WorksEngineering
Constant-Weight Sparse Vector Coded Augmented-AFDM
Abstract
Sparse vector coding (SVC) and its variants convey information through sparse vectors (SVs), or sparse support patterns and detecting information by solving sparse recovery problems, making them attractive for short-packet ultra-reliable low-latency communications (URLLC). However, as the required information rate increases, conventional SVC schemes or the enhanced SVC (ESVC) schemes require a rapidly growing number of SVs, resulting in significant increase of complexity and receiver processing latency. On the other hand, when communicating in doubly selective channels, orthogonal frequency division multiplexing (OFDM) suffers from the Doppler spread induced inter-carrier interference, while affine frequency division multiplexing (AFDM) experiences the same peak-to-average power ratio (PAPR) problem as OFDM, although it allows diversity gain from channel's double selectivity. To address these issues, this paper proposes a constant-weight sparse vector coded augmented AFDM (CW-SVC/AFDM ) scheme, along with a corresponding joint support and symbol detection algorithm. In this scheme, SVs are represented by constant-weight codewords that feature a fixed Hamming weight and a large Hamming distance, resulting in relatively small SV sets. Simulation results compare various SVC/ESVC schemes and diverse coding schemes employed by the CW-SVC/AFDM, including Hamming, BCH, RS, and polar codes, for QAM symbol protection. The results demonstrate that the proposed scheme provides a distinct reliability advantage over conventional SVC/ESVC schemes, achieving a gain of up to 14 dB. Furthermore, the study indicates that the CWC's weight and minimum Hamming distance must be carefully selected to balance detection reliability, QAM modulation order, and sparse recovery feasibility, thereby maximizing the overall system's performance.
Introduction
Ultra-reliable low-latency communications (URLLC) and massive machine-type communications (mMTC) represent two distinct service categories in modern mobile communication networks, each characterized by different communication requirements [1]. URLLC is designed to support the applications that are highly sensitive to reliability and latency [2], while mMTC targets mostly the large scale machine-to-machine communications [3]. Moreover, these two scenarios often share some overlapping application domains, including vehicle-to-vehicle (V2V) communications [4], autonomous driving [5], internet of things (IoT) [6], and remote sensing and control [7]. For example, in autonomous driving, ensuring passenger safety and maintaining proper vehicle operation require highly reliable and low-latency communications both between vehicles and between vehicles and base stations. These requirements are closely aligned with both mMTC and URLLC. Such applications typically require the rapid exchange of real-time data, including sensing information (such as temperature, humidity, traffic signals, and road conditions) and control commands (such as start/stop, acceleration/deceleration, and power switching).
Sparse vector coding (SVC) has emerged as a potential solution in URLLC by transmitting data through mapping information bits into sparse support patterns [8]. Accordingly, its decoding process can be formulated as a sparse recovery problem under the compressed sensing (CS) framework [9]. Besides, CS also plays an important role in enabling low-overhead access and joint detection in mMTC [10], thereby rendering the SVC scheme applicable in both URLLC and mMTC applications. To achieve better performance, in [11], the authors proposed an enhanced SVC (ESVC) scheme, where information bits are jointly mapped to both the support patterns of the sparse vector and their conveyed QAM symbols. This design significantly reduces the number of the sparse support patterns required to represent the same amount of information in the SVC schemes, thereby enhancing both transmission efficiency and overall system performance. Building upon this, the authors of [12] introduced another novel short-packet transmission technique known as sparse superposition coding (SSC), which extends the ESVC framework by transmitting QAM symbols after pseudo-random spreading in the time-frequency domain. At the receiver, the decoding problem is reformulated as the detection of non-zero indices, which can be efficiently solved using the low-complexity energy detection algorithms. Many studies have demonstrated that SVC and its variants exhibit strong relevance in various application scenarios [12, 13, 14, 15].
However, as the majority of information bits in the SVC and its variant schemes are still conveyed through the combinations of support patterns, a slight increase in the number of bits per symbol results in an exponential growth in the number of support patterns. This dramatically increases the computational complexity of the sparse recovery process at the receiver, leading to a significant rise in decoding latency, thereby posing challenges to meet the strict latency requirement of URLLC.
As a class of nonlinear codes, the codewords of a constant-weight code (CWC) have the same Hamming weight [16]. Note that for binary vectors of an equal weight, the pairwise Hamming distance is always even, and increasing the weight of codewords does not by itself guarantee a larger minimum Hamming distance for an arbitrary CWC. For the constant-weight codebooks constructed in this paper, under the constraint that the codeword weight does not exceed half of the codeword length, the minimum Hamming distance is deliberately designed to increase with the codeword weight, thereby facilitating more reliable decoding. In contrast to activating random sparse supports for conveying information in the conventional SVC scheme, the CWC-derived sparse vector coding (CW-SVC) scheme yields less number of sparse vector (SV) candidates but a relatively higher weight, which results in a bigger number of supports per candidate SV. Consequently, the primary information carrier shifts from SVs to QAM symbols. Additionally, the use of higher code weights in CWC reduces the total number of SV candidates, which decreases the computational complexity required for the sparse recovery algorithms at the receiver.
Moreover, high-reliability communications in high-mobility environments have been becoming essential in many applications, such as in V2V communications, autonomous driving and unmanned aerial vehicles (UAVs) [17]. Classical multicarrier systems such as orthogonal frequency division multiplexing (OFDM) suffer from the inter-carrier interference (ICI) due to Doppler shifts. In contrast, affine frequency division multiplexing (AFDM), with its unique subcarrier waveform design, processes signals in the Doppler-delay domain and effectively mitigates Doppler-induced interference in addition to the delay-caused inter-symbol interference (ISI), thereby ensuring a robust performance in the doubly-selective fading scenarios [18]. However, similar to OFDM, AFDM suffers from high peak-to-average power ratio (PAPR) problem, which poses challenges for implementing the size- and power-limited devices, such as that commonly found in IoT or autonomous driving systems [19]. For these kinds of applications, a communication scheme with low PAPR while of high efficiency in the doubly-selective fading environments is critical.
To address these challenges, this paper proposes a novel CWC-aided ESVC, referred to as the CW-SVC scheme, for implementation with an augmented AFDM (FDM) system [20], referred to as the CW-SVC/FDM system. The main novelties of the proposed scheme include the following. First, in the proposed CW-SVC/FDM system, the support patterns of sparse vectors are formed by constant-weight codewords with large minimum Hamming distances, which shifts the primary information carrier of the system from the support patterns to the channel-coded QAM symbols. Second, with the FDTW method, the resulting constant-weight codebook turns support detection into a fixed-size candidate detection problem, which allows the candidate patterns to be detected in parallel and is, hence, suitable for low-latency applications. Third, the proposed scheme offers advantages in the transmission and detection of critical information, such as active users’ identities in mMTC scenarios. In addition, the low PAPR of FDM and its robustness in doubly selective fading environments can effectively enhance the practical utility of the CW-SVC/FDM scheme, including drone communications in IoT networks and autonomous driving.
The main contributions of the paper can be summarized as follow:
We propose the CW-SVC scheme, where the support patterns are the codewords of a constant-weight code designed using the “from dimension to weight (FDTW)” construction [23]. The resulting large Hamming distance among the support patterns enables high-reliability discrimination during the sparse recovery process.
We propose amalgamating CW-SVC with FDM to form the CW-SVC/FDM system, where the QAM
symbols conveyed on the supports are protected by conventional channel coding, yielding an enhanced system performance.
We propose a practically effective joint support and symbol detection scheme, where the detection of different candidate patterns can be carried out in parallel to reduce the receiver’s processing latency.
For protection of the information conveyed by QAM symbols, various linear channel coding schemes are studied and compared, including Hamming, Reed–Solomon (RS), Bose–Chaudhuri–Hocquenghem (BCH), and polar codes. Simulation results demonstrate that while all coding schemes allow system performance enhancement, polar code exhibits the best performance among the coding schemes and parameter configurations evaluated in our simulations, enabling an up to 6.5 dB gain when compared to the case without using coding for protection of QAM symbols.
Simulation results demonstrate that the proposed CW-SVC/AFDM system is capable of attaining a substantial improvement in terms of the overall performance. Compared with the conventional ESVC schemes, the proposed system exhibits a significantly enhanced reliability in SV detection, leading to the overall bit error rate (BER) primarily dominated by the errors arising from QAM demodulations, which can be protected by employing an appropriate channel coding scheme.
In Table 1, we explicitly contrast our contributions with the state-of-the-art schemes in the literature.
Table 1. Novelty comparison between our paper and the related references [13, 14, 16, 18, 19, 20, 21, 22].
The remainder of this paper is organized as follows: Section II describes the system model, including the CW-SVC scheme and the principles of AFDM. Section III introduces the properties of CWC and describes the FDTW construction of CWC. Section IV illustrates the proposed joint support and symbol detection, and compares the detection complexity and latency of the existing ESVC and proposed CW-SVC schemes. Then, in Section V, performance results are presented to compare the performance of various schemes. Finally, in Section VI, we conclude the research in this paper. An example of the CWC codewords used to obtain the results in this paper is presented in Appendix A.
System Model
The block diagram of the proposed CW-SVC/AFDM system is shown in Fig. 1, where the data stream is constructed using the principle of the CW-SVC and transmitted via the AFDM framework [20]. In this design, the information payload is divided into two parts, and . is used to select a support pattern from the codeword set formed by a CWC, while is first processed by a classical linear encoder (CLE), which is then modulated into QAM symbols, followed by their mapping onto the supports of the selected pattern. The resulting sparse vector is then compressed using matrix to form the data stream , which is matched to the number of subcarriers employed by the AFDM for further processing. As shown in Figure 1, following the AFDM related processing, the signal is transmitted over a doubly selective fading channel. At the receiver side, support and symbol detection are performed by the proposed detection schemes designed based on the properties of CS and CWC.
Figure 1. Transceiver diagram of the proposed single user system employing CW-SVC/A²FDM.
It is worth noting that the proposed CW-SVC scheme is not limited to the AFDM waveform. The CW-SVC operation maps information bits to the constant-weight support patterns and to the QAM symbols before the waveform modulation. Hence, it can in principle be combined with various other linear modulation waveforms, including AFDM and OFDM. In this paper, AFDM is adopted as the transmission framework because of its robustness in doubly selective fading channels and its low PAPR. Therefore, the detection algorithms are presented in this context.
In further detail, the data stream block with bits is separated into two sub-blocks of with bits and with bits, as shown in Figure 1. is used for support pattern selection and forms the input of the CLE. Hence, . According to the concept of index modulation (IM) [24], bits generate a unique index to select a support pattern from the CW-pattern set constituting CW-codewords selected from the constant-weight codebook. The construction of the codebook is introduced in detail in Section III. The pattern satisfies , representing a binary sparse vector of length with supports. The bits of are first encoded based on a conventional channel coding scheme, such as Hamming, BCH, RS or polar codes, which support different levels of error correction capability with different trade-off in implementation complexity and/or decoding delay [25]. The output of the linear encoder is denoted by . The parameters of these coding schemes in our simulations will be detailed in Section V. Then, the encoded bits in generate a set of symbols, expressed as , based on -QAM (or PSK) modulation. The symbols in are mapped onto the corresponding supports in , yielding a sparse vector , which contains all information of the transmitted bits in .
An example depicted in Figure 2 shows the processing of CW-SVC. As shown in Figure 2, bits of data are divided in two sub-blocks: one of bits on top is used for selecting a pattern and the remaining of bits form the input to the CLE (not drawn in the figure). Assume that a Hamming code is applied. Then, bits generate coded bits. These coded bits are then modulated to form QAM symbols, which are mapped to the two supports in the selected pattern to form .
Figure 2. An example for the CW-SVC with , and .
Once the sparse vector is obtained, as shown in Figure 1, it is compressed from length to length of the number of subcarriers, through a Gaussian compressing matrix , giving
The compressed symbols are then input to the AFDM for further processing.
Similar to the fast Fourier transform (FFT) in OFDM and discrete affine Fourier transform (DAFT) in AFDM [26], discrete augmented affine Fourier transform (DAFT) [20] is the basic building block of AFDM.
As shown in Figure 1, the fundamental component of , namely the inverse (), comprises three components: , the inverse FFT (IFFT) matrix , and . As previously discussed, the proposed scheme is designed for operation in the scenarios including high mobility and aimed to maintain a low PAPR. To achieve these objectives, in contrast to OFDM, the matrix is constructed based on the subcarrier mapping strategy in the single-carrier frequency division multiple access (SC-FDMA) [27] to reduce PAPR. Meanwhile, transforms subcarrier waveform from a sinusoidal signal to a chirp signal, causing subcarrier frequency to vary over time. Consequently, this time-varying characteristic enhances the resolution of subcarriers in the Doppler-delay domain, similar to that in AFDM, to achieve diversity gain.
Assuming again that the number of subcarriers is , as shown in Figure 1, the is expressed as
where is the IDFT matrix with entries , , with is a diagonal matrix, while is a unitary matrix to make a unitary matrix, i.e. satisfying , with being an identity matrix. To mitigate the PAPR of the transmitted signal, can be designed as [20]
by following the principles of SC-FDMA [27]. In (3), is a block-diagonal matrix consisting of number of component FFT matrices of -dimensional, i.e.
where is Kronecker product, is a normalized FFT matrix, is an integer, preferably equalling to for convenience of FFT implementation. In (3), is a mapping matrix obeying the interleaved mapping strategy, as that employed in the interleaved FDMA (IFDMA) in SC-FDMA [27]. Hence, we refer to our AFDM scheme as the interleaved AFDM (IAFDM) scheme. Note that the mapping matrix satisfies . Therefore, we can readily show that the DAFT matrix defined in (2) is a unitary matrix.
After the -point IDAFT on , the time-domain samples can be expressed as
where and . The collection of in (5) can be written in matrix form as
As in AFDM, AFDM introduces a prefix to mitigate the effect of the delay-spread resulting from multipath propagation. Considering the property of chirp waveform, which has the periodicity of [18]
where is the periodic index, the chirp-periodic prefix (CPP) in AFDM is given as
where is the length of the CPP.
After the parallel-to-serial (P/S) conversion and adding the CPP, as shown in Figure 1, the signal is transmitted over the wireless channel, which may be a doubly selective fading channel. The matrix form of the received signal can be expressed as
where is the additive white Gaussian noise (AWGN) vector, and the channel matrix is given by [18]
where is the number of propagation paths. In (10), , and are the complex gain, Doppler shift in digital frequency, which is the Doppler shift normalized by the sampling frequency, and the integer delay, respectively, associated with the -th path. Following [18], the terms in (10) are defined as follows: is an diagonal matrix given by
which for has the contribution of the phase shifts induced by the CPP. When , the CPP has no effect on the signal, and hence the corresponding diagonal elements in are equal to . accounts for the effect of the Doppler shifts, while models the effect of the delay of the -th path, where is a forward cyclic-shift matrix expressed as [26]
After discarding the CPP, as shown in Figure 1, support and symbol detections are executed to estimate the pattern index . The total bits are then recovered, where bits are derived from the index , and the remaining bits are obtained from the QAM symbols in the support of , after symbol demodulation and channel decoding.
In the following section, a method for constructing CWC is firstly presented. Then, the detection method is addressed in § 4.
Design of Constant Weight Sparse Vector Code
As mentioned in § 2, the support patterns are designed based on a CWC. Consider a , where denotes the codeword length, is the minimum Hamming distance between distinct codewords, and is the weight of the codewords. In the CWC codebooks designed for the proposed system, the codeword weight is constrained by , and the minimum Hamming distance is deliberately designed to increase with the codeword weight. In the proposed CW-SVC/AFDM system, CW-codewords are employed to determine the support patterns. Hence, it follows that . Furthermore, to facilitate detection achieving high reliability, for example in user activity detection in the grant-free multiple access (GFMA) scenario, a large is desired. Consequently, the desired codewords in the CWC are expected to have a relatively large weight in the CW-SVC with . This illustrates the fundamental difference between the proposed CW-SVC scheme and the other existing ESVC schemes [11, 28, 29]. Furthermore, in the existing ESVC schemes, the number of the candidate support patterns is from with the minimum Hamming distance as low as 2, whereas the number of codewords of is upper-bounded by the Johnson bound [30]. Specifically, for , the Johnson bound is given by
where denotes the maximum possible number of codewords of . Therefore, the number of candidates in a CWC-assisted support pattern set for our CW-SVC/AFDM system is significantly smaller than , in particular when and are relatively big.
Regarding the construction of constant-weight codebooks, when holds, the CWC can be constructed directly. In this case, if is divisible by , the positions are partitioned into disjoint support sets of size , with each set serving as one codeword, so that any two codewords have disjoint supports and the minimum Hamming distance is exactly . However, when this property is no longer satisfied, i.e. for , in this paper the CWC is generated by the “from dimension to weight (FDTW)” construction [23]. With the FDTW method, a constant dimension code (CDC) is first generated and then the codewords of the CWC are obtained from its cosets of each subspace. In detail, let us consider a vector space with vectors, where is the dimension of , i.e. the length of vectors in . The elements in these vectors belong to . According to [23], given a non-negative integer , the set of all subspaces of with dimension is known as a Grassmannian, which can be represented by . The distance between such two subspaces and is given by [31, 32, 33]
for all . Denote . If there exists a minimum dimension distance , allowing for all , we say that is a CDC set in and is the size of this set, i.e. the number of subspaces.
Let be a subspace of with dimension , which contains vectors of -dimensional. Then, the cosets of are given by [23]
Note that in (15), although can be chosen from the whole space , different choices of may generate the same coset. In fact, since and , there are only distinct cosets of , which partition . Note furthermore that the principal distinction between a subspace and its cosets is that a subspace always contains the zero vector, while the cosets do not.
Once the cosets of a subspace are obtained, their vectors are mapped to decimal numbers, which specify the indices of “1” in a length vector, and all other positions are filled with “0”, giving the supports of vectors. Consequently, each coset corresponds to a codeword of length and weight , i.e. a sparse vector of length and sparsity . By taking the union of the cosets of all subspaces in , a CWC with codewords is obtained. Note that, in the proposed CW-SVC system, only codewords are utilized as the candidate support patterns, where . The minimum Hamming distance of this CWC is given by [23]
where to ensure remains non-negative. The parameters of a general CWC constructed by the FDTW method is shown in Table II.
| CWC codeword length | |
| CWC codeword weight | |
| CWC’s minimum Hamming distance | |
| CWC size |
Table II. Parameter relationships between CDC and the CWC constructed from FDTW method.
To further illustrate the above CWC construction principle, an example is provided in the following. Note that the values of and are usually determined by the requirements of the sparse vectors in the proposed
CW-SVC/AFDM system. When these values are given, the minimum Hamming distance should be as large as possible to achieve a higher reliability, provided that the number of codewords in CWC is sufficient for supporting user and/or data rate requirements. Therefore, with this example, we assume that a CWC(16, 6, 4) is needed, where , and . Hence, consider a vector space , where and , which leads to , can be chosen to be , which results in . Then, a CDC is chosen for the FDTW construction. According to Table II, , giving . Thus, the minimum subspace distance . For simplicity, in this example, we select two subspaces of two-dimensional, generated over by two linearly independent vectors, given by
each of which contains vectors of length . As seen in (17), and share only the all-zero vector. Hence, . Therefore, the subspace distance between and is
As shown in (17), “1” in occurs at the third and fourth indices of the 4-length vectors in , indicating the activation of these indices. Therefore, for , the convenient choices of distinct coset representatives are obtained by varying the inactive indices, i.e. the first and second indices that are always zero in the 4-length vectors in . Note that the zero vector is always considered as a choice of , which returns the subspace itself. Hence, for , a set of distinct coset representatives can be chosen as , which results in
Similarly, for , the 4 distinct coset representatives are and the corresponding cosets of are
Then, the 4-length vector elements in these cosets are converted from binary to decimal to represent the support indices in the CW-codewords, giving a support pattern set containing the 8 codewords of the CWC(16,6,4).
Table III lists the support indices of the CW-codewords generated based on the cosets in (19) and (20). For example, the four vector elements in (0000) + are mapped to , the four vector elements in (0011) + are mapped to . Similarly, the mapping results in the other cases can be obtained.
| support pattern | support indices |
Table III. Support indices of the CWC(16,6,4) constructed from FDTW with cosets (19) and (20).
According to (16), the minimum Hamming distance of this CWC with eight codewords of length and is .
As shown in tbl:2 and Table III, the number of codewords in the CWC constructed from FDTW is limited, due to the introduced redundancy for reliability protection. Therefore, a detection algorithm with relatively high complexity can be employed for the support detection with low latency. Afterwards, the QAM symbol detection can be implemented using the principle of, for example the minimum mean square error (MMSE) estimation. The detailed detection algorithms will be discussed in the following section.
Support and QAM Symbol Detection
In this section, we first present the proposed detection algorithm for joint support-pattern identification and QAM symbol detection in the CW-SVC/AFDM system. Then, we analyze the detection scheme’s computational complexity and processing latency, which are compared with that of the conventional orthogonal matching pursuit (OMP) based detection algorithms employed by the ESVC schemes.
Detection Algorithm
It has been well-recognized that URLLC relies on relatively short packet transmission due to the strict limit on transmission latency, but requiring ultra high reliability, making it extremely challenging in practical implementation. In some of the existing ESVC schemes [8, 11, 22, 34], message transmission and detection are based on the CS principles [35, 36, 37, 38]. To guarantee the required reliability of message recovery, in CS-based detection, different pattern candidates must be sufficiently distinguishable. This requires that the mutual correlation between the columns of CS matrix is sufficiently low and the sparsity is small and limited. Furthermore, due to the strict reliability requirement of URLLC, the modulation schemes of high order, such as 64 QAM or above, may not be acceptable. This is because to transmit the same amount of information, while using a higher order modulation scheme increases transmission rate and hence reduces latency, but it results in a higher error rate, which penalizes the reliability of URLLC. To mitigate this dilemma, conventional ESVC schemes generally employ the modulation (typically QAM) of relatively low-order to maintain reliable QAM detection, meaning that only a small number of data bits are conveyed by QAM. Accordingly, for a given information payload and a sparsity level, a relatively big number of bits must be conveyed by the support patterns. Since candidate support patterns are required to convey bits per symbol, the complexity of sparse recovery increases exponentially with . Additionally, the modulation of QAM symbols on the supports of sparse vectors further enlarge the search space of signal detection. In this case, the employment of maximum likelihood (ML) detection that achieves optimal performance is impractical due to its extremely high complexity. Instead, the support detection is typically achieved by the suboptimal greedy CS algorithms, such as OMP [39] or multipath matching pursuit (MMP) [40], which, in comparison with ML detection, sacrifices some reliability performance while significantly reducing the detection complexity.
In contrast to the conventional ESVC schemes, the support patterns in the proposed CW-SVC/AFDM system are determined by CW-codewords that are well-regularized. Specifically for URLLC applications, the CW-codewords with large minimum Hamming distances are expected, which result in limited number of candidates, allowing to employ the detection schemes of relatively high-complexity in practice. Furthermore, when the Hamming distance of a CWC increases, its weight also increases, as demonstrated in § 3. Hence, compared to the conventional ESVC assisted schemes that rely on a small number of supports to convey high-order QAM symbols, using our CW-SVC scheme most of the data can be conveyed by the relatively low-order QAM symbols, with only a small portion of the data will be transmitted by the candidate support patterns. Therefore, in this paper, we propose to employ the detection schemes of a relatively high complexity but low latency. Specifically, we propose a candidate-wise MMSE-assisted residual detection. With this detection, the receiver first generates an effective sensing matrix for each of the candidate support patterns, from which an MMSE estimator is then obtained for detecting the corresponding QAM symbols. Subsequently, the received signal is reconstructed for each of the candidate support patterns, and then the residual between the reconstructed and actual received signals is calculated. Consequently, by comparing the residuals corresponding to all candidate support patterns, the receiver determines the transmitted support pattern’s index that corresponds to the support pattern yielding the minimum residual. After the support pattern is detected, the QAM symbols conveyed on the supports can be extracted for subsequent channel decoding. From the above description we can know that the detection for different candidate support patterns can be implemented in parallel to reduce the receiver processing latency.
To formulate the proposed detection, upon combining (1) and (9), and assuming that the -th pattern (CW-codeword ) is actually selected, the received signal in the CW-SVC/FDM system can be expressed as
where is the subvector of the sparse vector restricted by the support positions determined by , as seen in Table III for example, where each support conveys one QAM symbol, while is the submatrix of , consisting of the columns indexed by the elements in .
However, the receiver does not know that the transmitter selected the -th pattern. Therefore, in order to fulfill the detection, the receiver first assumes that the activated pattern is one of . Then, considering that the receiver has the full knowledge of and , it can implement the MMSE-assisted processing to mitigate the interference between different supports. Specifically, the MMSE estimation matrix corresponding to the assumed pattern can be calculated as
where is the effective sensing matrix. In (22), is the signal-to-noise ratio (SNR) and is an identity matrix of size . Consequently, when the receiver assumes that was transmitted (although the actually transmitted vector was , necessitating the clarification stage shown in (25)), the QAM symbols corresponding to the support of can be estimated as
To detect the QAM symbols, the bias from the MMSE processing is corrected by normalizing using , giving
Then, for each , a penalty is computed as the sum of the absolute errors between the received signal and the reconstructed signal . Consequently, the index of the selected sparse vector (CW-codeword) by the transmitter can be determined as
Accordingly, the data bits conveyed by the index of the sparse vector can be recovered as .
Once the index of the activated sparse vector is detected, the data bits conveyed by the QAM symbols conveyed on the supports of can be demodulated from . Afterwards, the bits of can be recovered from the corresponding linear decoding of . In particular, if polar coding is employed as the CLE, the inputs of the decoder need to be the log-likelihood ratio (LLR) values of the detected symbols, which can be expressed as
with , where is the th symbol in , , and is one of the possible constellation point of the -QAM modulation employed, where the th bit of a QAM symbol is either “0” or “1” with , and denotes the set of constellation points.
Overall, the detection algorithm is summarised in Algorithm 1. Note that, when applied to URLLC, the detection in terms of can be implemented in parallel to reduce latency.
Algorithm 1 Joint support and symbol detection.
Require: , , , ,
Ensure: , ,
for do
Form using the columns of corresponding to the supports in ,
Form the effective sensing matrix ,
Compute the MMSE estimator of (22) and estimate symbols as (23),
Correct the bias as (24),
Calculate the penalty
end for
Determine the best CW-SVC index as:
Recover from
Recover from
Recover by linearly decoding
Complexity and Latency Analysis
In the conventional SVC/ESVC schemes, usually the OMP is implemented to detect the active supports based on the whole sensing dictionary [41]. In each OMP iteration, the dominant operation is given by the correlation between the residual and all the columns in , which requires operations. In addition, a least-square aided update is performed over the identified support set, whose complexity is approximately at the -th iteration. Therefore, after iterations, the overall complexity of the OMP can be approximated as
In contrast, in the proposed detection scheme for CW-SVC/AFDM systems, the support patterns are restricted to a fixed CWC codebook with candidate codewords. For each candidate , the corresponding effective sensing matrix is formed, and a QAM symbol sequence is estimated by the MMSE estimator of (22). Since the matrix inversion in the MMSE estimator is performed for each candidate, it requires operations per candidate, while the subsequent symbol estimation and penalty calculation require approximately operations per candidate. Therefore, the overall complexity of the proposed detector is dominated by the per-candidate matrix inversions and is
which grows linearly with the number of candidates . When comparing (27) with (28), for a small , the complexity of the proposed detector is manageable, and the candidate-wise operations can be implemented in parallel to reduce the processing latency. For a large , e.g. , the detection complexity becomes high, which is consistent with the serial-detection results discussed in the performance section.
In addition to ultra-high reliability, low latency is another key requirement of URLLC. In a communication system, the end-to-end latency consists of several key component delays, including queuing delay, scheduling/access delay, transmitter-side processing delay, over-the-air transmission delay, propagation delay, receiver-side processing delay, and feedback/retransmission delay [42]. In the proposed scheme, the receiver-side processing latency can be significantly reduced in comparison with the OMP-based method. Specifically, in OMP-based scheme and most other sparse recovery algorithms, the received signal has to be processed through multiple iterations to achieve detection. In each iteration, the residual is updated and compared with all candidate codewords or atoms to determine the final detected signal. Since the computational results in the previous iteration is required for the subsequent iterations, this procedure must be performed in a sequential manner. Furthermore, while a larger number of iterations can improve detection performance, it also leads to the increase of receiver-side processing latency, which increases linearly with the number of iterations. In contrast, our proposed detection algorithm does not rely on iterative operations. The detection metrics associated with different candidates are independent of each other, and hence support both serial and parallel implementations, providing flexibility for different application scenarios. Specifically, when hardware resources are constrained while the latency requirement is relaxed, the candidate metrics can be evaluated in serial. In contrast, when the latency requirement is stringent while hardware resources are not the limiting factor, as in URLLC scenarios, the candidate metrics can be evaluated in parallel to meet the latency requirement.
Performance Results and Analysis
In this section, we present the simulation results to demonstrate the performance of our proposed CW-SVC/FDM system. Unless specified, in our simulations, the length of sparse vector is , the number of subcarriers is and there are information bits transmitted per symbol, where for CW-codewords selection and for classical linear encoding and further QAM/PSK modulation. Note that in the case of using no channel coding, namely uncoded CW-SVC, is implemented for the same bit throughput. The CW-codewords used in simulations are generated by the FDTW construction detailed in Section III and an example is given in Appendix A. The delays of doubly-selective fading channels normalized by the sampling duration are randomly distributed in the integer set , where denotes the maximum normalized delay-spread. The Doppler shifts normalized by subcarrier spacing are distributed in , set according to with uniformly distributed in . The value of is set as to achieve the maximum diversity order. Unless specified otherwise, is set to 4 for the generation of .
Figure 3 demonstrates the BER performance of the FDM systems aided by the conventional ESVC and the proposed CW-SVC without or with the polar CLE. In this comparison, the CWC(128, 64, 32) is employed and the number of CW-codewords is . The detailed parameters are listed in Table IV. The conventional ESVC uses OMP algorithm for sparse recovery. As shown in Figure 3, the BER of the conventional ESVC scheme exhibits an error floor at about the BER of , which is due to the employment of the greedy algorithm. In contrast, with a similar data throughput and channel state information (CSI), the proposed CW-SVC scheme without or with polar CLE outperforms the conventional ESVC. Specifically, the CW-SVC without CLE enables a BER of at about 16.5 dB, which is more than 14 dB lower than that by the conventional ESVC scheme. Furthermore, when the CW-SVC is enhanced by polar CLE, about a further 6 dB of SNR gain can be obtained, achieving a BER of at about 10.5 dB.
| weight | Modulation | |||
| Conventional ESVC | 4 QAM | |||
| Uncoded CW-SVC | BPSK | |||
| polar coded CW-SVC | 4 QAM |
Table IV. System parameters considered in Figure 3, Figure 4 and Figure 5.
Figure 3. BER performance comparison between the conventional ESVC, uncoded CW-SVC and the polar coded CW-SVC schemes operated with FDM.
In Figure 4, we analyze the BER performance for the SVC, QAM and overall, when the conventional ESVC scheme and the proposed CW-SVC scheme without CLE are compared. The legends in Figure 4 are defined as follow:
Figure 4. BER of SVC, QAM and overall of the conventional ESVC and the uncoded CW-SVC.
SVC error: errors generated solely in the support pattern detection;
QAM error: errors generated solely in the QAM demodulation after the support patterns are correctly detected;
Overall error: include the errors in both the above cases.
As shown in Figure 4, for the proposed CW-SVC scheme, the BER of SVC is significantly lower than the BER of QAM, while the overall BER is similar as that of QAM. This explains that in the proposed CW-SVC scheme, the overall BER performance is dominated by the errors generated from QAM demodulation. Therefore, in the proposed CW-SVC/ systems, appropriate CLE is required to protect the information conveyed by QAM, to reduce the overall BER. In contrast, for the conventional ESVC scheme, the BER of SVC is close to the overall BER, meaning that the overall system’s BER is dominated by the SVC errors. The results shown in Figure 4 are predictable, because in the proposed CW-SVC/ system, the support patterns of sparse vectors are protected by the CWC. However, in the conventional ESVC assisted systems, the support patterns of sparse vectors are not protected, contributing the majority of errors.
In Figure 5, the BER performance of the CW-SVC/ system with and applied in the matrix for is compared. Clearly, there is not much difference in these results. According to the principles analyzed in [20], with the number of subcarriers , when , the is reduced to a single-carrier system. Then, the A²FDM becomes a multicarrier system from and the number of distinguishable subcarriers increases with the growth of , until the system becomes the AFDM when . For all the values of , as long as the value of is appropriately selected to achieve the maximum diversity gain, the effect of is little on the BER performance of A²FDM. However, we should note that the PAPR of A²FDM signals increases with the increase of the value of , i.e. no PAPR at , while highest PAPR at , which is the PAPR of AFDM signals. Moreover, the low-PAPR property of A²FDM due to its embedded SC-FDMA-based IDA²FT was studied in [20]. Since the CW-SVC-generated signal is processed by the same A²FDM modulator, the employment of CW-SVC does not alter the low-PAPR property of the A²FDM signalling.
Figure 5. BER performance comparison of the CW-SVC/A²FDM systems, when , 4, 16 and 64 are respectively applied for the matrix in A²FDM.
Figure 6 compares the BER performances of the CW-SVC/A²FDM systems aided by BCH, polar, Hamming or RS CLE or no CLE. In addition to the CWC, the detailed parameter settings for the CLEs are given as follows:
Figure 6. BER performance comparison of the CW-SVC/AFDM systems with BCH, polar, Hamming, RS, or no CLE.
Uncoded: BPSK modulation, no CLE.
BCH code: 4-QAM, BCH (63, 36), which can correct up to 5 bit errors.
polar code: 4-QAM, 32 information bits are applied with an 11-bit cyclic redundancy check (CRC) code. The length of the mother code is 64 bits. The mother code is decoded at the receiver using a CRC-aided successive cancellation list (CA-SCL) decoding scheme with a 16 bits list.
Hamming code: 4-QAM, Hamming (63, 57), which can only correct 1 bit error.
RS code: 4-QAM, RS (14, 8) over GF(), which can correct up to 3 symbol errors.
Note that since different coding schemes require specific lengths of input bits, we uniformly adopt zero padding at the end of data to adjust the input data length for a fair comparison. The encoded bits are also padded with zeros to reach bits for QAM modulation. At the receiver, the padded zeros are removed before and after decoding to recover the detected data . In the BER calculation, only the original bits are taken into account, while the zero padding bits are excluded in counting the errors.
As expected, as shown in Figure 6, the uncoded case exhibits the worst performance when a BER below is demanded, despite it employs the BPSK modulation of the lowest order. As only one bit error can be corrected, the performance of the Hamming coded scheme is the second worst one, offering only about 0.5 dB performance gain at the BER of over the uncoded system. By contrast, the RS coded, BCH coded and polar coded scheme all achieve significant performance gain in comparison with the Hamming coded scheme. Specifically, among these three coded schemes, due to its symbol-level error correction, RS coded scheme shows the worse bit-level performance, but still provides a SNR gain of about 3 dB over the uncoded case at the BER of . The BCH coded scheme, which offers 5-bit error correction, results in a performance gain of about 4.5 dB compared to the uncoded system. Among the coding schemes and parameter configurations evaluated in our simulations, the best performance is achieved by the polar coded scheme, which, aided by the CRC coding and the CA-SCL decoding, enables a 6.5 dB performance gain over the uncoded system.
Unlike BCH, Hamming and RS codes studied in Figure 6, the error correction capability of polar codes cannot generally be characterized by the fixed number of correctable bit or symbol errors. Figure 7 compares the BER performance of the polar coded CW-SVC/AFDM systems with using , 128 and 256. For the polar coding and decoding, 11-bit CRC code and 16-bit CA-SCL are employed. Thus, the length of the input data of the polar encoder is always 43 bits. With this fixed number of input bits, a larger yields a lower coding rate, but requires a higher QAM modulation to transmit the total code bits on the same number of supports. Hence, as seen in Figure 7, 4QAM, 16QAM and 256QAM are employed for the cases of , 128 and 256, respectively. It is well-known that for the polar codes of the same length, a smaller coding rate offers a higher capability to mitigate the effects of noise and fading. On the other hand, given the same transmit power determined by the number of information bits, a QAM of higher order yields poorer reliability of detection than a QAM of lower order. Hence, as shown in Figure 7, the performance of the system with (16QAM) is better than that of the system with (4QAM), with a gain of about 1 dB. However, the performance of the system with (256QAM) is significantly worse than that of the system with either (4QAM) at a loss of 4.5 dB or 128 (16QAM) at a loss of 5.5 dB. Hence, from the results we can conclude that for a given information rate and the given number of supports, there is an optimum selection for and the order of QAM, which enables a polar coded CW-SVC/AFDM system to achieve the lowest BER within some given SNR range.
Figure 7. BER performance of the polar coded CW-SVC/AFDM systems with and , 128 and 256.
Fig. 8 demonstrates the BER performance of the proposed system with a fixed length of the polar-coded bits, but different values of , 15, 18, 30, 50, 70, 90, 110 and 117, respectively. According to [43], a 6-bit CRC is employed for , 15 and 18, while an 11-bit CRC is employed for the other cases. To ensure a fair comparison in terms of energy consumption, the transmitted signals are normalized such that the average energy per effective information bit is identical for all the considered values of . Hence, for a fixed polar-coded length of , increasing yields a higher coding rate and a lower redundancy, thereby reducing the error-correction capability. Therefore, as shown in Figure 8, the BER performance becomes worse as increases. The results verify that, given the energy per effective information bit, the decrease of coding redundancy has a significant impact on the BER performance. In the limit case of , the 117 information bits and 11 CRC bits occupy all the 128 inputs of the polar encoder, resulting in a polar code without frozen bits. Consequently, this configuration exhibits the poorest BER performance among all the considered cases.
Figure 8. BER performance of the polar coded CW-SVC/FDM systems with for communications at different information rates of , 15, 18, 30, 50, 70, 90, 110 and 117.
Figure 9 compare the BER performances of the CW-SVC/FDM systems with different CWC codebooks, including CWC(128, 128, 64), CWC(128, 64, 32), CWC(128, 32, 16) and CWC(128, 16, 8). Accordingly, for a similar throughput, BPSK, 4 QAM, 64 QAM and 256 QAM are used for modulation. No CLE is considered to avoid the effect from channel coding. There are two cases considered in Figure 9. In the first case, the full codebooks are utilized for SVC selection, which are marked as the red curves in Figure 9, corresponding to , 8 and 16. Here, the numbers of candidate patterns are different, caused by the different minimum Hamming distances of the CWC considered. In the second case, the number of candidate patterns is chosen for all codebooks regardless of the CWC considered, which are represented by the blue curves in Figure 9. From the results of Figure 9 it can be observed that the number of candidate patterns has no effect on the performance of systems. However, the performance improves as the weight increases from 8 to 16 and then to 32. This is because the minimum Hamming distance increases by following the relation of .
Figure 9. BER performance of the CW-SVC/A²FDM systems employing the CW-codewords with supports of and 256 QAM, 64 QAM, 4 QAM and BPSK modulation, respectively.
Moreover, a larger weight means a big number of supports and hence, more QAM symbols, which in turn means a lower modulation order for transmitting the fixed number of information bits, hence leading to performance improvement. However, when the weight is increased to , although the minimum Hamming distance is increased to and the modulation order is BPSK, the number of supports per sparse vector becomes too large to be recovered from the observed signal with a dimension of . Consequently, its resulted performance may be poor, especially, in the high SNR region where interference dominates the achievable performance.
In Figure 10, the BER performances of the polar coded CW-SVC/AFDM systems with three different CWCs are demonstrated. The CWCs considered include CWC(128, 32, 16), CWC(128, 28, 16) and CWC(128, 24, 16). Among these CWCs, CWC(128, 28, 16) and CWC(128, 24, 16) are generated by the FDTW construction as detailed in § 3, with and , respectively, whereas CWC(128, 32, 16) is constructed directly with as described in that section. As shown in Table V, the numbers of codewords of these codebooks differ significantly. For the three codebooks, equals 3, 6 and 11, respectively, so that support patterns are employed, namely , 64 and 2048. As the results in Figure 10 show, the overall error performance becomes better with the increase of . However, the situations for the systems with different codebooks are different. Specifically, when , the system errors are dominated by the QAM/PSK demodulation and its followed CLE decoding, the same as the previous results shown in Figure 9. When , the number of codewords of a codebook increases, resulting in the increase of the number of SVC detection errors, which is close to the number of errors by QAM demodulation. Accordingly, the overall BER is approximately equally contributed by both the SVC errors and QAM/CLE errors. In the third case when , the decrease of minimum distance leads to significant increase of the number of codewords, as shown in Table V, where the codebook of the CWC(128, 24, 16) contains codewords, of which are employed as the candidate support patterns. In this case, the CW-SVC scheme converges to the conventional ESVC scheme. Accordingly, the SVC errors overrun the QAM/CLE errors, making the overall BER performance dominated by the SVC errors. Note again that although in this case, the overall BER performance is still promising, about 1 dB worse than the other two cases, the complexity of SVC detection is extremely high due to the large number of codewords involved. Moreover, in our simulations, the detection is performed in serial, making the processing time growing with the number of candidate support patterns, which becomes significant when . If parallel implementation is employed, the critical-path latency can be significantly reduced at the cost of hardware resources that scale with .
| CWC | in FDTW | ||
| CWC(128, 32, 16) | - | - | 8 |
| CWC(128, 28, 16) | 3 | 72 | 64 |
| CWC(128, 24, 16) | 2 | 2328 | 2048 |
Table V. Number of codewords in the codebooks considered in Fig. 10
Figure 10. BER performance of the CW-SVC/AFDM systems where the CWCs with and , , and , respectively, are considered.
Finally, Figure 11 demonstrates the BER performance of the uncoded CW-SVC/FDM systems with the CWCs of different length of , 256, 384 and 512, while the weight is or 64. As shown in Figure 11, for both and 64, the code length of yields no effect on the system’s BER performance, which is in contrast to the conventional ESVC systems. The reason is that the number of candidates of support patterns in CW-SVC/FDM system is not directly related to the length of the sparse vectors. As shown in (21), the SVC detection only consider the supports in the sparse vectors, i.e. the weight of CWC. As long as the number of supports, i.e. weight, is fixed, the codeword length of does not affect the system BER performance.
Figure 11. BER performances of the CW-SVC/FDM systems with respectively , 256, 384, 512 and , 64.
Compared with the conventional ESVC scheme, the CW-SVC without CLE achieves the BER level of at an SNR about 14 dB lower, and the polar-coded CW-SVC further reduces the required SNR by about 6 dB. Under the same information throughput and transmission resources, the polar-coded scheme provides the strongest QAM protection, offering about a 6.5 dB gain over the uncoded case at the BER level of , outperforming the Hamming, RS and BCH coded schemes. For the polar-coded CW-SVC, with 16QAM offers about a 1 dB gain over with 4QAM, while with 256QAM is clearly inferior. Within the reliably recoverable range, increasing the CWC weight and minimum Hamming distance improves the BER performance, whereas the sparse-vector length has no effect on the BER for a fixed weight.
Conclusion
In this paper, we proposed a CW-SVC/FDM scheme that utilizes CWC to protect the information conveyed by SVs, while employing CLE to safeguard the QAM symbols transmitted on vector supports. This architecture enables ultra-reliable low-latency short-packet transmission in doubly selective fading environments. At the receiver, a joint SV support and QAM detection algorithm was designed, which can be efficiently implemented using parallel processing to reduce the processing latency. Our theoretical analysis and simulation results demonstrate that the proposed scheme yields a significant BER improvement over conventional ESVC schemes while maintaining low-latency parallel processing feasibility. Accordingly, a low block error rate (BLER) in short-packet URLLC applications is possible, as BLER and BER is related by the union bound: BLER BER, where is the number of bits per packet. The results reveal that in CW-SVC/AFDM systems, CWC provides ultra-reliable protection for SV-conveyed information, while QAM-conveyed information is robustly protected by an appropriate error-correction scheme. Evaluating a range of error-correction candidates for QAM protection demonstrated that polar codes are the most promising choice among the coding schemes and parameter configurations evaluated in our simulations. Furthermore, the results confirm that the CWC’s weight and minimum Hamming distance must be carefully selected to balance detection reliability, QAM modulation order, and sparse recovery feasibility, thereby maximizing overall system efficiency. Moving forward, our future research will extend this framework to multi-user CW-SVC/AFDM systems operating in diverse IoT environments, including mMTC and URLLC in device-centric cell-free scenarios.
Constant Weight Codebooks
There are six CWC codebooks considered in this paper, which are (128, 128, 64), (128, 64, 32), (128, 32, 16), (128, 16, 8), (128, 28, 16) and (128, 24, 16). CWC(128, 28, 16) and CWC(128, 24, 16) are generated by the FDTW construction with and . All codewords in the remaining codebooks possess the same property: their minimum Hamming distance satisfies . Specifically, each codeword is assigned a distinct set of indices in the sparse vector, such that all codewords in one codebook have mutually different support pattern. The support indices of the CWC(128, 32, 16) are shown in Table VI as an example.
| 0 | 16 | 32 | 48 | 64 | 80 | 96 | 112 |
| 1 | 17 | 33 | 49 | 65 | 81 | 97 | 113 |
| 2 | 18 | 34 | 50 | 66 | 82 | 98 | 114 |
| 3 | 19 | 35 | 51 | 67 | 83 | 99 | 115 |
| 4 | 20 | 36 | 52 | 68 | 84 | 100 | 116 |
| 5 | 21 | 37 | 53 | 69 | 85 | 101 | 117 |
| 6 | 22 | 38 | 54 | 70 | 86 | 102 | 118 |
| 7 | 23 | 39 | 55 | 71 | 87 | 103 | 119 |
| 8 | 24 | 40 | 56 | 72 | 88 | 104 | 120 |
| 9 | 25 | 41 | 57 | 73 | 89 | 105 | 121 |
| 10 | 26 | 42 | 58 | 74 | 90 | 106 | 122 |
| 11 | 27 | 43 | 59 | 75 | 91 | 107 | 123 |
| 12 | 28 | 44 | 60 | 76 | 92 | 108 | 124 |
| 13 | 29 | 45 | 61 | 77 | 93 | 109 | 125 |
| 14 | 30 | 46 | 62 | 78 | 94 | 110 | 126 |
| 15 | 31 | 47 | 63 | 79 | 95 | 111 | 127 |
Table 6. CWC (128, 32, 16)
References
- [1]3GPP, “Study on scenarios and requirements for next generation access technologies,” ETSI, Tech. Rep. 3GPP TR 38.913, May 2022, release 17.
- [2]M. U. A. Siddiqui, H. Abumarshoud, L. Bariah, S. Muhaidat, M. A. Imran, and L. Mohjazi, “URLLC in beyond 5G and 6G networks: An interference management perspective,” IEEE Access, vol. 11, pp. 54 639–54 663, 2023.DOI
- [3]E. Goshi, F. Mehmeti, T. F. La Porta, and W. Kellerer, “Modeling and analysis of mMTC traffic in 5G core networks,” IEEE Transactions on Network and Service Management, vol. 22, no. 1, pp. 409–425, 2025.DOI
- [4]H. Yang, K. Zheng, K. Zhang, J. Mei, and Y. Qian, “Ultra-reliable and low-latency communications for connected vehicles: Challenges and solutions,” IEEE Network, vol. 34, no. 3, pp. 92–100, 2020.arxiv.org/abs/1712.00537
- [5]D. P. M. Osorio, I. Ahmad, J. D. V. Sánchez, A. Gurtov, J. Scholliers, M. Kutila, and P. Porambage, “Towards 6G-enabled internet of vehicles: Security and privacy,” IEEE Open Journal of the Communications Society, vol. 3, pp. 82–105, 2022.
- [6]D. C. Nguyen, M. Ding, P. N. Pathirana, A. Seneviratne, J. Li, D. Niyato, O. Dobre, and H. V. Poor, “6G internet of things: A comprehensive survey,” IEEE Internet of Things Journal, vol. 9, no. 1, pp. 359–383, 2022.DOI
- [7]F. Liu, Y. Cui, C. Masouros, J. Xu, T. X. Han, Y. C. Eldar, and S. Buzzi, “Integrated sensing and communications: Toward dual-functional wireless networks for 6G and beyond,” IEEE Journal on Selected Areas in Communications, vol. 40, no. 6, pp. 1728–1767, 2022.DOI
- [8]H. Ji, S. Park, and B. Shim, “Sparse vector coding for ultra reliable and low latency communications,” IEEE Transactions on Wireless Communications, vol. 17, no. 10, pp. 6693–6706, 2018.arxiv.org/abs/1710.10780
- [9]R. Zhang, B. Shim, W. Yuan, M. Di Renzo, X. Dang, and W. Wu, “Integrated sensing and communication waveform design with sparse vector coding: Low sidelobes and ultra reliability,” IEEE Transactions on Vehicular Technology, vol. 71, no. 4, pp. 4489–4494, 2022.DOI
- [10]M. Alam and Q. Zhang, “A survey: Nonorthogonal multiple access with compressed sensing multiuser detection for mMTC,” Wireless Communications and Mobile Computing, vol. 2021, no. 1, p. 8840519, 2021.arxiv.org/abs/1810.05422
- [11]W. Kim, S. K. Bandari, and B. Shim, “Enhanced sparse vector coding for ultra-reliable and low latency communications,” IEEE Transactions on Vehicular Technology, vol. 69, no. 5, pp. 5698–5702, 2020.DOI
- [12]X. Zhang, D. Zhang, B. Shim, G. Han, D. Zhang, and T. Sato, “Sparse superimposed coding for short-packet URLLC,” IEEE Internet of Things Journal, vol. 9, no. 7, pp. 5275–5289, 2022.DOI
- [13]L. Yang and P. Fan, “Multiple-mode sparse superimposed code with low block error rate for short packet URLLC,” IEEE Communications Letters, vol. 28, no. 2, pp. 248–252, 2024.DOI
- [14]X. Zhang and D. Zhang, “Sparse superimposed coding based on index redefinition,” IEEE Communications Letters, vol. 27, no. 5, pp. 1467–1471, 2023.DOI
- [15]L. Yang and P. Fan, “Improved sparse vector code based on optimized spreading matrix for short-packet in URLLC,” IEEE Wireless Communications Letters, vol. 12, no. 4, pp. 728–732, 2023.
- [16]P. R. Ostergard, “Classification of binary constant weight codes,” IEEE Transactions on Information Theory, vol. 56, no. 8, pp. 3779–3785, 2010.DOI
- [17]M. El Zorkany, A. Yasser, and A. I. Galal, “Vehicle to vehicle “V2V” communication: Scope, importance, challenges, research directions and future,” The Open Transportation Journal, vol. 14, pp. 86–98, 2020. [Online]. Available: https://doi.org/10.2174/1874447802014010086
- [18]A. Bemani, N. Ksairi, and M. Kountouris, “Affine frequency division multiplexing for next generation wireless communications,” IEEE Transactions on Wireless Communications, vol. 22, no. 11, pp. 8214–8229, 2023.arxiv.org/abs/2204.12798
- [19]S. Kamal, I. A. Khan, A. Ali, A. H. Farooqi, A. Ahmad, and C. Estevez, “Novel Nyquist-I filter to reduce PAPR for SC-FDMA scheme,” Telecommunication Systems, vol. 88, no. 1, p. 34, 2025.DOI
- [20]Z. Lu, M. El-Hajjar, and L.-L. Yang, “Augmented affine frequency division multiplexing for low PAPR signaling and diversity gain protection,” IEEE Access, vol. 14, pp. 69 426–69 442, 2026.DOI
- [21]G. Choi and N. Lee, “Learning while transmitting: Pilotless polar coded modulation for short packet transmission,” arXiv preprint arXiv:2602.17979, 2026.arxiv.org/abs/2602.17979
- [22]M. K. Sinha and A. P. Kannu, “Generalized sparse regression codes for short block lengths,” IEEE Transactions on Communications, vol. 72, no. 5, pp. 2536–2551, 2024.
- [23]T. Etzion and A. Vardy, “A new construction for constant weight codes,” in 2014 International Symposium on Information Theory and its Applications, 2014, pp. 338–342.
- [24]X. Cheng, M. Zhang, M. Wen, and L. Yang, “Index modulation for 5G: Striving to do more with less,” IEEE Wireless Communications, vol. 25, no. 2, pp. 126–132, 2018.arxiv.org/abs/1712.06235
- [25]M. Rowshan, M. Qiu, Y. Xie, X. Gu, and J. Yuan, “Channel coding toward 6G: Technical overview and outlook,” IEEE Open Journal of the Communications Society, vol. 5, pp. 2585–2685, 2024.arxiv.org/abs/2405.07547
- [26]A. Bemani, N. Ksairi, and M. Kountouris, “AFDM: A full diversity next generation waveform for high mobility communications,” in 2021 IEEE International Conference on Communications Workshops (ICC Workshops). IEEE, 2021, pp. 1–6.arxiv.org/abs/2104.11331
- [27]L.-L. Yang, Multicarrier Communications. John Wiley & Sons, 2009.DOI
- [28]X. Zhang and K. Wang, “I/Q-modulated sparse vector coding for short-packet communications,” IEEE Transactions on Vehicular Technology, 2026, early access. [Online]. Available: https://doi.org/10.1109/TVT.2026.3667519
- [29]Y. Zhang, X. Zhu, J. Zheng, W. Yang, X. Yu, H. Zeng, Y. Liu, and Y. L. Guan, “Dual-mapping sparse vector transmission for short packet URLLC,” in GLOBECOM 2025 - 2025 IEEE Global Communications Conference, 2025, pp. 5091–5096.
- [30]S. Johnson, “A new upper bound for error-correcting codes,” IRE Transactions on Information Theory, vol. 8, no. 3, pp. 203–207, 1962.DOI
- [31]R. Koetter and F. R. Kschischang, “Coding for errors and erasures in random network coding,” IEEE Transactions on Information Theory, vol. 54, no. 8, pp. 3579–3591, 2008.
- [32]D. Silva, F. R. Kschischang, and R. Koetter, “A rank-metric approach to error control in random network coding,” IEEE Transactions on Information Theory, vol. 54, no. 9, pp. 3951–3967, 2008.
- [33]T. Etzion and A. Vardy, “Error-correcting codes in projective space,” IEEE Transactions on Information Theory, vol. 57, no. 2, pp. 1165–1173, 2011.
- [34]Y. Zhang, X. Zhu, Y. Liu, X. Fan, Y. Liang Guan, M. L. Dennis Wong, and V. K. N. Lau, “Block sparse vector codes for ultra-reliable and low-latency short-packet transmission,” IEEE Transactions on Communications, vol. 73, no. 10, pp. 9750–9766, 2025.DOI
- [35]R. G. Baraniuk, “Compressive sensing (lecture notes),” IEEE Signal Processing Magazine, vol. 24, no. 4, pp. 118–121, 2007.DOI
- [36]S. Qaisar, R. M. Bilal, W. Iqbal, M. Naureen, and S. Lee, “Compressive sensing: From theory to applications, a survey,” Journal of Communications and Networks, vol. 15, no. 5, pp. 443–456, 2013.DOI
- [37]S. Foucart and H. Rauhut, “An invitation to compressive sensing,” in A mathematical introduction to compressive sensing. Springer, 2013, pp. 1–39.DOI
- [38]J. W. Choi, B. Shim, Y. Ding, B. Rao, and D. I. Kim, “Compressed sensing for wireless communications: Useful tips and tricks,” IEEE Communications Surveys & Tutorials, vol. 19, no. 3, pp. 1527–1550, 2017.arxiv.org/abs/1511.08746
- [39]J. A. Tropp and A. C. Gilbert, “Signal recovery from random measurements via orthogonal matching pursuit,” IEEE Transactions on Information Theory, vol. 53, no. 12, pp. 4655–4666, 2007.
- [40]S. Kwon, J. Wang, and B. Shim, “Multipath matching pursuit,” IEEE Transactions on Information Theory, vol. 60, no. 5, pp. 2986–3001, 2014.
- [41]X. Zhang, H. Chen, D. Zhang, G. Qin, B. Davaasambuu, and T. Sato, “Uniquely decomposable constellation group-based sparse vector coding for short packet communications,” China Communications, vol. 20, no. 5, pp. 119–134, 2023.
- [42]M. Bennis, M. Debbah, and H. V. Poor, “Ultrareliable and low-latency wireless communication: Tail, risk, and scale,” Proceedings of the IEEE, vol. 106, no. 10, pp. 1834–1853, 2018.arxiv.org/abs/1801.01270
- [43]3rd Generation Partnership Project (3GPP), “NR; Multiplexing and channel coding,” 3GPP, Technical Specification TS 38.212, Oct. 2025, version 19.1.0, Release 19.










