Resource Allocation Design for Next Generation Multiple Access: A Tutorial Overview

The fundamental input-output relationship of NGMA in natural channels is given by

where $\mathbf{s}\in\mathbb{C}^{D\times 1}$ contains a total of $D$ data streams of all $K$ users such that $\mathbf{s}\sim\mathcal{CN}(\mathbf{0},\mathbf{I}_{D})$, as widely adopted in the literature[Wei2017]. Its subset $\mathbf{s}_{\Omega_{k}}\in\mathbb{C}^{D_{k}\times 1}$ contains $D_{k}=\left|\Omega_{k}\right|$ data streams of user $k$, $\forall k\in\{1,\ldots,K\}$, where $\sum_{k=1}^{K}D_{k}=D$ and $\cup_{k=1}^{K}{\Omega_{k}}=\left\{1,\ldots,D\right\}$. The channel matrix $\mathbf{H}\in\mathbb{C}^{M\times N}$ models the propagation between transmitter and receiver, where $N$ and $M$ are determined by the number of available system resource elements at the transmitter and receiver, respectively, such as the number of antennas, time slots, or subcarriers, etc. In practice, the channel matrix $\mathbf{H}$ needs to be defined in a proper domain such that it does not violate the intrinsic physics of wave propagation. For instance, if the channel matrix is defined in the spatial domain, we usually consider a narrowband assumption for the system [YijieRSMA], such that each entry of $\mathbf{H}$ can characterize the flat fading coefficient between each pair of transceiver antennas. Vectors $\mathbf{y}\in\mathbb{C}^{M\times 1}$ and $\mathbf{z}\in\mathbb{C}^{M\times 1}$ denote the received signal and noise at the receiver side, respectively, where the additive white Gaussian noise (AWGN) follows $\mathbf{z}\sim\mathcal{CN}(\mathbf{0},\sigma^{2}\mathbf{I}_{M})$ with the noise power of $\sigma^{2}$. If the channel matrix is defined in the frequency domain, a multicarrier communication system is usually considered with $M=N$ denoting the number of subcarriers, where $\mathbf{H}$ denotes the frequency domain effective channel between transceiver[Al-Imari2011LDSOFDM]. For concise notation in this paper, the notations $\mathbf{y}$, $\mathbf{H}$, $\mathbf{z}$, and $\mathbf{s}$ will be consistently employed to denote entities with analogous meanings, albeit with varying specifications and dimensions across different scenarios.

On the other hand, the channel matrix $\mathbf{H}$ can be defined in different domains depending on the adopted modulation waveform and on which domain NGMA is performed, as shown in Table II. Note that the channel matrix $\mathbf{H}$ can be defined in multiple domains, such as the frequency-spatial domain for MIMO-OFDM systems. In such a case, NGMA can be performed in the frequency domain, spatial domain, or cross the frequency-spatial domain, depending on the structure of $\mathbf{H}$. Despite the domain of $\mathbf{H}$, its rank $\mathrm{Rank}\left(\mathbf{H}\right)$ determines the maximum available DoFs of the considered communication systems, which is the theoretical maximum number of users that can be accommodated in an interference-free manner. Moreover, the eigenvalue distribution of $\mathbf{H}$ determines the channel disparity in a certain domain, facilitating the exploration of multi-user diversity for efficient resource allocation design. When $D\leq\mathrm{Rank}\left(\mathbf{H}\right)$, the system is known as underloaded [DerrickEEOFDMA] where the system resources are sufficient to accommodate all the data streams of all the $K$ users. In this case, OMA exhibits a handy approach to serve all the $K$ users, since each data stream can be allocated in at least one available system DoF and enjoys an interference-free transmission. In contrast, when $D>\mathrm{Rank}\left(\mathbf{H}\right)$, the system is overloaded [PerGainWei] and the users access the channel in a non-orthogonal manner, i.e., NOMA or RSMA. In such a case, advanced multi-user interference mitigation techniques need to be adopted to improve system performance, such as SIC[Wei2017] and message passing-based multiuser detection [BayestehSCMA].

In (1), the vector function $\mathbf{g}(\cdot):\mathbb{C}^{D\times 1}\xrightarrow{}\mathbb{C}^{N\times 1}$ maps the information symbols $\mathbf{s}$ of all the $K$ users to the $N$ system resources at the transmitter side. In particular, this map** function captures a specific approach of multiple access as well as resource allocation among multiple users, which plays a key role in determining the system performance. In other words, the joint consideration of multiple access and resource allocation is to design the map** function, $\mathbf{g}(\cdot)$, for a given channel matrix $\mathbf{H}$, such that the transmission and the channel properties match with each other so as to achieve certain purposes, e.g., interference management, rate maximization, and power minimization. In practice, advanced channel estimation technologies are generally required to acquire the channel matrix $\mathbf{H}$ [BigueshMMSE, zhao2017multiuser, Sun2019]. In certain scenarios, such as slow-fading channels, accurate CSI can be acquired at the transmitter with a reasonable amount of system overhead. In such cases, the estimated accurate CSI would be treated as perfect CSI to simplify the design of multiple access and resource allocation. In contrast, for the cases in which acquiring accurate CSI is not possible, only imperfect CSI is available. As such, different robust resource allocation design approaches should be performed such that the QoS of all users can be guaranteed to a certain extent, e.g., [WeiProceeding].

For a given $\mathbf{H}$, OMA schedules to serve one user in each DoF, while NOMA schedules to serve more than one user in each DoF. Also, hybrid NOMA strategically combines the approaches by serving a portion of the users through OMA while allocating the remaining users via NOMA [ZengHybridNOMA]. In practice, the formulation of resource allocation design problems can be tailored to specific system design goals, depending on the chosen multiple access scheme. Specifically, OMA imposes strict constraints to ensure resource orthogonality at the expense of underutilization of limited system resources. In contrast, NOMA is more flexible in terms of resource allocation design and can degenerate to OMA configuration if the power or rate allocated to certain NOMA users is set to zero. Nonetheless, the receiver in an OMA system benefits from a reduced complexity compared to that of a NOMA system[PerGainWei]. Meanwhile, hybrid NOMA synergizes the advantages of both NOMA and OMA, enabling IUI-free transmission for certain users while simultaneously supporting a larger number of users through non-orthogonal multiplexing. In contrast, RSMA multiplexes common and private data streams via precoding in the spatial domain based on a proper rate splitting strategy at the transmitter side[BrunoJSAC]. At the RSMA receiver, the common data stream is decoded by treating the interference as noise, and the private data streams are decoded successively based on the SIC framework. This structure offers high flexibility in exploiting the spatial DoF and resource allocation, which bridges spatial-division multiple access (SDMA) and NOMA in an efficient manner[BrunoJSAC].

We note that the input-output relationship in (1) can be applied to both uplink and downlink transmission. In downlink communication systems, the BS has the full information of $\mathbf{s}$ and $\mathbf{H}$ and thus the map** function $\mathbf{g}(\cdot)$ can be designed in a centralized manner at the BS. As in uplink systems, the map** function $\mathbf{g}(\cdot)$ needs to be designed in a distributed manner, since each user only has its own specific CSI (either perfect or imperfect CSI) and information symbols.

Consider an uplink system, where a single-antenna BS serves $K$ single-antenna users. All the transceivers are equipped with an OFDM modulator or demodulator [DerrickEEOFDMA] and all the users are accessed the channel concurrently based on OFDMA. In this case, $M$ in (1) is the total number of subcarriers adopted in OFDM modulation. Between user $k$ and the BS, there would be a frequency domain channel matrix $\mathbf{H}_{k}\in\mathbb{C}^{M\times M}$. Thus, we have $N=MK$ and the channel matrix in (1) can be rewritten as $\mathbf{H}=\left[\mathbf{H}_{1},\ldots,\mathbf{H}_{K}\right]$. The data streams $\mathbf{s}$ and the received signals $\mathbf{y}$ are also defined in the frequency domain. In this case, the total number of data streams $D\leq M$ as the system DoF is limited by the number of subcarriers adopted at the BS.

For the considered uplink OFDMA system, the input-output relationship in (1) can be specified as

where $\mathbf{\Lambda}_{k}=\operatorname{\mathrm{diag}}\{\sqrt{p_{k1}},\ldots,\sqrt{p_{kD_{k}}}\}$ and $p_{kd_{k}}\geq 0$ is the power allocation for the $d_{k}$-th data stream of user $k$. Note that the total transmit power of user $k$ should be limited by its power budget $P_{k,\mathrm{max}}$, i.e., $\sum_{d_{k}=1}^{D_{k}}p_{kd_{k}}\leq P_{k,\mathrm{max}}$. The binary matrix $\mathbf{\Pi}_{k}\in\mathbb{B}^{M\times D_{k}}$ is the user scheduling matrix. $\{\mathbf{\Pi}_{k}\}_{md_{k}}=1$ denotes that the $d_{k}$-th data stream of user $k$ is scheduled to the $m$-th subcarrier, otherwise $\{\mathbf{\Pi}_{k}\}_{md_{k}}=0$. Without loss of generality, we assume that each data stream is allocated to one subcarrier, i.e., $\sum_{m=1}^{M}\{\mathbf{\Pi}_{k}\}_{md_{k}}=1$, $\forall d_{k}$, and thus the total number of subcarriers allocated to user $k$ is $\sum_{d_{k}=1}^{D_{k}}\sum_{m=1}^{M}\{\mathbf{\Pi}_{k}\}_{md_{k}}=D_{k}$. For OFDMA, we need to limit $\sum_{k=1}^{K}\sum_{d_{k}=1}^{D_{k}}\{\mathbf{\Pi}_{k}\}_{md_{k}}\leq 1$ to guarantee that at most one data stream of a user can be scheduled to each subcarrier.

In practice, when a static multipath propagation environment and a proper cyclic prefix length are considered, the frequency domain channel of each user $\mathbf{H}_{k}$ is a diagonal matrix, which lays the foundation for interference-free parallel data stream transmission and facilitates low-complexity single-tap FDE at the BS. In contrast, in high-mobility channels when the Doppler is not well compensated, the frequency domain channel $\mathbf{H}_{k}$ is generally not diagonal, leading to severe inter-carrier interference (ICI) in OFDM and IUI in OFDMA systems[Zhiqiang_magzine]. In this case, multiple access schemes designed in the delay-Doppler domain might be more beneficial [Chong2022achievable], as will be detailed in the later sections. For now, we restrict ourselves to OFDM transmission in a static environment, i.e., ICI-free, in which case OFDMA is appealing for multi-user communication. In this case, the received signal in subcarrier $m$, $\forall m\in\{1,\ldots,M\}$, at the BS is given by

Due to $\sum_{k=1}^{K}\sum_{d_{k}=1}^{D_{k}}\{\mathbf{\Pi}_{k}\}_{md_{k}}\leq 1$, the multiple data streams of each user can be easily retrieved at the BS based on a simple single-tap FDE, and thus the achievable rate of user $k$ can be formulated as:

The resource allocation design challenge is to schedule the data streams of all users and allocate power to each data stream based on the available CSI, for achieving the design goal while taking into the system resource constraints. Note that the user scheduling matrices $\mathbf{\Pi}_{k}$, $\forall k$, need to be designed at the BS in a centralized manner[DerrickEEOFDMA], while the power allocation $\mathbf{\Lambda}_{k}$ for user $k$ can be designed distributively at each user, based on its individual CSI.

Consider a downlink system where a single-antenna BS serves $K$ single-antenna users. Both the BS and users still adopt OFDM modulation and demodulation by dividing the system bandwidth into $M$ subcarriers. In this downlink system, the channel matrix in (1) is constituted by $\mathbf{H}=\left[\mathbf{H}_{1},\ldots,\mathbf{H}_{K}\right]^{\mathrm{T}}\in\mathbb{C}^{MK\times M}$ and the received signal in (1) is given by $\mathbf{y}=\left[\mathbf{y}^{\rm T}_{1},\ldots,\mathbf{y}^{\rm T}_{K}\right]^{\mathrm{T}}\in\mathbb{C}^{MK\times 1}$, i.e., $N=MK$, where $\mathbf{H}_{k}\in\mathbb{C}^{M\times M}$ and $\mathbf{y}_{k}\in\mathbb{C}^{M\times 1}$ denote the frequency domain channel matrix between the BS and user $k$ and the received frequency domain signal at user $k$, respectively. In contrast to OFDMA, we consider an overloaded scenario and assume that the BS can schedule only one data stream to transmit for each user, i.e., $K=D>M$ and $D_{k}=1$, $\forall k$. Thus, power domain NOMA can be adopted to accommodate all the $K$ users to form multi-carrier NOMA[Wei2017]. For power-domain NOMA, superimposition coding is adopted at the transmitter [Wei2017, PerGainWei] to multiplex more than one user on each subcarrier while SIC decoding is employed at the user side to mitigate IUI. In this case, the input-output relationship in (1) can be specified as

where $\mathbf{s}\in\mathbb{C}^{K\times 1}$ collects the information symbols for all the $K$ users, $\mathbf{z}_{k}\sim\mathcal{CN}(\mathbf{0},\sigma^{2}_{k}\mathbf{I}_{M})$ denotes the noise at user $k$ with noise power of $\sigma^{2}_{k}$, and $\mathbf{\Pi}\in\mathbb{B}^{M\times K}$ denotes the user scheduling matrix at the BS. The diagonal matrix $\mathbf{\Lambda}$ is defined by $\mathbf{\Lambda}=\operatorname{\mathrm{diag}}\{\sqrt{p_{1}},\ldots,\sqrt{p_{K}}\}\in\mathbb{R}^{K\times K}$, where $p_{k}\geq 0$ is the power allocation for user $k$ at the BS. $\{\mathbf{\Pi}\}_{mk}=1$ denotes that user $k$ is scheduled to subcarrier $m$, otherwise $\{\mathbf{\Pi}\}_{mk}=0$. More explicitly, the received signal in subcarrier $m$ at user $k$ is given by

For power domain NOMA, since more than one user can be scheduled on each subcarrier, i.e., $\sum_{k=1}^{K}\{\mathbf{\Pi}\}_{mk}\geq 1$, achieving IUI-free transmission is not a practical expectation without exploiting extra DoF from other domains. Instead, the power domain NOMA relies on power allocation and SIC decoding by exploiting the channel discrepancies such that the inherent interference can be properly harnessed. According to the principle of power domain NOMA, on subcarrier $m$, the power allocation and the SIC decoding order should be determined by the channel gain order[Wei2017]. In particular, denote the user set scheduled on subcarrier $m$ as $\mathcal{K}_{m}=\{k|\{\mathbf{\Pi}\}_{mk}=1,\forall k\}$ and without loss of generality their channel gains are sorted $\{\mathbf{H}_{k}\}_{mm}\geq\{\mathbf{H}_{k^{\prime}}\}_{mm}$, $\forall k<k^{\prime}\in\mathcal{K}_{m}$. Accordingly, if both users $k$ and $k^{\prime}$ are scheduled on subcarrier $m$, a higher power is allocated to the user with a lower channel gain, i.e., ${p}_{k}\leq p_{k^{\prime}}$. Also, the user with a higher channel gain always first decodes the information of the user with a lower channel gain before attempting to decode its own information symbol [Tse2005]. The achievable rate for user $k$ is given by

where $\sum_{m=1}^{M}\{\mathbf{\Pi}\}_{mk}=1$ as each user can only be allocated to at most one subcarrier in the overloaded scenario. Note that user $k$ suffers from IUI from user $k^{\prime}<k\in\mathcal{K}_{m}$ having a stronger channel in subcarrier $m$. Particularly, for the strongest user in subcarrier $m$, it needs to perform $|\mathcal{K}_{m}|-1$ steps SIC and then decodes its own information. As a result, IUI-free decoding is realized for this user, and its achievable rate is given by

The resource allocation for downlink power domain NOMA is performed at the BS in a centralized manner and its task is to schedule users and to allocate power based on the CSI of all the $K$ users, i.e., $\mathbf{H}_{1},\ldots,\mathbf{H}_{K}$, for realizing a certain system design goal.

OTFS modulation is a recently proposed multi-carrier waveform that aims to provide reliable communication over time-varying channels [Hadani2017orthogonal, ZhiqiangLetterPartI, ShuangyangLetterPartII, WeijieLetterPartIII]. In contrast to the conventional OFDM waveform, OTFS modulates information symbols in the delay-Doppler (DD) domain [Yuan2023survey, Yuan2019simple], offering a distinct approach to signal processing and transmission. As a result, OTFS modulation can effectively exploit the DD domain channel properties, including quasi-static, compactness, potential sparsity, and separability [Zhiqiang_magzine], which enables numerous advanced designs in various wireless applications [Mengmeng2023P2PMIMO, Ruoxi2023Globecom, Shuangyang_ISAC, Weijie2021JSTSP, Chang2023Predictive]. Notably, the coupling between the DD domain channel and the information symbols is characterized by the “twisted convolution” [Shuangyang2023Globecom, LSY_THP, Zhiqiang2022off], a specialized form of convolution that incorporates an additional phase term. The convolution nature of the DD domain channel naturally provides new design challenges and opportunities for NGMA.

Let us consider an uplink multi-user transmission leveraging OTFS waveform with $K$ users and the channel between each user and the BS has $P$ independent resolvable paths. For the sake of illustration, both the BS and $K$ users are equipped with a single antenna. We define a DD grid of size ${\tilde{M}}\times{\tilde{N}}$ for accommodating information symbols, which corresponds to a time-frequency (TF) grid with ${\tilde{M}}$ subcarriers and $\tilde{N}$ time slots. Besides, we consider the case of DDMA [Chong2022achievable], where each user’s information symbols are placed along the whole Doppler dimension but with a specific delay, i.e., each user occupies a row in the DD grid, as shown in Fig. 2. Notice that the coupling relationship between channel and information in the DD domain is convolutional in nature[Hadani2017orthogonal]. Particularly, we shall notice that each user’s data will spread to $l_{\max}$ adjacent rows due to the channel delay. Here, $l_{\max}$ denotes the maximum delay index that is computed as the ratio between the maximum delay spread and the delay resolution $T/{\tilde{M}}$, where $T$ denotes the duration of one time-slot (corresponding to the duration of an OFDM symbol). Specifically, the symbol multiplexing for user $k$ can be characterized by the indicator matrix ${\bf\Pi}_{k}\in{\tilde{M}\tilde{N}\times\tilde{N}}$, $\forall k$, associated for each user [Chong2022achievable], which maps the information symbols to the $k$-th row of the DD grid. Thus, we have $D=\tilde{N}$ and $N=\tilde{M}\tilde{N}$. We adopt ${\bf H}_{k}\in{\mathbb{C}}^{\tilde{M}\tilde{N}\times\tilde{M}\tilde{N}}$ denoting the DD domain channel matrix between user $k$ to the BS such that $M=\tilde{M}\tilde{N}$. Thus, the input-output relationship in (1) is further specified by

where $\mathbf{s}_{k}\in{\tilde{N}\times 1}$, $\forall k$, is the symbol vector of user $k$, $p_{k}>0$ is the transmit power of user $k$, and the detailed structure of ${\bf H}_{k}$ is specified in [Chong2022achievable, li2021cross]. Assuming that the SIC detection is adopted at the BS, i.e., the BS always first decodes information of the stronger user with a higher received signal power while treating the IUI caused by weaker users as noise. By assuming that the users’ received signal power levels are sorted in descending order, i.e., $p_{k}\operatorname{Tr}\left(\mathbf{H}_{k}\mathbf{\Pi}_{k}\mathbf{\Pi}^{\rm H}_{k}\mathbf{H}^{\rm H}_{k}\right)\geq p_{k^{\prime}}\operatorname{Tr}\left(\mathbf{H}_{k^{\prime}}\mathbf{\Pi}_{k^{\prime}}\mathbf{\Pi}^{\rm H}_{k^{\prime}}\mathbf{H}^{\rm H}_{k^{\prime}}\right)$, $\forall k<k^{\prime}$, the achievable rate of user $k$ is given by[Chong2022achievable]

where the IUI faced by user $k$ is given by

It should be noted that the rate expression given in (10) is subjected to the specific realization of channel delay, Doppler, and fading coefficients, which is generally challenging to acquire at the user side in uplink transmissions. It was shown in [Chong2022achievable] that the achievable rate in (10) can be tightly upper-bounded by the rate determined by the effective signal-to-interference-plus-noise ratio (SINR) of each user, which is only related to the effective channel gains. More importantly, compared to uplink OFDMA with SIC, a modest achievable rate improvement can be observed thanks to the fact that OTFS is insensitive to delay and Doppler, while OFDMA is prone to substantial channel variability within the TF domain due to the presence of delay and Doppler [Li2020performance, Wei2020transmitter, Shuangyang2021hybrid]. This sensitivity leads to considerable fluctuations in the effective SINR of OFDMA, and consequently decreases the achievable rate.

RSMA offers a versatile and generalized non-orthogonal transmission, which presents a dynamic interference management framework that softly bridges SDMA and power domain NOMA, adapting to varying interference levels and specific system requirements[BrunoJSAC].

Here, we consider a narrow-band downlink single-layer linearly precoded RSMA scheme [YijieRSMA] where a multi-antenna BS serves $K$ single-antenna users. In this case, the channel matrix in (1) is defined in the spatial domain, the number of antennas at the BS is $N$, and $M=K$. Again, in overloaded scenarios, we assume that the BS schedules only one data stream to transmit for each user, i.e., $K=D$ and $D_{k}=1$, $\forall k$. The message $W_{k}$ for user $k$ is firstly split into two sub-messages, i.e., one common message $W_{\mathrm{c},k}$ and one private message $W_{\mathrm{p},k}$. The common messages of all the $K$ users $W_{\mathrm{c},1},\ldots,W_{\mathrm{c},K}$ are combined into one common message $W_{\mathrm{c}}$ and is encoded into a common data stream $s_{\mathrm{c}}$, which is decoded by all the users. The private sub-messages of all the $K$ users $W_{\mathrm{p},1},\ldots,W_{\mathrm{p},K}$ are independently encoded into private data streams ${s}_{\mathrm{p},1},\ldots,{s}_{\mathrm{p},K}$, which are decoded by the corresponding users, respectively. Then, the generated $K+1$ data streams intended for $K$ users are linearly precoded and transmitted to the channel. For the considered downlink single-layer RSMA system, the input-output relationship in (1) can be specified as

where $\mathbf{y}\in\mathbb{C}^{K\times 1}$ collects the received signal of all the $K$ users and $\mathbf{H}=[\mathbf{h}_{1},\ldots,\mathbf{h}_{K}]^{\rm T}\in\mathbb{C}^{K\times N}$ collects the channel vectors between the BS and all the users. $\mathbf{p}_{\mathrm{c}}\in\mathbb{C}^{N\times 1}$ and $\mathbf{p}_{\mathrm{p},k}\in\mathbb{C}^{N\times 1}$, respectively, denote the precoding vectors for the common data stream and the private data stream of user $k$, $\forall k\in\{1,\ldots,K\}$. The received signal at user $k$ is given by

At the receiver side, each user first decodes the common data stream $s_{\mathrm{c}}$ into the common message $\widehat{W}_{\mathrm{c}}$ and performs SIC to subtract its signal from the received signal while treating the interference from all the private data streams as noise. Then, user $k$ decodes its private data stream ${s}_{\mathrm{p},k}$ into its private message $\widehat{W}_{\mathrm{p},k}$ by treating the remaining interference from other private data streams as noise. After decoding the private message, user $k$ reconstructs the original message by extracting $W_{\mathrm{c},k}$ from $W_{\mathrm{c}}$ and combining $W_{\mathrm{c},k}$ with $W_{\mathrm{p},k}$ into $W_{k}$. Based on the above procedure, the instantaneous rates for decoding the common and private data streams at user $k$ are given by

respectively. Note that the achievable rate for the common message cannot exceed $\{\mathbf{R}\}_{\mathrm{c},k}$, such that all the users can decode $s_{\mathrm{c}}$ successfully, i.e.,

Without loss of generality, let $C_{k}$ denote the common rate allocation to user $k$ for $W_{\mathrm{c},k}$. The overall achievable rate for user $k$ includes both the common rate $C_{k}$ and the private rate $R_{\mathrm{p},k}$, i.e.,

The resource allocation for downlink single-layer RSMA is also performed at the BS in a centralized manner. The resource allocation aims to design the precoding vectors and the common rate allocation based on the CSI of all the $K$ users to achieve a specific system design objective.

In summary, the choice of domain for defining the channel matrix $\mathbf{H}$ significantly influences the specification of the signal generation function $\mathbf{g}(\cdot)$. Indeed, the domain of $\mathbf{H}$ sets the foundation for the overall NGMA framework, while $\mathbf{g}(\cdot)$ specifies the particular NGMA scheme to employ. More importantly, both $\mathbf{H}$ and $\mathbf{g}(\cdot)$ are indicative of the available system resources that can be allocated, which determine the problem formulation and establish the resource allocation design DoF for improving the system performance.

Recently, several promising techniques have been proposed to proactively configure wireless channels, aiming to mitigate the issues mentioned earlier. By customizing the radio propagation environment, these approaches facilitate the creation of favorable wireless channel conditions for multiple access, thereby inherently overcoming the obstacles faced by NGMA communication systems. Generally, the NGMA model for reconfigurable channels is described by

where $\mathbf{\Phi}$ is the channel reconfiguration variable and its dimension is determined by the specific type of channel reconfiguration techniques. In this scenario, the method of multiple access and the resource allocation among multiple users, i.e., the vector function $\mathbf{g}(\mathbf{s})$, can be jointly designed with the channel reconfiguration variable $\mathbf{\Phi}$. Particularly, the system resources can be utilized strategically to enhance the system performance based on specific QoS requirements. Different from the fixed conditions for natural channels, the reconfigurable channel condition enabled by $\mathbf{\Phi}$ introduces greater flexibility in designing NGMA schemes. Consequently, the focus of resource allocation optimization problems shifts from merely determining an efficient strategy, $\mathbf{g}(\mathbf{s})$, for a given fixed radio propagation environment to jointly designing both the strategy $\mathbf{g}(\mathbf{s})$ and the wireless channel $\mathbf{H}\left(\mathbf{\Phi}\right)$. From a mathematical perspective, the former scenario is essentially a subset of the latter case. With this concept in mind, it is crystal clear that the potential for performance enhancement is greater when the overall channel is properly reconfigured by smartly designing $\mathbf{\Phi}$. In particular, in practical scenarios where $D\leq\min\{N,M\}$, the original system resources are sufficient to accommodate all the $K$ users, OMA schemes remain to be an efficient approach to serve all the $K$ users. As such, the extra design DoF can be leveraged to establish more favorable radio propagation environments to achieve desired system design objectives. As such, resha** each user’s channel is possible such that the orthogonality among channels can be maintained, facilitating the mitigation of IUI. On the other hand, when the original system DoF of the natural channel is limited, i.e., $D>\min\{N,M\}$, the additional design DoF introduced by $\mathbf{\Phi}$ becomes crucial. It enables the creation of a more favorable radio propagation environment for NOMA or hybrid NOMA. In particular, by fine-tuning the channel gains, the user paring policy in NOMA or hybrid NOMA can be designed more flexibly and appropriately, potentially enhancing the system performance.

In the following, we will briefly introduce several promising emerging techniques in the literature that can be exploited to shape desired radio propagation environments.

Equip** UAVs with communication transceivers introduces the capability to establish air-to-ground NGMA links, which can significantly enhance system performance. Compared to conventional cellular systems that rely on fixed terrestrial infrastructure [wong2017key, you2021towards], UAV-assisted communication systems offer on-demand connectivity by deploying UAV-mounted wireless transceivers flexibly over a target area. This flexibility is particularly useful in emergency situations, such as natural disasters and major accidents, where UAVs can be leveraged as aerial base stations to establish temporary communication links quickly and cost-effectively [8663615]. Additionally, the high mobility and maneuverability of UAVs allow them to adapt their trajectories based on the environment and terrain variations, thus improving system performance. By properly optimizing the UAV trajectories, we can place new communication nodes at desired locations to facilitate certain optimal wireless channel conditions for NGMA schemes. Therefore, UAV-assisted communication systems offer a promising solution for providing connectivity in remote or hard-to-reach areas, emergency situations, and other scenarios where traditional infrastructure may be unavailable or insufficient.

Next, to illustrate the above advantages, we consider a toy example where a UAV-mounted BS is equipped with a uniform planar array (UPA) composing $N_{x}\times N_{y}$ antenna elements, i.e., $N=N_{x}N_{y}$, and serves a group of $K$ single-antenna users. In such a case, the reconfigurable channel between the UAV and user $k$ is given by [8445944]

Here, $\varrho=(\frac{c}{4\pi f_{c}})^{2}$ is a constant determined by the speed of light $c$ and the carrier frequency $f_{c}$. Moreover, variables $\mathbf{r}_{0}=[x_{0},~{}y_{0},~{}H_{0}]^{\mathrm{T}}$ and $\mathbf{r}_{k}=[x_{k},~{}y_{k},~{}0]^{\mathrm{T}}$ indicate the three-dimensional (3D) Cartesian coordinates of the UAV and user $k$, respectively, where $H_{0}$ is the fixed altitude of the UAV. Furthermore, the term $\sqrt{\varrho}\left\|\mathbf{r}_{0}-\mathbf{r}_{k}\right\|^{-1}$ and vector $\mathbf{a}_{k}\in\mathbb{C}^{{N}\times 1}$ represent the channel gain between the UAV and user $k$ and the corresponding antenna array response (AAR), respectively. Specifically, vector $\mathbf{a}_{k}$ is expressed as [8974403]