Communication Challenges in High-Density Deployments of Wearable Wireless Devices

Wearable wireless devices are very likely to soon move into the mainstream of our society, led by the rapidly expanding multibillion dollar health and fitness markets. Should wearable technology sales follow the same pattern as those of smartphones and tablets, these new devices (a.k.a. wearables) will see explosive growth and high adoption rates over the following five years. It also means that wearables will need to become more sophisticated, capturing what the user sees, hears, or even feels. However, with an avalanche of new wearables, we will need to find ways to supply them with low-latency, high-speed data connections, so as to enable the truly demanding use-cases such as augmented reality. This is particularly true for highdensity wearable computing scenarios, such as public transportation, where existing wireless technology may have difficulty to support stringent application requirements. In this article, we summarize our recent progress in this area with a comprehensive review of current and emerging connectivity solutions for high-density wearable deployments, their relative performance, and open communication challenges.

Keywords: Mobile wearable applications, communication challenges, high-density deployments, short-range radio protocols, ray tracing
Paper by:
Alexander Pyattaev, Kerstin Johnsson, Sergey Andreev, and Yevgeni Koucheryavy

To view the full version, please follow: this link.

Mobile wearables today

Mobile wearable devices are the pinnacle of miniature wireless technology, allowing one to carry inside a wrist-watch what is typically found in a smartphone. Today, all signs indicate that wearable computing (or, simply, wearables) will emerge as the ’next big thing’ within the mobile ecosystem across smartphone manufacturers, application developers, advertisers, and content creators [1].

Even though particular samples of wearable technology have been seen for already 20 years (primarily, in military applications, with soldiers wearing sensors and radios concealed in their uniforms, night-vision and helmet mounted cameras, etc.), they largely remained cumbersome, bulky, and unaesthetic for consumer applications [2]. Generally, only headphones have been truly successful in being adopted in the dress code so far. However, recent progress in mobile communications and battery technology, advanced miniature electronics, materials, and software is enabling increasingly capable, energy efficient, lightweight, and fashionable wearable computing products.

Already today, wearable fitness trackers are revolutionizing the ways people do sports, followed by other categories of wearables, such as smartglasses and smartwatches, often positioned as accessories to modern smartphones. Not limited to individual consumer applications, emerging wearable solutions include rings, healthcare monitors, and even some examples of smart textiles, which all have the potential to benefit multiple market segments.

Analysts predict that wearable technologies will soon create unrivaled market opportunities for a wide variety of players, from apparel manufacturers, IT and telecom companies, to brands, content providers, advertisers, and OEMs [3]. According to the recent predictions, the global wearable device market is expected to grow almost 4000% between 2012 and 2017 with the lion’s share of volumes initially given by fitness bands and then followed by smartwatches and smartglasses [4].

While currently the wearable technology market is dominated by the products released by two industry giants, Apple and Google, we soon expect an avalanche of new wearable devices, such as smartphone compatible watches, innovative healthcare solutions, and many variations of smartglasses [5]. As the result, by year 2018 the wearable industry foresees revenues on the order of 5B to 30B, making for an unprecedentedly large market.

Interestingly enough, the biggest barrier towards mass adoption of wearables does not lie in the field of technology. It appears that such factors as fashion and style are actually forming the customer’s desire to purchase a new device. Therefore, future wearables will need to become more stylish and fashionable, before people start considering them seriously. However, the moment this barrier is lifted, we will experience a huge influx of new devices flooding our daily lives. One curious example in this category is iWatch announcement from Apple, which might just do for watches what the iPhone did for mobile phones.

As more functional and fashionable devices hit the market [6], the main focus will shift from fashion to usability. At that point, it will be critical to deliver wire-equivalent connectivity for the wearable devices, sometimes at very high data rates [7]. For instance, smart glasses have huge potential for any workforce that could benefit from access to hands-free information flows, but those flows need to be streamed to the glasses continuously, and with reasonably low latency.

Here, Google is not the only player, followed by Vuzix, GlassUp, Recon Instruments, and Telepathy. Finally, real, working smartclothing might still be somewhat of a science fiction, but many companies, like OMSignal, Hexoskin, and Athos, are planning to make it happen soon. Again, wireless interaction would be the key for its seamless integration into our society.

In summary, several important changes are expected to occur in the smart wearable device market, partly as a result of developments in the application model, and partly due to the increasing density of wireless links. As new wearables get introduced, we will need to find ways to supply them with low-latency, high-speed data connections [8], so as to enable the truly demanding use-cases such as augmented reality [9]. This article summarizes our progress along these lines with a comprehensive review of current and emerging connectivity solutions for high-density wearable deployments, their relative performance, and open communication challenges.

Setting stage for high-density deployments

As current technology allows for increasingly smaller functional devices, we witness tighter integration of those devices into our daily lives. With modern sensing capabilities, we expect to receive more significant information about the world around us, while making it more intelligent and flexible [10]. Similarly, through wearable computing, we strive to enhance our own capabilities by allowing us to see and hear better, and interact with surrounding technologies easier [11]. However, similar to large-scale sensor networks, the vision of wearable computing requires serious engineering effort before it could be really used by everyone. One of the key issues is the density of the resulting network: if each person is supposed to have several wearable devices, it would be necessary to multiplex hundreds of connections in crowded areas, resulting in connection densities never seen before.

Already today, in any public transportation, one can observe a significant proportion of passengers to be reading an e-book, listening to the music, playing electronic games, or performing some other operations with their mobile devices. Not surprisingly, due to limited capacity of today’s mobile Internet access, most of those activities are happening offline. Furthermore, the vast majority of the audio-visual information is carried by wires, without involving wireless medium at all. Hence, hundreds of passengers can easily enjoy services without interfering with each other, but only as long as those are in no need of wireless links.

If we are to assume that some 25 out of 50 passengers in a conventional bus desire to use Bluetooth 2.0 communications technology for their Hi-Fi stereo headsets at the same time, we would face the simple fact that there is barely enough bandwidth for all of them. And that is just for the average of 0.5 wearable device per passenger, while we could soon be looking at around 5 devices per passenger. In the very near future, we will apparently have to consider the situation that what we can support today in terms of network density is simply not enough – we will need to efficiently multiplex many more wireless links.

Wireless engineers have already been facing similar scalability challenges in sensor networks. While sensor network densities are much lower, the coverage areas typically exceed those for wearable computing applications. As a result, future wearables may just do to the shortrange technology what massive machine-type communication scenarios have already done to the cellular networks. Indeed, in the past technology, the density of service has seldom been the main design target. By contrast, coverage and capacity have traditionally been driving wireless protocol development, and for the majority of use cases it seemed to be sufficient (even though frequent users of WiFi hot spots may argue otherwise).

In this paper, we take a closer look at the emerging challenging scenarios that wearable computing brings along, and investigate if a paradigm shift is needed in the design priorities of the wireless technologies aiming at wearable markets. We also consider the current state-ofthe-art in wearable device and IoT-oriented protocols, as well as several more recent solutions, as potential candidates for the future “wire replacement” in the realm of wearable computing. Finally, we evaluate attainable levels of spatial reuse as well as achievable peak bitrates for the most promising candidate technologies when used in high-density wearable computing scenarios. We conclude by detailing the appropriate design targets that might result in more advanced protocols for future wearable computing applications.

Candidate technologies for mass wearables

Every existing wireless protocol may be viewed as a compromise between simplicity, efficiency, and flexibility. Below we review some of the existing protocols from the point of view of highdensity wearable applications and investigate their relative position with respect to this balance point.

State-of-the-art wireless protocols

Today, WiFi is probably the dominant short-range connectivity solution. We find conventional WiFi interface, based on IEEE 802.11 technology, in close to any mobile device, with emerging IEEE 802.11ac and IEEE 802.11ah extensions targeting the important special cases of high-throughput and low-power applications. The WiFi MAC (medium access control) can usually multiplex around 5-10 users reasonably well, typically achieving spectral efficiency of over 90% [12]. However, when we consider the high-density application scenarios characteristic of future wearable computing, this protocol may have difficulty to support stringent application requirements.

For instance, as follows from our above survey, smartglasses could be one of the major wearable applications, but running a high-definition video signal without wires, albeit making a very attractive selling point, cannot be easily done without several tens of megabits of bitrate. Hence, in emerging protocols aiming to replace Bluetooth and WiFi at some point, this limitation has been addressed quite radically, by scaling up to several orders of magnitude in terms of throughput, from tens to thousands of megabits per second (such as WiGig products based on IEEE 802.11ad). In what follows, we review these novel mmWave solutions in more detail. Meanwhile, a summary of existing short-range radio protocols in wide use today is provided in Table 1.

Table 1: Comparison of current short-range radio technologies

Rate, Mbps

Range, m

Users per cluster

Bluetooth

0.5-20

10-15

8

IEEE 802.11g/n

50-120

10-50

5-10

IEEE 802.11ac

up to 400

10-30

10-15

IEEE 802.11ah

0.1

10-100

5-10

ZigBee

0.5

10-100

5-10

Indeed, our daily experience with WLAN (Wireless Local Area Network) technology confirms that it has difficulty to support some 20 devices attempting to access the channel simultaneously, and has near-zero efficiency with 50 active users. The reason behind this observation is that WiFi has been designed around a certain “typical” number of users that would end up contending for the channel. Any more than that provisioned number – and the network efficiency will degrade dramatically. Conventionally, for short-range radios, the underlying assumption is that there will be at most 10-20 devices within a single collision domain. What happens beyond that “limit” is typically left out of the discussion in respective standards and remains up to the researchers/developers to tackle. As we argue below, if we take a look at the future of short-range radios (namely, the 60 GHz band protocols [13]), we will see a similar situation.

Emerging short-range radio solutions

Today’s short-range radio developments focus mostly on the ultra high frequency (UHF) bands, primarily, the 60 GHz ISM (industrial, scientific and medical) spectrum. Currently, there are several emerging solutions on the market, but all of them are built around similar principles.

The first UHF solution, released in 2008, is commonly known as the Wireless HD standard. It implements a controller-based MAC with both random-access and scheduled operation within the same network. Each Wireless HD network has exactly one controller, that governs every single aspect of its operation, much like a cellular base station would, but only with timedomain multiplexing. However, unlike cellular systems, Wireless HD does not have dedicated procedures for resource negotiation between neighboring networks. It means that whenever multiple networks overlap in space, time, and frequency, they do not have any reasonable means to coordinate their effective schedules. Therefore, all scheduled transmissions that were supposed to occur at the same time would be very likely to fail due to disruptive interference.

Unfortunately, with the primary application of Wireless HD being in audio-visual equipment (e.g., in media centers and home theaters), this issue is not likely to be addressed any time soon, as normally one would not have conflicting media systems at home. This does, however, mean that Wireless HD technology cannot be immediately employed for most wearable applications, as it would only operate when there is no chance of overlapping coverage areas. As it has exactly four frequency channels, one can safely assume that there cannot be more than four networks running wireless HD at any given point of space.

Following Wireless HD, the WiGig standard (implementing IEEE 802.11ad technology) has been released in 2009. It offers a much greater flexibility to short-range wireless systems. In particular, WiGig allows for arbitrary data traffic to be efficiently exchanged between devices, and has much better capabilities to support low-cost devices that may opt out of supporting high-bitrate transmissions. However, when it comes to architecture, WiGig is very similar to Wireless HD: it is also designed around the concept of a central controller, which would assist in scheduling of all transmissions within the network. Just like in Wireless HD, WiGig assumes that there will never be a situation when more than four networks need to coexist, as it does not provide any means to resolve the resulting conflicts between controllers. Similarly, WiGig does not allow for a reliable coexistence of more than four networks at a time.

ECMA 387, released in 2010, is probably the best-suited technology for wearable applications. All the same with the previously discussed options, it utilizes a controller-based MAC. However, on top of that, ECMA 387 is the only short-range standard explicitly covering network mobility by providing mechanisms to remedy possible conflicts of interest that may happen as a result (see Section 16.5.3.11 of the standard for details). The individual subnetworks in the ECMA 387 network have two mechanisms to resolve conflicts of coverage areas: soft channel switch and coordination.

Whenever a conflict is detected, one of the controllers would have to look for alternative frequency channel to use, and if one is found, it will switch to the new channel together with all of the associated devices. However, if that is not possible, ECMA controllers employ a mechanism to keep their reference clocks aligned by adjusting their beacon transmission times. This allows for the beacons from all the overlapping networks to be transmitted one after another. Hence, the controllers have capability to keep track of each other’s scheduled transmissions and avoid unwanted conflicts.

All of that additional signaling, however, adds to protocol overheads, and consequently to the complexity and cost of the devices. Sadly, it also takes a fair amount of time for such a system of networks to converge, and thus an abrupt change in the network structure may be detrimental to all existing connections. In summary, ECMA does allow an arbitrary number of networks to coexist, but not without a cost.

This incurred overhead is at least 21.3 microseconds per superframe of 16.384 ms (about 0.1% of its duration) for each network within the collision domain, plus the fact that no spatial reuse can happen between any two pairs of devices which belong to thus coordinating controllers. In practice, even the raw overhead of extra beacons could not be neglected, as for, say, 200 networks it is already 26% of the superframe resource, and that is assuming that there are no other problems. That said, we can safely assume that ECMA 387 protocol could work reasonably well in public transportation scenarios, as one would not anticipate more than several hundreds of people in a single collision domain. Unfortunately, there are no vendors supporting ECMA 387 technology at the moment of writing this article.

As a result of our technology review, we conclude that emerging radio protocols to be used for short-range communication generally do not adequately address the problem of contention and resource allocation on the scale one would anticipate in the wearable computing scenarios. The original underlying assumption has been, that due to a very fast decay of wireless signal at higher frequencies, it would not matter how well the protocol reacts to conflicting coverage areas, as there would never be any meaningful number of users, not mentioning networks, in one collision domain. In what follows, we demonstrate that this assumption does not hold anymore even in a very conventional wearable scenario.

Image_0010

Figure 1: Motivating high-density scenario: a commuter train

Spatial reuse in short-range communications

In this section, we consider the achievable spatial reuse factors in a typical public transportation environment. Today, a significant fraction of the population takes daily commutes in public transportation on a regular basis. In our motivating public transportation scenario, we consider hundreds of people crowded into regular train cars, airplanes, or buses. There could easily be as many as 200 people in a subway train car, adding up to at least 300 mobile devices (as there are actually more mobile phones than people). To make matters worse, a commuter train could act as a decent waveguide at 2.4 or 5 GHz ISM frequencies (see Figure 1), when traveling inside a tunnel. Below we estimate the anticipated contention levels in modern public transportation by applying advanced radio coverage prediction methods.

Noteworthy, in public transportation scenarios the radio signal is reflected by the walls of the vehicle, reducing the effect of occlusions that would normally be created by people and other objects. To evaluate the effective coverage area of a given device in the vehicle, we utilize a variation of a ray tracing model, which has been specifically designed for the purposes of this research. In our model, we assume free-space propagation if no people occlude the path between the TX and the RX positions. In addition to conventional path loss, at 60 GHz there is also a noticeable contribution of atmospheric absorption, which is modeled conventionally as 15 dB/km in reasonable conditions according to ITU specifications [14].

If there are people occluding the line-of-sight (LOS) path between the TX and the RX, then we assume that the incident radio wave is scattered. This is due to the fact that humans have the reflection coefficient of roughly 0.4 (linear) at 5 and 60 GHz based on [15], and the respective transmission can thus only reach receiver via some other (indirect) paths. One of the effects to account for here is diffraction, which for an obstacle sized as a human being costs about 10-15 dB at both 5 and 60 GHz frequencies, based on the Universal Theory of Diffraction (interested reader is referred here to ITU recommendations on the subject of diffraction). Unfortunately, as no strict knowledge of passenger geometry can be obtained, we pessimistically add a 15 dB penalty in all of the cases as an upper bound on path loss, when the only path is that via diffraction.

Similarly, wireless energy can be reflected from walls, floor, and ceiling of the vehicle, which are typically made of metal, thus resulting in specular reflections with minimal scattering or attenuation. One can readily estimate the path loss in such cases based on the principles of geometric optics. Naturally, reflected signals take a longer route to propagate, and hence typically carry less energy. Additionally, if the path to a reflection surface is occluded, no reflected energy is received. However, in practice it rarely happens that all of the reflected paths are occluded simultaneously, as the passengers do not usually occlude the ceiling of the vehicle. In summary, the ray tracer employed in this study does not claim absolute accuracy, but serves to obtain the first-order understanding by establishing the lower bounds on how far would electromagnetic energy propagate.

As our deployment model for the target commuter scenario, we adopt the existing MOVIA subway train car, which is typically used in large metropolitan areas of San Francisco, London, Shanghai, and many others. We also assume that several cars compose a longer train, as they would normally do in real-world subways. For more technical data on the trains in question, the reader is referred to the manufacturer’s webpage1. In addition, we make an assumption of exactly five wearable devices per passenger, which may even become an underestimation in the future, but for now should provide us with a clear baseline scenario. Finally, we assume that all of the devices are always able to reach their theoretical maximum of throughput, which we set as 60 Mbps per 20 MHz at 5 GHz and 4.620 Gbps per channel at 60 GHz. As our study shows, these are feasible numbers for low-cost devices, whereas higher rates generally require complex beamforming procedures and thus do not fit into the low-cost paradigm.

With our ray tracing tool, we first investigate the signal propagation characteristics in the motivating commuter scenario. The heat map in Figure 2 clearly illustrates how far the radio signal propagates along the train on different frequencies. There are some non-linear effects observed due to occlusions, but we conclude that a general regression would adequately capture the overall dynamics of the signal propagation.

Further, studying in more detail path loss vs. distance, we clearly observe a distinct pattern of dependence. In Figure 3, we illustrate the two particular propagation mechanisms, reflection and diffraction, at longer ranges. What is also important to note here, is that if we assume a reasonable WLAN protocol in operation, it would require an isolation of around 80 dB between the receivers sharing the same frequency channel. The vertical lines in the plot confirm that in a half-empty train those path losses are achieved at distances of 11 and 110 meters for 60 and 5 GHz carriers, respectively. Therefore, our results suggest that the 5 GHz signal easily covers the train car completely and carries sufficient energy even beyond it. On the contrary, the 60 GHz signal is notably attenuated by people and dissipation in free space.

From our results, we also learn that interference with noticeable levels of well above the noise floor travels reliably over sufficient distances, such that multiple passengers with all of their devices become affected. One can estimate how much impact there would be, and the corresponding results are detailed in Figure 4. Clearly, we cannot simply disregard the fact that there could be as many as several hundreds of wearable devices trying to share the channel at the same time. Even with several orthogonal frequency channels (e.g., 4 at 60 GHz and 24 at 5 GHz), it is still barely sufficient to even approach the required numbers. In particular, during peak commuting hours (with up to 200 passengers per a train car), radio protocols at 60 GHz would need to deal with up to 40 devices trying to share the channel. Most importantly, those devices would all belong to different networks, complicating any feasible coordination. At 5 GHz, the situation is even worse, with up to 100 devices in a collision domain on the same channel, which is way beyond workable ranges for both WiFi and Bluetooth protocols.

Image_0030

Figure 2: Signal propagation inside the train, 100 people

Interestingly, we also learn how larger numbers of users inside the train car have a highly non-linear effect on achievable capacity. Even though we may anticipate that the neighboring people would absorb some radio energy, in fact it barely happens, and the increase in user densities has truly detrimental effects on available service quality. For instance, by analyzing Figure 4, we conclude that at already 20 people per train car the use of 5 GHz band for multimedia applications becomes somewhat complicated. Similarly, the anticipated hundreds of megabits rapidly degrade to tens for 60 GHz systems. It is also important to note that the bitrate calculations reported here are theoretical maximums, so in practice we expect even lower numbers due to MAC protocol overheads, as discussed above.

Summarizing, we discover that low-frequency radios face enormous challenges in high-density wearable scenarios. It may even be so that there are simply no means to make them serve the needed numbers of users at any reasonable cost in terms of frequency resource. In contrast, 60 GHz technologies could potentially deliver the desired rates of several tens of megabits per device in the target scenario. On the other hand, the actual protocols that exist today cannot efficiently handle tens of collocated networks operating simultaneously.

Image_0040

Figure 3: Path loss inside the train

Image_0060Image_0050

Figure 4: Network scaling in public transportation scenarios

Towards efficient wearable connectivity

As a conclusion to our evaluation of high-density wearable connectivity, we summarize the key envisioned challenges. First, the existing low-frequency communication protocols (such as WiFi and Bluetooth) simply do not scale well enough to support massive wearable computing deployments. There is a possibility that a new look at the radiated power levels could remedy this to some extent, but then the range and coexistence with other wireless systems would become a problem. Second, the existing 60 GHz protocols, while theoretically having the capacity to serve the desired user densities and enable massive wearable applications, lack efficient mechanisms to handle scenarios where there are tens of neighboring networks overlapping with each other.

One could, however, envision a future wireless system that would be specifically designed to serve high-density wearable deployments. For it to operate, all the necessary functionality would need to be in place to arbitrate competition for resources from hundreds of diverse systems (perhaps even without explicit negotiation due to security/privacy concerns). In addition, a pure random access based protocol may be required due to the fact that any sort of structure imposed by a person-specific coordinator is likely to be disrupted by some other people around.

In summary, wireless engineers currently face a serious challenge of designing an efficient wearable communication protocol that would operate in the very difficult high-density scenarios. To this end, the following requirements to such protocol may be distilled:

  • Densities of ˜4-5, up to 10 nodes per square meter, with some 50-100 devices in a single collision domain.

  • Average link lengths of 1 meter, expected interference range of 10 meters.

  • Highly repetitive, QoS-demanding access patterns typical for sensor networks.

  • Relatively low per-device bandwidth requirements.

  • Very high cluster densities (e.g., each 5 devices may form their own network), with potential cluster mobility.

Today, none of the existing state-of-the-art wireless protocols readily address the demanding high-density wearable scenarios. Whereas certain solutions have been designed to remain scalable under arbitrarily high loads, they still suffer from overheads proportional to the number of users. Unfortunately, when one deals with hundreds of devices in a collision domain, such overheads may turn out to be prohibitive. And that is on top of the fact that nearly all actually scalable protocols are random access based, and therefore cannot guarantee consistent access delays. To bridge this gap, additional efforts are required to engineer efficient radio protocols, based on more scalable approaches, that have known complexity irrespective of the number of active devices in the network. We expect this direction to become and important research trend in the very near future.

Acknowledgment

This work is supported by Intel Corporation, TISE, and the IoT SRA program of Digile, funded by Tekes. The work of the third author is supported with a Postdoctoral Researcher grant by the Academy of Finland.

Bibliography

[1] Juniper Research, Smart Wearable Devices: Fitness, Healthcare, Entertainment & Enterprise 2013-2018, 2013.

[2] M. Billinghurst and T. Starner, “Wearable devices: new ways to manage information,” Computer, vol. 32, no. 1, pp. 57–64, 1999.

[3] IDC, Wearables: Understanding the Global Opportunity.

[4] M. Wilkins, Global Wearable Device Sales by Type: 2012 to 2017. Strategy Analytics, 2013.

[5] ABI Research, Wearable Computing Technologies, 2013.

[6] G. Bailly, J. Muller, M. Rohs, D. Wigdor, and S. Kratz, “ShoeSense: A new perspective on hand gestures and wearable applications,” in Proc. of the SIGCHI Conference on Human Factors in Computing Systems, pp. 1239–1248, 2012.

[7] J. Foerster, J. Lansford, J. Laskar, T. Rappaport, and S. Kato, “Realizing Gbps wireless personal area networks – guest editorial,” IEEE Journal on Selected Areas in Communications, vol. 27, no. 8, pp. 1313–1317, 2009.

[8] B. Fong, P. Rapajic, G. Hong, and A. Fong, “Factors causing uncertainties in outdoor wireless wearable communications,” Pervasive Computing, vol. 2, no. 2, pp. 16–19, 2003.

[9] S. Feiner, “The importance of being mobile: Some social consequences of wearable augmented reality systems,” in Proc. of the 2nd IEEE and ACM International Workshop on Augmented Reality, pp. 145–148, 1999.

[10] R. Fletcher, K. Dobson, M. Goodwin, H. Eydgahi, O. Wilder-Smith, D. Fernholz,

[11] T. Klingeberg and M. Schilling, “Mobile wearable device for long term monitoring of vital signs,” Computer Methods and Programs in Biomedicine, vol. 106, pp. 89–96, 2012.

[12] S. Andreev, A. Pyattaev, K. Johnsson, O. Galinina, and Y. Koucheryavy, “Cellular traffic offloading onto network-assisted device-to-device connections,” IEEE Communications Magazine, vol. 52, pp. 20–31, 2014.

[13] C.-H. Park and T. Rappaport, “Short-range wireless communications for next-generation networks: UWB, 60 GHz millimeter wave WPAN, and ZigBee,” IEEE Wireless Communications, vol. 14, no. 4, pp. 70–78, 2007.

[14] “Recommendation ITU-R P.676-10,” tech. rep., ITU, 2013.

[15] O. Gandhi and A. Riazi, “Absorption of millimeter waves by human beings and its biological implications,” IEEE Transactions on Microwave Theory and Techniques, vol. 34, no. 2, pp. 228–235, 1986.