Cognitive Unmanned Aerial Vehicle-Aided Human Bond Communication System: Modeling and Performance Analysis

Effective networking over wireless media has become extremely essential today as communication between massive Internet of things (IoT) devices is on an increase, thereby leading to a limited spectrum resource for utilisation. Specifically, for healthcare infrastructure in a remote or critical situation, providing uninterrupted communication between the macro base station and IoT devices or user nodes is imperative. However, owing to their limited spectral capacity, unmanned aerial vehicles (UAVs)-based networks can provide an efficient solution and utilise both licensed and unlicensed bands for communication among users or devices. In this paper, our focus is on cache-enabled cognitive networking for secondary users (SUs) that accredits precise communication delivery for critical healthcare systems that are performed by the cognitive UAV (CUAV). In addition, we develop a caching strategy wherein a CUAV is capable of caching relevant information from high-power (HP) and moderate-power (MP) devices in its local and cloud storage by applying a non-orthogonal multiple-access method. In the downlink scenario, the CUAV proactively transmits the requested HP and MP information to the designated SUs considering this entire model over two states, namely effectual state and interference state, which we can realise by any presence or absence of interference. To maximise this system’s energy efficiency, we formulate an optimisation problem to minimise the transmission power and satisfy the target performance in terms of throughput for SUs. We solve the optimisation issue using the Lagrangian approach and the Karush-Kuhn-Tucker conditions. In all simulations, the energy efficiency during the effectual state renders an average performance of approximately 400% better than that of the interference state.

Recently, the scientific community has demonstrated the 91 opportunities and complexities of cognitive devices, UAV 92 communication, and caching tools with a possible solution 93 to alleviate most complexities introduced in some academic 94 research. A comprehensive analysis of the extensive imple-95 mentation of UAV systems in disaster control and health 96 management applications is provided in [13] and [14], 97 where the authors concentrated on ultrasonic sensing, neu-98 ral networks, and geotagging to accurately classify the 99 disaster-sensitive position, establish communication, and 100 assist primary response by a network. In [15], a robust system 101 architecture was presented to manage medical emergencies 102 through a UAV-based network, where one UAV coordinates 103 with another UAV based on its capabilities (i.e. existing 104 power, signal strength, reliability), and undertakes necessary 105 action to resolve the emergency. 106 The authors in [16] and [17] introduced a cognitive agent 107 (CA) that executes mobile edge computing cognitive features 108 in smart devices targeted at 5G mobile communication net-109 works. Every smart computer that owns a CA can interpret 110 intent, cache user information, contact data, and create activ-111 ity profiles based on the user interface in the Internet of things 112 (IoT) system. We see a related caching approach in [18], 113 where the authors jointly suggest an edge caching technique 114 that considers information awareness and a reward-based 115 algorithm to encourage UAV-focused content sharing of the 116 user equipment. Next, to realise the effectiveness of proba-117 bilistic caching, the authors in [19] investigated the service 118 success probability in a heterogeneous network by imple-119 menting a base station (BS) in a UAV with different caching 120 capacities and then maximising the probability caching place-121 ment. The practical implementation of any UAV-based com-122 munication is challenging owing to the endurance of the UAV; 123 to mitigate this issue, the study of authors in [20] presented 124 proactive caching as a viable solution. Additionally, in [20], 125 the authors proposed a model in which the UAV caches data 126 during the file caching process; subsequently, the ground user 127 may recover the file and store the information in their local 128 cache for every other ground user requiring similar mate-129 rial, thereby solving the endurance issue. A similar approach 130 is presented in [21], where the authors demonstrated four 131 major application scenarios of a UAV-supported ultra-dense 132 network and then presented a power management framework 133 based on network and spectrum state considerations. 134 The authors concentrated on a quality of experience system 135 by integrating UAVs with caching functionality in small-cell 136 networks to provide consumers with the most popular content 137 from their local cache to mitigate backhaul congestion 138 and then proposed a case study in network disturbance 139 management in [22]. With similar intentions of increasing 140 throughout, the authors of [23] provided a cache-enabled 141 UAV model over an IoT machine-type device network by 142 jointly optimising UAV deployment and formulating a con-143 cave problem for probabilistic caching placement. To guaran-144 tee the protection of UAV transmitted wireless networks, the 145 authors of [24] suggested a scheme under which a UAV with sensing for the limit of any PU's incursion throughput to 188 optimise the efficient transmission. However, the aforemen-189 tioned research [27], [28] neither observed nor formulated the 190 energy efficiency and optimisation issues for a cache-enabled 191 cognitive UAV (CUAV) network. In this paper, we inves-192 tigate the performance related to the energy efficiency of 193 a cache-enabled CUAV network as a possible solution to 194 proper spectrum utilisation and interference management for 195 HBC communication framework in a downlink scenario. 196 In the following, we summarise the key contributions of our 197 research.

198
• First, we propose a framework of cache-enabling cog-199 nitive network model that the CUAV caches from two 200 high-power (HP) and moderate-power (MP) devices in 201 the uplink scenario; it is then dispatched to serve the 202 SUs with the requested data in the downlink scenario. 203 In addition, we design a caching scheme concerning the 204 requested data by the SUs. Further, we study a spectrum 205 sensing scheme to identify idle channels and an inter-206 ference temperature technique to protect primary com-207 munication (licensed communication) for the proposed 208 model. Moreover, a power model and a channel model 209 are provided for CUAV communication, applicable to 210 both the uplink and downlink phases. 211 • Second, we derive the throughput and energy efficiency 212 expressions in the downlink scenario regarding the 213 requested information by SUs for effectual and interfer-214 ence states. In the considered network, a NOMA scheme 215 is used for HP and MP device information.

216
• Third, we formulate an optimisation problem for the 217 effectual and interference states, where we try to min-218 imise the transmission power of CUAV and MBS while 219 maintaining guaranteed throughput of HP and MP infor-220 mation. Subsequently, the formulated power-allocation 221 problem is solved by utilising the decomposition 222 method.

223
• Finally, we conduct simulations to evaluate the perfor-224 mance of the proposed NOMA scheme. By comparing 225 the performance of the NOMA system with the corre-226 sponding orthogonal multiple access (OMA) system in 227 the considered network, we show that the NOMA sys-228 tem outperforms its OMA system by achieving higher 229 energy efficiency. In addition, we demonstrate that 230 the proposed scheme outperforms the three benchmark 231 schemes in simulation results.

232
The remainder of this paper is organised as follows. 233 Section II covers the system model with a network, caching, 234 spectrum sensing, and channel design. In Section III, 235 we derive the expressions of throughput and energy effi-236 ciency. In Section IV, we solve the energy efficiency with 237 the corresponding formulation of the optimisation problem. 238 In Section V, we verify the effectiveness of our proposed 239 system with energy efficiency and compare the results of the 240 proposed system with benchmark schemes from the simula-241 tion. Finally, Section VI provides concluding statements and 242 future work.

II. SYSTEM AND CHANNEL MODELS
Consider a CUAV-enabled long-term-evolution communica-246 tion system, in which carrier aggregates between licensed 247 macro cells and unlicensed small cells [29]. As illustrated 248 in Fig. 1, the macro cell is the primary cell with which 249 PUs communicate and maintain their connection with the 250 MBS. One small cell is the secondary cell comprising a 251 CUAV (i.e. a small BS) connected to the SUs supporting an 252 unlicensed carrier with MBS. A CUAV is connected to the 253 MBS via a capacity-constrained backhaul link. Each PU is 254  The uplink or sensing phase is referred to as the caching 270 phase. In the uplink schema shown in Fig. 1 The downlink or serving phase is referred to as the deliv-288 ery phase. After the completion of the sensing service, the 289 CUAV 2 dispatches to serve SUs with the cached data over the 290 remote or critically located healthcare infrastructure, where 291 CUAV acts as a secondary transmitter and SUs are secondary 292 receivers. There are two types of SUs: SU-HP and SU-MP 293 in which the SU-HP device downloads the HP data, and 294 the SU-MP device downloads the MP data from the CUAV. 295 Each SU-HP device was paired with another SU-MP device. 296 The sets of SU-HPs and SU-MPs are denoted as N sh = 297 {1, 2, . . . , N sh } and N sm = {1, 2, . . . , N sm }, respectively, 298 where N sh and N sm denote the total numbers of SU-HP and 299 SU-MP devices, respectively. Therefore, SU is the sum of 300 the SU-HP and SU-MP devices. The set of SUs is denoted 301 by N s = N sh ∪ N sm . The set of PUs is denoted by N p = 302 1, 2, . . . , N p , where N p is the total number of PUs. CUAV 303 transmits the sensing data to SU-HP and SU-MP devices 304 using the same subcarrier with different power levels because 305 the NOMA technique is applied for the transmission of HP 306 and MP data. 307 2 CUAV can directly deliver the sensing data to the SUs; otherwise, the CUAV requests the sensing data from the MBS and then transmits it to the requested SUs. VOLUME 10, 2022 to the SU-HP and SU-MP devices via two types of links:

338
where δ is the skewness parameter (0 < δ ≤ 1). A similar 339 proof for (2) is given in [31] when δ = 1. The CUAV needs to first download the requested sensing data from the MBS and then transmits it to the requested SU-HP and SU-MP devices. Therefore the downlink transmission is comprised of two parts, the wireless backhaul links from the MBS to the CUAV and radio access links from the CUAV to the SU-HP and SU-MP devices. received SNR (ϑ mb ) from the MBS is lower than the threshold 353 SNR (ϑ t ), which we set for the HP/MP/CUAV devices; then, 354 we define a particular state as an effectual state. Conversely, 355 if the received SNR from the MBS is higher than the thresh-356 old SNR, we characterise that state as an interference state, 357 which implies that [32] 358 Here, we identify ω as a symbol of an effectual and inter-360 ference state. Additionally, ω = 0 indicates that the MBS is 361 inactive in communication with the receiver, causing no inter-362 ference (ϑ mb < ϑ t ) to active HP and MP signals. Conversely, 363 we symbolise any interference (ϑ mb ≥ ϑ t ) to the HP and MP 364 signals from the MBS as ω = 1.

365
In the cognitive environment, PUs and SUs utilise the same 366 carrier. Therefore, mutual interference can occur between 367 PUs and SUs for the imperfect detection of idle channels. 368 As CUAV may create harmful interference to PUs, CUAV 369 transmission needs to be monitored and regulated. Hence, 370 CUAV transmission is below a particular interference thresh-371 old (I th ) to defend the licensed PUs [33]. Mathematically, the 372 average interference power constraint of the pth PU is defined 373 as where m bo , m ba , and m pa are the masses of the CUAV body, 384 battery, and payload (in kg), respectively. g is the gravitational 385 constant (in m/s 2 ). F dr is the total drag force and is defined 386 as F dr = 0.5ρ a v 2 a (C bo A bo + C ba A ba + C pa A pa ), where C bo , 387 C ba , and C pa are the drag coefficient of the CUAV body, 388 battery, and payload, respectively; A bo , A ba , and A pa are 389 the projected area of the CUAV body, battery, and payload 390 (in m 2 ), respectively; ρ a and v a are the air density (in kg/m 3 ) 391 and the velocity in air (in m/s), respectively. n r and D m are 392 the number of rotor and diameter (in m), respectively.  Therefore, the total CUAV power consumption is written  (c ab ) between nodes a and b as follows:

437
where K denotes the Rician K-factor.c LoS denotes the where ψ and denote the environmental constants for the 452 LoS and NLoS links, respectively. Moreover, θ represents 453 the angle of elevation θ = tan −1 between the a and b nodes. Hence, the CUAV-to-MBS/PU 455 pathloss (P o ) for both LoS and NLoS links in decibels (dB) 456 can be written as [35]: where λ = c f ca . ρ is the pathloss exponent, λ is the wave-460 length, f ca is the carrier frequency, and c is the speed of light. 461 Moreover, α LoS and α NLoS are the factors of attenuation due to 462 the LoS and NLoS channels, respectively. The mean pathloss 463 of outdoor environment considering the probabilities for both 464 LoS and NLoS is computed as follows: The pathloss for outdoor-to-indoor environment is computed 468 as follows [36]: where PL u (d ax ) denotes the pathloss between a and external 471 wall of building (x) over the distance d ax . d xb is the distance 472 between x and b. PL o u (d ax ) is calculated using Eq. (11). α wall 473 is the internal wall attenuation factor (in dB/m), n wall is the 474 total number of penetrated internal walls, and L wall is the 475 internal wall attenuation loss (in dB).  the nth SU-HP (∀n ∈ N s ) as follows: where t e = t t t s +t t ρ p (ω = 0) 1 − ρ fa . t t and t s denote the 511 transmission and sensing times, respectively. B represents the 512 system bandwidth. P u,hn + P u,mn ≤ P t,u , where P t,u repre- as follows: where T u,mn denotes the throughput between the CUAV and 542 SU-MP devices, and T m,mu denotes the throughput between 543 MBS and CUAV for the MP signal. The channel gain between 544 CUAV and SU-MP is denoted as c u,mn , and the channel gain 545 between CUAV and MBS for the MP signal is denoted as 546 c m,mu . represents the SIC coefficient, which varies between 547 0 and 1. = 0 implies perfect SIC, and = 1 implies 548 imperfect SIC. Alternatively, in this case, if the MBS is active in communi-551 cation alongside CUAV, we recognise the throughput as an 552 interference throughput. Hence, we can develop the interfer-553 ence throughput of the nth SU-HP (∀n ∈ N s ) as follows: . I umn = 559 P u,mn c u,hn 2 + P mn c m,hn 2 . P mn = P m,hn + P m,mn . P mn 560 represents the transmission power of the MBS, and c m,hn 561 represents the channel gain from the MBS to SU-HP. As for 562 the probability function, ρ p (ω = 1) symbolises the active 563 state of the MBS communication. Distinctively, in (3), ρ de 564 signifies the probability of detecting a target signal when 565 the resultant probability is greater than a particular threshold 566 value, implying that noise is present in the system. Hence, the 567 entire term ρ p (ω = 1) (1 − ρ de ) is an inadequate probability 568 of detection for any SUs that face interference from the MBS. 569 Similarly, the interference throughput of the nth SU-MP 570 (∀n ∈ N s ) can be derived as follows:    (23)  .

704
As mentioned in (20), we formulate the optimisation problem 705 for the interference energy efficiency when the MBS is active. 706 Here, we attempt to maximise the throughput of the HP 707 and MP signals while maintaining the collective transmission 708 power of the HP and MP signals less than P t,u and P t,m . 709 Hence, we can express the SOP at the nth SU (∀n ∈ N s ) 710 as 711 min P u,hn ,P u,mn , P m,hn ,P m,mn P u,hn + P u,mn + P m,hn + P m,mn (26a) 712 where T 1 u,hc , T 1 m,hc , T 1 u,mc , and T 1 m,mc signify the intended 722 interference throughputs between the nodes CUAV and 723 SU-HP, CUAV and MBS for the HP signal, CUAV and 724 SU-MP, and CUAV and MBS for the MP signal. The 725 objective function (26a) is a collective sum of the power 726 transmitted by both the CUAV and the MBS. In addition, 727 the transmission power of the CUAV in (26a) is satisfied 728 by the constraints (26b), (26d), (26f), (26g), and (26i). 729 Equivalently, MBS transmit power is satisfied with (26c), 730 (26e), (26h), and (26j), and similarly with the effectual 731 state, the entire problem (26) has been decomposed into 732 two sub-problems namely: (i) MBS power allocation and 733 (ii) CUAV power allocation. Finally, we propose a TPA algo-734 rithm owing to the solution of two sub-problems, as given in 735 Algorithm TPA 2. 736 According to (26) .

814
Proof: The proof of the Theorem 4 is similar to that 815 in Appendix C. 816

851
In Fig. 3, we obtain the average energy efficiency against 852 the CUAV height over the suburban, urban, and dense urban 853 areas in the effectual state. To start with, in the suburban 854 area, the energy efficiency value gradually increases until 855 81 m, where the energy efficiency value is at its peak with 856 12463 bits/J, and then decreases gradually with the increasing 857 height of the CUAV. As there are fewer obstacles and building 858 comparatively, the CUAV needs less power to communicate; 859 thus, the energy efficiency level decreases with increasing 860 height. Next, in urban areas, as height increases, the effi-861 ciency rate steadily increases because the density of the 862 building is greater than that of the suburban areas. At 150 m,     Similarly in Fig. 4, Fig. 8 shows the average energy effi-  over similar regions. All energy efficiency plots increase with 941 increasing cu d,n at initial and then decrease with increasing 942 cu d,n . In the suburban areas, at 0.5 cu d,n , the efficiency value 943 is 1455.33 bits/J, which is again significantly lower than 944 the value achieved in the effectual state. Next, in urban and 945 densely urban regions at 0.5 cu d,n , the energy efficiencies 946 are 1426.68 bits/J, and 1379.07 bits/J. Specifically, the urban 947 energy efficiency level is higher than the energy efficiency 948 of the densely urban region, but both have overall lower 949 efficiency than the suburban areas.

950
Likewise Fig. 5, Fig. 9 shows the energy efficiency at 951 different I th in the interference state over suburban, urban, 952 and dense urban regions. As the MBS will interfere with 953 the CUAV communication, all efficiency levels will be 954 lower. The energy efficiency values are 1493.81 bits/J, 955 1433.29 bits/J, and 1353.8 bits/J at I th = −50 dBm in 956 suburban, urban, and dense urban areas, respectively. 957 Fig. 10 shows the average interference latency at differ-958 ent SIC coefficients. The interference latency is defined as 959   We evaluate our proposed NOMA system with the OMA sys-1006 tem to demonstrate its efficacy. This design is identical to the 1007 proposed design except that it caches and dispatches using the 1008 OMA scheme rather than the NOMA scheme. The bandwidth 1009 is evenly shared and allotted to SU-HP and SU-MP devices 1010 under the OMA system. Moreover, each device is assigned 1011 an orthogonal bandwidth allocation, ensuring that SU-HP and 1012 SU-MP devices do not interfere with one another.

1013
In Fig. 11, we plot the average energy efficiency at various 1014 CUAV heights for NOMA and OMA methods over subur-1015 ban, urban, and dense urban regions in the effectual state. 1016 It is worth noting from the figure that the performance of 1017 NOMA scheme CUAV networks outperforms that of tradi-1018 tional OMA scheme CUAV networks. From the numerical 1019 point of view, we can see that over the height of 81 m the 1020 maximum energy efficiency achieved by the CUAV with 1021 NOMA method is 12463 bits/J for the suburban region. With 1022 the OMA method, the CUAV reaches up to 9129.26 bits/J 1023 at 80 height, which is significantly lower than the NOMA 1024 method. Moreover, for the urban region at 150 m height, 1025 the CUAV can perform up to 9383.78 bits/J in the NOMA 1026 method, which is significantly better than the OMA method 1027 at similar height. Likewise, for the dense urban region where 1028 the NOMA method still manages to outperform the OMA 1029 method. Overall, as CUAV height grows, the performance 1030 deterioration of the OMA method becomes more significant 1031 than that of NOMA method, confirming the benefits of the 1032 NOMA transmission system. 1033 VOLUME 10, 2022   respectively. We also see that the energy efficiency for both 1072 fixed power is high compared to fixed power CUAV. This is 1073 due to the fact that the optimum power of MBS is less than 1074 its fixed power.   Now, the second-order derivative of (A.1) with respect to 1134 P m,hn and P m,mn , the mathematical expression is written as 1135 Now, the second-order derivative of (A.4) with respect to 1148 P m,hn and P m,mn , the mathematical expression is written as 1149 where δ e , δ f , δ g , and δ h are positive variables for the con-1171 straints given in (23b)-(23d).

1172
The Karush-Kuhn-Tucker (KKT) conditions are necessary 1173 to obtain optimal solution, and the following KKT conditions   where δ i , δ j , δ k , δ l , and δ m are positive variables for the 1203 constraints given in (25b)-(25d).

1229
where P 12 u,hn = max P 1 u,hn , P 2 u,hn , P 34 u,hn = max P 3 u,hn , P 4 u,hn , P 12 u,mn = max P 1 u,mn , P 2 u,mn , and P 34 u,mn = max P 3 u,mn , P 4 u,mn . Engineering. His current research interests include 1415 interference analysis between wireless cellular 1416 and satellite service and wireless communica-1417 tion system structure (physical layer and channel 1418 modeling).