This paper builds a Near-Field Communication (NFC) based localization system that allows ordinary surfaces to locate surrounding objects with high accuracy in the near-field. While there is rich prior work on device-free localization using far-field wireless technologies, the near-field is less explored. Prior work in this space operates at extremely small ranges (a few centimeters), leading to designs that sense close proximity rather than location. We propose TextileSense, a near-field beamforming system that can track everyday objects made of conductive materials (for example, a human hand) even if they are a few tens of centimeters away. We use multiple flexible NFC coil antennas embedded in ordinary and irregularly shaped surfaces we interact with in smart environments—furniture, carpets, and so forth. We design and fabricate specialized textile coils woven into the fabric of the furniture and easily hidden by acrylic paint. We then develop a near-field blind beam-forming algorithm to efficiently detect surrounding objects, and use a data-driven approach to further infer their location. A detailed experimental evaluation of TextileSense shows an average accuracy of 3.5 cm in tracking the location of objects of interest within a few tens of centimeters from the furniture.
This paper seeks to build a Near-Field Communication (NFC) MIMO beamforming system that can accurately localize objects, with or without NFC capability in the near-field. While there has been rich prior work on the device and device-free localization in the far-field, for instance, using technologies such as Bluetooth,1 mm-wave,5 ultrasound,3 RFID,15 and visible light,16 much less exploration exists in the near-field. However, near-field technologies have significant advantages that are worth exploring: (1) their shorter range raises less privacy implications compared to the far-field counterparts; and (2) technologies such as NFC are ubiquitous in our smartphones as well as battery-free everyday objects (for example, credit cards, ID cards, and so on). The few systems that do explore near-field localization in prior work are limited, however, in one of two key ways: (1) first, they only operate at extremely close ranges (for example, few cm), at which point, localization reduces to proximity sensing.11 This excludes ranges of tens of centimeters—an interesting region where both near-field and far-field effects are in play, and (2) second, most prior near-field localization systems require rigid coils that can only be mounted on regular and flat surfaces.6,8
This paper aims to use NFC multi-coil beamforming to detect the presence and location of certain objects of interest within tens of centimeters—where both near-field and far-field effects interplay. We seek to sense certain classes of objects made of conductive material (for example, objects containing metal or human hands) in close proximity (for example, few tens of centimeters). Our system senses these objects with flexible textile-friendly coils attached to existing irregularly shaped home surfaces like a couch or a carpet. We demonstrate how this opens up several applications for human-computer interfaces, gesture, and posture sensing (see Figure 1).
Figure 1. TextileSense can be integrated into ordinary furniture (for example, couch, bed, or carpet). Multiple textile coils sense the presence and location of conductive objects within a few tens of centimeters. This opens up many applications: (a) Body posture sensing; (b) Lost and found: TextileSense notifies the user where he/she left a wallet; and (c) User interface: the blanket serves as a touchless screen to control home appliances.
We present TextileSense—a near-field sensing system for flexible textile surfaces that senses surrounding objects made of conductive materials. TextileSense's core includes multiple textile coil antennas that can be embedded in the furniture covering and readily hidden by puffy paints. The coils together operate as a near-field multiple-in multiple-out (MIMO) system and can manipulate the near-field to recognize objects of interest at unknown locations across farther distances. We consider two classes of objects: (1) Tagged non-conductive objects, such as NFC-enabled credit cards and key fobs, whose identity and location can be obtained; and (2) Untagged conductive objects, such as human hands and metallic objects, whose presence and location can be identified. A detailed experimental evaluation on TextileSense shows a 3.6 cm accuracy in locating NFC tags with various distances and orientations and a 2.9 cm accuracy in locating human hands. We further show that TextileSense can operate accurately at a distance of 20.3 cm, a fourfold improvement over the range limit of 5 cm of commercial NFC.
TextileSense's secret sauce is a mechanism to develop a textile NFC reader with the ability to accurately sense the location of tagged/untagged objects, within a few tens of centimeters of the furniture, regardless of their position or orientation. To do so, we rely on a familiar wireless technology: MIMO. In far-field wireless technologies (such as Wi-Fi, cellular, and so forth), MIMO uses multi-antenna radios to collectively beam signal power toward different spatial directions so as to improve radio coverage and enable advanced location tracking. However, developing the analogous multi-coil MIMO in the near-field poses several new challenges. First, to beam energy accurately and sense an object, one needs to know the direction; yet, for tag-free objects at unknown locations, this is unknown a priori, leading to a chicken-or-egg problem. Second, unlike in the traditional far-field propagation model, even if we do find the optimal beamforming direction, it may not directly relate to the location of objects. The rest of this paper describes how we solve these key challenges to enable near-field MIMO for NFC.
1.1. Detecting objects using near-field beamforming
The first challenge is to address the chicken-or-egg problem: how to know the direction to beam RF energy without knowing the object's location. A natural approach to do so is to apply a variety of beamforming weights and beam energy to different subsets of the space within proximity of the NFC furniture. These beamforming weights must be carefully chosen to fully cover the space around the furniture, yet minimizing the overlap between them to speed up the search. This calls for precise models on how beam-forming weights in the near-field translate into the spatial patterns of energy. While such models have been explored in the far-field (for example, in the RFID context13), doing so for the near-field communication is more complicated. The energized pattern is not only determined by the set of phase shifts applied across the array of NFC readers, but also impacted by the location, orientation, and impedance of the unknown objects.
To address this challenge, TextileSense develops a blind near-field beamforming algorithm to sense objects in the near-field with unknown locations, orientations, and impedance. At a high level, TextileSense uses the magnetic coupling between the object and the reader to infer the optimal beam-forming weights to detect its presence, regardless of whether this object has NFC coils or is made of conductive materials. Specifically, once the object couples with the reader, it can be seen as a high-impedance load to the transmitter circuit (NFC reader). In other words, the transmitter circuit should notice a voltage variation if a load is introduced into the circuit. The voltage variation will change when the load (for example, NFC card, metal, human body, and so forth) harvests more energy from the NFC readers, and this variation can be measured from the transmitter side. By leveraging this physical principle, TextileSense uses a gradient-based approach favoring the set of beamforming weights that introduce large voltage variation into the transmitter circuit. By repeating this process, our algorithm converges to the optimal set of beam-forming weights that sense the presence of objects in proximity. Section 4 further details our solution to detect objects.
1.2. Locating objects in the near-field
Once detected, the next challenge TextileSense must address is to locate the objects of interest. A key challenge here is the limited bandwidth of NFC as well as the non-applicability of traditional far-field MIMO location tracking solutions that rely on the distance between the objects of interest and the reader being farther than one wavelength. To address this, our approach relies on the fact that unlike the far-field, the near-field experiences significant voltage shifts at readers due to coupling that can be reliably measured. TextileSense develops a detailed empirical model of the locations of the object based on the beamforming weights as well as the amplitude of object-related voltage response as perceived by the readers.
1.3. Building NFC-enabled flexible textiles
Finally, TextileSense should integrate textile-compatible coils to build NFC-enabled furniture. In collaboration with material science researchers, we present a novel solution to fabricate coils with conductive fabric, which can be woven into the furniture covering and allows for flexibility and stretchability to ensure user comfort. Section 5 describes how TextileSense is informed by experimentation and analysis to ensure the robustness of its textile coils when subject to bending and crumpling, and how it models the consequent resonant frequency shift and antenna gain degradation.
TextileSense opens up several applications which we briefly explain below and evaluate in Section 6:
We emphasize a few important limitations of TextileSense (1) TextileSense cannot deal with extremely small spacing between multiple objects that need to be simultaneously discerned (within 1.5 mm) due to the strong coupling among them; (2) TextileSense's performance can be degraded by extreme folds or wrinkles of textiles, and our approach explicitly designs solutions to minimize this degradation; and (3) TextileSense NFC readers require a power source; however, it can readily piggyback on access to wall power commonly available in configurable furniture (for example, reclinable couches).
We implement TextileSense on four software-defined radios, each connected to an 18 × 18 cm custom conductive Nylon-based square coil attached to the couch. Our results show:
We propose a localization system design of a MIMO-enabled NFC reader which locates surrounding NFC tags as well as untagged conductive objects. Our system achieves few centimeter level location tracking of nearby tagged and untagged objects and an overall detection range of 20.3 cm from the textile NFC reader.
This section describes the basics of the NFC protocol and the mechanics of near-field magnetic coupling for both tagged and untagged objects.
2.1. NFC protocol
According to the ISO 14443 NFC protocol, an NFC reader initiates communication by periodically broadcasting a universal query command to wake up nearby tags (if any) and solicit responses (standard acknowledgment and unique IDs). Meanwhile, it inevitably experiences magnetic coupling with nearby conducting objects, even if they do not contain NFC coils.
2.2. Magnetic coupling in NFC
The underlying communication principle of NFC is based on magnetic coupling. The NFC reader operates in the 13.56MHz ISM band. The current flowing through the coil antenna of the NFC reader generates a magnetic field that couples with nearby NFC tags or conductive untagged objects. In the rest of this discussion, we use object to denote either a tagged or untagged object that magnetically couples with the NFC reader—the underlying physics remains largely the same. The magnetic coupling effect transfers energy from the NFC reader to the object owing to an induced current in the object. In the near-field, since the strength of magnetic fields decreases rapidly with distance by its inverse 3rd power,10 the communication range of commercial NFC systems is around 5 cm.
Figure 2(a) shows a simplified circuit diagram for a single pair of reader and object, with the latter shown by its equivalent circuit. Due to the magnetic coupling between the reader and the object, the current IT in the reader will induce a voltage VR on the object:
Figure 2. (a) The magnetic coupling between the active NFC reader and an object of interest can be quantified as the mutual inductance m. It induces a voltage VR across the object's equivalent circuit; and (b) A team of reader coils couple with a tagged or untagged object.
where m is the mutual inductance between the antenna of the object and the reader at the resonant frequency; and ZR and IR are the impedance and current induced in the object's equivalent circuit.
As the object perceives an induced voltage from the reader, the current induced also generates its own magnetic field which changes the voltage across the reader antenna. The voltage VT across the reader antenna is written as:
where V'R is the voltage introduced by the object; and V0 is the original voltage on the reader antenna without any object in range. In effect, the object functions as a voltage divider. Hence, we conclude that when there is less energy delivered to the nearby object, the voltage on the corresponding NFC reader antenna will be larger. In the paper, we show how TextileSense leverages this basic property of NFC to detect the presence of nearby objects without knowledge of their orientation, location, and impedance.
TextileSense aims to detect and locate objects in the proximity of a multiple-coil textile NFC reader. It specifically aims to beamform electromagnetic waves in the near-field to detect the influence of conductive objects.
TextileSense's system design is as follows: TextileSense applies different beamforming weights across multiple textile coils of an NFC reader, which can alter the magnetic field to maximize the influence of conductive objects (tagged or untagged) in the near-field. It infers the optimal set of beamforming weights by measuring the voltage across multiple reader coils. The underlying principle relies on the weak magnetic coupling between the object and the reader coils. As we gradually measure the voltage across multiple coils corresponding to different beamforming weight vectors, we can learn the environment and improve the searching of optimal beamforming vectors to discover various objects in the near-field. Once TextileSense discovers an object, it leverages the voltage measurement on the object's influence across reader coils with a data-driven model to locate the object.
We address the key challenges in designing three main aspects of TextileSense:
TextileSense provides a near-field MIMO solution that detects the presence of conductive objects whose location, impedance, and orientation are a priori unknown. We call this near-field blind beamforming, where blind denotes the fact that neither do we have prior wireless channel measurements from the objects nor are we aware of their existence or location. This leads to a chicken-or-egg problem: to beam energy to an object, we need its location, which is precisely what we are aiming to find. Unlike the far-field,13 beamforming weights in the near-field under the NFC context are heavily influenced by the environment, the reader itself, and the presence of conductive objects. In this section, we illustrate how this fundamentally changes our approach to perform blind beamforming.
4.1. Indirect channel measurements
In this section, we describe our approach to detect the presence of passive conductive objects. In the far-field, without prior knowledge of the object's location or wireless channels, the reader would struggle to detect if the object is present or otherwise. In the near-field, however, a reader may detect the presence of a conductive object with no energy source. This is because the object and the reader can magnetically couple with each other. This coupling effect is captured by the mutual inductance between the object and the reader, which is a function of the impedance and location of both the reader and the object. Thus, the mutual inductance plays a role in near-field magnetic channels which is similar to the wireless channel state information in the far-field. We seek to use this information to find the optimal beamforming vector that maximizes the amount of energy delivered to the object. This is critical in improving our location-tracking algorithm given that the amount of energy absorbed by the object gives us important cues about the location of the object.
To obtain the optimal beamforming vector to an object, TextileSense needs to measure its near-field magnetic channel (we deal with multiple objects in Section 4.3). Consider a team of reader coils (see Figure 2). Mathematically, let us assume that a team of N reader coils collaboratively beam energy to one nearby object. The voltage induced by the object at the ith reader coil can be written as:
where mTi is the mutual inductance between the nearby object and the ith reader coil, ITt is the current in the kth reader coil; and ZR is the unknown object's impedance as in the equivalent circuit in Figure 2(b). In this equation, is the voltage introduced to the object by all N reader coils, and can be represented by VR. In other words, the voltage induced at the NFC reader coil is proportional to the mutual inductance, mTi. Indeed, the magnetic channel mTi is critical in performing optimal beamforming of energy toward the object. This is because, for optimal beam-forming, one needs to apply a set of weights to the current of the transmitted signal ITi which can add up the induced signal VR constructively at the object. Based on the channel reciprocity, if we know the channel between the object and each reader coil mTi, one can write the optimal beamforming vector B* as:
where † is the conjugate operator.
However, obtaining the magnetic channel mTi directly from the measured voltage at the reader coil VTi is not straight-forward. The reason is twofold: (1) The voltage induced at a certain reader coil is also influenced by the magnetic channels of other coils. Ideally, one can measure the channel by making all other reader coils open-circuit, then use a known impedance of the object with Equation (3) and apply B* to beamform optimally to the object.2 However, turning off the coils would reduce the voltage and decrease the system's effective range, and (2) we typically do not know the impedance of the object a priori, which influences mTi. The following section details how TextileSense infers the object's voltage with unknown impedance, location, and orientation.
4.2. Finding optimal beamforming vectors
As explained in the previous section, measuring the precise voltage induced at the objects purely from the voltage at a reader coil is challenging due to several unknowns, such as the impedance of the object and the influence of other coils. However, even in absence of these quantities, we can make the following intuitive observation: if the optimal beamforming vector is used across coils to maximize energy delivered to a specific object, the sum of voltage measured across all reader coils should reach a minimum. At a high level, this is because transferring higher net energy to the object will reduce the net energy available to the readers.
To mathematically see why, we revisit Equation (2) and rewrite it by including the mutual inductance between the reader coils. We write the voltage at the ith reader coil as:
where V0i is the voltage of ith reader coil when other reader coils are open-circuit and no other objects are present in the near-field; and VTik is the voltage introduced by nearby reader coils (VTik = mTik ITk, where mTik is the mutual inductance between the ith reader coil and the kth reader coil). These two components can be calculated as known priors, and we use VT0i to represent the sum of them. Therefore, it is easy to see that the voltage at the object is maximized when the voltage at the reader is minimized.
At this point, we can formulate an optimization problem that finds the beamforming weights that minimize the net reader voltage. Assume the space of beamforming vectors has J discrete elements Bj (j = 1, …, J), and is the voltage of the ith reader coil when the beamforming vector Bj is applied. Let , be the initial voltage of the ith reader coil when the beamforming vector Bj is applied without any object present. Specifically, we write . Our objective is to find the beamforming vector that delivers a maximum amount of energy to the nearby object. Given that we assume only one object is in the near-field for now, we can obtain the optimal beamforming vector as follows:
Our analysis shows that for arrays of coils, the space of beamforming weights is locally convex. For example, we analyze a three-coil system with one nearby object while applying various beamforming weights across two coils. Figure 3 plots the sum of the voltage measurement on three coils. It shows a global minimum that delivers maximized energy to the object (see the zoomed-in version in Figure 3(b) ). Hence, we use Stochastic Gradient Descent to perform the optimization.
Figure 3. (a) Sum of voltage (normalized) measured across three coils when two of them apply different beamforming weights with a step of 5° from 0° to 360°. The global minimum represents the beamforming vector that delivers a maximum amount of power to the nearby object; (b) Zoomed-in version in the proximity of the global minimum; and (c) Voltage measurements when another object is present in the proximity of the target object.
4.3. Beamforming to multiple objects
While our discussion so far considers only one object in the near-field, this section deals with the case of multiple objects. In traditional far-field beamforming, multiple objects do not pose a problem, since they do not influence each other. However, in the near-field, multiple objects can potentially couple with each other at the same time. TextileSense, therefore, has to consider multiple objects—if not, the voltage measured from the reader coils will not optimally beam energy to all objects. To see why, we revisit our example in Figure 3(b), add another object, and measure again the sum of the voltage across three coils corresponding to different beamforming weights, as shown in Figure 3(c). We notice that the consequent voltage map varies considerably from the single-object case.
While prior work in the near-field in wireless charging9 can charge multiple mobile phones, it does not guarantee to deliver optimized energy to individual receivers; Hence, it cannot guarantee to detect all objects in the near-field.
As a result, TextileSense must account for multiple objects and decouple their influence on the voltage across reader coils. It then finds the optimal beamforming vector for each object in the near-field. We further note that a simple exhaustive search is too time-consuming to be practical. Therefore, TextileSense needs to maximize the total number of objects found under a limited overall time budget.
TextileSense's high-level approach to do so relies on the voltage measurements from multiple reader coils, and it progressively detects objects in the near-field. It then uses this information to update its optimization algorithm.
Discovering objects. Our approach to discover objects initializes by assuming the presence of a single object in hope of finding a response. We then utilize any response we receive, particularly from nearby objects to infer the presence of other objects. Specifically, we leverage the fact that the responses from nearby objects are impacted by the coupling between objects that are farther away.
To model the coupling among multiple objects, we revisit Equation (3) and rewrite the voltage induced by the object r (r = 1, …, Q) at the reader coil i when the beamforming vector Bj is applied as:
where Zr and Zq are the unknown impedance for the object r and q (in their equivalent circuit representations); is the current in the kth reader coil when the jth beamforming vector is applied; is the corresponding current in the object q; mTi, is the mutual inductance between the ith reader coil and the object r; and mRq, is the mutual inductance between the object r and the object q. Now, we can write the voltage at the ith reader coil as when the beamforming vector Bj is applied.
At this point, we aim to estimate the channel information for each potential object. We set an upper bound Q for the number of potential objects in the near-field. For an N-coil system and R potential objects in the near-field, there are N * Q unknown mutual inductance between the objects and the reader coils, unknown mutual inductance among the objects, and Q unknown impedance of the objects. While there are unknown parameters, we can resolve them by applying different sets of beamforming weights since we obtain N equations from each reader coil every time we apply one beamforming vector. For example, with four coils and five potential objects, we need to apply nine different sets of beamforming weights. While the equations are non-linear, we use Powell's hybrid algorithm7 to solve them.
Choosing beamforming vectors. There are many possible combinations of beamforming vectors to be applied for estimating the channel of potential objects. TextileSense needs to favor the beamforming vector which delivers a larger amount of energy to these objects. In Section 4.2, we formulate an optimization problem to find the beamforming weights that minimize net voltage. TextileSense leverages the beamforming weights along the gradient to estimate the magnetic channels by solving the non-linear equations.
Improving object count estimates. A key to accurately estimating the potential conductive objects in the near-field is to set an appropriate upper bound of the number of them. TextileSense adaptively tunes the upper bound Q based on the responses from tagged objects in the environment, if available, which provide accurate channel information. We always start estimating the number of objects with an initial Q. If there is no response from a tagged object when we apply the estimated channel for potential objects, we increase Q by one. As we gradually receive responses, we can progressively fine-tune our estimates of these parameters with increasing accuracy. In our experiment, we set the initial value of Q to be 5. With this approach, we can decouple the influence of multiple objects on the voltage of the reader coils and calculate the optimal beamforming vector for each object.
4.4. Tagged vs. untagged objects
Telling apart tagged vs. untagged objects. Untagged objects that are conductive and close to the reader will also couple with our coil antennas. Note that TextileSense models the magnetic channels for both tagged and untagged objects in an identical way. In Section 4.3, TextileSense estimates the magnetic channels for all potential objects. With optimal beamforming, TextileSense can discover them in the near-field. Of these objects, NFC tags actively harvest energy in the near-field and can therefore provide a response. We treat the non-responsive objects as untagged conductors.
How well can we detect untagged conducting objects? An important factor that decides how well TextileSense can sense a conductive object is how effectively it resonates with the NFC frequency of operation. We note that different shapes, volumes, and materials of conductors lead to various resonant frequencies. For example, water, mobile phones, computer monitors, and even the human body have distinct resonant frequencies. Our NFC signal is at 13.56MHz, which may not resonate equally well with all classes of objects. Any mismatch lowers the mutual inductance between the reader coil and the object, leading to a small voltage variation at the reader.
Modeling fleeting conductors. While our optimization problem models objects that are static, objects that were computed in the past may no longer exist at the same location in the future. To account for this, TextileSense tracks the magnetic channel of discovered objects. Note that as conductive objects couple with nearby objects, their movement changes the channel of these objects. Thus, TextileSense adaptively tracks the optimal beamforming vectors of the objects based on the voltage feedback from the reader coils. Specifically, TextileSense monitors the variation of the measured voltage across the reader coils, which indicates that the magnetic channels have been changed.
Once TextileSense detects the object using the proposed near-field blind beamforming, it leverages an efficient data-driven algorithm to localize the object.12
This section describes our methods to design and fabricate textile coils. Specifically, we discuss: (1) the design of a coil pattern that maximizes the radiation characteristics within the constraints of the available area, while remaining robust to bending and crumpling; and (2) fabrication methods that integrate textile coils on the furniture.
5.1. Textile coil material and fabrication
Textile coil material. There are primarily two types of conductive fabric: (1) intrinsically conductive fibers; and (2) non-conductive substrates, which are then coated with an electrically conductive element such as copper and silver. A key trade-off that dictates our choice of conductive fabrics to build our coil antennas is the balance between high conductivity and low parasitic capacitance. Intrinsically conductive fibers have better conductivity; yet, woven conductive fibers tend to have large parasitic capacitance due to the spacing between individual threads of fibers that is negative to the performance of coils. In this case, we choose Nickel-Copper fabric as the conductive textile. This conductive textile sheet is made of copper and nickel-coated nylon rip-stop fabric and has an acrylic adhesive layer for better transfer.
Textile coil fabrication. The conductive textile sheet is attached to a 0.4 mm flexible acrylic sheet as the flexible substrate. A laser prototyping system (LPKF U3) is then used to cut the textile sheet into the desired coil shape. The laser scanning parameters are carefully selected to cut through the textile sheet without damaging the acrylic substrate.
5.2. Textile coil design
Our objective is to design a coil geometry with an optimized antenna gain within the furniture's limited area. In this paper, we particularly consider one side of the couch as the designed area to deploy our system (see Figure 5(d)). To achieve an optimized antenna gain, TextileSense needs to consider the trade-off between the trace width and the number of loops. We model the Q-factor of an inductor to capture the efficiency of our coil antenna. Specifically, the Q-factor can be represented as the ratio of the inductance L to the resistance R of a coil at a given frequency. Note that the inductance and the resistance of the coil antenna is a function of the trace width and the number of loops. We then use the trace width and the number of loops as the unknown parameters to empirically optimize for the Q-factor. Our evaluation shows that the optimal design of the textile coil uses nine turns of loops, 8mm trace width of each loop and 2mm gap between loops. We note that the available deployment area depends on the furniture. Our approach can be used to design the optimal configuration for various sizes of furniture.
5.3. Textile bending and crumpling
Figure 4(b) shows the simulated magnetic field of our textile coil without bending (flat). We note that it has high radiation strength and its radiation pattern is perfectly symmetric. However, when the textile coils are deployed on the furniture like a couch, it may not always remain flat. This section describes the effect of bending and crumpling on TextileSense's performance, as measured by degradation in antenna gain.
Figure 4. (a) TextileSense's localization algorithm with four coils; (b) Magnetic radiation pattern of a flat TextileSense coil. The bottom small figures show the radiation strength from Top and Right view; and (c) Magnetic radiation pattern of a curved TextileSense coil with 90° bending angle.
Bending and crumpling effect. We first study the impact of bending on TextileSense's performance when deployed on a couch. We use a bending angle to model the bending effect. The bending angle can be represented as , where W is the length of the square coil, and R is the radius of an imaginary cylinder to which the antenna is bent. We notice that the overall radiation strength suffers from degradation due to bending. This is because when we bend the coil antenna, the resonant frequency of the antenna shifts toward a higher frequency, hence the gain of the coil decreases. We then evaluate the resonant frequency shift and the antenna gain degradation across different bending angles from 20° to 110°. We notice that the resonant frequency shift and antenna gain degradation is quasi-linear with different bending angles. We see a 0.2MHz resonant frequency shift and a 6dB antenna gain decrease with a 110° bending angle. Also, we model the crumpling of a coil using multiple cylinders with different bending angles. We show that our coil antenna has a 9dB antenna gain degradation and 0.24MHz resonant frequency shift when curved by two imaginary cylinders, both with 110° bending angles, from below and above, respectively. TextileSense mitigates the gain degradation by using our near-field beamforming algorithm.
6.1. Object tracking
We show the cumulative distribution function (CDF) of the tracking accuracy of tagged and untagged objects in Figure 5(a) and (c), respectively. TextileSense is able to locate a tagged object within a median accuracy of 2.84 cm. Consider the case where a wallet/watch is accidentally left on the couch. TextileSense is able to quickly detect this situation through its algorithm and notify the user. Further, our system can potentially support gaming such as augmented and virtual reality where the location of objects needs to be known. We show that TextileSense can detect the location of an NFC-tagged plush toy on the TextileSense couch.
6.2. User interface
TextileSense furniture can be potentially used as a touchless screen, and we evaluate this possibility in Figure 5(b). TextileSense can locate a human hand with a median error of 1.53 cm when the user puts his/her hand in close vicinity, making it a promising candidate for touchless screen interfaces. With its ability to locate a human hand, TextileSense can thus track a user's hand once its presence is detected. The user can move his/her hand to form fine-grained gestures, and TextileSense is expected to perform consistent localization to keep tracking and analyzing. We show the user can finely adjust the TV volume by waving the hand over different locations on top of the furniture. Here's a video of our system in action: https://youtu.be/Ieil0NQlk_M.
6.3. Pose estimation
TextileSense can also be used to sense the user when the user sits on the couch. Specifically, it can track the location of the user and also the posture of the user. We demonstrate that our system can sense the user's posture—lying or sitting on the couch with 91.3% accuracy.
This paper designs TextileSense, an NFC-based system that locates objects (tagged or untagged) in the surroundings using multiple textile coils. TextileSense senses the voltage variation of its transmitter coils induced by proximate objects to detect them and identify their location. We optimize the geometry of the coils and fabricate them to remain robust to fabric bending and crumpling. Through extensive experiments, we demonstrate cm-level localization of both tagged and untagged objects in the near-field.
2. Jadidian, J., Katabi, D. Magnetic mimo: How to charge your phone in your pocket. In Proceedings of the 20th Annual International Conference on Mobile Computing and Networking (2014), ACM, NY, 495–506.
3. Knauth, S., Kaufmann, L., Jost, C., Kistler, R., Klapproth, A. The iloc ultrasound indoor localization system at the EVAAL 2011 competition. In International Competition on Evaluating AAL Systems through Competitive Benchmarking (2011), Springer, Berlin, Heidelberg, 52–64.
8. Ranganathan, V., Waters, B.H., Smith, J.R. Localization of receivers using phased-array wireless power transfer systems. In 2015 IEEE Wireless Power Transfer Conference (WPTC) (2015), IEEE, Boulder, CO, USA, 1–4.
15. Yang, L., Chen, Y., Li, X.-Y., Xiao, C., Li, M., Liu, Y. Tagoram: Real-time tracking of mobile RFID tags to high precision using cots devices. In Proceedings of ACM MobiCom (2014), ACM, NY, 237–248.
To view the accompanying Technical Perspective, visit doi.acm.org/10.1145/3615451
The original version of this paper was published in Proceedings of the 2021 ACM/IEEE Int. Conf. on Processing in Sensor Networks.
This work is licensed under a Creative Commons Attribution International 4.0 License. https://creativecommons.org/licenses/by/4.0/
The Digital Library is published by the Association for Computing Machinery. Copyright © 2023 ACM, Inc.
No entries found