UAV-Aided Transceiver Design for Secure Downlink OW-DFTs-OFDM System: A Multi-User mmWave Application

Unmanned aerial vehicle (UAV)-based communication system design has already attracted substantial interest due to UAV flexibility in deployment, cost effectiveness, and in-built line-of-sight air-to-ground channels. However, there is a persistent issue of security threats associated with the broadcast nature of UAV: physical layer security (PLS) can be introduced to enhance the secrecy performance in that regard. In the system proposed here, the combined effect of 3D fractional-order Liu chaotic system and 3D fractional-order Li chaotic system is introduced to enhance the PLS of the UAV-to-ground communication network. In addition, to reduce multi-user interference (MUI), efficient orthogonal variable spreading factor (OVSF) codes are integrated with a zero forcing (ZF) scheme. Moreover, to improve bit error rate (BER), a combined MUI signal discrimination and ZF signal detection scheme is introduced, strengthened by various channel coding techniques with multi-user beamforming weighting. The numerical outcomes establish the efficacy of the proposed system in terms of PLS and improvement of data rate with signal-to-interference-plus-noise ratio. This approach achieved an out-of-band emission reduction of 119 dB. Furthermore, an improved BER of $1 \times 10^{-4}$ is achieved in 16-QAM for a signal-to-noise ratio of 6 dB.


I. INTRODUCTION
As a promising element for future generation mobile networks, millimeter-wave (mmWave) communication with astounding features is emerging. Leveraging very large bandwidths, mmWave communication can accommodate the ever-growing demand for wireless data [1]. However, mmWave signals are highly sensitive to a blockage, which means that communication by ground users (devices) can be intercepted by hills or tall buildings. This challenge can be addressed through the use of easily deployable and mobile unmanned aerial vehicles (UAVs) as an alternative to The associate editor coordinating the review of this manuscript and approving it for publication was Bilal Khawaja . terrestrial systems. In addition, UAVs are not vulnerable to line-of-sight (LOS) links and are more suitable for mmWave signal transmission. Therefore, deployment of UAVs as a relay network makes seamless connectivity and a wide range of coverage possible. In that light, the effect of UAV-based relay network deployment in mmWave communication is very encouraging [2], [3].
Massive antennas can be accommodated within a small UAV as a result of the short wavelength of mmWave, since small versions of large-scale antennas can be created for shorter wavelength. UAVs can easily shift to more suitable locations to circumvent obstacles and are capable of utilizing mmWave transmission. Furthermore, the major drawbacks of mmWave, such as path loss and blockage, can be reduced by VOLUME 10, 2022 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ careful beamforming design [4]. However, the high beamforming gain in massive antenna systems can improve the signal energy in limited-scattering regions [5].
Although the use of UAVs as a relay network promises significant gains, secured transfer of information is very challenging due to the open nature of air-to-ground wireless channels. Leakage of information in signal transmission over wireless LOS channels is always risky and demands substantial attention. Hence, the addressing of security issues is a most important area of concern in UAV communications [6]. However, may not be appropriate for incorporation in UAV systems, since computational complexity in traditional encryption schemes consumes a huge amount of energy, and UAV eavesdroppers can contaminate the transmission and initiate more aggressive attacks on terrestrial networks. Therefore, the development of advanced techniques such as enhancing physical layer security (PLS) to alleviate attacks by malicious UAVs demands serious attention [7]. PLS is actually a useful alternative for securing wireless communication networks. In estimations of the performance of PLS, secrecy capacity and secrecy outage probability are the two basic parameters which can be modeled by a wiretap channel. Applications of advanced 5G technologies such as non-orthogonal multiple access (NOMA), 3D beamforming, and mmWave are introduced in UAV communication system to further upsurge PLS [6]. To intercept the received signal qualities of a malicious eavesdropper, PLS utilizes the intrinsic characteristics of wireless communication channels, including noise, interference, and fading. Furthermore, PLS ensures keyless secure signal transmission by leveraging various signal designs and signal processing methods [8]. UAV communication networks could obtain both secure transmission and energy efficiency by utilizing proper PLS technologies with differential design of the legitimate channel and eavesdropping channel. At present, UAV-aided wireless communication with PLS has gained substantial attention from the research community. In accordance with the different functions of UAVs in secure communication, UAVs can be classified as UAV-involved secure communications and UAV-enabled secure cooperation [9], [10].
In recent times, two fractional order chaotic systems, threedimension (3D) Liu chaotic system and 3D Li chaotic system (LLCS), have been introduced as PLS measures [11]. It has been verified that those systems exhibit chaotic behavior over a wide range of the fractional-order. Compared to other existing PLS schemes, these newly introduced fractional-order chaotic systems achieve greater applicability of this particular realization for different applications such as dynamic switching, synchronization, and encryption. They also exhibit better cryptographic capability and are highly suitable for provision of an efficient PLS encryption technique.
Cases of application of UAVs, e.g., as flying base stations (BS) and aerial nodes, are considered with attention to issues of security of confidential information. Various case studies have investigated application of the PLS mechanism [8]. In [10], the authors develop a technique for securing transmission at the physical layer and optimizing degrees of freedom by leveraging some unique features of UAVs. In [12], the authors adopt transmit jamming strategy where some of the UAVs in an array direct jamming signals to mislead eavesdroppers. That work also introduces the use of 3D antenna gain in the air-to-ground links. [13] explores the effectiveness of 5G access in drones for such purposes as enhancement of drone corridor radio coverage; optimization of uplink communication; optimization of network lifetime; and precoding to increase PLS. In [14], the authors investigate the operation of multiple UAV-assisted relay and jammers to secure mmWave communication; and derive new closed-form expressions of secrecy outage probability utilizing models of 3D-antenna gain. By incorporating fullduplex UAVs in terrestrial cellular networks, it is possible to achieve multi-antenna transceiver for zero-padded orthogonal frequency division multiplexing (OFDM) systems [15].
Regarding the prevention or countering of both active and passive eavesdropping in UAV communication, the application of various techniques to PLS, e.g., resource allocation, trajectory design, and cooperative UAVs, to fight against both active and passive eavesdropping in UAV communication has been discussed in [6]. To enhance the secrecy rate of non-orthogonal multiple access transmission (NOMA), a technique for optimization of protected zone shape has been proposed [16] at UAV-BS. In [2], the authors explore secrecy communication in simultaneous wireless information and power transfer UAV-assisted relay system at mmWave. In [1], a power allocation technique is proposed, achieved by separating transmit power between precoder and finger-print embedded authentication tag. The authors of [17] propose precoding strategies that can incorporate the channel state information of an eavesdropper and eliminate the need to transmit artificial noise in underloaded situations. As a means to increase secrecy performance, [4] presents a cooperative jamming technique that utilizes destination and an external UAV for interrupting eavesdroppers.
In light of the above studies and to the best of our knowledge, to date no study has worked to develop an orthogonal variable spreading factor (OVSF)-encoded overlapwindowed discrete Fourier transform spreading orthogonal frequency division multiplexing (OW-DFTs-OFDM) system. This paper presents a multi-antenna configured UAV-assisted OVSF-encoded OW-DFTs-OFDM system that leverages integrated UAVs-terrestrial mobile network with various channel coding. The specific contributions of the paper can be summarized as follows-• To reduce multi-user interference (MUI) along with computational complexity of the precoding technique, we propose a normalized zero forcing (ZF) precoding technique incorporating efficient OVSF codes. The coverage of mmWave signals get reduced due to extensive path loss associated with mmWave frequencies while passing through obstacles. In addition, the enhanced sensitivity of signals in mmWave frequency may also produce large interference. • To achieve the target of minimizing multipath fading channel effects in UAV communication, we introduce null subcarriers in combination with cyclic prefixed (CP) and cyclic suffixed (CS) samples. Compared to the cellular and satellite systems, UAV communication has separate channel characteristics. In general, multipath fading effects are associated with the communication links used by UAVs, which can severely degrade the signal and thus impair communication.
• To minimize out-of-band (OOB) power and at the same time combat spectral leakaging and improve bit error rate (BER) performance, we propose the use of a raised cosine window and a channel coding scheme. One of the major limitations of the OFDM system is OOB interference, which is caused by the sidelobes present in the spectrum of an OFDM symbol. Existing techniques cannot completely reduce the OOB power leakage and thus introduce tremendous degradation of BER performance at the receiver.
• To incorporate the PLS encryption method in our system, we utilize fractional-order based LLCS technique.
• To prove the effectiveness, our proposed system verifies the significant enhancement of user data rate, BER performance, and OOB reduction at reasonably acceptable peak-to-average power ratio (PAPR) by means of a numerical study.
The rest of the paper is organized as follows. Section II presents a system model including network architecture and a signal model. The numerical results are presented in Section III, and concluding remarks and future directions for research are suggested in Section IV.
Notations: In the present study, the notations |·| 2 , ||·|| 2 , (·) T , (·) H are used to denote the square of the absolute value of complex matrix; the square of the Euclidean norm of complex matrix; the non-conjugate transpose of the complex matrix; and the conjugate transpose of the complex matrix, respectively. The notation (·) −1 denotes the inverse of the complex matrix variable.

II. SYSTEM MODEL A. NETWORK ARCHITECTURE
A UAV-based mmWave relaying network is considered in Fig. 1(a), where a UAV is deployed as a relay to aid transmission from the single next generation base station (gNB) to the U user equipment (UE) and E eavesdropper (ED). The gNB, UEs (legitimate users) and EDs are located on ground and the UAV is located in the sky. Assume that the UAV 1 works in full-duplex 2 using amplify-and-forward (AF) relaying. The total transmission is divided into two time slots. In the first time slot, gNB transmits signals to the UAV; in the second time slot, the UAV forwards the signals to the UEs and EDs.
The gNB would be composed of N T transmitting antennas and it would serve UEs (U = 1, 2, . . . . . . ., U ) via N T × N T a multi-antenna configured UAV. The set of EDs is denoted as E = 1, 2, . . . . . . ., E. The set of total equipments (i.e., UEs and EDs) is denoted as is the total number of equipments (devices). Each UE and ED would be equipped with N R receiving antennas. Fig. 1(a) shows the network architecture, composed of (b) Transmitter and (c) Receiver, described below.

1) TRANSMITTER
On the transmitter side, the extracted binary data from the audio signal of each individual UE are processed with a variety of channel coding schemes such as low density parity check (LDPC), repeat and accumulate (RA), and 1 2 -rated convolutional channel coding schemes [21], [22]. Forward error correction (FEC)-encoded binary data are subsequently encrypted using an LLCS (hybrid PLS) system addressed at [11]. The encrypted binary data are digitally modulated to generate complex symbols [23], which are then encoded with OVSF spreading codes in order to discriminate between the orthogonal spreading sequences of one user and another. Then 1024−point DFT is performed on OVSF spreaded signals and further processed with 4096−point inverse fast Fourier transform (IFFT) to generate DFTs-OFDM signals. Both CP and CS are added to those DFTs-OFDM signals. After that, overlap-windowed (OW) processing is performed using raised cosine window to smooth the transition of amplitude and phase between adjacent DFTs-OFDM blocks [24], [25]. Later the OW-DFTs-OFDM signals are power scaled, processed with a zero forcing precoding scheme [26], and transmitted from the antennas after fixing identical channel power.

2) RECEIVER
On the receiver side, the user's own transmitted signal for each ground UE is detected using respective beamforming weight. The detected signals are power scaled for the purpose of restoring them to their original signal power and then processed to remove cyclic prefixing and suffixing. Then those signals are operated with 4096−point FFT. The FFT operated signals are further processed with 1024−point inverse DFT (IDFT). Later the signals are OVSF despreaded, digitally demodulated, and channel decoded by exploiting hybrid PLS decryption technique to eventually recover audio signals.

B. SIGNAL MODEL
According to [11], the 3D fractional-order Liu chaotic mapping system can be written as where q 1 , q 2 and q 3 are the fractional orders. a, b, c, e, d, and m are the constant parameters. Using the initial values of (x, y, z) variables, the primary key K 0 of 2049K U × 1 matrix in size can be generated and written as where [·] indicates the rounding operation to the nearest integer. (50/3) is a multiplication factor value used to confine K 0 values within the desirable range, between 0 and 255. The primary key K 0 is disseminated into U encrypted keys (K 1 , K 2 , . . . . . . .., K U −1 , K U ) with each key containing 2049 elements and each element providing eight binary bits.
Using the MATLAB function repmat, the length of the binary data of each of the encrypted keys is made identical to the data length of the channel encoded binary data extracted from the audio signal. The encrypted channel encoded binary data vector b k of data lengthN for user k can be written as where ⊕ indicates the XOR operation and K k is the ultimately generated key in binary form using 3D fractional-order Liu chaotic mapping system for user k. The binary data vector b k is further encrypted using 3D fractional-order Li chaotic mapping system based on [11]. That system can be written as where a 1 , b 1 , c 1 , d 1 , and e 1 are constant parameters. With consideration of the initial values of (x, y, z) variables, the primary key K 0 of 2049K U ×1 matrix in size can be generated and written as where [·] indicates the rounding operation to the nearest integer; (20/3) is a multiplication factor value used to confine K 0 values in the desirable range; the primary key K 0 is disseminated to U users and in the case of such keys in binary form K 1 , K 2 , . . . . . . . . . , K U −1 , K U , the second encrypted binary data vector b k of data lengthN for user k can be expressed asb With added zeros, the doubly encrypted binary data vector b k of data lengthN is processed for audio to image conversion and the size of the image isÑ ×Ñ × 3 pixels. The extracted binary data vector c k of data lengthÑ (=Ñ ×Ñ × 8) is converted into a digitally modulated complex symbol vector c k with data lengthN . Regarding the applicability of OVSF spreading, the new form of the complex data vector is d k with data lengthN (= 8N ). The data vector d k is rearranged into a data matrixD of size M ×L, where M (= 1024) is the number of samples of 1024−point DFT blocks in each column and L is the number of columns. The column data vectord k,l is operated with M −point DFT and the new data vectord k,l of sample length M for user k can be expressed as The DFT operated signal vector presented in Eq. (7) is rearranged into a data matrixD of sizeÑ ×L, whereÑ is the number of data samples in each column data vectord k,l andL is the number of 4096−point IFFT blocks. In each data vectord k,l ,Ñ (= 0.5(N −Ñ )) zeros are added at each end and the modified form of the column data vector would bed k,l in a new data matrixD of size N ×L. The column data vector d k,l is operated with N −point IFFT and the new data vector ... d k,l of sample length N for the user k can be written as where ...  [24]. The overlap-windowed-DFTs-OFDM signal vector is represented byd k,l , such that E d k,l 2 = 1 and the power rescaled signal vector of unity power is represented byd k,l . On concatenating all the signal vectorsd k,l for user k, a data matrix D k is formed, which isÑ ×L in size.
The 3D distance d (0,k) between the UAV and the UE k with its 3D coordinates [x 0 , y 0 , z 0 ] and [x k , y k , z k ] can be written as: On the basis of elevation angle, carrier frequency and environmental factors, the LOS probability from UAV to UE communication can be written, for the case of user k, as [27] where and β 0 are the function of environment characteristic. The path loss between the UAV and the UE k can be expressed as [28]: where η 1 and η 2 are the excessive path loss coefficients for LOS and non-LOS links. f c is the carrier frequency, c is the speed of light and α is the path loss exponent. The average path loss between UAV and UE k can be written as [28] where P 0k NLOS = 1 − P 0k LOS . In uplink transmission from gNB with its 3D coordinates [x g , y g , z g ] to the UAV, the path loss L 0 for the multipleinput multiple-output (MIMO) flat fading channel H 0 can be expressed as [28] As each UE is equipped with N R receiving antennas, and beam forming weighting requires reshaping of data matrix D k : all of its elements are stacked into a single column vector and then converted into a matrixD k of size N R × N C , where N C = (Ñ ·L)/N R . The transmit precoders/beam forming weights (W k ) of k UE can be estimated from the probabilistic path loss model based 3D geometrical MIMO fading channels existing between the UAV and the UEs. For ZF precoding/ beamforming weighting, the ZF precoder can be written as [26] where W 1 , W 2 , W U −1 , and W U are the estimated transmit precoders, each of which is an N T × N R matrix. It is noteworthy that at this stage, application of the transmit precoder to input data presents a large amount of signal power; in this paper, we normalize each transmit precoder such that the reestimated transmit precoders for the UE k can be expressed as VOLUME 10, 2022 The transmit precoded signal without power scaling from gNB can be written as Due to precoding, signal power varies from gNB-UAV channel to UAV-UEs channel. The transmit precoded signal X 0 is to be multiplied by a signal power P x assigned in N T × N T diagonal dominated matrix such that, transmitted power of all transmitting channels of gNB will be identical and equal to P g . However, the signal from the gNB received at the UAV can be written as where n 0 ∼ CN 0 N T , σ 2 0 I N T denotes the additive white Gaussian noise (AWGN). On application of ZF signal detection technique addressed at [18], the decoded transmitted signal can be written as ...
The signal model presented in Eq. (18) is rescaled to make it compatible with desirable UAV transmission power P 0 of matrix size N T × N T , by multiplying with the signal power P y assigned in an N T × N T diagonal dominated matrix, and the power scaled transmitted signal from the UAV to the UE can be written as ... ... X 0 = P y P x X 0 = P x P y X 0 .
The signal received at the UE k can be expressed as where n k ∼ CN 0 N R , σ 2 k I N R denotes the AWGN and H keq = H k P x P y is the equivalent channel and it is an N R × N T matrix. The equivalent channel matrix H keq is found to be generally sparse in nature and due to the sparsity of its matrix, its pseudo inverse operation for achieving an exact solution to signal model presented in Eq. (21) cannot be obtained. Using regularized ZF equalization technique to invert the effect of the channel H keq [29], the noise contaminated detected signal for UE k can be written as where nk ∼ CN 0 N T , σ 2 k I N T denotes the AWGN and = 1 × 10 −22 is the regularization parameter. The signal model of Eq. (22) is further multiplied by user's own channel matrix H k and we get ...
where nk ∼ CN 0 N R , σ 2 k I N R denotes the AWGN. In Eq.
(23), H k W k 2 is the instantaneous total signal power for the UE k and H k W i 2 is the total instantaneous interference signal power for the UE k and N R σ 2 k is the AWGN. For the case of ED, the signal received with its considered channel H e can be written as where n e ∼ CN 0 N R , σ 2 e I N R denotes the AWGN and H eeq = H e P x P y is the equivalent channel. The noise contaminated detected signal for ED can be written as where n ... e ∼ CN 0 N R , σ 2 ... e I N R denotes the AWGN. In Eq. (26) the value of H e W k X k is very much negligible and the ED would only receive noisy signal.
The received signal-to-interference-plus-noise ratio (SINR) for the UE k after ZF precoding based decoding can be written as The data rate for UE k can be written as [30] R k = B log 2 (1 + SINR k ), where B is the channel bandwidth.

C. BER CALCULATION ALGORITHM
The main steps of calculating BER in our proposed system are presented in Algorithm 1 provided below: Algorithm 1 BER Calculation 1. Input: Initial parameters provided in Table 1 and BER = 0; 2. for all k ∈ U do 3. Generate random data b k for UE k; 4. Generate primary key K 0 using Eqs. (1) and (2); 5. First encryption: Encrypt binary data b k = b k ⊕ K k using Eq. (3); 6. Generate primary key K 0 using Eqs. (4) and (5) In this section, we discuss the computational complexity of the proposed system, as determined by the total number of complex multiplications employed in transmission and reception.
• The number of multiplications and additions for LLCS encryption, ϑ LLCS = 8 N e K U + 8 N e K U , where N e and K U are the number of elements contained in each key and total encrypted keys, respectively.
• For OVSF spreading, ϑ OVSF = 8N , whereN is the data length of the digitally modulated complex symbol. • For IFFT operation, ϑ IFFT =Ñ ×L, whereÑ is the data samples in each column data vector andL is the number of N -point IFFT block.

III. SIMULATION RESULTS AND DISCUSSION
In this section, the numerical results are presented for the PLS encrypted multiuser mmWave UAV-assisted OVSF-encoded OW-DFTs-OFDM system under the simulation parameters presented in Table 1. Here, a special case of four UEs and one ED is considered. The height of four UEs and single ED is considered to be 1.5 m and are located at a distance of 138 m, 129 m, 122 m, 131 m, and 135 m respectively away from the UAV, and the distance between the gNB (at a height of 50 m) and the UAV (at a height of 120 m) is 158 m. The height of ED and The location of the UAV and the considered number of UEs are assumed to be recognizable to the associates. Generally, it is required that the channel state information (CSI) be estimated by the UEs. However, in practice the transmitting UAV and UEs have different CSI. In such cases, CSI has been obtained from the estimated MIMO flat-fading channel in specific cases of downlink transmission. In this proposed system, we emphasis secure transmission with utilization of LLCS.
In Figure 2, the bifurcation diagrams of the fractional-order Liu system are presented with fractional orders, initial (x, y, z), and parameter (a, b, c, e, d, m) Figure 3.
In Figure 4, the bifurcation diagrams of the fractional-order Li system are presented with fractional orders, initial (x, y, z), and parameter (a 1 , b 1 , c 1 , d 1 , e 1    To evaluate the LLCS, the LLCS encrypted binary data vectorb k and its non-encrypted binary data vector b k for user k can be converted individually into eight bit decimal integer (symbolic) values and represented by a LLCS encrypted symbolic data vectorb PLS−k and a non-encrypted symbolic data vectorb NPLS−k . The values ofb PLS−k andb NPLS−k can be determined by using Eq. (6). The effectiveness of the LLCS encryption for ensuring secrecy against passive ED for user k can be written in terms of an estimation of the correlation betweenb PLS−k andb NPLS−k as [31]:    andb NPLS−k , respectively. cov(X 1 , X 2 ) and corr(X 1 , X 2 ) represent the covariance and correlation of two random variables X 1 and X 2 , respectively. SD(·) is the standard deviation. Using Eq. (29), it is observable that LLCS encryption scheme has a significant impact on generated symbolic integer values in Figure 6. The estimated correlation value with and without implementation of the LLCS encryption scheme for user k is found to have a value of 0.0486, which is clearly indicative of significant dissimilarity between the symbolic data. Figure 7 shows that the data (information) of the four users are totally hidden and encrypted by means of an audio-toimage transformation scheme utilizing simultaneous implementation of LLCS based PLS encryption techniques.
It can be seen in Figure 8 Figure 9 shows the PAPR reduction performance of the proposed system, presented by means of the complementary cumulative distribution function (CCDF). It is noteworthy that the PAPR of the transmitted signal with DFT spreading is about 2.5 dB lower than in the system without DFT spreading technique at a CCDF value of 1 × 10 −3 .
In Figure 10, graphical representations of transmitted audio signals of different users are presented without utilizing the LLCS scheme. Due to implementation of the LLCS scheme, waveforms of the individual users are totally changed, as depicted in Figure 11.
In Figure 12, the retrieved audio signals contain a significant amount of noise at E b /N 0 value of 0 dB; such signals   do not have any similarities to those presented in Figure 10. In Figure 13, the retrieved audio signals at E b /N 0 value of 10 dB have significant and satisfactory similarities with the VOLUME 10, 2022  transmitted signals presented in Figure 10, with the adoption of LDPC channel coding and 16-QAM digital modulation technique at an estimated BER of zero.
In Figures 14 through 16, the power spectra of transmitted and retrieved audio signals at E b /N 0 values of 0 dB and 10 dB are presented for different users. The estimated power at different frequency components is found to be identical in the case of transmitted audio signals and retrieved signals at E b /N 0 value of 10 dB (Figures 14 and 16). In case of noise contamination at E b /N 0 value of 0 dB (Figure 15), the scenario is completely changed and that effect dominates throughout the whole frequency band. In the proposed system, multiple antennas are preferably used at both the transmitter and the receiver, in order to exploit the spatially diverse communication link. In addition, the implementation of precoding based MIMO beamforming technique combined with OVSF spreading codes presents uncorrelated received signals and also improves SINR and data rate. The estimated received SINR values are presented with respect to the E b /N 0 values in Figure 17, which indicate that the quality of the received   signals is good and the estimated data rate is also reasonably acceptable, as can be seen in Figure 18.    Figure 19. At 0.1% BER, an E b /N 0 gain of 7 dB is achieved in 16-QAM with LDPC, in contrast to 16-PSK with RA channel coding.
In Figure 20, it can be seen that for user 2, the BER varies from 4.80% to 0.20% in the case of 16-PSK and 16-QAM utilizing LDPC channel coding technique for assumed typical E b /N 0 values of 2 dB. For the customarily acceptable E b /N 0 values of 5 dB, the estimated BER values are 0.00014 and 0.02478 in the case of 16-QAM with LDPC and 16-PSK with RA, respectively, which verifies a system performance improvement of 22.48 dB.
It is noteworthy that in Figure 21 the BER varies from 10.73% to 0.37% in the case of 16-PSK with 1 2 -rated convolutional channel coding and 16-QAM with LDPC channel  coding at an E b /N 0 value of 2 dB for user 3. For the considered E b /N 0 value of 5 dB, the estimated BER values are 0.0002 and 0.0460 respectively in the case of 16-QAM with LDPC and 16-PSK with RA channel coding, which correspond to a significant system performance improvement of 23.62 dB.
In the case of user 4, it can be seen in Figure 22 that the BER varies from 5.46% to 0.17% for 16-PSK with RA channel coding and 16-QAM with LDPC channel coding at an E b /N 0 value of 2 dB. For the customarily acceptable E b /N 0 value of 5 dB, the estimated BER values are 0.0001 and 0.0290 respectively for 16-QAM with LDPC and 16-PSK with RA channel coding, which verifies a system performance improvement of 24.63 dB.
In Figure 23, the BER performance of the proposed system is compared with that of various other schemes. For further details, a brief discussion is included on utilized schemes presented by the authors in [32]- [35]. The authors of [32] generalized the Nyquist criteria for ordinary linear modulation to the discrete-Fourier transform (DFT) spread OFDM  of constellation-rotated pulse amplitude modulation symbols. In [33], the authors propose an OFDM scheme which incorporates index modulation, spread spectrum, and C-transform techniques and strongly emphasizes the BER performance of the proposed OFDM, conventional OFDM systems, and several other similar systems. The authors of [34] propose a new precoded filter bank system based on the pruned DFT spread filter bank multicarrier modulation scheme utilizing only the notion of double rate transmission in combination with a data allocation strategy. In that scheme, both BER performance and PSD comparison between filters are presented. In [35], the authors combined a generalized frequency division multiplexing (GFDM) system with a power-domain NOMA system, and present numerical results indicating improved BER and higher achievable sum rate with lower PAPR and OOB power radiation.
The proposed scheme clearly outperforms all the other schemes under 16-QAM and LDPC channel coding technique, as shown in Figure 23. With the exception of user 3, the BER of all the users became 0 after E b /N 0 value of 5 dB.   The BER of users 2 and 4 are almost identical throughout the E b /N 0 range of 0 − 5 dB. As a consequence, it can be stated with confidence that in terms of BER performance, the proposed multi-user OW-DFTs-OFDM system under 16-QAM and LDPC channel coding technique is more robust than the other similar systems presented above. Even the BER of user 1 demonstrates relatively better performance for the case of the proposed system adopting 16-PSK digital modulation. Considering the typical assumed value of E b /N 0 = 6 dB, the BERs of the proposed scheme for arbitrarily chosen cases of user 3 and all other systems is shown in Table 2.

IV. CONCLUSION AND FUTURE WORK
We have investigated an UAV-assisted mmWave downlink OVSF-encoded OW-DFTs-OFDM system for secured data transmission. In this study, the implementation of OVSF spreading code and ZF precoding in UAV-aided mmWave OW-DFTs-OFDM system resulted in the reduction of MUI and the enhancement of the user data rate. Our designed system, with null subcarriers inserted at both ends of an OFDM symbol and with utilization of raised cosine filter at both ends of a OW-DFTs-OFDM symbol, has made achieved a satisfactory reduction of OOB spectrum power and elimination of the effect of fading channel at mmWave. We studied the system performance in terms of BER, using various channel coding schemes including LDPC, RA and 1 2 -rated convolutional with 16-QAM and 16-PSK digital modulations. Moreover, the findings show that LDPC channel coding with 16-QAM digital modulation can achieve low BER performance. As for further research on the proposed system, there are still many issues to be studied, notably: (i) the effect of multiple eavesdroppers; (ii) deployment of massive MIMO [36]; (iii) enhancement of the effectiveness of PLS technique [37]; and (iv) incorporation of deep neural networks and other innovative tools [38].