1.2. Problem Statement and Literature Review
1.2.1. Anti-collision protocols/algorithms
The principle of anti-collision protocols is to reduce collision timeslots caused by the simultaneous transmission from tags or readers. They are classified into the three main approaches, which are (i) tag anti-collision, (ii) reader scheduling, and (iii) signal processing algorithms/technologies.
1.2.1.1. Tag anti-collision
In the tag anti-collision approach, tags are controlled by the readers to transmit their data in an organized and efficient manner, thereby minimizing collisions and optimiz- ing the tag identification process. Many different communication algorithms have been investigated. They are mainly based on different multiple-access techniques, which are Frequency Division Multiple Access (FDMA), Space Division Multiple Access (SDMA), Time Division Multiple Access (TDMA), Code Division Multiple Access (CDMA), and hybrid systems. FDMA-based algorithms use different frequency bands for different tags’ responses [44, 45], while SDMA-based algorithms utilize either directional anten- nas or multiple readers to spatially separate the channel and distinguish between tags [44, 45, 46]. They, thus, might have a higher performance, yet are extremely complex and lead to the higher cost as well [44]. TDMA-based anti-collision algorithms utilize separate timeslots for the transmission of signals according to a predefined mechanism.
They are divided into two broad groups, namely deterministic and probabilistic.
The deterministic anti-collision algorithms: The reader splits collided tags by send- ing request commands using the tags’ IDs. These methods are based on tree-based anti-collision algorithms such as Binary Tree (BT) [1, 26, 39] and Query Tree (QT) [1, 47] algorithms. BT algorithms continue to split each subset of collided tags into two smaller subsets by a unique binary code, and query each subset until all tags have been identified. QT Algorithms are similar to the BT ones, but instead of assigning binary codes to the tags, it uses a query tree structure to group tags based on certain attributes, such as their EPC prefixes. Different improved version of these BT and QT protocols have been also studied for years. In particular, an advancement of BT algorithm i.e., Adaptive Binary Tree Splitting (ABTS) is proposed in [48] in which not only collisions but also needless idle slots are decreased. In [49], Chen et al. introduce a modified version of the ABTS, known as Enhanced Binary Tree Splitting (EBTS).
To reduce the receiving time at the reader, EBTS determines where the collided bits are and truncates unnecessary data bits. In addition, extensive studies for the QT algorithm have been proposed in [50, 51], such as Adaptive QT (AQT). In order to
minimize collisions and expedite tag identification, the protocol utilizes information obtained from the previous reader readings. In [52] another novel tag anti-collision algorithm based on M -ary query tree scheme (MQT) is investigated. This work can achieve two objectives: firstly, eliminating unnecessary queries, and secondly, splitting collided tags into multiple smaller subsets, allowing for efficient utilization of the af- fected bits. The results of this study demonstrate an out-performance of the existing QT-based algorithms.
The probabilistic anti-collision algorithms: control tags’ responses in a probabilistic (random) manner in timeslots [53, 54, 55]. The FSA, which is known as the most ef- fective probabilistic algorithms and widely used in RFID standards, organizes a frame of multiple timeslots, with each tag transmitting its ID only once per frame to mini- mize collisions. This process is repeated until timeslots with the signal collision are no longer detected. The principle of FSA is that if the frame size is equal to the total num - ber of tags, the number of singleton slots, and thus, the identified tags in the frame is maximum. It refers to a fact that if the tag cardinality is estimated accurately, the per- formance of FSA protocol can be improved. Therefore, most current protocols/works that are based on FSA try to estimate the total number of tags using observations of timeslots during frames, which are, for examples, Vogt method [56, 57], Maximum a posterior (MAP) [58], and Bayesian inference [59]. In particular, two estimation meth - ods were introduced Vogt, named Vogt-I [56] and Vogt-II [57]. While Vogt-I method calculates the lower bound for the estimate as (S + 2C), where S and C represent the observed numbers of singleton and collision slots, Vogt-II employs Chebyshev’s inequality to minimize the Euclid distance between the observed and expected vectors of empty, singleton, and collision slots. Also, in [58], the author try to maximize the posterior probability of the expected slots using their observations in each frame.
Although these works have proved their performance improvement, it is noted the estimation accuracy might be heavily affected by wireless fading environments with the presence of the so-called capture effect (CE) [60, 61, 62, 63, 64, 65, 66] and detection error (DE) [9, 10, 11] or when the number of tags is significantly increased. Specifically, CE refers to a phenomenon in which a tag might be identified in a collision slot since its received SINR is higher than the reader’s sensitivity threshold. In addition, in DE, a tag might not be detected even in a singleton slot because its received SNR in this case, might be less than the threshold. Figure. 1.11 describes a simple example of FSA- based protocol with the CE and DE. The request f = 5 is broadcasted to all tags, and each tag selects a slot at random to respond. Here, tag 5 is supposedly detected in the
19
Tags’ response
DE QUERY with f 5 Reader
Observations
CE
Timeslot status E: Empty S: Singleton C: Collision
Figure 1.11: An illustration of FSA-based communication protocol with CE and DE.
collision slot 1, while tag 4 is not observed in the singleton slot 3. These phenomena result in erroneously observed slot states. In other words, the timeslot observations may not accurately reflect the actual number of responses. Tags can be hidden during the identification process by either another tag or an unsuccessful detection [67]. As a result, the estimation accuracy of conventional methods may be degraded. Extensive research on these phenomena has been conducted in the RFID literature, covering both theoretical [60, 63, 68] and experimental aspects [67, 69, 70]. To cope with the cardinality estimation in the presence of the CE, several works have been proposed [9, 65, 66, 71, 72]. In [65], by assuming the existence of the CE, authors try to find a capture probability and the tag cardinality by minimizing the norm-2 distance as in Volt-II method, which is known as capture-aware backlog estimation (CMEBE) method. This work is then improved by [9] and [71] with maximum likelihood (ML) and Bayesian approach, respectively. Furthermore, the work in [66] addresses the issue of cardinality estimation in the presence of both the CE and DE using the Expectation - Maximization approach.
Besides, in dense RFID systems with a huge number of tags, the estimation usually inaccurate due to hardware constraints of the frame size. To cope with this problem, another anti-collision algorithm, called Probabilistic Dynamic Framed Slotted Aloha (PDFSA) has been proposed in [73]. PDFSA divides the total number of tags into sub-groups by a power control approach, while DFSA is applied for each sub-group during reading rounds.
1.2.1.2. Reader scheduling
Reader scheduling refers to designed processes of scheduling and coordinating the activities of multiple readers to minimize the reader collisions. The objective is to
Tag 1 Tag 2 Tag 3 Tag 4 Tag 5
1 2
frame
3 4 5
Tag 1 Tag 5
Tag 4 Tag 2 Tag 3
1 2
frame
3 4 5
S E E C E
Σ
reduce the number of required frequencies while minimizing the time/energy needed for all readers to communicate with their respective tags within their interrogation zones.
1.2.1.3. Communication and Signal processing algorithms/technologies
Communication and Signal processing algorithms plays an important role in mit- igating the signal collision issue in large-scale/dense RFID systems. Different works have been studied, which are based on different approaches such as CDMA-based, non- orthogonal multiple-access (NOMA), Compressed Sensing (CS), Multiple Input Multiple Output (MIMO) RFID, and hybrid systems.
CDMA-based approach is well known as one of the best solutions to cope with the signal collision in conventional multiple access wireless networks. Thus, it can be utilized in designing RFID tag anti-collision protocols/algorithms [45, 74, 75, 76, 77].
In these designs, each CDMA tag is assigned a distinct signature waveform, also known as a pseudo-random (PN) code (usually, Gold code) in which each waveform cq(t) is represented as follows
Lc
cq(t) = aq(n)p(t − nTc), (1.3)
n=1
where aq(n) is a PN code sequence consisting of Lc chips that take values (±1). p(t) is a pulse of duration Tc, and Tc is the chip interval.
The characteristics of PN codes are commonly defined by their auto-correlations and cross-correlations. Auto-correlation is a measure of how similar a code is to time shifted versions of itself, and cross-correlation is a measure of how similar a code is to time shifted versions of other codes in a code-set [74]. The cross-correlation function ρi,j of a PN code, with a period of Ts are given by
∫ Ts
When the cross-correlation ρi,j is zero, the codes are called orthogonal.
To detect signal transmitted from CDMA-based tags, an appropriate CDMA detec- tor is implemented at the reader side. The CDMA detector might consist of a matched filter bank, which is depicted in Fig. 1.12 [78] where each matched filter corresponds to one signature waveform. It is assumed that Q CDMA-based tags with different waveforms/PN codes are simultaneously transmitting signals to the reader. Then, the received signal r(t) at the reader can be represented as follows
0 ci(t)cj(t)dt. (1.4)
ρi,j =
21
Matched Filter Bit decision
z1
𝑐1(𝑡)
r t Matched Filter
z 2
𝑐2(𝑡)
Matched Filter
zq
𝑐𝑞 (𝑡) 𝑇𝑏
∫ ( ) 𝑑𝑡
0 𝑇𝑏
∫ ( ) 𝑑𝑡
0 𝑇𝑏
∫ ( ) 𝑑𝑡
0
Σ
Σ
qq Sync. signals
Figure 1.12: CDMA detector: A matched filter bank [78].
Q
r (t) = Aqbqcq(t) + n (t) , (1.5)
q=1
where Aq is the carrier amplitude, while transmitted bit bq ∈ {−1, 1}, n (t) is the noise.
The matched filter output corresponding to the k-th code is written as
zq = Aqbq + Ajbjρjq + nq. (1.6)
j̸=q
The reader, then, can uses the knowledge of the PN codes to recover the transmitted information from each tag. Nevertheless, the PN codes are usually not orthogonal in practice, which results in the so-called the multiple access interference (MAI) (the second term of (1.6). Several methods have been proposed to mitigate the effects of MAI in which Decorrelating Detector (DD) [74] is well known as one of the most efficient solutions. The structure of DD is plotted in Fig. 1.13, which is composed of a filter bank that matches the tags followed by an inverse correlation matrix R−1. Then, the signal vector after MAI elimination denoted by bˆ can be presented as
bˆ = R−1r = Ab + R−1n, (1.7)
where R is a Q × Q correlation matrix. A is the diagonal matrix containing corre- sponding received amplitudes. Here, it is noted that R+ = (R−1)qq > 1 in general. It refers to a fact that although DD can completely eliminate MAI, it also enhance the background noise that might degrade the system performance.
… … ...
z2
R-1
r t
z1
Sync. signals
Bit decision
b‸1
b‸2
Figure 1.13: Decorrelating detector.
On the other hand, NOMA allows multiple tags to be served/responded at the same time/frequency resources [7] thanks to the principle of successive interference cancellation (SIC) at the receiver of the reader side. The SIC technique is based on removing the interfering signal from the received signal, one at a time as they are decoded. In particular, tags that respond to the reader at the same time are required to have different transmitted power levels, which is known as power-domain NOMA.
The reader first decodes the strongest signal, while treating the weaker ones as the interference. After the strongest signal has been decoded, its information is used to subtract the signal of the particular tag from the received signal. The decoding can only be successful if the signal-to-interference-and-noise ratio (SINR) of the considered signal satisfies a predefined threshold of the reader. The reader is equipped with SIC is shown via a simple model as in Fig. 1.14, where it tries to decode received signal from two tags, i.e., Tagm and Tagn. Here, Tagm is assumed to experience a better channel gain than Tagn. In this case, the reader first decodes the signal of Tagm, removes the signal by SIC, and then decodes the signal of Tagn.
Recently, in [8], authors provide a design guideline for the BackCom systems (that include RFID) using a hybrid TDMA and power domain NOMA. In [79], a mechanism to reduce the collisions in NOMA-aided BackCom systems is introduced by designing efficient transmitted power levels for tags. The authors in [80] address the performance of NOMA-aided BackCom systems with multiple antennas.
b‸
q
… …
… ...
Matched Filter
𝑇𝑏
∫ ( ) 𝑑𝑡
𝑐 (𝑡) 0
1
Matched Filter
𝑇𝑏
∫ ( ) 𝑑𝑡
0 𝑐2(𝑡)
Matched Filter
𝑇𝑏
∫ ( ) 𝑑𝑡
𝑐 (𝑡) 0
𝑞
zq
23
Reader
Time/Frequency
Superimposed
Decode Tagm’s signal signals received at
the reader
Decode Tagn’s signal
Tagn’s signal
Figure 1.14: A RFID system using NOMA.
Compressed sensing is a signal processing technique that enables the efficient recov- ery of sparse signals from fewer measurements. In RFID systems, compressed sensing can be used to address the challenge of reading a massive number of tags with multiple readers. This approach utilizes a fact that the number of tags in practical systems is sparse (much smaller) in comparison with the whole tags’ ID space. Several outstand- ing contributions have been presented in [81, 82, 83].
Multiple Input Multiple Output (MIMO) involves the use of multiple antennas at the reader and the tags to transmit and receive signals. By using multiple antennas, MIMO can exploit the spatial diversity of the signals, enabling the identification of multiple tags simultaneously. The use of MIMO in RFID systems to cope with collision has been also an active research area in recent years [84, 85, 86]. Nevertheless, the design of the readers and tags becomes more complex and expensive, which might be unpractical/challenging in large-scale/dense RFID systems.