System Development of Calibration-Free 20.7-Bit Chipless RFID Without Iterative Optimization of Tag Configurations

Earlier studies have improved the data capacity of frequency-coded chipless radiofrequency identification (RFID) to 100 bits, but these high-density tags may suffer from two limitations, including the design complexity for billions of tags and an additional procedure for measuring clutter. In this paper, a calibration-free chipless RFID system with 20.7-bit capacity is proposed; moreover, the design of the 1.68 million tags is highly efficient, preventing individual and iterative optimization for such large numbers of configurations. These distinct features are obtained by integrating the signal processing of reader and the resonance synthesis of tags. The signal processing employs short-time Fourier transform (STFT) and enhanced filtering to achieve calibration-free detection. The resonance synthesis features minimized numbers of geometric parameters, reduced mutual coupling between resonators, and response surface models for the resonant frequency to realize noniterative optimization. The proposed system is validated by three approaches. First, residual plots indicate that the range of residuals is confined to the average bandwidth of frequency slots. Second, 36 IDs are sampled and automatically transformed into chipless tags. The measured reliability of four bands is as high as 100%, 95.0%, 98.3%, and 93.9%, respectively. Finally, the parameterization of signal processing is validated by time-frequency analysis and reliability.


94
The goal of this paper is to develop a high-capacity and 95 calibration-free system without performing iterative opti-96 mization for the tags. In addition to our preliminary study 97 that encodes 22.1 bits only in simulation [15], this paper 98 constructs the highest overall data capacity with calibration-99 free detection; more importantly, the tags are designed by 100 the noniterative synthesis. 1.68 million tags can be deter-101 mined by mathematical expressions, instead of iterative full-102 wave simulation. The proposed technique consists of reader 103 signal processing based on STFT and specific tags created 104 using response surface models (RSMs) [22]. By reducing 105 the number of geometric parameters of one resonator and 106 mutual coupling between resonators, the resonant frequen-107 cies can be synthesized independently. The capability of the 108 proposed system will be demonstrated through residual plots, 109 the measurement of a 20.7-bit system over 2.0-6.0 GHz, and 110 the time-frequency analysis. The impact of the study is to 111 advance the chipless RFID to real-world implementation with 112 two-fold performance enhancements. 1.68 million IDs can be 113 converted to tag configurations in an efficient way, and their 114 information can be extracted without performing calibration 115 runs.

117
The proposed chipless RFID is illustrated by the setup 118 of communication system, hardware, and software. In par-119 ticular, the setup of communication system is explained 120 by the frequency spectrum and a new encoding scheme 121 that improves guardbands. The hardware and software are 122    In addition, the frequency separations for the third and 156 the twenty-seventh codes are 353 MHz and 273 MHz, 157 respectively. In fact, the frequency separations shown 158 in Figure 2 guarantee a frequency separation of more 159 than 200 MHz. These frequency separations serve the coding 160 as a guardband, which extends the range of adjacent reso-161 nances; accordingly, the risk in frequency aliases is reduced. 162 Following this rule, the second sub-frequency band, and it utilizes frequency resource more efficiently. The fre-180 quency separation is ensured to be greater than 200 MHz, 181 which contributes to sufficient guardbands. In contrast, the 182 conventional frequency shift encoding requires 7 frequency 183 slots for 8 resonators (8 7 = 21 bits); accordingly, adjacent 184 frequency slots are separated by only 45 MHz. This indicates 185 that the proposed encoding reduces the potential aliases due 186 to closely adjacent resonances.

188
To serve the revised frequency shift encoding, a chipless 189 tag employs 8 resonators. The resonator selected is an 190 L-shaped slot, the topology of which is shown in 191 Figure 3. The L-shaped slot is fabricated on a 1.524-mm-thick 192 VOLUME 10, 2022  The reader signal processing employs STFT featuring 211 enhanced filtering. While STFT has been considered as an 212 effective calibration-free technique [10], [11], [12], [13],  The process starts by transmitting frequency modulated con- where S(f , t) is the output in terms of a time-frequency 230 spectrogram, τ is delay time across the window function w(t), 231 and t and f are time and frequency, respectively. This STFT 232 operation leads to the time-frequency spectrogram, as shown 233 in Figure 5(b).

234
The window function and its length are the most cru-235 cial factors in the time-frequency analysis. The window 236  Following the STFT process, the proposed calibration-254 free technique performs the enhanced filtering to distinguish 255 non-resonance components. First, the maximum magnitude 256 is detected, and those signals with a magnitude that is one 257 tenth of the maximum are removed. Afterward, the antenna 258 reflection of the input port is dependent on both time and 259 frequency. The reflection at a lower frequency lasts longer; 260 in contrast, the reflection at a higher frequency continues 261 for a shorter time. When averaging the signal over τ , this 262 bias incurs decreased reliability. Thus, the proposed filtering 263 mechanism identifies this influence and rejects those reflec-264 tion components. This results in an enhanced time-frequency 265 spectrogram, as shown in Figure 5(c). This time-frequency 266 plot provides clarified resonances; more importantly, taking 267 the average of S(f , t) over t is not sensitive to the selected 268 region of t, and thus the subjective judgment for the interval 269 of delay time can be prevented.

270
With the enhanced time-frequency spectrogram at hand, 271 the calibration-free technique computes average frequency 272 responses over signal durations, which are output as the final 273 frequency signature, as shown in Figure 5(d). As such, even 274 though the clutter is not calibrated, the resonances of the 275 chipless tag can be identified clearly.

277
Although the online detection issue is resolved by the 278 calibration-free technique, the offline design problem, 279 namely, constructing the 20.7-bit system and synthesiz-280 ing 1.68 million tags, requires a noniterative synthe-281 sis method. Figure 7 shows the crucial elements of the 282 proposed method. First of all, there are three design 283 parameters for one resonator. More numbers of parameters 284 introduce higher-order interactions, which increase model 285 complexity. Second, arranging 8 resonators together further 286 increases the overall number of design parameters, unless 287 they have insignificant mutual coupling that leads to inde-288 pendent synthesis. Third, the goal of the proposed technique 289 is to build up RSMs, which are the function of the soli-290 tary design parameter. These models should fully explain 291 the relationship between a resonant frequency and that 292 design parameter. Only by integrating the three elements, the 293 1.68 million tags can be designed without iterative tuning or 294 optimization. 295 VOLUME 10, 2022   The test of optimum layout is carried out using 5.2-GHz 322 y-polarized plane waves, which impinge on the two L-shaped 323 slots. Figure 9 shows the simulated x-polarized electric field   To verify the results, the 6 arrangements are further illumi-339 nated by broadband plane waves, and the RCS spectrums are 340 simulated and shown in Figure 10. The resonator originally 341 operated at 4.9 GHz has frequency shifts of 16 MHz, 2 MHz, 342 and 197 MHz, for the three horizontal arrangements, respec-343 tively, whereas the frequency shifts for the three vertical 344 arrangements are 11 MHz, 7 MHz, and 310 MHz. Accord-345 ingly, the open ends are required to be set apart as far as 346 possible.

347
Based on these design rules, the proposed tag geometry 348 with minimized mutual coupling is depicted in Figure 11

388
The 8 RSMs and the training dataset are shown in 389 Figure 13. Good agreement can be observed between the 390 results predicted by (2) and the training data, confirming the 391 significant R 2 obtained (99%). Thus, the resonant frequency 392 can be predicted using an expression (2), instead of iterative 393 optimization or tuning from full-wave simulation.

394
Integrating the 3 characteristics, 8 resonators can be 395 designed independently, and their solitary design parame-396 ter, l, is determined by (2). An ID can be converted into a tag 397 configuration almost instantly. Even though 1.68 million tags 398 need to be coped with, the proposed noniterative synthesis 399 method greatly reduces the computational time from iterative 400 optimization, and this makes 20.7-bit chipless RFID feasible 401 in the real world.

403
The capability of the proposed calibration-free 20.7-bit chip-404 less RFID is examined through three aspects, including model 405 prediction, reliability, and the parameterization in signal 406 processing. 407 VOLUME 10, 2022    were tested over 3 read ranges, denoted by R, including R = 449 10 cm, R = 15 cm, and R = 20 cm. 10 measurements 450 are repeatedly conducted at each distance. The measured 451 results are evaluated as the read reliability along with the 95% 452 confidence interval.

453
First of all, the capability of calibration-free signal process-454 ing is demonstrated in Figure 18. The coding information for 455 the 10 selected IDs is shown in Table 2. The results include 456 the proposed calibration-free technique and the calibrated 457 normalized RCS, determined by    Next, the environment is modified as the carton, and the 486 36 (tags) × 10 (repeated tests) × 3 (read ranges) = 1080 487 experiments were conducted. The reliability in this environ-488 ment is illustrated in Figure 20. As compared to the clutter-489 free scenario, the carton causes slight frequency shifts for the 490 chipless tags. While the reliability is still above 92.2% at the 491 first two sub-frequency bands for R = 10 cm, the results 492 are reduced as 81.3% and 72.9% at the third and fourth sub-493 frequency bands, respectively.

494
Finally, the environment is changed into the plastic bas-495 ket loaded with the sundry items. The clutter responses 496 are more significant than those in the previous scenarios. 497 The measured reliability in this clutter-rich environment is 498 shown in Figure 21. When R = 10 cm, the first and second 499 sub-frequency bands still present reliability of 92.3% and 500 88.0%, respectively. The lowest reliability at this read range 501 occurs at the fourth sub-frequency band (57.1%), indicating 502 that large clutter responses and the loading effect deterio-503 rate the detection. When the read range increases, low SNR 504 reduces the reliability more notably. Considering the pro-505 posed system addresses a 20.7-bit system without implement-506 ing error correction codes, these results validate the proposed 507 system for high SNR scenarios. Future research directions 508 can integrate error correction codes [26], at the expense of 509 data capacity, or assemble an array for each resonator to 510 improve the SNR.     First, the window is varied over Hamming, Gaussian, 520 and rectangular functions, w(t), respectively. Following the 521  In this paper, a calibration-free and high-capacity chipless 551 RFID system has been presented. As compared to the earlier 552 calibration-free chipless RFID, the proposed system achieves 553 the highest data capacity of 20.7 bits; more importantly, the 554 1.68 million chipless tags can be designed without iterative 555 optimization using full-wave simulations, once the RSMs 556 have been established. The noniterative design technique is 557 validated by the R 2 of models, the residuals of prediction, 558 and the reliability of measurement. The R 2 of the 8 RSMs 559 is greater than 99%, and the residuals are confined to the 560 average bandwidth of the frequency slots. The measured 561 reliability of the four bands ranges from 93.9% to 100%. The 562 proposed technique is expected to bridge the gap between 563 the theoretical high-capacity chipless RFID and real-world 564 applications with improved design efficiency.  Since 2013, he has been a Faculty Mem-688 ber with the Department of Electronic Engineer-689 ing, National Taipei University of Technology 690 (NTUT), Taipei, where he is currently a Professor. 691 He has participated in a wide range of research projects, including chipless 692 RF identification, transparent antennas and metasurfaces, mm-wave anten-693 nas and circuits, RF energy harvesting, antenna array failure correction, 694 antennas for body centric communications, microwave reconfigurable com-695 ponents, and multiobjective optimization techniques. His current research 696 interests include chipless sensor networks, high-gain antennas, and inkjet 697 printing technology.