Design and performance evaluation of communication protocols in rfid systems

ia = i− a and Ka = K − a. We now determine Ps(a|i) in (2.20) by analyzing the effect of background noise on the tag detection. Let’s denote Ps(j) a probability that the j-th tag is successfully detected, we have Ps(j) = ( 1− PQDDe (j) )MID , (2.25) where PQDDe can be obtained from (2.19). Here, it is also worthy to mention that all MID bits of the tag’s ID are required correctly received for a successful detection. As 33 a result, Ps(a|i) can be found as follows Ps(a|i) = ∏ j∈{a} Ps(j), (2.26) where a tags indexed by ai. Finally, the system efficiency η can be obtained by simply substituting (2.22) and (2.26) into (2.20). 2.4. Performance Evaluation and Discussions To analyze the performance of RFID systems with QDD, we focus on three perfor- mance metrics: (i) system efficiency defined in Section. 2.3.2, (ii) false alarm rate, and (iii) false detection rate that will be defined in the Section. 2.4.2. The performance is evaluated under different system parameters, as shown in Table. 2.1. A detailed flowchart describing the approach to compute average BER and system efficiency us- ing the Monte-Carlo simulations in given in Fig. 2.5. The results are obtained with 1000 iteration runs using Matlab software. They are compared with those of DD-based ones to show the effectiveness of the proposed scheme. Table 2.1: Simulation parameters for RFID system. Symbol Description Values in Sec. 2.4.1 Values in Sec. 2.4.2 N Number of CDMA tags 1÷ 19, 500÷ 1500 1000 K Number of Gold codes 30 15 L Length of register 4 4 Lc Gold code length 30, 31 31 f Frame size 32 512 SNR Signal-to-Noise Ratio 5dB, 7dB −10÷ 10 [dB] ϵ Nb. of feed-forward stage matrix 3 3 2.4.1. System Efficiency In Fig. 2.6, we describe the bit-error-rate (BER) performance (both theoretically and simulation) of the QDD and DD with respect to different numbers of tags, for given SNR of 7 dB. For the BER computation, we have considered that each tag transmits a total of 10000 bits. Here, it is noted that the SNR can be also set by any other values that can illustrate the effect of noisy channel on the detectors’ performance. The Gold code length denoted by Lc is set by 31 chips, while the number of stages in QDD i.e., ϵ is set by 3. It is observed that the simulation results match with the theoretical ones, validating the analysis. The BER also increases with respect to the increasing of the number tags due to interference. However, the performance of the QDD is better than that of the DD, when the number of tags is large enough (≥ 10). The reason is that 34 TRANSMITTER (READER) Generate family Gold codes Randomly generated bits CHANNEL RECEIVER (TAG) bN Generate AWGN channel Generate AWGN noise samples Set output according to Eq.(2.2) Generate transmitted signal from tags according to Eq.(2.3) (Tags respond to reader) Check all tags involved in a timeslot error_count = error_count + 1 Quasi-Decorrelating Detector succ_count = succ_count + 0 System efficiency = succ_count 4(Assuming: 10 )bN = ERROR CALCULATION SYSTEM EFFICIENCY CALCULATION DETECTION No Yes _BER . b error count N N = succ_count = 0 For each tag Check all tags in the system error_count = 0 For each tag One of bit is in error? bN error_count = error_count + 0 NoYes Check successfully detected tag? succ_count = succ_count + 1 N Figure 2.5: Flowchart of simulation process to calculate BER and system efficiency. the noise has been enhanced in DD under the effect of the code correlation matrix, while in QDD it is mitigated thanks to the stage truncation of the matrix. We also compare the BER performance of DD and QDD with respect to different values of SNR in Fig. 2.7. The number of tags are initialized as 30. The Gold code length and the number of stages in QDD are set to Lc = 31 chips and ϵ = 3, respectively. As observed from the figure, QDD demonstrates a better BER performance than DD in almost cases. This is because QDD mitigates the noise enhancement caused by the 35 Number of Tags 1 4 7 10 13 16 19 B ER 0.0125 0.013 0.0135 0.014 0.0145 DD-Theoretical DD-Simulation QDD-Theoretical QDD-Simulation Figure 2.6: BER performance of QDD and DD detectors with respect to a number of tags, given Lc = 31, SNR = 7 dB ϵ = 3. inverse transformation of the correlation matrix in DD. SNR (dB) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 B ER 10-4 10-3 10-2 10-1 100 DD QDD Figure 2.7: BER comparison between DD and QDD by varying values of SNR. We now validate our analysis in the subsection 2.3.2 by showing the theoretical and simulation results of the considered system efficiency η with QDD, for a given number tags in Fig. 2.8. The frame size f and the code length Lc are supposedly to be 32 and 36 31, respectively. We can see that the results are matched to each other, that confirms the correctness of the analysis. Number of tags 500 700 900 1100 1300 1500 Sy st em e ffi ci en cy 9 9.5 10 10.5 11 11.5 12 QDD Theoretical QDD Simulation Figure 2.8: System efficiency with respect to the number of tags, given f = 32, K = 30, Lc = 30, SNR = 7 dB. Under the effect of the detection error and code collision, the system efficiency η with both QDD and DD is evaluated with respect to the number of tags and the number of codes in Figs. 2.9 and 2.10, respectively. Here, it is noted that if one among 96 bits of a tag’s ID is wrongly detected, it is not decoded successfully. We can see in both figures that the system efficiency with QDD is larger than that with DD (especially when the number of codes increases as in Fig. 2.10). The reason, which has been mentioned above, is the noise enhancement in DD caused by the code correlation matrix R−1, and the noise mitigation in QDD thanks to the stage truncation of the inverse of the correlation matrix Mϵ. The results confirm the advantages of QDD-based detector in the structure of RFID readers in comparison with the DD-based one. The system efficiency with respect to different values of the frame size is re-plotted in Fig. 2.11, for given N=1000, K = 30, Lc = 31, and different values of SNR (SNR=5 dB in Fig. 2.11(a) and 7 dB in Fig. 2.11(b)). It is interesting to see that for a given value of SNR, we can choose an optimal frame size that maximize the system efficiency. In our examples, the optimal frame sizes are 35 and 30 timeslots. This fact might suggest a suitable selection of system parameters for the identification process in practical systems, which we believe very useful for system designers. 37 Number of tags 500 700 900 1100 1300 1500 Sy st em e ffi ci en cy 2.6 2.7 2.8 2.9 3 3.1 3.2 DD QDD Figure 2.9: System efficiency with respect to the number of tags, given K = 30, f = 32, Lc = 31, SNR = 7 dB. 2.4.2. False Alarm and False Detection We now evaluate the performance of the previous proposed missing-tag algorithm proposed in [77] with QDD. Two performance metrics i.e. false alarm rate and false detection rates, denoted by Rfa and Rfd, respectively, are presented. In particular, false alarm occurs when an available tag in system is notified missing, and thus, the rate is defined as follows Rfa = Nfa N , (2.27) where Nfa is the number of available tags detected as missing ones. On the other hand, false detection occurs when an actual missing tag is confirmed to be present in systems and thus, the rate Rfd is defined as Rfd = Nfd N , (2.28) where Nfd is the number of actual missing tags detected as available ones. We specifically plot in Fig. 2.12(a) the false alarm rate and Fig. 2.12(b) the false detection rate with respect to different values of SNR. In the figures, N , f , K, L, and threshold for detection are set to be 1000, 512, 15, 4, and 0.3, respectively. Here, it is important to note that the threshold is used to detect the transmitted binary bit (0/1) from the tags. We can see that, thanks to the efficiency of QDD in coping with noisy 38 Number of codes 25 28 31 34 37 40 43 46 49 Sy st em e ffi ci en cy 2.5 3 3.5 4 4.5 DD QDD Figure 2.10: System efficiency with respect to the number of codes, given K = 30, f = 32, Lc = 31, SNR = 7 dB. Frame size 20 25 30 35 40 45 50 Sy st em e ffi ci en cy 0.21 0.22 0.23 0.24 0.25 DD QDD (a) SNR = 5 dB Frame size 20 25 30 35 40 45 50 Sy st em e ffi ci en cy 2.6 2.7 2.8 2.9 3 3.1 3.2 DD QDD (b) SNR = 7 dB Figure 2.11: System efficiency with respect to frame size, given N=1000, K = 30, Lc = 31. channels, the rates with QDD are lower than those with DD, and they will be most the same when the SNR keeps increasing. This is because when SNR increases, the detection error decreases, and thus, the performance of the protocol is more reliable regardless of the detector. Finally, we plot in Fig. 2.13 the rates versus the threshold where N = 1000, f = 512, K = 15, L = 4, and SNR=0 dB. Again, we observe that the reliability of missing- tag detection protocol with QDD is mostly better than that with DD. Based on this 39 SNR (dB) -10 -5 0 5 10 Fa ls e al ar m ra te 0 0.04 0.08 0.12 0.16 0.2 DD QDD (a) False alarm rate SNR (dB) -10 -5 0 5 10 Fa ls e de te ct io n ra te 0 0.2 0.4 0.6 0.8 1 DD QDD (b) False detection rate Figure 2.12: False alarm and false detection rate with respect to the SNR in the conventional missing- tag detection protocols with DD and QDD, given N=1000, K = 15, f = 512, L = 4, Threshold = 0.3. Threshold 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Fa ls e al ar m (F A) ra te 10-6 10-5 10-4 10-3 10-2 10-1 100 Fa ls e de te ct io n (F D) ra te 0.1 0.15 0.2 0.25 0.3 0.35 FA DD FA QDD FD DD FD QDD Figure 2.13: False alarm and false detection rates with respect to the threshold in the conventional missing-tag detection protocols with DD and QDD, given N=1000, K = 15, f = 512, L = 4, SNR = 0 dB. simulation results, system designers might have an optimal protocol parameters setting depending on practical transmission environments. 40 2.5. Summary In this chapter, the performance of a hybrid ALOHA/CDMA RFID system with both QDD and DD have been investigated. The structure and the performance of QDD and DD were re-studied in the context of RFID. The system efficiency was then re-analyzed in practical environments with the presence of both code collision and de- tection error. Computer simulations were performed, which showed that the analytical efficiency matched with the simulation one. It was also observed that identification and missing-tag detection protocols with QDD outperformed those with DD in practical noisy channels, which we believed useful for system designers. The content of this chapter has been published in the paper: Tuyen T. Hoang, Hieu V. Dao, Vu X. Phan, and Chuyen T. Nguyen, Performance Analysis of Hybrid ALOHA/CDMA RFID Systems with Quasi-decorrelating Detector in Noisy Channels, REV Journal on Electronics and Communications, Vol. 9, No. 1–2, January–June, 2019. 41 Chapter 3 ON THE DESIGN OF NOMA-ENHANCED BACKSCATTER COMMUNICATION SYSTEMS 3.1. Introduction A BackCom system [79, 103] consists of two main components, i.e., a reader (or controller) and backscatter nodes (BNs). During the communication process, each BN tunes its antenna impedance and communicates with the reader by modulating and reflecting the incident radio frequency (RF) signal via reflection coefficients. A portion of the incident RF signal power is harvested to supply the power for BN’s circuit, while the remaining incident RF signal power is reflected back to the reader thanks to the reflection coefficients. The BackCom, therefore, can achieve information transmission via surrounding radio signals from ambient RF sources, without requiring a dedicated energy supply for BNs [104]. These energy-saving features make the BackCom become a prospective candidate for IoT applications in future wireless networks. 3.1.1. Related Works and Motivation To exploit the benefits of the above-mentioned technologies, the combination of NOMA and BackCom has recently attracted research efforts worldwide. The NOMA- aided BackCom systems offer high spectral/energy efficiency and cost-effectiveness for collecting massive low-power IoT devices. This makes such systems a candidate prime for green IoT networks. 3.1.1.1. Related Works Driven by the potential and popularity of NOMA-aided BackCom systems, exten- sive studies have recently addressed the design and performance evaluation for such systems [8, 105, 106, 107, 108, 109, 110, 111]. Particularly, Jing et. al. provided the design guideline for NOMA-aided BackCom systems, where the reflection coeffi- cients for the multiplexed BNs from different NOMA groups are set to different values to utilize the PD-NOMA [8]. In [105], the authors analyzed the backscatter-NOMA integrated systems of cellular and IoT networks in terms of outage probability per- formance. The authors in [106] studied the unmanned aerial vehicle (UAV)-assisted BackCom systems using the PD-NOMA scheme, in which the network throughput 42 was maximized by the optimal UAV’s altitude. The resource allocation problem of NOMA-aided bistatic BackCom systems was formulated in [107], with NOMA and dynamic time-division multiple-access (TDMA). In [108], the authors addressed the cognitive-enabled backscatter network using NOMA, where the sum rate of the BNs was maximized under the multi-slot energy causality constraint. The authors in [109] investigated the NOMA-aided BackCom systems, where signals from at most two BNs were multiplexed on the frequency resource block using NOMA in each timeslot. Most recently, the effective capacity of a downlink NOMA-aided BackCom systems was stud- ied in [110]. 3.1.1.2. Motivations It is worth noting that the key idea in NOMA-aided BackCom systems is to utilize different backscattered power levels from different BNs, in which they are controlled to backscatter their data at the same time. To support this approach, a design frame work has been proposed in [8], in which the framework is understood as a set of technical designs including communication rules as well as hardware architectures of BNs [112]. It provides a criteria for choosing power reflection coefficients for which BNs are classified into different regions depending on their power levels. Nevertheless, there are two main drawbacks in this design. First, the BNs are chosen from different regions for NOMA grouping in a random manner. This results in a high probability that the signal from the selected BN is not decoded successfully because of adverse issues on wireless channels. Second, the design framework in [8] was for the static NOMA-aided BackCom systems only. In practical systems, BNs may enter and/or leave the reader’s coverage area frequently, in which dynamic schemes need to be addressed. From such limitations, it is necessary and important to provide novel schemes for the performance enhancement of conventional NOMA-aided BackCom systems, which motivates us to focus on this study. 3.1.2. Major Contributions and Organization The primary objective of this chapter is to offer novel schemes for the performance improvement of conventional NOMA-aided BackCom. Also, both static and dynamic BackCom are investigated. Here, it is noted that our previous study in [113] was the first to tackle the aforementioned limitations of the conventional approach. Particu- larly, a new user pairing scheme was proposed to improve the system performance in a static setting, where the weakest signal node from the near region is grouped with the weakest one from the far region. In this chapter, we provide a comprehensive design 43 framework for both static and dynamic NOMA-enhanced BackCom systems, where the initial work in [113] has been substantially extended as follows: C1: We present a design framework for static NOMA-enhanced BackCom systems, considering two-node pairing (TNP) scheme [113] and novel adaptive power re- flection coefficient (APRC) scheme. Instead of randomly choosing BNs for NOMA groups as in [8], the TNP scheme se- lects NOMA groups based on the possibility of successful decoding. The TNP scheme is expected to predict and prevent unsuccessful transmissions from NOMA groups. In addition, the APRC scheme could increase the possibility of successful decoding in NOMA groups by adjusting BN’s power reflection coefficients depending on their channel conditions. C2: We introduce a design framework for dynamic NOMA-enhanced BackCom sys- tems, considering the novel dynamic-sized pairing (DSP) and hybrid APRC/DSP schemes. The conventional approach is applicable for static systems, in which the NOMA group size is fixed. To support dynamic systems, the DSP scheme is introduced, where its goal is to increase the number of successful NOMA groups in dynamic NOMA-enhanced BackCom systems. Moreover, we also present the hybrid APRC/DSP scheme, which combines the APRC and DSP schemes to further enhance the performance of dynamic NOMA-enhanced BackCom systems. C3: We provide insightful results in terms of the number of successful backscatter nodes and the number of bits that can be successfully decoded by the reader to highlight the outperformance of our proposed schemes compared to the conventional ones. The analysis of the TNP scheme regarding the number of successful backscatter nodes is provided. Monte-Carlo simulations are also performed to validate the correctness of the theoretical analysis. The rest of the chapter is organized as follows. Section II describes the considered system model and conventional NOMA-aided BackCom systems. The proposed schemes for both static and dynamic NOMA-enhanced BackCom systems are presented in Section III, including TNP and APRC schemes for static systems as well as DSP and hybrid APRC/DSP schemes for dynamic systems. The simulation results are given in Section IV. Finally, we conclude the chapter in Section V. 44 Antenna Apply SIC ( )ix t ( )jx t BNi 1G 2G( )y t Sensor & Controller 1ξ Mξ Time-slot duration of sT B Mini-slot for a -node NOMA group sTb B Mini-slot for single-nodeb 2ξ... ... sT (a) (b) 2G MG r 1ξ 1G BN j IR OR (c) Backscatter signalIncident RF signalReader Backscatter Node Figure 3.1: Illustration of (a) system model, (b) time-slot structure, and (c) NOMA-aided BackCom system with M = 2. 3.2. System Model and Conventional Approach 3.2.1. System Description Our considered system, as shown in Fig. 3.1 (a), consists of a reader and B backscat- ter nodes (BNs), which can be sensors, Internet of Things (IoT) devices, and radio fre- quency identification (RFID) tags. The BNs are assumed to be uniformly distributed within an annular coverage area determined by an inner radius RI and an outer radius RO. Then, the distance from a BN to the reader, denoted by r, can be modeled as the binomial point process, where its probability density function (PDF) is expressed as fr (r) = 2r R2O−R2I [114]. The reader collects data from BNs using backscatter communications (BackCom). Particularly, the reader initially sends a request to specific BNs. Upon receiving the request, the BNs backscatter their data to the reader within the mini-slots of a time-slot duration1. Each timeslot with a duration of Ts is partitioned into multiple mini-slots as depicted in Fig. 3.1 (b). A mini-slot accommodates the data from either a single 1In this chapter, we focus on uplink communications, where the hybrid time-division multiple-access (TDMA)/power- domain non-orthogonal multiple access (NOMA) scheme is employed [115]. 45 BN or multiple BNs supported by the NOMA technique [116]. As a result, the time allocated to a mini-slot is defined as bTsB , where B is the total number of BNs, and b is the number of BNs multiplexed by the NOMA technique. Here, b = 1 for a single BN, while 2 ≤ b ≤M for NOMA-aided multiple BNs with M the NOMA group size. It is worth noting that systems using the power-domain NOMA technique require a considerable difference in the channel gains among users to decode data successfully [117]. To facilitate the NOMA-aided BackCom systems, each BN in a NOMA group is able to switch its power reflection coefficient of ξ in a range of values, i.e., 1 ≥ ξ1 ≥ ξ2, · · · ,≥ ξM > 0. This is controlled by the reader to make a significant difference in channel gains among the BNs. As a result, the received power at the reader from the i-th BN with the reflection coefficient of ξk can be expressed as Pri = Pξkr −2ρ i , (3.1) where i ∈ {1, 2, · · · , B} and k ∈ {1, 2, · · · ,M}. Additionally, P is the reader’s trans- mitted power, and ρ is the path-loss coefficient. 3.2.2. Conventional Approach The conventional NOMA-aided BackCom system was reported in [8], where the hybrid TDMA/NOMA scheme was employed for uplink transmissions. Notably, to facilitate the BackCom systems using the power-domain NOMA scheme, the reader virtually divides its coverage area intoM sub-regions, i.e., G1, G1, · · · , GM , as depicted in Fig. 3.1 (a). Here, a sub-region Gb, with b ∈ [1,M ], is an annular region defined by the the radii Rb and Rb+1 (Rb < Rb+1 and RM+1 = R). Based on the training broadcast message along with a unique identity (ID) for each BN, the reader can obtain the channel state information (CSI), which is supposed to be reliable and up-to-date, and then classifies the BNs into different sub-regions. This depends on the signal power level of BNs received by the reader, which is estimated in (3.1). The reader randomly selects one BN per sub-region for NOMA grouping. It is worth noting that, if the M -size NOMA group is not feasible, the reader might repeat this process with (M −1) BNs, (M − 2) BNs, and the rest. Similar to our considered BackCom system, the backscattering transmission of NOMA groups of multiple BNs as well as single BNs are taken within mini-slots in a time-slot duration as depicted in Fig. 3.1 (b). Different NOMA groups selected in a random manner by the reader are first transmitted in the mini-slot duration of bTsB , while individual BNs respond later in the mini-slot time of TsB . At the receiver side of the reader, the NOMA decoding is performed via the successive interference cancella- 46 tion (SIC) technique, which is assumed to be perfect. The decoding order is from the strongest signal to the weakest one. In other words, for each mini-slot of the NOMA group, the reader first detects and decodes the strongest signal, while treating the weaker ones as the interference. As transmission errors are unavoidable, the strongest signal can only be successfully decoded and extracted from the received signal if its signal-to-interference-and-noise ratio (SINR) satisfies a predefined threshold of γth. Assuming that the signal from i-th BN is the strongest one in a NOMA group size of M , where i ∈ [1,M − 1]. The condition for successfully decoding the i-th strongest signal, in which other signals from j-th BNs are treated as interference, can be expressed as SINRi = Pξir −2ρ i(∑M j=i+1 Pξjr −2ρ j +No ) ⩾ γth, (3.2) where No is the noise power. If the condition in (3.2) is satisfied, the reader then decodes the second strongest signal and the rest. Otherwise, the strongest signal could not be decoded successfully, leading to the failed decoding of the remaining weaker ones. Example: An example of the conventional NOMA-enhanced BackCom system is illustrated in Fig. 3.1 (c). Also, the NOMA group size is M = 2 corresponding to the two-BN pairing case. We assume that two BNs, i.e., BNi and BNj , forming a NOMA group belong to two different sub-regions, i.e., G1 (near) and G1 (far), respectively. Consider that BNi and BNj are paired using NOMA in a mini-slot of t. The received signal in the mini-slot t is, then, expressed as y(t) = hixi(t) + hjxj(t) + n(t), where hi and hj are the channel gains, while xi(t) and xj(t) are the reflected signals from BNi and BNj , respectively. Additionally, n(t) is the Gaussian noise. Assuming that the BNi experiences a better channel gain than that of the BNj , i.e., hi > hj . Thus, the reader first decodes the signal of BNi, i.e., xi(t), removes the signal by SIC, and then decodes the signal of BNj , i.e., xj(t). Limitations: There are two critical limitations of the conventional approach. Firstly, it is the random selection of BNs for NOMA groups. This might lead to a high probabil- ity of unsuccessful decoding and significant deterioration of the system’s performance. Secondly, the conventional approach is applicable to static NOMA-aided BackCom systems. Nevertheless, many practical Back