40] D. Hegde, D. C. Mowery, S. J. H. Graham, Pioneers, submariners, or thicket-builders: which firms use continuations in patenting? NBER Working Paper No. 13153, 2007. [41] P. Criscuolo, B. Verspagen, Does it matter where patent citations come from? Inventor vs. examiner citations in European patents. Res. Pol. 37 (10) (2008) 1892–1908. [42] E. Giuliani, The role of technological gatekeepers in the growth of industrial clusters: evidence from Chile, Reg. Stud. 45 (10) (2011) 1329–1348. [43] S. Wasserman, K. Faust, Social Network Analysis: Methods and Applications, Cambridge University Press, Cambridge, UK, 1995. [44] W. de Noory, A. Mrvar, V. Batagelj, Exploratory Social Network Analysis with Pajek, Cambridge University Press, Cambridge, 2005. [45] Y.V. Sipapin, A.P. Kolesnikov, Patent equivalents: problems of retrieval and patent family tracing in databases, World Pat. Inf. 11 (3) (1989) 139–146. [46] B. H. Hall, A. B. Jaffe, M. Trajtenberg, The NBER patent citations data file: lessons, insights and methodological tools, NBER Working Paper No. 8498, 2001. [47] Martinelli, Ö. Nomaler, Detecting dual patterns of cumulativeness and discontinuity: a genetic approach to patent citation network”, DIME Workshop: Measuring, Mapping and Modelling Technical Change, 6–7 May 2010, Pisa, 2010.

Rudi Bekkers is an assistant professor of Economics of Innovation and Technical Change at the Eindhoven University of Technology (The Netherlands). He earned a PhD. from the same institution with a multidisciplinary study on the development and diffusion of mobile telecommunications standards in Europe. He holds a M.Sc. in Technology and Society and a B.Sc. in Electric Engineering. His research interests focus on: standardisation and intellectual property rights (IPRs) in mobile and other ICT industries, telecommunication policy (especially broadband and mobile) and university–industry knowledge transfer. He also holds a position of senior researcher at Dialogic Innovatie en Interactie, a policy research firm in The Netherlands.

1216

R. Bekkers, A. Martinelli / Technological Forecasting & Social Change 79 (2012) 1192–1216

Arianna Martinelli is a post-doc researcher at LEM (Laboratory of economics and management) at the Scuola Superiore Sant'Anna. She holds a PhD from the Eindhoven University of Technology where she investigates the use of patent citation networks for the investigation of technology dynamics. She holds a B.A. in Economics from Bocconi University (Italy) and a M.Sc. in Industry and Innovation Analysis from SPRU (United Kingdom). He research interests mainly focus on the relation between technological change and industrial dynamics (in particular in telecommunications). Furthermore, she is interested in the empirical representation of technological change by means of patent citation networks and any use (and misuse) of patents data.