Joint Beamforming, Energy Management, and Trajectory Optimization for Figure-Eight Loitering in Solar-Powered HAPS-Enabled ISAC Systems

Solar-powered HAPSs have attracted significant attention owing to their capability of achieving long-endurance operation through solar energy harvesting and onboard energy storage [22]. Existing studies have investigated energy-aware communication design, trajectory optimization, and resource allocation for solar-powered aerial platforms [35, 19, 37, 6, 20, 16]. For example, [35] optimized aircraft trajectories to balance harvested and consumed energy, while [19] developed an interdisciplinary framework integrating aerodynamics, communications, and renewable energy management. Nevertheless, these studies primarily focus on communication and energy management without explicitly considering integrated sensing functionalities.

The emergence of integrated sensing and communication (ISAC) has motivated extensive research on the joint optimization of communication and sensing performance for aerial platforms. Existing works have investigated communication-sensing beamforming design, including energy-efficient beamforming and distributed/bistatic multiple-input multiple-output (MIMO) ISAC architectures [15, 14]. Furthermore, the mobility of aerial platforms provides an additional degree of freedom for improving ISAC performance, leading to extensive studies on the joint design of trajectory optimization, beamforming, and resource allocation for UAV-enabled ISAC systems [24, 23, 28, 11, 38]. Representative examples include the energy-aware UAV-ISAC framework in [24], the limited-backhaul UAV-enabled ISAC framework in [23], the control-based trajectory optimization approach in [28], and the joint beamforming and trajectory optimization schemes proposed in [11, 38]. These studies demonstrate the benefits of mobility-aware communication-sensing co-design.

Compared with low-altitude UAV platforms, HAPS-enabled ISAC systems operate in the stratosphere and are expected to provide persistent wide-area coverage over extended periods. Consequently, renewable energy harvesting, battery dynamics, and propulsion energy consumption become fundamental design considerations. Recent studies have begun extending ISAC to HAPS platforms through joint communication-sensing optimization [21, 51]. For example, [21] investigated a HAPS-UAV integrated ISAC architecture, while [51] jointly optimized HAPS deployment and beamforming under SAR imaging constraints. In parallel, joint communication and SAR imaging (JCASAR) has emerged as a promising paradigm for enabling communication and imaging systems to share spectrum, hardware, and energy resources. Existing studies have mainly focused on waveform design and dual-function radar-communication architectures [45, 44, 48, 30, 54], demonstrating the feasibility of integrating communication and SAR imaging.

Despite these advances, existing studies generally optimize communication, sensing, trajectory control, or energy management over a single operational period. The long-term coupling among communication performance, SAR imaging, mobility, renewable energy harvesting, battery dynamics, and day-night operation has not yet been systematically investigated. This gap motivates the unified communication-sensing-energy co-design framework proposed in this paper, which jointly integrates communication transmission, SAR imaging, trajectory control, solar energy harvesting, and battery energy management to enable sustainable day-night operation for solar-powered HAPS-enabled ISAC systems.

This paper develops a solar-powered HAPS-enabled ISAC framework that jointly integrates communication, SAR imaging, figure-eight loitering, and energy management for sustainable day-night operation. The main contributions are summarized as follows.

• •

  We propose a figure-eight loitering architecture for solar-powered HAPS-enabled ISAC systems to provide persistent ISAC services over two geographically separated service regions while maintaining continuous fixed-wing flight.

• •

  We develop a unified communication-sensing-energy model that jointly characterizes solar energy harvesting, battery dynamics, propulsion power consumption under steady circular flight (SCF), communication transmission, and SAR imaging, thereby characterizing the coupling among communication, sensing, mobility, and energy sustainability.

• •

  We formulate coupled optimization problems for daytime operation (DTO) and nighttime operation (NTO), where the battery state explicitly links the two operational phases through a long-term energy budget.

• •

  During DTO, the flight speed profile, transmit beamforming, and sensing covariance matrix are jointly optimized to maximize the communication sum rate subject to sensing, mobility, power, and energy constraints. During NTO, the flight speed and altitude are jointly optimized to minimize propulsion energy consumption under the available battery energy. Efficient iterative algorithms are developed to solve the resulting non-convex optimization problems.

• •

  Numerical results demonstrate that, for the considered dual-region deployment scenario, the proposed framework achieves a higher communication sum rate than the benchmark trajectories while satisfying the prescribed sensing and energy constraints. Moreover, the proposed NTO strategy significantly reduces propulsion energy consumption, thereby supporting sustainable day-night operation.

Outline: The remainder of the paper is organized as follows. Section II introduces the solar-powered HAPS-enabled ISAC system model. Section III formulates two optimization problems for the DTO and NTO phases. Sections IV and V develop efficient solution methods for the DTO and NTO problems, respectively. Numerical results are presented in Section VI to evaluate the performance of the proposed framework. Finally, Section VII concludes the paper.

Notations: The vectors (matrices) are denoted by lower-case (upper-case) boldface characters. $\mathbf{I}_{M}$ stands for the $M\times M$ identity matrix, and $\mathbf{0}$ represents a zero matrix with appropriate dimension. $\dot{\mathbf{x}}(t)$ denotes the derivative of the time-dependent function $\mathbf{x}(t)$ with respect to the time $t$. Superscripts $T$, and $H$ represent the transpose and conjugate transpose, respectively. $\mathbb{E}\{\cdot\}$ represents the expectation operator. The trace of $\mathbf{A}$ are represented by $\mathrm{tr}(\mathbf{A})$. $[\mathbf{A}]_{i,j}$ stands for the $(i,j)$ entry of $\mathbf{A}$. $\mathbf{A}\succeq\mathbf{0}$ is equivalent to $\mathbf{A}$ being positive semidefinite. $|\cdot|$ stand for the magnitude of a complex number, and $\|\cdot\|$ is the Euclidean norm of a complex vector. $\mathcal{O}(\cdot)$ denotes the convolution Big-O notation.

Owing to the high operating altitude of the HAPS, a strong LoS link typically exists between the platform and each ground user. However, in urban or partially obstructed environments, scattering and reflection may contribute additional non-line-of-sight (NLoS) components. To account for both effects, the downlink channel is modeled as a 3D Rician fading channel composed of a dominant LoS component and a diffuse NLoS component. Accordingly, the channel vector between the HAPS and the $k$th user at time slot $n$ is expressed as [2, 40]

	$\displaystyle\mathbf{g}(\mathbf{p}[n],\mathbf{u}_{k})=\frac{1}{\sqrt{\ell}}$
	$\displaystyle\quad\times\Big(\sqrt{\frac{K_{u}}{K_{u}+1}}\mathbf{g}_{k,\mathrm{LOS}}[n]+\sqrt{\frac{1}{K_{u}+1}}\mathbf{g}_{k,\mathrm{NLOS}}[n]\Big),$		(8)

where $K_{u}$ is the Rician factor, $\ell=d^{2}(\mathbf{p}[n],\mathbf{u}_{k})/\rho_{0}$ denotes the free-space path loss with $\rho_{0}=(\frac{4\pi}{\lambda})^{2}$, and $d(\mathbf{p}[n],\mathbf{u}_{k})=\sqrt{\|\mathbf{p}[n]-\mathbf{u}_{k}\|^{2}+H^{2}}$ being the distance between the HAPS and the $k$-th user at time slot $n$. $\mathbf{g}_{k,\mathrm{NLOS}}[n]\sim\mathcal{CN}(\boldsymbol{0},\mathbf{I}_{M})$ represents the NLOS component, whose elements are independently drawn from a complex Gaussian distribution with zero mean and unit variance, and $\mathbf{g}_{k,\mathrm{LOS}}[n]=\mathbf{a}_{k}(\mathbf{p}[n],\mathbf{u}_{k})$ denotes the LoS component, represented by the steering vector associated with the $k$th user. Owing to the adopted ULA deployment and constant HAPS attitude, the steering vector depends only on the instantaneous relative geometry between the HAPS and the user, and is given by

		$\displaystyle\mathbf{a}_{k}(\mathbf{p}[n],\mathbf{u}_{k})=$
		$\displaystyle\Big[1,e^{j\pi\cos(\theta(\mathbf{p}[n],\mathbf{u}_{k})}),\ldots,e^{j\pi(M-1)\cos(\theta(\mathbf{p}[n],\mathbf{u}_{k}))}\Big]^{T}.$		(9)

Here, $\theta(\mathbf{p}[n],\mathbf{u}_{k})$ is the angle of departure (AoD) towards the $k$th user, given by

$\displaystyle\theta(\mathbf{p}[n],\mathbf{u}_{k})=\arccos\Big(\frac{H}{\sqrt{\|\mathbf{p}[n]-\mathbf{u}_{k}\|^{2}+H^{2}}}\Big).$

(10)

Accordingly, the received signal at the $k$th user during time slot $n$ can be expressed as

$\displaystyle y_{k}[n]=\mathbf{g}_{k}^{H}(\mathbf{p}[n],\mathbf{u}_{k})\mathbf{x}[n]+n_{k}[n],$

(11)

where $n_{k}[n]\sim\mathcal{CN}(0,\sigma_{k}^{2})$ denotes the additive white Gaussian noise with noise power $\sigma_{k}^{2}$.

The HAPS is assumed to employ horizontal solar panels mounted on its upper surface. As illustrated in Fig. 2, the solar-powered HAPS operates continuously by harvesting solar energy during the daytime and utilizing the stored electrical energy during the nighttime. During the DTO, a portion of the harvested solar power is directly consumed for station-keeping and communication, while the surplus energy is stored in rechargeable batteries to support NTO. To ensure energy-neutral operation over a full 24-hour cycle, this section develops a comprehensive energy model consisting of three components: DTO for energy collection, NTO for energy consumption, and energy storage dynamics. These three models collectively describe the energy flow within the HAPS system and form the analytical foundation for subsequent trajectory and energy optimization.

The DTO phase focuses on energy-aware communication and sensing optimization for the HAPS. During daylight hours, the HAPS simultaneously serves ground users and performs SAR imaging while harvesting solar energy. The objective of the DTO problem is to maximize the average achievable sum-rate by jointly optimizing the flight speed profile, transmit beamforming vectors, and sensing covariance matrix, subject to sensing, mobility, power, and energy constraints. Since the figure-eight trajectory is predetermined, the HAPS position $\mathbf{p}[n]$ is uniquely determined by the flight speed profile through the angular recursions in (1) and is therefore not treated as an independent optimization variable. The optimization problem is formulated as follows:

	$\displaystyle\mathcal{P}1:$	$\displaystyle\mathop{\max}\limits_{V[n],\mathbf{w}_{k}[n],\atop\mathbf{R}_{s}[n]\succeq\boldsymbol{0},\forall n}\frac{1}{N}\sum_{n=1}^{N}\sum_{k=1}^{K}\mathcal{R}_{k}[n],$
	$\displaystyle\mathrm{s.t.}$	$\displaystyle\,\,\mathbf{a}^{H}(\mathbf{p}[n],\mathbf{t}_{q})\left(\sum_{k=1}^{K}\mathbf{w}_{k}[n]\mathbf{w}_{k}^{H}[n]+\mathbf{R}_{s}[n]\right)$
		$\displaystyle\qquad\quad\times\mathbf{a}(\mathbf{p}[n],\mathbf{t}_{q})\geq\Gamma d^{2}(\mathbf{p}[n],\mathbf{t}_{q}),\,\,\forall q,$		(31a)
		$\displaystyle\,\,\mathrm{SNR}_{\psi}[n]\geq\mathrm{SNR}_{\mathrm{min}},$		(31b)
		$\displaystyle\,\,\sum_{k=1}^{K}\|\mathbf{w}_{k}[n]\|^{2}+\mathrm{tr}(\mathbf{R}_{s}[n])\leq P_{\mathrm{max}},$		(31c)
		$\displaystyle\,\,(\ref{trajectory_cons3}),\,\sum_{n=1}^{N_{1}-1}V[n]=\frac{2\pi R_{1}}{\Delta_{t}},\,\sum_{n=N_{1}}^{N-1}V[n]=\frac{2\pi R_{2}}{\Delta_{t}},$		(31d)
		$\displaystyle\,\,P_{a}-P_{\mathrm{SCF}}[n]-P_{\mathrm{ave}}[n]\geq 0,$		(31e)
		$\displaystyle\,\,E_{\mathrm{total}}[N]\geq e_{\mathrm{req}},$		(31f)

where constraint (31a) ensures that the sensing beampattern gain toward each target exceeds the prescribed threshold, thereby providing sufficient illumination for SAR imaging. Constraint (31b) guarantees that the received SAR SNR satisfies the minimum imaging requirement, whereas constraint (31c) limits the total communication and sensing transmit power. Constraint (31d) specifies the feasible flight-speed range and ensures that the HAPS completes one full revolution on each circle of the prescribed figure-eight trajectory. Constraint (31e) enforces the instantaneous power balance during the DTO phase by requiring the harvested solar power to simultaneously support propulsion and communication/sensing transmissions, thereby preventing battery discharge. Finally, constraint (31f) guarantees that the battery energy remaining at the end of the DTO phase is no smaller than the minimum energy required for the subsequent NTO phase, denoted by $e_{\mathrm{req}}$, thereby coupling Problems $\mathcal{P}1$ and $\mathcal{P}2$. The solution to Problem $\mathcal{P}1$ is presented in Section IV.

After completing the DTO phase, the battery energy available at the beginning of the NTO phase is denoted by $e_{\mathrm{start}}=E_{\mathrm{total}}[N]$. Since solar energy is unavailable during the NTO phase, the HAPS relies exclusively on the stored battery energy to support propulsion, communication, and SAR imaging. To maximize the nighttime endurance, the total propulsion power consumption is minimized subject to the flight and battery-energy constraints. The resulting optimization problem is formulated as

	$\displaystyle\mathcal{P}2:$	$\displaystyle\mathop{\min}_{H,V[n],\forall n}\sum_{n=1}^{N}P_{\mathrm{SCF}}(V[n],H),$
		$\displaystyle\quad\,\,\,\mathrm{s.t.}\,\,H_{\rm{min}}\leq H\leq H_{\rm{max}},$		(32a)
		$\displaystyle\qquad\quad\,V_{\rm{min}}\leq V[n]\leq V_{\rm{max}},$		(32b)
		$\displaystyle\qquad\quad\,\sum_{n=1}^{N}\eta_{d}(P_{\mathrm{SCF}}[n]+P_{\mathrm{ave}}[n])\Delta_{t}\leq e_{\mathrm{start}},$		(32c)

where constraints (32a) and (32b) specify the allowable flight altitude and speed ranges, respectively. Constraint (32c) ensures that the total battery energy consumed during the NTO phase does not exceed the available battery energy at its beginning. Since $e_{\mathrm{start}}=E_{\mathrm{total}}[N]$ and constraint (31f) guarantees $E_{\mathrm{total}}[N]\geq e_{\mathrm{req}}$, the battery energy available at the beginning of the NTO phase always satisfies its minimum energy requirement. Consequently, Problems $\mathcal{P}1$ and $\mathcal{P}2$ are coupled through constraints (31f) and (32c), which govern the battery energy transferred from the DTO phase to the NTO phase. The solution to Problem $\mathcal{P}2$ is presented in Section V.

Given the HAPS trajectory $\mathbf{p}[n]$, which is uniquely determined by the flight speed profile $V[n]$, the objective is to jointly optimize the communication beamforming vectors $\mathbf{w}_{k}[n]$ and the SAR imaging covariance matrices $\mathbf{R}_{s}[n]$. The corresponding optimization problem is formulated as

	$\displaystyle\mathcal{P}3:$	$\displaystyle\mathop{\max}\limits_{\mathbf{w}_{k}[n],\forall k,\atop\mathbf{R}_{s}[n]\succeq\boldsymbol{0},\forall n}\,\frac{1}{N}\sum_{n=1}^{N}\sum_{k=1}^{K}\mathcal{R}_{k}[n],$
		$\displaystyle\quad\,\,\,\,\mathrm{s.t.}\quad(\ref{P1}a),(\ref{P1}b),(\ref{P1}c),(\ref{P1}e)\,\mathrm{and}\,(\ref{P1}f).$

Since $\mathbf{p}[n]$ and $V[n]$ are fixed, the optimization variables $\mathbf{w}_{k}[n]$ and $\mathbf{R}_{s}[n]$ across different time slots are mutually independent. Consequently, problem $\mathcal{P}3$ can be decomposed into $N$ parallel subproblems, each corresponding to a specific time slot $n$, as

	$\displaystyle\mathcal{P}4.n:$	$\displaystyle\mathop{\max}\limits_{\mathbf{w}_{k}[n],\forall k,\atop\mathbf{R}_{s}[n]\succeq\boldsymbol{0}}\,\sum_{k=1}^{K}\mathcal{R}_{k}[n],$
		$\displaystyle\quad\,\,\mathrm{s.t.}\quad(\ref{P1}a),(\ref{P1}b),(\ref{P1}c),(\ref{P1}e)\,\mathrm{and}\,(\ref{P1}f).$

For problem $\mathcal{P}4.n$, we introduce the auxiliary matrix variable $\mathbf{W}_{k}[n]=\mathbf{w}_{k}[n]\mathbf{w}_{k}[n]^{H}$, which satisfies $\mathbf{W}_{k}[n]\succeq\boldsymbol{0}$ and $\mathrm{rank}(\mathbf{W}_{k}[n])=1$. Accordingly, problem $\mathcal{P}4.n$ can be equivalently reformulated as

	$\displaystyle\mathcal{P}5.n:$	$\displaystyle\mathop{\max}\limits_{\mathbf{W}_{k}[n]\succeq\boldsymbol{0},\forall k,\atop\mathbf{R}_{s}[n]\succeq\boldsymbol{0}}\,\sum_{k=1}^{K}\hat{\mathcal{R}}_{k}(\{\mathbf{W}_{k}[n]\},\mathbf{R}_{s}[n]),$
		$\displaystyle\,\,\mathrm{s.t.}\,\mathbf{a}^{H}(\mathbf{p}[n],\mathbf{t}_{q})\left(\sum_{k=1}^{K}\mathbf{W}_{k}[n]+\mathbf{R}_{s}[n]\right)\mathbf{a}(\mathbf{p}[n],\mathbf{t}_{q})$
		$\displaystyle\qquad\qquad\qquad\qquad\qquad\geq\Gamma d^{2}(\mathbf{p}[n],\mathbf{t}_{q}),\,\,\forall q,$		(33a)
		$\displaystyle\,\,\sum_{k=1}^{K}\mathrm{tr}(\mathbf{W}_{k}[n])+\mathrm{tr}(\mathbf{R}_{s}[n])\geq\frac{\mathrm{SNR}_{\mathrm{min}}V[n]}{C_{\mathrm{SAR}}\sin(\psi)^{2}},$		(33b)
		$\displaystyle\,\,\sum_{k=1}^{K}\mathrm{tr}(\mathbf{W}_{k}[n])+\mathrm{tr}(\mathbf{R}_{s}[n])\leq P_{\mathrm{max}},$		(33c)
		$\displaystyle\,\,\mathrm{rank}(\mathbf{W}_{k}[n])\leq 1,$		(33d)
		$\displaystyle\,\,(\ref{P1}e)\,\mathrm{and}\,(\ref{P1}f),$		(33e)

where $\hat{\mathcal{R}}_{k}(\{\mathbf{W}_{k}[n]\},\mathbf{R}_{s}[n])$ is given by

	$\displaystyle\hat{\mathcal{R}}_{k}(\{\mathbf{W}_{k}[n]\},\mathbf{R}_{s}[n])$
	$\displaystyle=\log_{2}\Big(\sum\limits_{k=1}^{K}\mathrm{tr}\left(\mathbf{G}_{k}\mathbf{W}_{k}[n]\right)+\mathrm{tr}\left(\mathbf{G}_{k}\mathbf{R}_{s}[n]\right)+\sigma_{k}^{2}\Big)$
	$\displaystyle\,\,\,-\log_{2}\Big(\sum\limits_{p=1,p\neq k}^{K}\mathrm{tr}\left(\mathbf{G}_{k}\mathbf{W}_{p}[n]\right)+\mathrm{tr}\left(\mathbf{G}_{k}\mathbf{R}_{s}[n]\right)+\sigma_{k}^{2}\Big).$

with

	$\displaystyle\mathbf{G}_{k}=$	$\displaystyle\mathbb{E}\{\mathbf{g}(\mathbf{p[n],\mathbf{u}_{k}})\mathbf{g}^{H}(\mathbf{p[n],\mathbf{u}_{k}})\}$
	$\displaystyle=$	$\displaystyle\frac{1}{\ell(K_{u}+1)}(K_{u}\mathbf{a}_{k}(\mathbf{p}[n],\mathbf{u}_{k})\mathbf{a}^{H}(\mathbf{p}[n],\mathbf{u}_{k})+1).$		(34)

Because the objective function in (33a) is non-concave and the rank constraints in (33d) are non-convex, problem $\mathcal{P}5.n$ remains a challenging non-convex optimization problem. To tackle this issue, we adopt the SCA method [12, 49, 50], which iteratively replaces the non-concave objective with a concave surrogate function, i.e., a locally tight and tractable approximation that preserves the original function’s first-order behavior. Specifically, at each iteration $o\geq 1$, we have $\hat{\mathcal{R}}_{k}(\{\mathbf{W}_{k}[n]\},\mathbf{R}_{s}[n])\geq\bar{\mathcal{R}}_{k}^{(o)}(\{\mathbf{W}_{k}[n]\},\mathbf{R}_{s}[n])$, where $\bar{\mathcal{R}}_{k}^{(o)}({\mathbf{W}_{k}[n]},\mathbf{R}_{s}[n])$ denotes the concave surrogate of $\hat{\mathcal{R}}_{k}$ at iteration $o$, which can be expressed as in (IV-A) with

$\displaystyle\bar{a}_{k}^{(o)}=\log_{2}\Big(\sum\limits_{p=1,p\neq k}^{K}\mathrm{tr}\Big(\mathbf{G}_{k}\mathbf{W}_{p}^{(o)}\Big)+\mathrm{tr}\left(\mathbf{G}_{k}\mathbf{R}_{s}^{(o)}\right)+\sigma_{k}^{2}\Big).$

$\displaystyle\bar{\mathbf{B}}_{k}^{(o)}[n]=\frac{\log_{2}(e)\mathbf{G}_{k}}{\sum\limits_{p=1,p\neq k}^{K}\mathrm{tr}\left(\mathbf{G}_{k}\mathbf{W}_{p}^{(o)}[n]\right)+\mathrm{tr}\Big(\mathbf{G}_{k}\mathbf{R}_{s}^{(o)}[n]\Big)+\sigma_{k}^{2}}.$

Furthermore, following the semidefinite relaxation (SDR) technique in [32, 33], the rank constraints in (33e) are relaxed, transforming problem $\mathcal{P}5.n$ into a convex semidefinite program (SDP), denoted as (SDR5.n). By iteratively solving (SDR5.n) using CVX [13], we obtain a sequence of solutions ${\mathbf{W}_{k}^{(o)}[n]}$ and ${\mathbf{R}_{s}^{(o)}[n]}$. The objective value of $\mathcal{P}5.n$ is guaranteed to increase monotonically across iterations, thereby ensuring the convergence of the overall optimization for problem $\mathcal{P}3$.