Thin and flexible printed antenna designed for curved metal surfaces

This paper presents a flexible low-profile antenna suitable for tagging onto metal curved surfaces. The novelty of the proposed antenna lies in the design of an extended ground plane and folding it to realize the bottom layer to shield against metal surfaces. This method reduces the manufacturing complexity as no vias are required in the fabrication process. In addition, the antenna impedance for metal and non-metal surfaces could be obtained through a slight tuning of the slot length. The antenna was realized using a novel low-cost thermal ink transfer process and integrated with a radiofrequency identification chip to successfully demonstrate passive wireless communication at 915 MHz. From the wireless results, a maximum read distance of 1.86 m was obtained for the proposed antenna on metal, compared to a distance of 0.31 m for a reference dipole antenna on metal. Furthermore, the return signal strength indicator pattern obtained show a good correlation to the radiation pattern. The effect of different metal sizes and different metal curvatures were investigated, and results suggest that the effect of these parameters on the impedance mismatch and corresponding power transfer coefficient is not significant. The results are promising for the use of antennas in tagging metal surfaces in the airplane, automotive and unmanned aerial vehicle industries.


Introduction
With the prevalence of the Internet of Things (IoT), there has been growing interests to develop flexible electronics for various applications such as remote health monitoring, supply chain monitoring and agriculture monitoring.In particular, IoT sensing using radiofrequency identification (RFID) tag technology has been demonstrated by several authors using battery-less or passive devices [1,2].Due to a rising demand to employ passive RFID tags for tagging onto the metal surfaces of metal containers, airplane parts and unmanned aerial vehicles [3][4][5], it has become increasingly important to use antennas which are not detuned when placed directly onto metal surfaces.A typical real-life scenario would involve a metal container with automotive parts which is tagged on the surface.The container would be lifted by a forklift fitted with RFID equipment to read the container and driven through an RFID portal which also reads the RFID tag [3].In the aviation industry, the use of RFID tags on plane parts shortens the time needed for parts inventory, where the life span of components could be inspected wirelessly without the need to open the access panels to perform a visual inspection.RFID tag antennas could also be used for maintenance purposes in power plants [6,7].Specifically, conditions in plant operations could be monitored and maintenance performed for a specific component or equipment which has deteriorated and requires repair.
For RFID applications, the antennas need to be flexible for attachment onto different surfaces as well as be low cost.To realize flexible antennas, fabric [8] and film based substrates [9][10][11][12] have been considered.Textile antennas operating at near field communication and Bluetooth applications have been demonstrated [8,13,14].However, the use of conductive threads results in significant losses at higher frequencies due to the skin depth phenomenon.This is since the plated metal thickness is restricted to several hundred nanometers to maintain the fabric properties [15].Furthermore, the surface of a fabric is not smooth, which necessitates the use of surface modification to reduce the surface roughness for printing [15].Film-based substrates [9][10][11][12] which are commonly used for RFID tags include polyethylene terephthalate (PET), polyimide (PI) and polyethylene naphthalate.As the substrate surfaces are very smooth, it facilitates the printing of fine conductive traces using inkjet printing [9,[16][17][18][19][20][21][22][23][24][25].Other deposition techniques for realizing RFID antennas include screen printing [9,[26][27][28][29][30][31], spray coating [32,33] and gravure printing [34,35].At the ultra high frequency (UHF) range considered for RFID applications, inkjet printing is not very feasible to obtain the required thickness due to the small thickness of a single printed layer [36,37], which necessitates the printing of many layers to account for the skin effect phenomenon.Moreover, the metal thickness is non-proportional to the number of overprinted silver layers [38], which makes it challenging to estimate the minimum number of layers required.In addition, the cost of silver ink in inkjet printing is significantly higher compared to copper films.Similarly, the use of silver paste in screen-printing is also not cost-effective.
Typical RFID antennas of the dipole configuration have a known disadvantage in that the performance is detuned when the tag is placed directly onto metal or curved surfaces.From literature, researchers have worked on antennas for either metal or curved surfaces, however there is little information on antennas suitable for both metal and curved surfaces.RFID metal tag antennas have been designed on printed circuit board (FR4) [39][40][41] which are rigid, or on high dielectric constant substrates [42][43][44] which are more costly or not as commonly available.Researchers have also investigated the use of foam spacers to separate the tag antenna from the metal surface [45][46][47] to mitigate the influence of the metal surface.Alternative antenna configurations suitable for metal surfaces such as the microstrip patch [39][40][41] and the inverted-F [44] have also been proposed.However, these configurations involve vias used either as a vertical feeding structure [44] or to connect two metal layers together [39], which increases the manufacturing complexity and cost compared to a single layer design.
The present work focuses on the design and realization of UHF RFID antennas on flexible substrates, for direct placement onto metallic and curved surfaces.Another goal is to have a simple and costeffective process for prototyping the RFID antenna which would also be suitable for large area printing (>100 × 100 mm).In this case, the focus is not on the substrate used, but on a cost-effective approach to fabricate the flexible antennas.As such, a film substrate PI is used in this work to demonstrate the feasibility of the proposed fabrication process.Apart from the high chemical stability of this substrate for etching processes, PI is preferred for its superior thermal stability (up to 400 • C) [8,32], which facilitates the soldering of UHF RFID chips with the proposed antenna to demonstrate passive wireless communication in this work.Furthermore PI is widely used in flexible electronics [12,16,17,19,23,24,32], is commercially available and could be purchased in small quantities for prototyping purposes.Consequently, a cost-effective ink transfer approach is introduced to realize copper patterns on PI substrates.This simple approach could also be applied to other flexible substrates such as copper clad PET substrates.
In this paper, a novel antenna suitable for curved metal surfaces is proposed for RFID tagging applications.The novelties of this work are as follows.Firstly, the antenna is designed with an extended top ground plane layer which is folded to realize the bottom metal layer to shield against metal surfaces.In this way, no vias are required and the manufacturing complexity could be reduced.The novel configuration resembles a slot-dipole configuration and has a low profile with a smaller footprint compared to a typical dipole antenna, which makes it ideal for tagging applications.Furthermore, as the antenna structure and feed line are on the top layer, the antenna impedance could be easily tuned for different surfaces by trimming the key antenna dimensions.
Secondly, a novel thermal ink transfer process suitable for flexible substrates is proposed to fabricate the antenna.From literature, researchers have proposed different transfer printing techniques [48,49] which involve either a printing or retrieval process.In particular, the interaction between the transfer substrate (stamp)/ink interface and the ink/substrate interface determines if a printing or retrieval process occurs.In this work, an alternative method to realize flexible electronics circuitry without the use of a stamp is proposed.This low-cost method is suitable for rapid prototyping and has the potential for large area patterning.Using this novel transfer printing approach, a passive wireless tag operating at 915 MHz was demonstrated with the proposed antenna, which showed minimal influence when placed directly on metal surfaces with different dimensions and varying curvatures.

Proposed design
The proposed antenna design is as shown in figure 1, which is a symmetrical structure designed on PI substrate.This configuration is similar to an embedded T-match [50].However this design involves two metal layers where an extended top ground plane layer is proposed to realize the bottom layer.This is obtained through the folding of the one layer extended ground plane to form the bottom layer, which acts as a shield against metal surfaces.At the same time, the connection between the two layers is realized through the folding of extended ground plane, without the need for vias to join the top and bottom metal layers.In free space, the current distribution of the proposed antenna resembles that of a dipole antenna, and yields a similar radiation pattern (figure 2(a)).With the introduction of a metal surface underneath the bottom layer, the antenna is observed to resemble a patch radiator as shown in figure 2(b).From the simulated results shown in figures 10(b) and (c) of section 4.3.1, it can also be observed quantitatively that the gain and radiation efficiency values of the proposed antenna are not significantly degraded with the introduction of a metal surface, as compared to a dipole antenna.Furthermore, meandering is introduced in the design to facilitate antenna bending.The impedance of the proposed antenna could be adjusted with bending, through the trimming of the slot length dimensions (l).This allows the impedance to be adjusted for different curvatures.
From figure 1, the main antenna parameters considered are the length of the antenna (L), the antenna width (W), the slot width (m) and feed width (w feed ).For the chip integration, the RFID chip is connected to the feed line of the antenna.To use the proposed antenna in free space, the slot length l is set to 13 mm in order to obtain a conjugate match with a RFID chip package of impedance (12.7 + j199)Ω at 915 MHz (SL3S4011, NXP).For the proposed antenna placed on metal, l is tuned to 14 mm.The tuning is realized by trimming the slot length l on both sides of the antenna.The antenna has an overall size of 84 × 15 mm, with a thickness of 1 mm.

Analysis of antenna parameters
The antenna was simulated using a 3D electromagnetic simulator (CST Studio Suite®) as shown in figure 2. The design was simulated using a time domain solver in CST Studio Suite®.The solver uses hexahedral meshing to simulate the structures.The accuracy set for the simulation to terminate is −40 dB and the structure is simulated from 0.1 to 2.1 GHz.Manual meshing was applied to the smaller features to ensure that the structure was sufficiently meshed in those regions.The dielectric constant and loss tangent of PI are defined as 2.7 and 0.02, respectively.The effect of different antenna parameters (L, W, m, w feed ) on the resonant frequency and impedance matching were investigated and the results analyzed.

Dipole design
A dipole antenna was designed as a reference antenna in this work, to directly match the impedance of the NXP chip at 915 MHz.A schematic of the antenna design is shown in figure 3, which is designed on a thin PI substrate of 40 µm thickness with a single copper layer.The length of the antenna is designed to tune the antenna to an operating frequency of 915 MHz while the inner loop is designed to transfer power to the dipole through inductive coupling.Meandering was used for the antenna design to reduce the footprint of the antenna.The impedance value obtained from simulation is (19.7 + j201.8)Ω and the final dimensions of the antenna are 10 cm × 2.4 cm.The impedance of the antenna was characterized using a test fixture [51].

Integration with NXP chip
For comparison, the proposed and dipole antennas were integrated with a UHF RFID chip (SL3S4011, NXP), as shown in the schematic in figure 4. Pins 1 and 2 of the packaged chip are connected to the antenna feed using a thermally stable solder paste (CHIPQUIK TS391AX), which is dispensed onto the copper pads using a syringe.After dispensing, heat is applied to the paste using a soldering iron at 300 • C to enable the wetting of the solder paste onto the pads.

Thermal ink transfer process
A simple and low cost thermal transfer printing method was developed to realize the antenna patterns as shown in figure 5.The advantage of this non-lithographic method lies in that no photomasks, exposure or development steps are required, making the method attractive for rapid prototyping.The process starts with cleaning the single copper layer of the PI substrate with acetone followed by isopropanol to remove substrate contamination.The substrates are then adhered to a transfer substrate on four sides using PI tape.Next, the substrate is sandwiched between plastic laminator sheets and placed into a laminator at 125 • C. The lamination is repeated 20 times to ensure maximum ink transfer.After lamination, the PI substrate is separated from the transfer substrate via peeling.In this work, the ink transferred onto the flexible substrate was observed to behave like a photoresist to protect the copper beneath.Utilizing this property, the copper/PI substrate could be etched in a copper etchant directly after printing.After etching the print layer is removed using acetone, revealing the etched copper pattern underneath.Layers of adhesive (50 µm) and PI (40 µm, 125 µm) are then adhered to the flexible substrate to realize a total thickness of 1 mm.To connect the top and bottom layers, the top copper layer was folded to the bottom layer.Solder was applied to the bottom copper layer to ensure electrical connection across the entire copper layer.

Antenna characterization 3.2.1. Impedance and radiation pattern
The impedances of the two prototype boards were extracted from the S-parameter measurements [51]  using a vector network analyzer (VNA).The measurement frequency range was set at 0.1-2.1 GHz with an averaging factor 128. Before measurement, a twoport electronic calibration was first performed at the SMA connectors of the two RF cables.To characterize the antenna impedance, a test fixture comprising of two semi-rigid cables is attached to the antenna, as shown in figure 6.The ports of the RF cables connected to the test fixture are defined as port 1 and port 2, respectively.
Before measuring the S-parameters of the antenna, the electrical delay of the test fixture has to be accounted for such that the reference point for the S-parameters is shifted from the SMA connectors to the feed point of the antenna [51], as shown in figure 6.After removing the effect of the test fixture, the antenna impedance could be calculated from the following equation using the measured S-parameters [51]: Here R 0 represents the characteristic impedance of the coaxial cables.The radiation patterns were measured in an anechoic chamber at 915 MHz, using an output power of +8 dBm from the VNA and with the antenna rotated from 180 • to −180 • .The radiation patterns were then extracted from the S-parameters using the equation [51]: where E t is the radiation electric field with a cancelled current on the exterior of the coaxial cables and α = −1 for a symmetrical antenna excited by a differential feed.E 1 and E 2 refers to the respective radiated electric field when either port 1 or port 2 of the test fixture is fed.In measuring E 2 the cable was disconnected from port 1 and connected to port 2. During measurements, the disconnected port of the antenna was terminated with a 50 Ω load.

Read distance and return signal strength indicator (RSSI) measurements
The measurement setup to determine the maximum read range and the RSSI is shown in figure 7. The RSSI measurements are performed using a RFID electronic product code (EPC) reader (Unitech RS200-75G5S2G) in an anechoic chamber environment.The unit comes with a patch antenna with a measured gain of 6.9 dBi at 925 MHz, which is used as the reader antenna.The total transmitted power is calculated to be around +31 dBm effective isotropic radiated power (EIRP).To determine the maximum read range, the transmit power of the reader was set at +24 dBm within the frequency range of 916.8-923.4MHz.The read distance from the proposed antenna to the reader was then varied until the EPC code of the RFID chip could no longer be detected by the reader.For the RSSI measurements, a higher transmit power of +30 dBm at the reader was used.
In addition the distance between the reader and the proposed antenna was set at 40 cm in order to obtain reliable measurements of the RSSI across the different angles of rotation.The measurements are obtained for the XZ-plane and YZ-plane, which correspond to the E-plane and H-plane, respectively.

Effect of varying antenna parameters
The antenna impedance could be controlled by varying the antenna length (L), antenna width (W), slot width (m) and feed width (w feed ).A conjugate match between the antenna and RFID chip could be obtained by adjusting the above key parameters.The effect of these parameters was investigated by performing a parametric analysis sweep in the simulation model, with the results shown in figures 8(a)-(d).
Specifically figures 8 and 9 (in section 4.2) provide one with insights on how to design the antenna and determine the appropriate dimensions from the results of the parametric analysis.
From figure 8 as the antenna length L is increased, both the resistance and reactance values are affected by the corresponding shift in the resonant frequency.Figures 8(a  keeping the other dimensions fixed at W = 15 mm, w feed = 0.5 mm and m = 3 mm.The corresponding impedance varies from (34.2 + j180.8)Ω for L at 80 mm to (5.1 + j373.1)Ω for L at 100 mm.The increase in inductance with the increase in L results in a decrease in the resonant frequency.An optimum length of 84 mm is selected for the final design, which corresponds to an impedance of (6.9 + j200.5)Ω at 915 MHz.This yields an impedance match of 90% to the RFID chip.A similar trend in the resonant frequency is observed with an increase in the antenna width (W).Figures 8(c) and (d) show the shift in the resonant frequency in increasing W from 15 mm to 20 mm while keeping L = 84 mm, w feed = 0.5 mm and m = 3 mm.The corresponding impedance changes significantly from (6.9 + j200.5)Ω to (15.2 + j292.7)Ω.In particular as both parameters L and W affect the resonant frequency, it is important not to design the antenna near the resonant frequency as that would result in unstable impedance values near the transition region.A value of W = 15 mm is observed to be a good compromise between the desired impedance and the resulting resonant frequency.On the other hand, varying the slot width m does not affect the resonant frequency but the antenna reactance.A reduced slot width of 1 mm yields an impedance of (5.6 + j182.8)Ωcompared to (6.9 + j200.5)Ω for a slot width of 3 mm with L = 84 mm, W = 15 mm and w feed = 0.5 mm.As the inductance is dependent on the slot length and width [52], a reduction in the slot width would result in a lower inductance value due to a reduction in the overall length of the meander slot.Similarly, the dependence of the inductance on the slot length (l) implies that l could be easily adjusted to tune the inductance (antenna reactance) for different surfaces such as from air to a metal surface.Trimming the slot length (l) rather than adjusting the slot width would allow for a faster tuning of the antenna impedance upon placement onto different surfaces.In this work, l is trimmed from 13 mm (free space) to 14 mm to adjust the antenna impedance for a metal surface.Lastly, in terms of adjusting the feed width (w feed ), increasing w feed from 0.5 mm to 1 mm results in a slight decrease in the antenna reactance, from (6.9 + j200.5)Ω to (5.8 + j187.9)Ω.

Effect of dielectric constant on correlation
For the proposed antenna, the effect of the material properties on the antenna impedance is significant, as most of the fields are concentrated within the 1 mm thickness of the substrate.This is in contrast with the single layer dipole reference antenna designed on a small substrate thickness of 40 µm, where the material properties are not expected to significantly influence the antenna performance.As the material properties of the PI and adhesive are not provided by the manufacturer and the thickness of the PI comprise of 70% of the total thicknesses, the dielectric constant of PI within the range of 2.7-3.5 [53] was considered and simulated.The results are correlated with measurements as shown in figure 9. From the results in figure 9, it is observed that a higher dielectric constant of 3.5 results in a lower resonance.A value of 2.7 for the dielectric constant is observed to yield a good correlation with measurement results.The results are consistent with those reported previously [54], where a change in the dielectric constant was noted to have a significant change on the resonance frequency.The results imply the importance of a knowledge of the dielectric properties towards an accurate design, where literature abounds in this area.This is in particular for the case of a two-layer antenna where the majority of the fields are concentrated within the substrate thickness.

Antenna characteristics 4.3.1. Power bandwidth, gain and radiation efficiency
For an RFID tag antenna, the chip and the antenna both have complex impedances.While antennas are typically designed to match 50 Ω loads, an RFID tag antenna is designed to directly match the complex impedance of the chip instead.Direct matching is necessary to maximize the tag performance without the need for matching circuits.As such, the power reflection [55] rather than the return loss for 50 Ω is important for a passive RFID tag design as it indicates the tag characteristics [46,56].The power reflection coefficient (PRC) [55] is used for the case where complex impedances are directly connected together, and it indicates the amount of maximum power from the generator which is not delivered to the load.For an RFID tag antenna design, the goal is to minimize the PRC between the antenna and the chip.The PRC could be expressed as: where Z ant is the impedance of the antenna and Z chip is the impedance of the RFID chip.The measured PRC results for the proposed antenna on metal (300 × 200 mm) without curvature and on a metal with curvature (minimum diameter of 100 mm) were investigated, with the results shown in figure 10.From figure 10, the resonant frequency is observed to shift slightly from 0.89 GHz to 0.9 GHz for the antenna placed on a flat metal (300 × 200 mm) surface and on a metal surface with curvature of diameter 100 mm (hemi-spherical).From the PRC results in figure 10, the −3 dB threshold indicating the half power bandwidth is not significantly changed for the antenna on different surfaces.For the antenna placed on a flat metal, a bandwidth of 8.2% (0.86-0.93 GHz) is observed, compared to a bandwidth of 10.9% (0.86-0.96GHz) for a curved metal surface.This shows that the proposed antenna maintains a good tolerance when placed from a metal surface onto a curved surface.
Figures 10(b) and (c) show the simulated antenna gain and radiation efficiency in free space and on metal.From the figure, the simulated gain of the proposed antenna in free space is observed to be −5.3 dBi at 915 MHz with a radiation efficiency of −7.5 dB.On a metal surface, the corresponding gain is −8.7 dBi with a radiation efficiency of −14.9 dB.It should be noted that while the degradation in radiation efficiency for the proposed antenna is about 7.4 dB, the gain reduction is only 3.4 dB.This is in contrast to the dipole reference antenna, which has a simulated gain of 2.2 dBi and a radiation efficiency of −0.4 dB in free space.From the Friis equation, the read range is affected by the gain of the tag antenna.Compared to the dipole reference antenna (2.2 dBi gain), the proposed antenna in free space has a shorter read range in part due to a lower gain (−5.3 dBi) as described in section 4.5.Yet when the dipole reference antenna is placed on metal, a degradation of 26 dB is observed for the radiation efficiency, along with a gain reduction of around 39 dB.As the degradation of the dipole reference antenna is significantly worse, the results suggest that the proposed antenna could alleviate the degradation in radiation efficiency when placed on a metal surface.This is in addition to the minimal mismatch loss expected as the proposed antenna is designed for a good impedance match on metal.This phenomenon could also be observed from the read distance results shown in figure 13  From figure 10(b), the simulated gain of the proposed antenna was observed to be negative across the frequency range, which has been reported previously [47] and is common for electrically small UHF tag antennas since the gain of an antenna has been suggested to be limited by its electrical size [57,58].From literature, other UHF RFID tag antennas on metal have reported negative gains ranging from −6.4 dBi to −17 dBi [39,46,47].For an RFID tag antenna, the gain of the antenna is considered to be of less significance compared to the radar cross-section (RCS) [59,60].As such, future work would consider the design of a tag with modulated transmissions.With this function, the tag could send data during one of the continuous CW periods by switching its input impedance between two states, effectively changing its RCS and thus modulating the backscattered field [61].

Radiation pattern
The radiation results in terms of the co-polarization and cross-polarization in the E-plane and H-plane are shown in figures 11 and 12.A good correlation could be observed for the co-polarization results.From figure 11, a difference of ⩽−30 dB is observed between the co-polarization and cross-polarization in the E-plane orientation at 0 • .Similarly from figure 11(b), a difference of around −30 dB could be observed between the co-polarization and crosspolarization levels in the H-plane orientation at 0 • .Some discrepancies were observed between the simulation and measurement results, which could be due to the effect of the cable and the fixture, where the phase patterns of E 1 and E 2 may not have been captured accurately [51] during measurements.This would in turn result in the occurrence of variations during the extraction process using the S-parameter method in [51].To alleviate the variations observed, the S-parameter data from only one of the two ports was used to obtain figures 11(a) and 12(a), based on a symmetry of the antenna structure.In order to obtain results with improved accuracy using the S-parameter method in [51], fine adjustments in the measurement setup and phase correction procedure would be recommended.
The results in figure 11 are observed to resemble that of a dipole antenna (shown in figure 12) particularly in the upper plane.While the dipole is a symmetrical structure with similar radiation pattern, the presence of a metal for the proposed antenna results in a smaller back radiation pattern.For the dipole design, the difference between the co-polarization and cross-polarization in the E-plane and H-plane is ⩽−20 dB at 0 • .In figure 11(b), the simulated H-plane cross-polarization values for the proposed antenna are not shown as they are in the range of −106.3 dB to −120.8 dB, which are too small to be visibly plotted using the software.This is similarly the case for figure 12(b), where the simulated H-plane cross-polarization values lie in the range of −54.2 dB to −98.1 dB.

Maximum read distance on metal
From the radiation results shown in figures 11 and 12, the maximum read distance occurs at an angle of 0 • .At this orientation, the read distance was measured for the proposed antenna on metal compared to free space (i.e. proposed antenna with l = 13 mm), and the dipole reference antenna on metal.It should be noted that the results shown in figure 13 considers not only the effect of the radiation efficiency but also the effect of the impedance mismatch.From the results in figure 13, the proposed antenna on metal yields the longest distance of 186 cm, compared to 105 cm for the proposed antenna in free space and 31 cm for the dipole reference antenna.The variation in the read distance between the proposed antenna on free space compared to metal could be due to a difference in the impedance mismatch.The results correspond well to the theoretical maximum read distance [61], where λ is the wavelength, P t represents the output power of the RFID reader, G t is the gain of the reader antenna and P t G t is the EIRP.G represents the gain of the tag antenna, ρ is polarization mismatch between the tag and the reader antennas (assumed negligible), and τ is the power transmission coefficient.The chip sensitivity P th is −18 dBm.For the dipole reference antenna placed on metal, τ reduces from 0.95 to 0.04, which yields a calculated read distance of 25 cm from equation ( 4).This correlates well to the measured read range.

Effect of ground plane size on antenna impedance and read distance
The effect of different metal sizes on the power transmission coefficient and the corresponding read distance was also investigated.The power transmission coefficient (τ ) is also known as the impedance matching coefficient between the RFID chip and antenna [62], and is calculated from the equation below [63]: where Z c = R c + jX c is the complex chip impedance and Z a = R a + jX a is the complex antenna impedance.The better the impedance match between the antenna and the chip at the frequency of interest (915 MHz), the closer τ is to 1. From figure 14(a), the impedance mismatch is observed to be less significant for the proposed antenna on metal compared to the dipole reference antenna.For different metal sizes considered, τ lies in the range of 0.53-0.80 for the proposed antenna.This contrasts with a significantly lower range of 0.09-0.13for the dipole reference antenna on metal.For the proposed antenna on metal, the impedance match is observed to be relatively low in free space as the dimensions of the antenna are optimized for metal surfaces at l = 14 mm.The parameter τ is also an indicator of the expected read distance r tag , which is proportional to the square root of τ and could be estimated using equation (4).  to 5-31 cm.The results in figure 14 show that the proposed antenna has a good tolerance against metal surfaces of different sizes, and that the power transmission coefficient suffices as a good indicator of the expected read distance due to the proportional relationship between the two parameters.

Angle dependence in wireless measurements
The performance of RFID tags are typically characterized using the read range and RSSI values.From literature, researchers have represented the radiation patterns in terms of the read distance measured at different angles of rotation [46,59,64], with omnidirectional patterns obtained for dipole-type antennas [46,59].Instead of characterizing the maximum read range [42] in this work, the RSSI of the tag is characterized with respect to different angles.The RSSI parameter indicates the power received from the backscattered signal from an RFID tag during the interrogation by a reader, and this parameter is considered to be more suitable for a few reasons.Firstly, it is challenging to determine the maximum read distance due to the small threshold power level to be detected by the reader (∼75 dBm), which gives rise to extremely sensitive readings [42].In addition, the axial ratio for the circularly polarized patch reader antenna is observed to have an axial ratio of around 5 to 6 dB.Due to a variation in the E-field components between two perpendicular planes, this implies that the read distance may not be repeatable, as the reader antenna is not perfectly polarized.Lastly, around 270 bits is required for a full communication between the tag and the RFID reader, which involves the protocol of Select, Query, Query Adjust, Acknowledgement followed by the sending of the EPC from the tag.At the threshold power level, there is a possibility that one may not be read correctly by the reader antenna, resulting in a non-detection of the EPC code.The RSSI patterns obtained are shown in figure 15(a).
Comparing the RSSI patterns with the radiation patterns in figure 15, a good correlation in the trend could be observed for both the XZ and YZ-planes.The variation observed in the RSSI values between the XZ and YZ-planes could be attributed to the axial ratio of the reader antenna, which is observed to be in the range of 5-6 dB from measurements.

Effect of curvature on proposed antenna impedance
From literature, little information is available on the effect of different curvatures on the antenna performance [65,66].Researchers have studied the change in reflection coefficient and the radiation pattern [66] from the bending of 50 Ω antennas [65,66].For RFID antennas, the performance degradation from bending the antenna has been studied through simulation [67].To the authors' best knowledge, an experimental evaluation on the effect of metal curvature on RFID antennas have yet to be investigated.In this work involving a non-50 Ω antenna, the effect of metal curvature was investigated by studying its influence on the power transmission coefficient (τ ).τ is a useful parameter as it indicates the degree of impedance matching for the proposed antenna with the RFID chip and is proportional to the read range.
The effect of metal curvature on the proposed antenna was investigated by adhering the antenna onto different metal planes.These are the XZ-plane, YZ-plane and both XZ and YZ-planes (hemi-sphere) planes.The definition of the XZ-plane and YZ-planes are shown in figures 16(a) and (b).The diameter of curvature ranges from 150 mm to 300 mm (XZplane, YZ-plane) and from 100 mm to 300 mm (both XZ and YZ-planes) with the results shown in figure 16.As the diameter of curvature decreases, τ is observed to decrease correspondingly.Considering the proposed antenna on a flat metal plane, τ = 0.8 and this decreases to 0.65-0.75 in the XZplane.For the YZ-plane, τ is in the range of 0.54-0.56.For bending in both the XZ and YZ-planes, τ is also observed to decrease but remains in the range of 0.5-0.67.The results imply a degradation of the impedance mismatch between the antenna and chip with smaller radius of curvatures [67].The resonant frequency from the antenna impedance plot is observed to decrease slightly with bending, from a value of 1.11 GHz (metal without curvature) to values of 1.07-1.10GHz upon bending in the different planes.This could be attributed to a change in the effective length the antenna upon which results in a of the inductance and the corresponding impedance match with the chip.In terms of the maximum read distance, a ratio of τ = 0.5 from bending in both planes represents a 20% decrease in the read distance compared to the distance obtainable on a flat metal plane (i.e.τ = 0.8).
While simulations were performed to estimate the antenna performance on different curvatures, the obtained results in terms of the expected impedance values did not yield a good prediction for the actual measurement results.This could be due to the stiffness of the actual sample, since the simulation model is assumed to be fully flexible (i.e.zero stiffness) when it is modeled bent at a certain radius of curvature.However, the fabricated sample used for measurements (figure 4) has a certain stiffness due to the stacking of a few PI and adhesive layers (figure 5(i)) to obtain the desired thickness.As a result, the fabricated sample does not conform to the same extent as the simulation model for a particular radius of A comparison of UHF RFID antennas designed for metal surfaces is shown in table 1, with the focus on antennas with small thicknesses.From table 1, while the antenna by Koo et al [46] has a much smaller thickness and is flexible, the read range reported is based on a 1 mm separation from the metal surface.This is also the case for a high impedance surface antenna reported by Chen et al [39].While the antenna is smaller than the proposed antenna, it is designed on a rigid printed circuit board substrate, with the use of vias to realize the electrical connection between two metal layers.On the other hand, a miniaturized antenna with a smaller footprint was demonstrated by Boo et al [47].Yet this is at the expense of a shorter read range compared to the proposed antenna.In addition, the miniaturized antenna [47] was realized on 1.6 mm thick foam, which is less flexible compared to PI or PET substrates commonly used in flexible electronics.A patch-type tag antenna proposed by Mo et al [41] yielded a slightly longer read range compared to the proposed antenna.However it should be noted that the antenna is fabricated on a rigid FR4 substrate with a larger antenna size.Lastly, it could be observed that the proposed antenna has the smallest thickness from the metal surface compared to the existing works presented in table 1.

Conclusion
In this paper, a flexible low-profile antenna suitable for curved metal surfaces has been proposed for RFID applications.The two-layer antenna is designed with an extended ground plane in the top layer, which is then folded to form the bottom layer.The bottom layer acts to shield against the metal surface underneath.In this design, no vias were required for the electrical connection, which simplifies the manufacturing complexity.The antenna was realized using a novel low-cost thermal ink transfer process and integrated with a RFID chip to successfully demonstrate passive wireless communication at 915 MHz.From the wireless test results, the proposed antenna has a maximum read distance of 1.86 m metal, compared 0.31 m for a reference antenna on metal.The measured RSSI patterns were shown to yield a good correlation with the overall radiation pattern.The effect of different metal sizes and varying curvatures was investigated, and the results suggest that the effect of these parameters on the power transfer coefficient is not significant.A minimum value of 0.5 was obtained for the power transfer coefficient, which implies a decrease of 20% in the read distance.The results are promising for the use of the proposed antenna in various industries where tagging onto metal surfaces are required.

Figure 2 .
Figure 2. (a) Simulated antenna structure in free space (b) with a metal surface underneath.

Figure 4 .
Figure 4. Integration of proposed antenna to chip (a) design, (b) fabricated sample.

Figure 7 .
Figure 7. Measurement setup for determining the read range and RSSI.
) and (b) show the simulated antenna impedance in varying L from 80 mm to 100 mm while

Figure 8 .
Figure 8.Effect of varying antenna length L on the (a) real part, (b) imaginary part of impedance and width W on the (c) real part, (d) imaginary part of impedance.

Figure 9 .
Figure 9.Effect of dielectric constant on the antenna impedance for l = 14 mm (a) real part, (b) imaginary part.

Figure
Figure (a) Measured PRC (power bandwidth) of proposed antenna, (b) simulated gain versus frequency, (c) simulated radiation efficiency versus frequency.
of section 4.4.In considering the proposed antenna on a curved metal surface of diameter 100 mm, the simulated antenna gain in figure 10(b) was observed to degrade slightly to −11.2 dBi while the radiation efficiency remains almost unchanged at −15.3 dB.

Figure 11 .
Figure 11.Co-polarization and cross-polarization of proposed antenna in (a) E-plane and (b) H-plane.

Figure 12 .
Figure 12.Co-polarization and cross-polarization of dipole antenna in (a) E-plane and (b) H-plane.

Figure 13 .
Figure 13.Maximum read distance from wireless

Figure 14
(b)   shows the measured read distance with respect to different metal sizes.From figure 14(b) the read distance is observed to yield a similar trend to figure14(a).For different metal sizes, the read distance for the proposed antenna on metal remains in the range of 120-190 cm.In comparison, the range for the dipole reference antenna is reduced significantly from 458 cm

Figure 14 .
Figure 14.Effect of different metal sizes on (a) power transmission coefficient and (b) maximum read distance.

Table 1 .
Comparison with reported UHF RFID tag antennas on metal.Calculated using equation (4) based on measured results at 1.2 W EIRP. a