Report on the development of three traceable channel sounding testbeds for wide-bandwidth radio propagation channel measurements up to 750 GHz in complex real-world scenarios

21NRM03 MEWS Deliverable 3 (D3): Report on the development of three traceable channel sounding testbeds for wide-bandwidth radio propagation channel measurements up to 750 GHz in complex real-world scenarios Organisation name of the lead participant for the deliverable: Queen Mary University London Authors: Qamar Muhammad, Lawrence Carslake, Ivan Bonev, Ben Chen, Fengchun Zhang, Ke Guan, Tian Hong Loh, Akram Alomainy Due date of the deliverable: 30 June 2025 Actual submission date of the deliverable: 25 June 2025 Confidentiality Status: PU - Public, fully open (remember to deposit public deliverables in a trusted repository). Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or EURAMET. Neither the European Union nor the granting authority can be held responsible for them. Deliverable Cover Sheet The project has received funding from the European Partnership on Metrology, co-financed from the European Union’s Horizon Europe Research and Innovation Programme and by the Participating States. Metrology for Emerging Wireless Standards (MEWS) Deliverable D3 Abbreviations (Part 1) Acronym Description AB Alpha-Beta model ABG Alpha-Beta-Gamma model AiP Antenna-in-Package AoA Angles of Arrival AoD Angles of Departure ADC Analog-to-Digital Converter BS Base Station BW BandWidth B5G Beyond 5G CATR Compact Antenna Test Range CDF Cumulative Distribution Function CE Channel Emulator CFR Channel Frequency Response CIR Channel Impulse Response DAC Digital-to-Analog Converter DFT Discrete Fourier Transform DDSS Double-Directional Scanning Sounding scheme Dir-VAA Directional Antenna-Based VAA DP Digital Pre-distortion process DSS Directional Scanning Sounding E/O Electrical-to-Optical unit EVM Error Vector Magnitude FIBF Frequency-Invariant Beam-Forming FIR Finite Impulse Response FSPL Free-Space Path Loss HF High Frequency HPBW Half Power Beamwidth IF Intermediate Frequency IFBW Intermediate Frequency Bandwidth LNA Low Noise Amplifier LO Local Oscillator LoS Line of Sight LTE Long Term Evolution 1 Abbreviations (Part 2) Acronym Description MIMO Multiple-input-multiple-output mmWave Millimeter Wave frequency band MCBF Modified Classical Beamforming Algorithm MMSE Minimum Mean Square Error MPC Multi-Path Components NF Near-Field NLOS Non-Light of Sight NR New Radio NRMSE Normalized Root-Mean-Square Error Omni-VAA Omnidirectional Antenna-Based VAA O/E Optical-to-Electrical unit OTA Over the Air Testing PADP Power Angle Delay Profile PAS Power Angular Spectrum PDP Power Delay Profile PDSCH Physical Downlink Shared Channel PL Path Loss RT Ray Tracing RF Radio Frequency RF1 Radio Frequency 1 band covers (410-7125) MHz RF2 Radio Frequency 2 band covers (24.25-52.6) GHz RF3 Radio Frequency 3 band covers (7.125-24.25) GHz RoF Radio-over-Fiber Rx Receiver SIC Successive Interference Cancellation SISO Single-Input Single-Output SNR Signal to Noise Ratio SGH Standard Gain Horn Sub-THz Sub-Terahertz band Tx Transmitter UCA Uniform Circular Array UWB Ultra-WideBand 2 Abbreviations (Part 3) Acronym Description VAA Virtual Antenna Array VDT Virtual Drive Test VSA Vector Signal Analyzers VNA Vector Network Analyzer WP Work Package 3GPP Third Generation Partnership Project 4G 4th Generation Wireless Systems 5G 5th Generation Wireless Systems 6G 6th Generation Wireless Systems Contents Metrology for Emerging Wireless Standards (MEWS) 1 DeliverableD3.......................................................... 1 Abbreviations(Part1) ..................................................... 1 Executivesummary....................................................... 4 1.Introduction ......................................................... 5 2. Novel channel sounder design at sub-THz band. Channel measurement in various scenarios . . . . . . . . . 6 2.1 Novel channel sounder designs at sub-THz band . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2ChannelmeasurementswithOmni-VAA,Dir-VAAandDSSschemesat100GHzandat300GHzinvarious deploymentscenarios.............................................. 21 2.3 Double DSS measurements for channel characterization in a large hall scenario at 300 GHz . . . . . . 51 2.4 Channel measurements between 330 and 500 GHz for indoor and corridor scenerios . . . . . . . . . . 57 2.5 Channel measurements between 500 and 750 GHz for indoor, corridor and outdoor scenarios . . . . . 61 3. Development of band-stitching algorithm. mmWave 8 × 8 Antenna-in-Package (AiP) Array Platform - experimentalvalidationandapplications.......................................... 73 3.1 Development of band-stitching algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 3.2 Experimental Validation and Applications of mmWave 8 × 8 AiP Array Platform . . . . . . . . . . . . . . 91 4. Terahertz (THz) Channel Modeling in 6G Deployment Scenarios via Ray Tracing (RT) . . . . . . . . . . . . . . . 100 4.1Introduction......................................................100 4.2 Channel characteristics of THz communications systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.3 Measurement-calibrated ray tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4.3.1 Ray tracing validation for 330 to 500 GHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 4.5 Ray tracing validation for 500 to 750 GHz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.6Conclusion.......................................................110 5.Conclusion .......................................................... 113 6.Appendix ...........................................................116 3 Executive summary This reportis theseconddeliverableD3of theMetrology for EmergingWireless Standards(MEWS) project, covering topics included in Work Package 2 (WP2) and it merges the knowledge of the newly developed channel sounder designs at Sub-Terahertz (sub-THz), conducted sub-THz channel measurements (by using a virtual antenna array made up by Omnidirectional antennas (Omni-VAA), a virtual antenna array composed of Directional antennas (Dir-VAA) schemes and Directional Scanning Sounding (DSS) scheme) in various scenarios (entrance, corridor, big hall), double DSS (DDSS) sub-THz measurements in a hall scenario, antenna in package (AiP) based channel sounder for dynamic scenarios, development of a band-stitching algorithm and performed ray-tracing (RT) simulations at sub-THz band. The deliverable is organized as follows: Chapter I is the introduction part, while Chapter II contains three main topics from WP2, i.e. an improved channel sounder designs developed for 100 GHz and 300 GHz, channel measurements with Omni-VAA, Dir-VAA and DSS schemes at 100 GHz and at 300 GHz in different scenarios and DSSS channel measurementsat300GHz. ChapterIIIdescribesa developmentofa band-stitchingalgorithmandanAiPbasedchannel sounder for dynamic scenarios. RT simulation results at sub-THz are presented and discussed in Chapter IV. Conclusions are drawn in Section V. Finally, the most common equations related to estimation of channel parameters and characteristics are included in an Appendix. 4 1. Introduction This deliverable document aims to present the achievements and contributions obtained within the channel sounder design, channel measurements and RT modeling for sub-THz band. It encompasses detailed measurements and modeling efforts across various relevant propagation scenarios, offering insights into the channel parameters and characteristics. The main objectives of the deliverable can be summarized as follows: • Development of time-domain channel sounder design for 100 and 300 GHz. A Radio over Fiber (RoF) scheme is proposedtoextendthemeasurementdistance. Aphasecompensationschemeisfurtherimplementedtosupport the Virtual Antenna Array (VAA) channel measurements. • Extension of the phase compensated radio over fiber technique for channel sounding systems operating between 300-500 GHz and 500-750 GHz. Additionally their implementation on different Vector Network Analyzer (VNA), RoF, and extender head manufactuters has been considered. • Channel measurements (by using Omni-VAA, Dir-VAA and DSS schemes) have been conducted at 100 GHz and at 300 GHz in various scenario (entrance, corridor, big hall). The results have been presented and discussed. • Channel measurements have been conducted between 300-500 GHz and 500-750 GHz for various scenarios (indoor, corridor, outdoor) for distances up to 100 meters. The results have been presented and discussed. • Channel measurements by using DDSS scheme have been conducted at 300 GHz and the results have been presented and discussed. • Development of band-stitching algorithm • Development of an AiP based channel sounder for dynamic scenarios. • RT simulations results at sub-THz have been presented and discussed. 5 2. Novel channel sounder design at sub-THz band. Channel measurement in various scenarios 2.1 Novel channel sounder designs at sub-THz band 2.1.1 Channel sounder structure, link budget and system calibration The presented work in this subchapter is based on [1] and [2]. Two channel sounder systems based on VNA, operating respectively at frequencies between 75 and 110 GHz and 220-330 GHz, are proposed and discussed. For both designs, the channel sounder structure, link budget and system calibration are studied. In the case of the channel sounder at 220-330 GHz, RoF technique and a novel phase compensation scheme to stabilize the phase variations have been implemented, enabling long-range phase-coherent measurements. The schematic diagrams of the the VNA based channel sounders are shown in Figure 1. (a) (b) Figure 1: The schematic diagram the VNA based channel sounder. (a) 75-110 GHz; (b) 220-330 GHz As illustrated in Figure 1a, the transmitter (Tx) side of the 100 GHz channel sounder sends signals from 12.50 to 18.33 GHz from port 1, which, after going through a six-times multiplier, produce sub-THz signals from 75 to 110 GHz. A coupler splits the signals into two signals, one of which is used to provide 100 MHz reference signals and the other to transmit through a horn antenna. Keep in mind that LO signals from 12.48 to 18.32 GHz are sent by port 3. After passing via a signal splitter, the LO signals are multiplied six times to reach the frequency range of 74.9 to 109.9 6 GHz. The received signal is demodulated to 100 MHz, amplified by an low noise amplifier (LNA), and transmitted to port 2 at the receiver (Rx) end. The result of dividing the received signal at 100 MHz by the reference signal is used to estimate the S21 parameter. The sounding structure of a channel sounder operating in the 220-330 GHz frequency range is comparable to that of a sounder operating in the 75-110 GHz frequency range. The primary distinction is that the multiplier coefficients are 24 for the LO side and 18 for the Radio Frequency (RF) side. The link budget analysis at 92.5 GHz and 275 GHz is presented in Figure 2. (a) (b) Figure 2: The link budget of the VNA based channel sounder a) 100 GHz channel sounder at 92.5 GHz b) 300 GHz channel sounder at 275 GHz The VNA was connected to the extenders at the Tx and Rx sides by using 5 m RF cables. The antennas are not 7 accounted for in the provided link budget study. To make sure the Tx extender functions properly and to prevent damage, the transmitted power from port 1 for both channel sounders is 9 dBm. The 100 GHz channel sounder has a dynamic range of 95.6 dB, which is 8.9 dB greater than the 300 GHz channel sounder’s. Because of the 100 MHz received signal, the RF cable loss is incredibly low at the Rx side as opposed to the high loss at the Tx side. Due to cable limitations, these two channel sounders can provide channel measurements at distances less than 10 meters. By lengthening the RF cable at the Rx side, the measurement range can be expanded. Channelmeasurementsrequirecalibration. Inordertode-embed thesystemresponsein theback-to-backmeasurement, a normalization procedure is typically performed prior to the channel measurement. Verifying the efficiency of the absorbers positioned around the antennas is the aim of our measurement. To calibrate the system response, a normalization procedure and a back-to-back measurement (without the antennas) were carried out first. Then, using antennas, comparative channel measurements were made at a distance of 0.4 m with and without absorbers, as indicated in Figure 3. More information about the measurement setup is presented in Table I. (a) (b) Figure 3: The photos of the channel measurement setup: a) Without absorber; b) With absorber The use of the absorbers can significantly reduce the spurious peaks, as shown in Figure 4a of the comparison of the channel impulse response (CIR) at 100 GHz. However, the red area showed a few weak unexpected spurious peaks. Thelargebandwidthand the non-optimalabsorbersforthe 300 GHzband maybe the reasonwhythe calibration result with absorbers of the 300 GHz channel sounder is worse than that of the 100 GHz channel sounder. By selecting the proper bandwidth for the measurements and the calibration progress method in the post-processing, those small spurious peaks can be removed. 8 (a) (b) (c) (d) Figure 9: Amplitude and phase of the proposed channel sounder with cable bending a) Compensated amplitude b) Phase of the forward link c) Phase of the feedback link d) Compensated phase In practical directional scanning measurements, at the Rxside, the mechanical stresses on the optical fiber and Rx coaxial cables are unavoidable owing to the movement of the antenna positioning stages. In order to characterize the cable bending, a 28 cm cylinder is used to bend the cables, as it was illustrated in Figure 7b. The channel responses are measured for a total of 5 twists around the cylinder. As shown in Figure 9b, the phase of the forward link is observed to be unstable and varying over 400°. Compared with Figure 9b and Figure 9c, the phase of the forward link is approximately twelve times higher than the phase of the feedback link. After the phase compensation procedure, the phase due to mechanical stress is observed to be maintained within 15°at the frequency range of 220-288 GHz and within 30°at the frequency range of 288-330 GHz, respectively, as shown in Figure 9d. 2.1.2.5 Conclusion In this subsection, a novel sub-THz VNA-based long-range phase compensated channel sounder at 220-330 GHz is proposed. The dynamic range of this channel sounder could be improved to 106.7 dB for the back-to-back connection at 275 GHz to support the long-range channel measurements. By using the RoF technique and phase compensation scheme, the long-range phase-coherent channel measurement could be conducted. The robustness of this novel channel sounder is validated and specified for the frequency range from 220 to 330 GHz by back-to-back measurements. The phases after compensation at 220-288 GHz and 288-330 GHz are within 15°and 37°, respectively, compared to over 400 °phase change introduced by cable effect at 220 −330GHz. 15 Figure 10: Block diagram of the proposed RoF VNA-based sub-THz channel sounder(330-500 GHz) 2.1.3 Application of a phase compensatedRoFsystem to500 GHz channel sounding This subsection extends the previous work by applying the phase-compensated ROF technique tochannel sounding across the 330–500 GHz frequency range. It introduces a novel, dynamic-range-adjustable design for a long-range sub-THz channel sounder based on a VNA. The system integrates power-adjustable RoF technologies with a phase compensation mechanism, both of which are analyzed in detail. Furthermore, link budget performance has been evaluated for optical fiber lengths of up to 600 meters. 2.1.3.1 Channel sounder structure Figure 10 presents the block diagram of the proposed sub-THz channel sounder. The short-range VNA-based sounder, designed for measurements up to 0.23 meters and operates between 330 GHz and 500 GHz. This system features a Keysight Technologies four-port PNA-X VNA (operating natively up to 67 GHz) paired with Virginia Diodes Inc. R&S®ZC500 VNA Extenders, expanding the frequency range to 330-500 GHz, and includes Standard Gain Horn (SGH)-26-WR2.2 horn antennas. The VNAX units perform frequency conversion to sub-THz bands via integrated multipliers, mixers, and filters, but the measurement range is constrained by signal losses in the coaxial cables connecting the VNA to the extenders. A dynamic range adjustable optical-power phase-compensated electro-optical setup is employed at the receiver to address phase drift in long-range sub-THz frequency-domain RoF systems. This includes power-adjustable electro-optical(EO) and optoelectrical (OE) converters, with 600 m fiber optic cable acting as the transmission medium for the LO signal. The phase compensation mechanism uses two three-port optical circulators based on bidirectional signal transfer (forward and feedback) through the same optical fiber. This feedback link helps de-embed impairments caused by the cable’s thermal variations or mechanical stress. The output amplitude from the OE converter in the LO forward path requires amplification to match the input power requirements of the frequency extender’s LO input. To achieve this, a Minicircuits ZX60-06183LN+ amplifier, which operates between 16 6 GHz and 18 GHz with a typical gain of 20 dB, has been selected. Additionally, the IF measurement signal must be transmitted over a 600-meter distance. At its frequency of 10 MHz, this signal can be transmitted over coaxial cable with manageable signal loss. An RG58U cable, introducing about 25 dB of attenuation over 600 m, was chosen for this purpose. To further compensate for the signal loss and maintain the system’s dynamic range, a second amplifier, the Mini-Circuits ZFL-500LN+ (operating from 0.1 MHz to 500 MHz with a typical gain of 28 dB), is employed. 2.1.3.2 Channel Sounder Link Budget Assessment A link budget analysis is required for two key links in the system to validate the performance of the proposed long-range sub-THz channel-sounding setup. The first key link starts at VNA Port 1, runs through the transmitting extender head, passes into the receiving extender head, and loops back to the VNA through a measurement RF amplifier and 600 meters of coaxial cable. The transmit power at VNA Port 1 is set to 7.2 dBm, ensuring it falls within the VNAX drive range of 10 ±3 dBm. After factoring in the amplification and losses from the cable, the power received at the VNA input port is measured to be -18 dBm. A 10 MHz intermediate frequency (IF) was selected as a compromise between reducing signal loss over the 600 m coaxial link and maintaining the optimal performance of the PNA-X and extender heads, which are optimized for 279 MHz operation. However, using frequencies below 50 MHz can lead to a lower dynamic range in the PNA-X VNA due to increased trace noise, which can affect the precision of the measurements. The other link LO signal path to the remote Rx extender head over the RoF system. At the 425 GHz test frequency, the LO signal is 11.71 GHz. VNA Port 3 is set to 10 dBm, ensuring sufficient power after the frequency divider connects the LO of Tx frequency extender. The optical system, with a link loss of 12.8 dB, includes EO and OE converters. The photo-receiver gain has been adjusted to work with the amplifier to meet extender head specifications, resulting in a higher-than-minimum link loss. An RF amplifier with 20 dB gain and a 6 GHz–18 GHz bandwidth has been chosen. Despite the long-range link, the system meets the LO drive power requirements and provides an additional 3 dB gain for the measurement IF signal, demonstrating successful operation over the 600 m Link. 2.1.4 Application of a phase compensated RoF system to 750 GHz channel sounding This sub-section builds on the work in the previous section by applying the phase compensated RoF technique to channel sounding in the 500 - 750 GHz frequency range. This comprises of a novel dynamic range adjustable design of a long-range sub-THz VNA-based channel sounder system using both power adjustable RoF technologies [7] and a phase compensation mechanism which will be analysed. Additionally, the operating link budget with up to 600 m of optical fibre has been assessed. 2.1.4.1 Channel sounder structure Using the RoF technologies, the implemented design for a dynamic range adjustablelong-rangesub-THzfrequency-domainphased-compensatedVNA-basedchannelsoundersystemoperating between 500 GHz and 750 GHz is shown in figure 11. This consists of a power-adjustable electro-optical (EO) converter, opto-electrical (OE) converter and fibre optic cables as the transmission path for the LO signal. The targeted channel sounding rangeis up to 600 m, hence an optical fibre length of 600 m is employed. For the frequency upconverter unit at the Tx end, four connections are required, namely, the radio frequency (RF, 9.26 – 13.89 GHz), local oscillator (LO, 9.26 GHz – 13.89 GHz), IF measure (7.438 MHz) and IF reference (7.438 MHz). For the frequency downconverter unit at the Rx end, this is reduced to two connections, LO (9.26 GHz – 13.89 GHz) and IF measure (7.438 MHz). These connections must be routed back to the VNA to form the measurement system. For the Tx side, the fre17 Figure 11: Block diagram of the NPL long-range 750 GHz sub-THz VNA-based channel sounder quency extender head is located nearby to the VNA allowing for short coaxial RF cables to be used for the connections. As the Rx frequency extender head could be located up to 600 meters away from the VNA, a RoF solution is used as the loss in coaxial RF cables would be too great to allow for measurements to be performed. However due to the phase-stability issues with the RoF solution, phase compensation must be performed to correct for this effect. Phase compensation techniques have been included in the system using two, three-port optical circulators. From the EO unit, the 1550 nm optical signal is passed forward through an optical circulator and then into the extended 600 m fibre optic cable towards the remote Rx antenna. With the optical signal at the Rx end of the measurement, it is againpassedthroughanopticalcirculatorandthenequallydividedintotwosignalsbyanopticalsplitter. Thefirsttoan OE converter to provide the LO signal to the remote Rx frequency extender unit. The second signal from the splitter is fedtothe opticalcirculatorclosesttothe Rxend whereis itused tofeedbackthe signalthroughthe600 mof fibreoptic cable. The feedback signal is isolated from the forward signal by the initial optical circulator and through the feedback OE converter the phase of the feedback signal can be monitored on the VNA. Such a setup enables bidirectional signal transfer through the same 600 m optical fibre cable to allow monitoring at the VNA of the phase drift along the fibre’s length. Assuming the phase shift imparted by each propagation along the cable is equal regardless of the direction, the feedback signal will exhibit twice the phase change exhibited at the end of the forward path. This can then used to de-embed the phase change of the cable over time in post-processing, maintaining a sufficiently accurate calibration over the measurement duration. The output amplitude of the LO receiver end OE converter requires an amplifier to meet the input power level specifications of the frequency extender LO input. A MinicircuitsZX60-06183LN+ amplifier operating between 6 GHz – 18 18 Figure 12: Link budget and power measurements (b) with a simplified system diagram from the Long-Range sub-THz VNA-Based Channel Sounder for the (a) LO signal paths and (c) measurement signal path correlated vertically GHz with a typical gain of 20 dB has been selected. The measurement IF signal is also required to be routed over 600 m. The7.438MHz frequencyofthis signalallowstransmissionover an appropriatelengthof coaxialcablewith manageable attenuation. An RG58U cable, providing approximately 25 dB of attenuation over 600 m, has been selected. A second amplifier, a Mini-Circuits ZFL-500LN+ (0.1 MHz – 500 MHz with typical 28 dB gain) is connected to reduce the effect of the cable losses on the system measurement dynamic range. 2.1.4.2ChannelSounderLinkBudgetAssessment Tovalidatetheproposedlong-rangesub-THzchannel-sounding system, a link budget analysis is requiredfor two primary signal paths. The analysis is based on an operating frequency of 625 GHz (mid-band), with the frequency extender heads configured in a through connection - their waveguide test ports are directly connected. Thefirstsignalpathofinterestisthemeasurementsignalpath,illustratedinFigure12(c). Thispathisestablished from VNA Port 1, passing through the transmitting frequency extender head and into the receiving extender head. The signal then returns to the VNA via a measurement RF amplifier and a 600-meter coaxial cable. The transmit power from the VNA Port 1 is configured at 10 dBm to remain within the VNAX system’s drive specifications, of 7 dBm to 13 dBm. After amplification and accounting for losses incurred through the coaxial cable, the received power at the VNA is measured at -18 dBm. A 7.438 MHz IF signal is selected for the measurement, representing a compromise between attenuation over the 600-meter coaxial link and the operational characteristics of the PNA VNA and extender heads, which are optimized for operation at 279 MHz. Additionally, the selection of 7.438 MHz avoids the reduceddynamicrangeassociatedwith IF signalsbelow 50MHz of theKeysightPNA, asthese canbe directlyconnected 19 to the PNA IF inputs. The second signal path under consideration, shown in Figure 12 corresponds to the LO signal for the remote receiving frequency extender via the RoF system. At the 625 GHz operating frequency, the corresponding LO signal is 11.57 GHz. The power at VNA Ports 3 and 4 areset to 13 dBm to ensure adequate signal strength after the coaxial cables. The first port connects to the local transmitting extender via a one-meter coaxial cable. The second port supplies the remote receiving extender via the RoF system, which comprises an electro-optical (EO) converter, 600 meters of optical fibre, additional optical components, and a photo-receiver. The cumulative link loss through the optical path, including the effects of EO and optical-to-electrical (OE) conversion, amounts to 12.8 dB. To comply with the input power requirements of the extender head, the gain of the photo-receiver has been intentionally reduced to function effectively in combination with a following RF amplifier. This results in an optical link loss that is greater than the minimum achievable. The RF amplifier used in this configuration typically provides 20 dB of gain, with a 1 dB compression point of 16 dBm and a bandwidth spanning 6 GHz to 18 GHz. When compared to a conventional short-range configuration in which the LO signal is supplied directly from the VNA, the RoF setup meets the required LO drive power specifications for the frequency extender heads. Moreover, the IF signal experiences an additional 3 dB of gain across the measurement link, resulting in equivalent operating conditions between the non-RoF and RoF configurations. 20 2.2ChannelmeasurementswithOmni-VAA,Dir-VAAandDSSschemesat 100GHz andat300 GHzin various deployment scenarios 2.2.1 Short-range channel measurements at 100 and 300 GHz The presented work in [1] serves as a basis for this section. The sub-THz channel sounders presented in section 2.1.1 were used in the measurements. As illustrated in Figure 13, directional measurements were carried out in a rich scatterers laboratory scenario. Figure 13: Photo of the laboratory scenario Details about the measurement setup are presented in Table I. Two sub-THz frequency bands 90 to 110 GHz and 250 to 300 GHz were used for the tests. Two identical horn antennas, one for the Tx side and one for the Rx side, were used in a directional measurement scheme at the Rx side during the measurements at each frequency band. The Tx antenna was stationary, while the bore-sight of the Rx was configured to rotate in azimuth in steps of 1 degree, or 360 steps, within the range of [−180, 179]°. Figure 14 presents the Power Angle Delay Profile (PADP)s from the measured data. (a) (b) Figure 14: PADPs of the directional measurements at a) 90-110 GHz b) 220-330 GHz Figure 15 illustrates the identified multi-path component (MPC)s and the scenario. 21 (a) (b) Figure 15: Relation of the main MPCs to the scenario a) 90-110 GHz b) 220-330 GHz Only the LOS path and the main first-order reflections are shown in both figures. The distance 2.25 m of the LoS path corresponds to 7.5 ns in the measured PADP plots. As shown in both figures, the estimated distances and angles of the main MPCs matched good enough with the scenario. The power is focused in the line of sight (LoS) direction at 100 GHz, as reflections from the rich scatterers, such as the devices’ screens and walls, are identified. There are obviously fewer MPCs at 275 GHz than at 100 GHz, which could be because of the increased PL and lower antenna half-power beamwidth (HPBW). Nonetheless, at 275 GHz, the platform and wall exhibit comparable strong reflections to those at 100 GHz. 2.2.2 Virtual Antenna Array Based Channel Sounding at 300 GHz: Implementation and Field Measurements This section is based on the work in [8] and presents the first VAA implementation and field measurements at 300 GHz. The validation measurements are performed in a empty entrance room scenario at 295-305 GHz. The VAA configuration is a virtual Uniform Circular Array (UCA) with 1200 array elements, which can be considered as an extremely large antenna 22 array (ELAA). A modified classical beamforming (BF) algorithm is used to acquire extremely high angular resolution. The measured narrowband power-angle-spectra (PAS) at 300 GHz from the DSS and VAA schemes are compared to confirm the high angular resolution obtained by the VAA scheme. The wideband PADPs for the DSS and VAA schemes are presented and analyzed. Finally, the the extracted from the PADPs MPCs are then compared with the room geometry. It was found that the parameters of the MPCs, i.e., delay and angles of arrival (AoA), are consistent with the room geometry. Initially, the phase fluctuation caused by the movement of the turntable is investigated by conducting a simple field measurement in a chamber at 300 GHz. The phase is recorded for a period of time 8 s after one turntable movement. After normalization of the system and the channel responses, the phase fluctuation results were directly obtained (see Figure 16). Figure 16: The phase fluctuation results The results show that the phase fluctuates within a range with a span over 135°. However, the phase becomes stable after 3.2 s. Therefore, 3.5 s has been used as a waiting time (in the VAA measurements) after the turntable rotates. For the field measurements, the measured frequency band is 295-305 GHz with the number of frequency points of 5001. The Intermediate Frequency Bandwidth (IFBW) is set to 500 Hz to obtain good dynamic range. The maximum detectable delay is 500 ns and the maximum detectable path length is 150 m. The simplified diagram and photos of the measurement scenario are depicted in Figure 17. 23 (a) (b) Figure 17: The diagram and photo of the measurement scenario a) Diagram of the scenario and Tx and Rx deployment b) Photo of the scenario Themeasurementsareconductedatanentrancehallwiththesizeof8.20mx4.78mx8m. Thedistancebetween the Tx and the Rx is 6.5 m. The height of the Tx and Rx is 1.25 m. The position of the Tx is fixed, while the Rx is mounted on a turntable to form a virtual UCA with a radius r of 9.5 cm and a rotation step of 0.3°, which corresponds to 1200 antenna array elements. The Fraunhofer distance equation (see eq. (5.15 in Appendix)) defines the distance for the Far-Field (FF) case of this VAA dfar and it is equal to 72.2 m. The distance between the adjacent array elements is 0.497 mm, which is close enough to the half-wavelength of the highest measured frequency 305 GHz, i.e., 0.491 mm. The antenna used on the Tx side is a SGH antenna with a wide azimuth HPBW of 36°and antenna gain of 15 dBi. Two types of SGH antenna are used to form the virtual UCA, i.e., wide-HPBW case and narrow-HPBW case. In the wide-HPBW case, the same antenna used at the Tx is utilized as the array element, while in the narrow-HPBW case, the used antenna has gain of 25.5 dBi and azimuth HPBW of 8°. The measurement time for a virtual UCA is approximately 8 hours. The VAA setup and measurement configurations are listed in Table III. TABLE III VAA AND SCENARIO CONFIGURATIONS Parameter Value Parameter Value Frequency 295 −305GHz Frequency bandwidth 10 GHz Number of points 5001 IF bandwidth 500 Hz Transmitted power 10 dBm Tx & Rx height 1.25 m VAA Radius 9.5 cm Rotation step 0.3◦ Number of elements 1200 Tx-Rx distance 6.50 m Typ. Tx antenna gain 15 dBi Typ. Tx HPBW 36◦ Wide-HPBW case Typ. Rx antenna gain 15 dBi Typ. Rx HPBW 36◦ Narrow-HPBW case Typ. Rx antenna gain 25.5 dBi Typ. Rx HPBW 8◦ From the VAA measurements, CFR for each VAA element can be directly obtained, and the measured CIRs can be estimated by using the inverse discrete Fourier transform (IDFT). In the current work, the CFRs are processed further 24 Figure 24: Photo of the back-to-back measurements using the turntable to mimic the mechanical bending of the fiber The forward link of this channel sounder is similar to the one in the RoF structure in the conventional RoFenabled VNA-based channel sounders [15], [16], [17], [18], [19]. The VNA sweeps the frequency range 75–110 GHz with 1001 frequency points and 500 Hz IFBW. The Tx and Rx antennas are not available. Therefore, the Tx and Rx frequency extenders are connected directly in the back-to-back measurements. To mimic the bending of the fiber during the channel measurement using the VAA or DSS scheme, the fiber is mounted on the turntable. The turntable rotates in an azimuth range of [0°, 359.5°] with a step of 0.5°, which corresponds to 720 virtual array elements. The duration of the back-to back measurement was 2 hours, which was close to the time of the actual channel measurements. Before the measurements, for the forward and feedback links, a calibration procedure performing the normalization in the VNA setting the system response to 1 for all swept frequency points is performed. Therefore, the amplitude and the phase change in the channel sounding system can be measured directly. 31 (a) (b) Figure 25: Phase performance of the phase-compensated channel sounder in back-to-back connection a) Forward link b) Compensated phase The phase measurements of the channel sounder for a few spatial locations are presented in Figure 25a. The cable bending during the rotation causes variation in the phase from −121.6°to 60.9°. The feedback response has also been measured in the back-to-back measurements. Therefore, the compensated phase can be calculated in the postprocessing stepbyusing (6)in [10]. Afterthe compensationis applied, the phasevariation is maintainedwithin ±5°, as it is observed inFigure25b, which demonstratesclearly thatthe proposed phasecompensation schemeworks effectively. 32 (a) (b) Figure 26: Measurement scenario of the VAA at 100 GHz a) NF and UWB validation b) Long-range validation 2.2.3.5 Channel measurement validation Omni-VAA-based channel measurements are carried out to validate the proposed channel sounder and algorithm at sub-THz bands. The scenario for the NF measurements is illustrated in Figure 26a. It contains a large hall and a long corridor. The photo of the scenario is presented in Figure 27. Detailed information about the measurement configuration is presented in Table IV. 33 TABLE IV MEASUREMENT CONFIGURATION Meas. Near-field Long-range UWB type Omni-VAA Dir-VAA DSS Omni-VAA DSS Omni-VAA VNA settings Frequency range [GHz] 99-101 99-110 Bandwidth [GHz] 2 11 Frequency point 1001 7701 IFBW [Hz] 500 Transmitted power [dBm] 10 Antenna settings Tx antenna type Omnidirectional Tx antenna gain [dBi] 4.5 Rx antenna type Omnidirectional SGH Omnidirectional SGH Omnidirectional Rx antenna gain [dBi] 4.5 20 4.5 20 4.5 Rx antenna HPBW [°] 360 16 360 16 360 Virtual UCA settings Radius of the UCA [cm] 6.7 6.7 0 6.7 0 6.7 Rotation step [°] 1 Number of elements 360 Scenario settings Antenna height [m] 1.3 2.1 1.3 Tx-Rx distance [m] 7.3 84.5 3-58 Figure 27: Photo of the measurement scenario in the NF case (i.e., Tx-Rx 1.1) The size of the area is 62.2 m × 42 m. On one side of the corridor there is a window with a metallic structure, while on the other side there are seven stone pillars. In the NF case (i.e., Tx-Rx1.1), the Tx-Rx distance is chosen to be 7.3 m, which is much smaller than the FF distance (see eq. (5.15) in Appendix). The VNA sweeps in the frequency range from 99 to 101 GHz with 1001 frequency points, and the IFBW is set to 500 Hz. 34 The delay resolution is 0.5 ns (corresponding to the distance resolution of 0.15 m) and the maximum detectable distance is 150 m. An omnidirectional antenna with a gain of 4.5 dBi [20] is used as the Tx and Rx antenna to validate the Omni-VAA. The height of the Tx and Rx antennas is 1.25 m. The turntable is rotated in the horizontal plane within a range of [0°, 359°] with a step of 1°, which forms a virtual UCA with a radius of R = 6.7 cm. Additionally, the DSSbased measurements conducted in the same Tx and Rx locations are performed to serve as reference. In the DSS measurements, a SGH antenna with an antenna gain of 20 dBi and an HPBW of 16°[21] is located at the center of the turntable as the Rx antenna, while the position of the Tx antenna is kept unchanged. Prior to the measurements, the antennas are removed and the Tx and Rx waveguides are connected to perform the back-to-back calibration, i.e., normalization procedure to calibrate the system response. The measurement configurations for the NF measurements are listed in Table IV. (a) (b) (c) Figure 28: The measurement results in the NF case a) PADP with the DSS b) Recorded CIR with Omni-VAA over array elements c) PADP with Omni-VAA with the FIBF method Figure 28a shows the PADP for the DSS scheme in the NF case. Ten distinct paths can be identified. Although the antenna pattern is embedded in the DSS results, the paths can be identified clearly when looking at angle and delay domain. Figure 28b and Figure 28c show the recorded CIR with Omni-VAA and the PADP result with Omni-VAA using the FIBF algorithm, respectively. From the recorded CIR it was also found 10 paths. However, the use of the omnidirectional antenna at the Rx led to a lower dynamic range compared to the PADP with the DSS (as seen in Figure 28b). After the FIBF algorithm is applied, a high angular resolution is obtained, though strong sidelobes can also be observed, as one would expect. The LoS path with an AoA of −33°and a delay of 24 ns (which corresponds to the LoS propagation 35 distance of 7.2 m) is estimated in the Tx-Rx 1 case, which matches the physical distance between the Tx and Rx. The calculated power of the LoS path is −90.5 dB, which is close to the calculated FSPL (89.7 dB) (see eq. (5.13) in Appendix) for distance dLoS = 7.3 m. Furthermore, an improvement in the Signal to Noise Ratio (SNR) of the PADP after the FIBF compared to the raw CIR (presented in Figure 28b and in Figure 28c) of 20.5 dB can be observed. The theoretical value for the array gain is 10log10(K) = 25.6 dB. The measured array gain is a bit lower than the theoretical one owing to the different non-idealities in the measurement system. (a) (b) (c) (d) Figure 29: The estimated results in the NF case (i.e., Tx-Rx 1.1) a) Estimated path results b) The relation of the detected MPC trajectory to the room geometry c) Synthesized CIR using estimated results d) The PADP results with Dir-VAA after the FIBF The FIBF with the SIC procedureis applied to the measured data. A signal threshold is set to be 10 dB higher than the noise floor. Note that for the long-range and UWB channel validation, the signal thresholds are the same as that in this NF case. Figure 29a shows the estimation results using the FIBF algorithm with the SIC. The detected ten dominant paths are mainly concentrated in the AoA range of [−90°, 0°]. Figure 29b illustrates the paths trajectory. Path 2, 3, 6, and 9 are found to be the reflections from the pillars. Paths 5 and 10 are due to the reflections from the elevator. Path 7 is observed to be the reflection from the back wall of the Rx, while paths 4 and 8 are respectively the first-order and second-order reflections from the metallic trash box. Both the propagation distance and angle of the estimated paths match the geometry of the hall. The estimation results in Figure 29a are compared with the DSS results in Figure 28a. A good agreement between the estimation results and the DSS results is found. Figure 29c presents the synthesized CIR on the virtual UCA elements using the estimated paths. It is compared with the recorded CIR on the virtual UCA 36 elements in Figure 28b and it can be concluded that most specular paths are detected accurately. Dir-VAA-based channel measurements are also carried out in the NF case (i.e., Tx-Rx1.1) using the same Tx-Rx deployment and measurement configurations. Comparing the DSS with Dir-VAA schemes, the measurement system and antenna configurations are the same with only one difference that the employed directive antenna is centered on the turntable for the DSS case (according to the definition of the DSS scheme), while for the Dir-VAA case an off-set with a radius of 6.7 cm (with respect to the center of the turntable) is used. Comparing the Dir-VAA and the Omni-VAA, the only difference is the used antenna, i.e., the omnidirectional antenna is used as the array element in the Omni-VAA, while the directive antenna is utilized as the array element in Dir-VAA. Figure 29d presents the PADP with the Dir-VAA after the FIBF. The LoS power after the FIBF is found to be −90.9 dB, which is close enough to the calculated FSPL value. Compared with the DSS PADP presented in Figure 28a, the MPCs are distributed in the same locations. Moreover, the angular resolution is significantly improved by the FIBF algorithm (the main beamwidths of the PADP with Dir-VAA after the FIBF and with the DSS are respectively 3°and 16°). The improvement in the SNR of the Dir-VAA is 15.0 dB, which is less than that of the Omni-VAA. The reason is that less effective array elements in Dir-VAA are used compared to those in Omni-VAA. The Dir-VAA has a high element gain (20 dBi at 100 GHz) compared to 4.5 dBi at 100 GHz for Omni-VAA. The lattercanbeusedtoincreasethedynamicrangeofthechannelsounder tosupportlong-rangechannelmeasurements. Figure 30: Photo of the measurement scenario in the long-range case) Channel measurements are also performed to confirm that the proposed sounder can be used for long-range channel measurement. Although in the state-of-the-art long-range measurements directive antennas to improve the system dynamic range are generally used [15], [22], in the current measurements, omnidirectional antennas are intentionally utilized to demonstrate the capability of the developed channel sounder. The long-range validation measurements are performed on the ground floor at the Innovation Building at Aalborg University, as shown in Figure 31c. The measurement scenario is a typical hall scenario with a size of 94 m × 30 m. The frequency range is set to be 99–110 GHz with a bandwidth of 11 GHz and the number of frequency points of 7701 resulting in a delay (distance) resolution of 0.09 ns (0.03 m) and a maximum measurable distance of 210 m, respectively. The IFBW is 500 Hz and transmitted power is 10 dBm. Two identical omnidirectional antennas [20] are used for the Tx and Rx antennas. The distance between Tx and Rx is 84.5 m. Because of the available stairs, there is a height difference of 0.73 m between the Tx and the Rx. 37 Therefore, to align the antennas, the heights of the Tx and Rx antennas are respectively 2.05 m and 1.32 m. As it was for the NF case, DSS measurements are also carried out for a reference. Figure 30 and Table IV present respectively the photo of the long-range measurement scenario and the list with measurement configurations. (a) (b) (c) (d) (e) Figure 31: The measurement results in the long-range case a) PADP with the DSS b) Recorded CIR with Omni-VAA c) PADP with Omni-VAA after the FIBF d) Estimated path results e) Synthesized CIR using estimated results Figure 31a shows the PADP result with the DSS in the long-range case. Six paths can be identified from the PADP plot. However, it is difficult to detect the paths from the recorded CIRs over array elements (presented in Figure 31b) owing to the low dynamic range of the Omni-VAA scheme in the long-range case. A high angular resolution can be observed after the FIBF algorithm is applied, as visible in the PADP after the FIBF in Figure 31c. The LoS path has an 38 AoA of 2°and a delay of 281.7 ns (which corresponds to the LoS propagation distance of 84.5 m), which is equal to the deployment distance between the Tx and Rx. The LoS power is −112.9 dB, while the theoretically calculated FSPL value is 111.4 dB. The small difference may be dictated by error in the alignment between the Tx and Rx antennas for such a long distance. Moreover, a strong sidelobe can be observed in Figure 31c, as it could be expected for the FIBF algorithm. A SNR improvement of the PADP after the FIBF in Figure 31c compared to the raw CIRs in Figure 31b of 16.0 dB is found. Figure 31d and Figure 31e show respectively the estimated paths and the synthesized CIR on the virtual UCA elements after using the FIBF with the SIC. Owing to the high PL and the low antenna gain of the used omnidirectional antennas, only 6 paths are detected in the measurement results using the FIBF with the SIC. The trajectory of the dominant paths isshowninFigure31c. Path2isfoundtobethereflectionfromthebackwalloftheTx,whilepaths3and4arereflections from the metallic structure of the back wall at the Rx side. Path 5 is the reflection from the outdoor lamppost, while path 6 is found to be the reflection from the wooden furniture in the middle rest area. Moreover, when comparing the reconstructed CIR in Figure 31e with the recorded CIRs over virtual array elements, a good agreement can be observed, which is a proof that the dominant paths in these two measurements fit well, validating the measurement results using the VAA concept. In the subsection, only few explanations will be given about the performed UWB channel measurements and characterization. However, the reader can refer to Fig. 15 - Fig. 17 in [10]. The measurement scenario and the Tx and Rx deployments are the same as in the NF case and is shown in Figure 31b. The Tx is situated close to the wall to simulate a base station (BS). Nine Rx locations (i.e., Rx 2.1-Rx 2.9) with the link distance ranging from 4–58 m are used. The antenna heights of the Tx and Rx are 1.25 m. The measurement settings are the same as those in the long-range measurements. Therefore, 360 × 9 = 3240 CFRs were obtained in total (number of UCA elements × number of measurement locations). Each measurement took approximately 4 hours and the entire measurement took around 40 hours. The detailed measurement specifications are listed in Table IV. The FIBF with the SIC is used to extract the parameters (amplitude, delay, and AoA) of the MPCs for each measurement location, and then the channel characteristics are estimated using those estimated channel parameters. The PL is calculated by using (5.14) in Appendix. Fig. 15 in [10] presents the measured PL compared with the fitted model. A good agreement between the fitted PL model and the FSPL model is observed. The delay spread is shown in Fig. 16(a) in [10]. The delay spreads is in the range of [5.67, 47.61] ns. Fig. 16(b) in [10] presents the AoA spread result. The angular spreads vary from 15.63°to 72.11°in this scenario. The PASs with the estimated results compared with the mean AoA ˆ ϕℓin the Tx-Rx2.2 and Tx-Rx2.9 are shown in Fig. 17 in [10]. It was found that the MPCs in Tx-Rx2.2 are concentrated on the mean angle leading to a low AoA spread. In the Tx-Rx2.9 case, the AoAs of the MPCs are in the range [−123, 180]°defining a high angular spread. The k-factor is found to decrease with the increase of the distance. These findings are expected and can be explained by the fact that with increasing of the distance, the LoS power decreases and the power levels of the MPCs becomes comparable to the LoS power(it can be also observed from the comparison of the power of the MPCs between the Tx-Rx2.2 and Tx-Rx2.9 in Fig. 17 in [10]). 2.2.3.6Conclusion This section presented the design, implementation, and validation of a phase-compensated channel sounder, which is able to support the VAA scheme and long-range measurements at W bands (75–110 GHz). The phase instability is a critical problem at sub-THz bands making the existing channel sounder solutions inappropriate for applying advanced array signal processing algorithms to extract the channel path parameters. It was proved by simulation that the VAA scheme cannot be directly used in the conventional sub-THz channel sounder, because the 39 unstable phase performance may cause BF loss up to 3.1 dB. Therefore, a phase compensation scheme is proposed to address effectively this issue and a phase variation up to ±5°is achieved in the back-to-back connection in the W-band. Virtual UCA-based channel measurements in the NF (with an LoS distance of 7.3 m) and long-range (with an LoS distance of 84.5 m) cases were carried out at 100 GHz to validate the proposed channel sounder. Moreover, a high-resolution channel parameter estimator (i.e. FIBF with the SIC) was applied to extract the MPC parameters. The purpose of the NF measurements was to validate the generalized high-resolution channel parameter estimator, while in the long-range case, it was aimed to illustrate the capability of the proposed channel sounder with a large dynamic range via intentionally employed omnidirectional antennas on both the Tx and Rx sides. In the long-range cases, six specular paths were found including only the LoS path and only the first-order reflections, while in the NF case, ten paths were identified including the LoS path, the first-order and second-order reflections. Both measurement results demonstrated the sparse property of the W-band channel. The measurements using theconventionalDSSschemewereusedasareferencetovalidatetheestimatedpathparameters. ComparedtotheDSS results, consistent path parameters in terms of path power, angle, and delay were obtained by using the FIBF with the SIC for both cases. These extracted path parameters were utilized to reconstruct the CIRs over the UCA elements. The comparison between the reconstructed CIR and the recorded CIR demonstrates a good level of agreement. Moreover, the FIBFalgorithm was also extendedtoa generic algorithm, whichis abletofunction forNF/FF, narrowband/wideband propagation scenarios using Omni-VAA and Dir-VAA schemes. Omni-VAA-based UWB measurements in the frequency range from 99 to 110 GHz were carried out in a hall scenario at 9 Rx locations covering the link distances from3 to 58 m. By applying the FIBF with the SIC algorithm, the power, delay, and AoA of the MPCs were extracted from the obtained measurement results. Additionally, the basic channel characteristics, such as PL, delay spread, angular spread, and k-factor, were presented and analyzed. 2.2.4Largevirtualantennaarray-basedempiricalchannelcharacterizationforsub-THzindoorhallscenarios Despite significant efforts have been dedicated to sub-THz channel measurements and modeling, important aspects remain insufficiently investigated in the literature. One crucial aspect is that the existing channel measurements and models have not covered the entire sub-THz bands for different scenarios. Another important point is that the recent measurements are mainly using the DSS scheme, which provide low spatial resolution and high sidelobes. Therefore, the multipath parameters estimated from the PADPs derived using the DSS method will be impacted, leading to unprecise modeling and characterization results. Third crucial aspect is that many previous studies have not explored the channel spatial consistency regarding MPC trajectories at sub-THz frequencies. The latter is of an significant importance for evaluation the performance of 6th Generation (6G) systems, particularly in the aspect of the beam management. This section, based on the work [23], presents the channel measurements and characterization of the sub-THz channel measurement campaign conducted in an indoor hall scenario at sub-THz bands. The channel measurement campaign was carried out with 55 Tx-Rx deployment locations. The Tx-Rx distance was in range [3, 58] m and the measured bandwidth (99-101) GHz. The used long-range channel sounder includes a phase-compensated scheme. The MPCsparametersare extractedby usinga channel parameterestimation algorithm. Furthermore, theMPCstrajectories intwo of the routesaretracedbythe use of a MPCstrajectorytracking algorithm. Theproposed100 GHz channel model for the indoor scenario was also compared with the existing channel models. A RoF-enabled phase-compensated VNA-based channel sounder (see Figure 22 in section 2.3.3.1), which can sup40 (a) (b) Figure 38: Results for a) delay spread versus Tx-Rx distance b) angular spread versus Tx-Rx distance The delay spread changes in the delay range of [5.9, 67.5] ns. It is worth it to note that in Figure 38a, the delay spread data is divided into four groups, i.e., Route 1, Route 2, Rx 26-35 data (in the hall) and Rx 48-55 data (in the corridor). Obviously, the delay spreads increases in both Route 1 (represented by red color) and Rx 26-35 (in green) as the Tx-Rx distance increases. The latter can be explained by the decrease in the LoS power with increasing the Tx-Rx distance, while the variation in the Non-Light of Sight (NLoS) power remain comparatively small. For Route 2 (in black) and Rx 48-55 (shown in blue), the delay spreads decrease with an increase of the Tx-Rx distance. This reduction may be related to the diminishing relative difference between the LoS and the NLoS path lengths as the Rx moves closer to the scatterers, such as walls and windows, and further away from the Tx. Figure 38b presents the relation between the azimuth AoA spread and the Tx-Rx distance. The azimuth AoA spread is in the range [9.2, 52.58]°. Similar to the delay spread, the AoAspreadsis alsodivided into four groups. It can beseen thatthe azimuth AoAspreadsin allgroups show an increasing trend as the Tx-Rx distance increases. The same reason as for the delay spread, in our understanding applies for the azimuth AoA spreads, i. e. it is also due to the decrease of the LoS power and the significance of the power of the NLoS paths with the increase of the distance. The reader is advised to refer to Fig.7 (b), Fig.7 (c), Fig.8 (b) and Fig.8 (c) in [23], where the cumulative distribution function (CDF) of the calculated delay spread (in a logarithmic scale), an exemplary Power Delay Profile (PDP) comparison for two Rx locations, the CDF of the azimuth AoA spread in logarithm scale and exemplary PAS for low and high AoA spread cases are respectively presented. The CDF of the calculated delay spread matches to a great extend with a log-normal distribution. Moreover, several high delay spread cases over 60 ns (e.g. Rx 49) are observed. For the exemplary PDP comparison, it was found that in the Rx 13 case (with low spread), MPCs are mainly distributed in the delay range [14, 105] ns, while in the case of Rx 49 (high spread), a prominent MPC with a long delay of 340 was noticeable. This MPC is the reflection from the rear wall of the Rx 49. Furthermore, the delay spread does not provide much about the structure of the PDP, because a single MPC with long delay can drastically impact the delay spread. The CDF of the empirical AoA spread matches well with the log-normal distribution. Furthermore, there are several Tx-Rx cases with high azimuth AoA spread, such as Rx 45. For the Rx 13 case, the MPCs are mainly distributed in the AoA range [−107, −39]°. The average azimuth AoA (in the case of Rx 13) is −55.16°, closely matching with the AoA of the LoS path, which is −39°, resulting in a relatively low azimuth AoA spread. In the case of Rx 45, the estimated MPCs are located in two distinct AoA intervals, spanning from −126°to −73°and from 58°to 100°. The calculated mean azimuth AoA is −18.28°and is significantly different from both MPC azimuth AoA intervals. 47 The K-factor is a parameter evaluating channel characteristics on the small-scale [25], [26]. It is presented in Figure 39. (a) (b) (c) Figure 39: K-factor results calculated using the estimation results a) K-factor versus Tx-Rx distance b) CDF of the Kfactor c) An exemplary PDP comparison of the low K-factor case (Rx 35, k = 0.6 dB) and the high K-factor case (Rx 13, k = 11.4 dB) The K-factor changes in the interval [0.62, 12.75] dB, as it is found that it decreases with the increase of the distance. The CDF results of the K-factor is in good agreement with the normal distribution function. In the Rx 13 case, due to the short Tx-Rx distance and spacious surrounding scenario, the LoS power is much higher than other MPCs. In the Rx 35 case, it is found that the power of the NLoS path is comparable to the power of the LoS, because the Rx 35 is in vicinity to the wall and to several metallic scatterers. TABLE VI COMPARISON OF OUR PROPOSED CHANNEL MODEL WITH THE EXISTING ONES. Reference Scenario Path loss [dB] Delay spread log10(ns) Angular spread log10(◦) K-factor [dB] Our model @ 100 GHz Indoor hall α= 1.77 χAB σ= 74.40 N(1.26,0.07) N(1.36,0.05) N(6.36,10.12) [27] @ 100 GHz Indoor hotspot α= 1.73 χAB σ= 72.40 N(1.28,0.03) N(1.40,0.14) N(7,16) [28] @ 28 GHz Indoor hall α= 1.81 χAB σ= 62.36 N(1.53,0.19) N(1.96,0.10) — 48 Table VIpresentsa comparison ofour model andtwo other models. In [23], the readercan findmore information about comparison between the proposed channel and other channel models. In the Third Generation Partnership Project (3GPP) 38.901 model [27], the channel models in indoor hotspot office (1 ≤ d ≤ 150 m) cover the frequency bands upto100 GHz. It isquitesimilar toourmeasurementcasepresentedhere(hallscenariowith theTx-Rxdistanceranging from [3 − 58] m at 100 GHz). Compared the proposed channel model with the 3GPP model, it can be observed that the fitted model of the delay spread, angular spread, and K-factor match closely to those in the 3GPP model. In order to demonstrate the channel sparsity at 100 GHz bands, the number of the extracted MPCs for Route 1 and 2 (by using a dynamic range of 30 dB) is presented in Figure 40 and analyzed. (a) (b) Figure 40: Number of MPCs detected using 30 dB dynamic range a) Route 1 b) Route 2 In Route 1, the number of detected MPCs is between 8 and 14. In Route 2, we can identify more MPCs (between 16 and 24). The latter is due to the fact that Route 1 is deployed in a spacious scenario with only several pillars in vicinity, while Route 2 is deployed in a scenario close to windows, walls and several metallic scatterers, providing more reflections and MPCs. The channel at 100 GHz shows sparsity and specularity. The number of trajectories per Rx location for both routes is illustrated in Figure 41. (a) (b) Figure 41: Number of trajectories in each Rx location a) Route 1 b) Route 2 For Route 1, the number of trajectories per Rx location varies from 4 to 10, while in the case of Route 2, it varies from 6 to 9. 49 In this section, sub-THz channel measurements were conducted in an indoor hall scenario at the frequency range (99-101) GHz covering the Tx-Rx distance in the range 3-58 m. The Tx was fixed, while the Rx was deployed in 55 positions, including 2 predefined routes. For each Tx-Rx measurement, the virtual UCA scheme was used with a raduis of 6.7 cm and rotation step of 1°, corresponding to 360 VAA elements in each Tx-Rx measurement. By applying the FIBF algorithm with the SIC procedure, it was possible to eccurately estimate and extract the channel parameters from the measurement data. The number of the detected MPCs (using a dynamic range of 30 dB) was within the range from 8 to 24, showing the sparsity of the 100 GHz channel. The PL, shown in dB over distance, exhibits a strong fit to the linear model. The delay and angular spreads are observed to match the log-normal distributions, while the K-factor fits well to the normal distribution. Furthermore, the MCD-based trajectory tracking algorithm was applied to trace the observed MPC paths within the measured channel based on the estimated MPC results. The geometry-based analysis of the dominant NLoS trajectories was made. The delay and azimuth AoA of these dominant trajectories closely match with the spatial characteristics of the hall. In addition, the proposed 100 GHz channel model was compared with the existing models. The comparison between our model and 3GPP 38.901 model, covering up to 100 GHz, demonstrates the correctness of the proposed model in the current work. Future work would include outdoor 100 GHz channel measurement and modeling and MCD-based clustering studies at different frequency bands. 50 2.3 Double DSS measurements for channel characterization in a large hall scenario at 300 GHz Future wireless communication systems, i.e., beyond fifth-generation (B5G), have been attracting increasing interest worldwide as the need for high data-rate in various applications continue to grow [28], [29]. Sub-terahertz (sub-THz) technologies, defined at the frequency bands in 100-300 GHz, are envisioned to be one of the core radio technologies for B5G communication systems, due to their large unused and unused frequency resource [30], [31]. However, as it was discussed in [32] the atmospheric attenuation at sub-THz bands is much higher compared to that at lower frequency bands. Moreover, the wavelength of the sub-THz frequencies is close to the size of the dust and the rain drops, which makes the channel at sub-THz bands to behave much differently than that of the conventional low frequency spectrum [33]. Although, numerous sub-THz measurement campaigns have been proposed in the literature, they are either focused on characterization of the sub-THz channel in the short-range indoor scenarios, e.g., office and meeting rooms [34], [35], [36], [37], [38], [39], [40], or only a few of long-range measurements have been reported in the long-range scenarios and mainly at 140 GHz [22], [41]. Therefore, the lack of experimental studies on channels in long-range scenarios at 300 GHz has motivated the work in [42], which serve as the basis for the content in this subchapter. The high fidelity of the long-range channel sounder has been validated in [34]. However, the channel characteristics and channel models at 300 GHz are not further analyzed in [34], because the measurement was conducted only for one measurement location. Therefore, in this chapter (based on the work [42]), channel measurements have been conducted using a DDSS to explore 300 GHz channel characteristicsin alarge hall scenario. Furthermore, thechannel measurements areconducted with8 different Tx-Rx distance ranging from 3-15.6 m. The investigation was based on the largeamount of the directional measurement data, i.e., 33120 CFR, which helped the calculation and analysis of the PADPs and power-angle spectra (PAS). Moreover, analysis of the key measurement-based channel parameters including PL, and dispersion spread in delay and angle domain is made. The cost-effective multi-link phase-compensated channel sounder proposed in [2] (please refer to Figure 5 in section 2.2.2.1) is used in the sub-THz channel measurements. To limit the high PL, directive antennas are used in measurements. The antenna gain for the antennas on both sides is 26 dBi. The used dynamic range is 50 dB in these measurements with the IFBW of 1 kHz. Figure 42 illustrates the the large hall measurement deployment scenario. 51 Figure 42: Top view of the large hall scenario The channel measurements are conducted in a large hall with a size of 42 m × 25 m. More specifically, the size of the corridor is 43.4 m × 3.7 m. Figure 43 show the pictures in the real scenario. 52 (a) Tx-Rx 2 (b) Tx-Rx 3 (c) Tx-Rx 4 Figure 43: Photos of the measurement scenario In these measurements, the Tx is deployed near the wall to mimic the BS, while four Rx locations are used to investigate the LoS scenario (i.e Rx 1 and Rx 3) and the NLoS scenarios (i.e., Rx 2 and Rx 4). Note that Rx 4 is positioned underneath the stairs and the LoS is completely blocked, while the NLoS case of Tx-Rx 2 is formed by adding a metallic plate with the dimension of 56 m × 85 mm, as illustrated in Figure 43a and in Figure 43c. Moreover, four channel measurements with different Tx-Rx distances in the LoS cases are also conducted, i.e., Rx 5 - 8, to model the channel parameters. To obtain the channel information in the spatial domain, the DDSS was used in the measurements. To eliminate the effect of the system, an Over-the-Air (OTA) calibration is performed, by using 1 m LoS channel measurement results as a reference to calibrate the system and antenna responses. The Tx and Rx extenders are mounted on two turntables. The Tx and the Rx are programmed to rotate the azimuth in the range of [−90°, 90°] and [−180°, 176°], respectively. The rotation step is 4°at both sides. According to [34], the 4°rotation angle step is found to be sufficient for wideband channel spatial profile reconstruction. Thus, the number of CFR obtained for each measurement point is 45×90 = 4140. In total, 33 120 CFR are obtained (i.e., number of Tx-Rx pairs × number of measurement locations). For more information, the detailed measurement specifications can be found in Table 3 in [42]. Furthermore, equations (1)-(10) in [42] give the relations for all channel characteristics and parameters estimated in the current investigation. 53 (a) Tx-Rx 1 (LoS) (b) Tx-Rx 3 (LoS) Figure 44: Exemplary DDPAS for the LoS measurement locations (a) Tx-Rx 1 (LoS) (b) Tx-Rx 3 (LoS) Figure 45: Exemplary DDPAS for the LoS measurement locations Figure 44show the exemplary DDPASsof themeasuredLoSchannel. Furthermore, Figure 45presents therelation of the main paths in the channel to the room geometry. In the Tx-Rx 1 scenario, the LoS path has the delay of 20 ns, 54 the angles of departure (AoD) of −46°, and the AoA of −48°, which matches well the geometry of the Tx-Rx deployment scenario. In the Tx-Rx 3 LoS scenario, the LoS component has the delay of 38 ns, the AoD of 0°, and the AoA of 0°. As it would be expected, the power is observed to be highly concentrated around the LoS direction, while also several additional reflection contributions from various directions could be identified. Moreover, with the measurement distance increasing from 5.9 m to 11.4 m, the LoS path power decreases from −40.4 dB to −45.3 dB. The number of MPCs in these two LoS scenarios are found to be respectively 2 and 8. Path 9 is observed to be the reflection from the window structure of the corridor. From the measured results, the multipath is found to be specular and sparse from the measured results. AoAs of the main paths in the LoS scenarios match the room geometry. The DDPASs and PADPs in the Tx-Rx 2 NLoS scenario are presented in Figure 46 and in Figure 47, respectively. (a) Tx-Rx 2 (b) Tx-Rx 4 Figure 46: DDPAS for the NLoS measurement locations (a) Tx-Rx 1 (ϕT x = −46°) (b) Tx-Rx 2 (ϕT x = 50°) Figure 47: Exemplary PADP for LoS and NLoS locations at the LoS direction The LoS path can be observed with the delay of 21 ns (matching well the Tx-Rx 2 distance of 6.2 m) and power of −77.2 dB, which may be caused by the diffraction from the edge of the metal obstacle. Moreover, the MPCs are mainly concentrated in the AoD range of [0°, 90°]. In the Tx-Rx 4 NLoS scenario, the LoS path is blocked totally by the stairs, and the MPCs in this case are mainly concentrated in the AoA range of [−180°, 0°]. Furthermore, compared with the DDPAS from the LoS scenario, the MPCs in the NLoS cases are richer compared 55 to those in the LoS cases, because of the fact that the dominant LoS components are blocked and thus, the difference in power between various MPCs is less. MoreresultsarepresentedinFigure10toFigure15in[42]. Figure10in[42]presentsanexemplaryomnidirectional PDP comparison. In the Rx 1 and Rx 2 cases, it is found that the LoS path in the OLoS case is blocked and the power decreases from −40.4 dB to −67.7 dB compared to that in the LoS case. Moreover, in both cases, the MPCs are rich. The MPCs are obseved to appear in the delay range of [21, 250] ns, which corresponds to the propagation distance range of [6.3, 75] m. Furthermore, in both cases, few weak paths over 400 ns, which corresponds to the propagation distance of 120 m, can be identified. In the Rx 3 and Rx 4 cases, the LoS path in the NLoS case (i.e., Rx 4) is blocked completely resulting in power reduction of 58.5 dB compared to LoS case. The MPCs in NLoS case are mainly seen in the delay range of [50.5, 270] ns, which corresponds to the propagation distance range of [15, 81] m), while the omnidirectional PDP in the LoS case is found to have a longer decay in the delay range of [37.5, 446.5] ns. Figure 11 in [42] compares the empirical and modeled PL-es. The antenna gain is not taken into an account in the omnidirectional PDPcalculation. ComparativelygoodmatchbetweenthedifferenttheoreticalandmeasuredPL-eshavebeenobserved. For more details, the reader can refer to Figure 11 in [42]. Figure 12 in [42] presents the comparison of the empirical delay spread. The empirical delay spreads in the LoS cases is within the range [7.5, 35.6] ns. Moreover, the LoS delay spreads increase with the increase in the link distance between the Tx and the Rx, which is caused by the decreased LoS power and the fact that the power of MPCs becomes more comparable to that of the LoS path. In the NLoS cases, the delay spread of the Rx 2 is 14.8 ns higher than that of Rx 1, while the delay spread of the Rx 4 is 7.6 ns lower than that of Rx 3. The relation of the empirical angular spread to the link distance between the Tx and Rx is shown in Figure 13 in [42]. The LoS AoA spreads are within the interval [18°, 43°] and increases with the distance. Moreover, the LoS AoD spreads are found to increase with the measurement distance range of [3, 8.1] m, and decrease in the Tx-Rx distance range of [8.1, 15.6] m. Figure 14 in [42] shows the PAS comparison of the different Rx locations - Rx 3, Rx 7, and Rx 8. The Rx 7 has a larger AoD difference between the LoS direction and the mean AoD compared to those of the Rx 3 and Rx 8, which can be explained by the higher contribution of the MPCs on the angular spread in the Rx 7 case. For the NLoS scenarios, the AoA spreads of the two NLoS cases, i.e., Rx 2 and Rx 4, are 88.0°and 115.2°, respectively, which are observed to be much higher than those AoA spreads at the same distance. Figure 15 in [42] presents the comparison of the PAS in azimuth and the mean azimuth for the LoS and NLoS cases. It is found that the mean AoA appears near the LoS peak in Rx 1 and Rx 3 (LoS cases) and that the LoS power contributes dominantly to the angular spread, which leads to a small angular spread around the mean AoA (i.e., LoS) direction. The opposite is observed for the Rx 2 and Rx 4 (NLoS cases), the LoS path is blocked and the MPCs contributed more significantly to the angular spread. The AoD spread of the Rx 4 is 43.6°higher than that of Rx 3 and the AoD spreads of Rx 1 and of Rx 2 are observed to be similar. The measured channels are found to be specular and sparse. In this chapter, the measurement results of the double-directional sub-THz channel measurements at 300 GHz bands in a large hall scenario are presented. The distances between the Tx and the Rx are 3-15.6 m. Both LoS and NLoS cases are considered in the measurements. Channel characteristics, i.e., DDPAS, PADP, PL, delay spread, and angular spread, are estimated and discussed. The DDPAS results show a good agreement with the room geometry, which demonstrates the accuracy of the performed measurements. The DDPAS results in the LoS scenario validate also the sparsity of the multipaths at 300 GHz. The empirical PL related to the measurement distance in dB is well-fitted by the theoretical model and found to be close to the theoretical FSPL model. 56 points was increased to extend the time resolution. The resulting peaks correspond closely with theoretical line-ofsight delays, 4.64 ns for 1.39 m and 14.9 ns for 4.48 m. Confirming the system’s validity at longer distances. (a) (b) Figure54: Atriummeasurementlocationwith(a)receivetrolleylocatedinpositionRx2and(b)thereceivemeasurement trolley from the perspective of the transmit antenna Table 2: Measurement Parameters for the Atrium Scenario Measurement Parameter Value IF Bandwidth 50 Hza Points 25,001 Start Frequency 500 GHz Stop Frequency 750 GHz Meas. cable length (Coaxial) 100 m LO cable length (Fibre) 100 m Channel sounding range 0 m – 30 m Sweep duration ca. 400 seconds aCalibrated at 1 kHz. 2.5.1.2 Atrium Atrium, a large indoor circulation space on the NPL campus used for gatherings, seating and conferences. This is a suitable indoor environment for channel sounding measurement, where sub-THz signals could potentially be used for medium-range (c.30 meters) high bandwidth communications. The space measures approximately 15x20x6 meters in length, width and height respectively. The construction was completed in approximately 2020 and as such the structure uses modern material such as multi-layer glass glazing, plaster, brick walls and concrete flooring. There is also furniture within the space, which has been moved such that it has a minimal effect on the expected propagation paths of the channel sounding measurements. A set of measurement locations have been chosen to sample the space around the room. These have been placed to test different effects of the room structure, such as parallel and perpendicular walls, mid-points and corners. A fixed transmit location (Figure 55 Tx1) has been selected for the measurements due to the complexity of moving the VNA measurement instrument. This transmit locationis still able to be rotated around a fixed point such that it aligned to face the receive antenna. The remote receiver head can be easily moved as it is battery powered and lightweight 63 Figure 55: Positions of measurement locations within the NPL Atrium for Tx locations 1 and Rx locations 1 - 7 64 aiding repositioning. It is thereceiver head thatis movedbetween measurement locations, shownin Figure 55Rx1Rx7. Measurement locations were measured using a tape and then marked on the floor to allow repeated measurements with different systems. The antenna is then positioned over the taped mark and alignment is then performed to ensure it is directly facing the transmit antenna. At the start of each measurement, a system normalisation and check is performed. This is used to remove the effects of the measurement system from the results, as such successful normalisation will remove any phase or attenuationcausedbythe either the RoFor extenderheads. Thenormalisation isperformed totheend of the extender head waveguide test port and is measured for the direct through case with the extender heads directly connected together. This moves the measurement reference plane to the end of the wave guide test port before the antennas are connected. It is currently a limitation of this experimental setup that the antennas are not part of the normalisation. It can be seen in the literature that antenna de-embedding is possible however for the results presented the effects of the antenna are significantly small compared to the measured channel. Normalisation of the phase feedback system is performed simultaneously so that the phase change of the measurement system can be tracked from the time the normalisation was performed. A check of the normalisation is undertaken before measurements commence. The checkisperformedbymeasurementofashort(1”or2”)waveguidesectionthathasbeenwellcharacterisedbyprevious measurements, the post normalisation results are then checked against the catalogue of previous measurements to ensure the system response is accurate. This allows for identification of normalisation errors that tend to come from inaccuracies in the connection of the sub-THz waveguide components. The measurement system used is the same as the laboratory RoF case presented in 2.4.1.1. This includes the directional VDI waveguide horn antennas, phase compensated RoF system and Keysight PNA. ResultsfromtheAtriummeasurementcampaignareshowninFigure56. Thesefiguresshowthedifferentchannel responses for each measurement location. Each figure shows a strong LOS peak that is correlated with the distance between the Tx and Rx locations, increasing in time delay as the separation increases. After this first peak there are additional peaks present in the measured results, which are from the Non-LOS effects such as reflections, scattering and diffraction created by the environmental interaction of the sub-THz signals. Table 3: Measurement Parameters for the Raceway Scenario Measurement Parameter IF Bandwidth 100 Hz Points 50,001 Start Frequency 500 GHz Stop Frequency 750 GHz Meas. cable length (Coaxial) 100 m LO cable length (Fibre) 100 m Channel sounding range 0 m – 60 m Sweep duration c. 400 seconds 2.5.1.3 Corridor (Raceway) Raceway, a long, enclosed corridor-like indoor space on the NPL campus primarily designed for movement between laboratories. This environment is chosen for channel sounding measurements for evaluating sub-THz signal propagation in confined, modern indoor settings. The raceway measures approximately 30 meters in length, 2 meters in width, and 3 meters in height, making it representative of a typical high-traffic corridor 65 (a) Tx1–Rx1 (b) Tx1–Rx2 (c) Tx1–Rx3 (d) Tx1–Rx4 (e) Tx1–Rx5 (f) Tx1–Rx6 (g) Tx1–Rx7 Figure 56: CIRs measured in the NPL atrium with a fixed transmitter (Tx) and the receiver positioned at seven distinct locations. Each figure (a–g) corresponds to a unique receiver position (Rx1–Rx7), showing the CIR observed at that point. 66 (a) (b) Figure 57: Raceway measurement location with (a) transmit trolley located in position Tx2 and (b) the transmit measurement trolley facing the metal reflector plate in institutional or commercial buildings. Constructed as part of a recent development around 2020, the space features modern building materials including metallic wall panels, multi-layer double-glazed windows spanning one side of the raceway, and a concrete epoxy-coated floor. The ceiling consists of exposed infrastructure partially concealed by a linear baffle system, which incorporates integrated lighting. The corridor is free from obstructions, with only minimal presence of equipment during measurements to ensure accurate assessment of signal behaviour in a static scenarios. This setting supports realistic propagation environments for high-frequency wireless communication testing, particularly over shortto medium-range distances in line-of-sight and reflective multipath conditions. The measurement locations in Figure 58 Rx2a – d, represent increasing distances along the corridor, with distances of 5, 10, 20 and 30 meters respectively from the Tx antennas. The Tx and Rx trolleys are located centrally in the corridor and the antennas are aligned so that the Tx and Rx antennas are bore-sight to bore-sight. The Tx2 position is 3 meters from the glass windows behind, along the corridor in the long direction. Transmit position 3, Tx3, is located around the corner to exhibit a Non-LOS case for the propagation channel when the Rx antenna is located at any of the Rx2 positions. For this case, the channel was augmented with and without a metal plate, approximately 0.6x1.3 meters, to provide a reflected path around the corner. The plate was placed at 45º to the Tx antennas directly in the path of propagation. The measurement process from the atrium measurements was repeated, including normalisation and checking with a waveguide through section. Additionally, the measurement setup was repeated from the atrium measurements, shown in Table 3 , although the number of measurement frequency points was increased to allow for a longer channel distance to be assessed, to include longer reflected signal path components. To keep the measurement, sweep period similar to the atrium measurements an increased IF bandwidth was selected. The raceway results in figure 59 show greater multipath components than the results presented for the atrium. This is especially evident for the channel measurements of greater distance such as figure59c (Rx2c) and 59d (Rx2d). This is attributed to the material properties 67 Figure 58: Positions of measurement antennas for the NPL Raceway measurement scenario. Distance in units of centimetres. of the measured environment; it is expected for there to be significant reflected signal from the metal clad wall panels andpossiblythewindowglassormetalframework. Theceilingdesign,madeofmetalplates,willalsoprovidemultipath effects shown in the results, due to reflection off the surface and diffraction or scattering from the edges. In comparison between both the channel measurements ’around a corner’ there is significant different with, (figure 59f) and without (figure 59e) a metal plate. The position of the metal plate greatly aids the reflection of the signal around the corner and provides an approximate 45 dB increase in signal amplitude when it is used. Without the metal plate there is are two weak reflected path around the corner however these are close to the noise floor of the measurement system. Table 4: Measurement Parameters for the Outdoor Scenario Measurement Parameter IF Bandwidth 300 Hz Points 100,001 Start Frequency 500 GHz Stop Frequency 750 GHz Meas. cable length (Coaxial) 100 m LO cable length (Fibre) 200 m Channel sounding range 0 m – 120 m Sweep duration c. 00 seconds 2.5.1.5 Outdoor Measurements were performed in an outdoor space, along the footway adjacent to Essen Way on the NPL Campus. This open-air environment offers a relatively uncluttered and extended line-of-sight range up to 100 meters for sub-THz channel sounding measurements. The pavement measures approximately 110 meters in length and 3 meters in width, bordered on one side by a tall red-brick wall and on the other by a wide road surface, shown in figure 60, providing a mix of reflective and absorptive surfaces for multipath analysis. The site features minimal obstructions and is surrounded by mature trees and some low foliage, which contribute to realistic scattering effects typical of semi-urban or landscaped environments. The brick wall introduces periodic shadowing and potential for diffraction, making it ideal for evaluating high-frequency signal behaviour in partially obstructed outdoor corridors. The test equipment, including channel sounding and measurement apparatus, is deployed on mobile trolleys as seen in the foreground of figure 60a. A temporary blue tent is situated at the far end of the track to support continuous data collection and environmental control. This location enables long-range, controlled outdoor experiments to evaluate signal attenuation and multipath characteristics for future sub-THz wireless systems. 68 (a) (b) (c) (d) (e) (f) Figure 59: CIRs measured in the NPL Corridor (Raceway) for two transmitter locations (Tx2 - 3) and the receiver positioned at seven locations. Each figure (a–f) corresponds to a receiver position showing the CIR observed at that point. figures (e) and (f) show the same channel with and without a metal reflector plate 69 (a) (b) Figure 60: Outdoor measurement location with (a) the receive trolley in the foreground and (b) the transmit trolley in the foreground for the 100 meter separation case. Four receive antenna measurement positions were used, at 25, 50, 75 and 100 meters separation from the fixed Tx antenna placed near the blue tent are shown in Figure 61. The same measurement procedure detailed in previous sections was used, including normalisation and checks of a known waveguide through with comparison to previous results. Again, the measurement parameters have been maintained from the previous measurements, however the number of measurement points has been increased to 100,001 to capture channels up to 120 meters. Currently there is a limitation within the Keysight PNA that a single measurement can have up to 100,003 points, limiting the maximum possible distance in a single sweep for the 250 GHz bandwidth. An additional 100 meters of fibre is used, bringing the total to 200 meters of optical fibre within the system. The IF bandwidth is also increased to 300 Hz providing a sweep time of approximately 10 minutes. As expected, Figure 62 shows as the length of the measured channel increases, the peak of the CIR moves further in time and decreases in magnitude. The measurement at the 100 meter case, figure 62d, exhibits that the increase in attenuation of the channel at these distances is sufficient to bring the LOS signal path within 15 dB of the system noise floor. This limits the ability of the system to identify reflected and other multipath signals as these will be of a lower magnitude than the LOS path and therefore in the system noise. The shape of the peaks at distances of 25, 50 and 75 meters are broadened, this may be due to system non-linearity and inability for the phase compensation mechanism to correct for all phase drifts within the system. The 100 meter measurement was performed first and does not exhibit this effect, although this could be masked by the noise floor. The effect is especially strong due to the long lengths of fibre and outdoor temperature fluctuations affectingnot just the fibre but also the extender heads and RF components of the system which are uncompensated. With an increased measurement sweep time, an decreased RF bandwidth can be used which would improve the system dynamic range by approximately 10 dB. Upgraded components may also offer an additional 30 dB dynamic range improvement at this frequency range allowing for measurement of longer lossy channels. 70 Figure 61: Measurement positions for the NPL outdoor channel sounding case. Each distance is in units of meters. (a) (b) (c) (d) Figure 62: CIRs measured outdoors for one transmitter locations (Tx1) and the receiver positioned at four locations. Each figure (a–d) corresponds to a receiver position showing the CIR observed at that point. 71 2.5.2Discussionofpracticalmeasurementtechniques Table 5 providesa comparisonof thecomponentsusedin both measurement systems operating in the 330–500 GHz and 500–750 GHz bands. The primary difference lies in the use of different frequency extenders for each range. The measurement setups—particularly in the indoor laboratory and atrium environments—were kept identical across both systems. The results show consistent behavior, and successful channel sounding was achieved. In the following sections, ray-tracing analysis will be presented for frequencies up to 750 GHz to support the validation of the measured channel characteristics. Table 5: Equipment used across different frequency bands Equipment 330–500 GHz 500–750 GHz VNA Rohde & Schwarz /Keysight Keysight Frequency Extenders Rohde & Schwarz ZC500 VDI WR1.5 E/O Converter Thorlabs MX40G Thorlabs MX40G Amplifier 1 ERZ-LNA-1000-2700 Minicircuits ZX60-06183LN+ Amplifier 2 Mini-Circuits ZFL-500L Mini-Circuits ZFL-500LN Horn Antennas SGH-26-WR2.2 DI WM-380 Photo Receiver Thorlabs RXM40 Thorlabs RXM40 Coaxial Cable 100 m 100 m Coaxial Cable type RG58U RG58U 72 Figure 68: The frequency responses of the emulated channel over the stitched band with the conventional calibration method and the DP method. hT RG idenotes the vector of the tap coefficients of the original target channel, and hDP 1 and hDP 2denote the vector of the tap coefficients of the fading channel filter with the pre-distortion method for the upper and lower sub-bands, respectively Serving as a reference, the band-unlimited frequency response of the target two-ray channel is also shown. In terms of both magnitude and phase deviations, the result from the pre-distortion method clearly outperforms that from the conventional method. The pre-distorted coefficients hDP 1and hDP 2for the two branches are presented at the bottom of Figure 68). The considered two-ray channel can be emulated with high fidelity over a roughly doubled bandwidth by loading [hDP 1, τDP 1]and [hDP 2, τDP 2]to the corresponding fadingchannel filters of a CE. High-fidelity ultrawideband emulation can be implemented by using the same principle and extending for band-stitching with more digital channels. 3.1.1.5 Conclusion In this subchapter, the conventional calibration method for the band-stitching for ultrawideband channel emulation is explained. However, to overcome its disadvantage of reshaping the intrinsic response of each sub-band to raised-cosine, a DP method is proposed by exploiting the user-accessible fading channel filter in the CE. The suggested technique can be explained as the convolution between the original target fading channel and an equalization filter, where the raised-cosine is set as the equalization target. The coefficients of the equalization filter were solved with the MMSE method. The proposed method is realized by using an extra number of taps of the fading channel filter, which is a limited resource of a CE. 79 The original target fading channel and the raised-cosine were considered as a whole as the equalization target and the equalization coefficients were calculated over the confined taps in order to limit the number of taps of the resultant pre-distorted channel coefficients. The resultant equalization coefficients correspond directly to the final pre-distorted channel coefficients. The proposed method and the conventional method were numerically evaluated with the measured intrinsic responses of a commercial CE for a two-band stitching scenario. As the figure of merit for the band-stitching performance, the variations of the magnitude and phase were used. It was found that for the conventional method, the variations of the magnitude and phase were ±1.35 dB and ±29.13°, compared to ±0.26 dB and ±1.34°for the proposed method. 3.1.2 On Band Stitching for Wideband Vector Measurements With Vector Signal Analyzers 3.1.2.1 Introduction It is stated in the standard [62] of the 3GPP that each carrier of the 5G NR signal in the Frequency Range 2 (FR2) could occupy a bandwidth of up to 400 MHz per carrier, compared to only 20-MHz bandwidth per carrier for 4G LTE system. While the wider bandwidth increases significantly the throughput of the communication systems, it sets also a challenging task for the measurement instrumentation utilized for the development or testing. In the case of vector measurements, such as waveform or demodulation measurements, the analysis bandwidth of the instrument needs to be wider than the bandwidth of the input signal. However, with the increase of the bandwidth of signals, the analysis bandwidth of many existing commercial instruments becomes insufficient. To address this problem, several measurement bandwidth extension methods have been proposed in [63], [64], [65], [66], [67], [68], [69], [70]. The proposed solution enable instruments of insufficient bandwidth to be utilized without any hardware modification. Those methods can be divided into two groups, i.e., time-interleaved methods [64] and frequency-interleaved methods [66], [67], [68], [70]. The work in [71] is used as a basis for the content in this subchapter. The band stitching, which is the fundamental principle in use, is discussed initially. It was widely used in other applications, where bandwidth extension is considered, such as channel sounding [72], channel emulation [55], [58], and signal generation [73]. The general procedure of band stitching for vector measurements involves several steps as follows: 1) spectrum division; 2) sub-band sampling; 3) interpolation and mixing; 4) sub-band impairments calibration; and 5) sub-band stitching. Depending on whether the first two steps the spectrum division and the sub-band sampling are done sequentially or simultaneously for all sub-bands, the practical measurement setup can be configured in a sequential structure [65], [66], or a parallel structure [67], [68], respectively. The different sub-band data are recorded typically separately either in time (the sequential case) or at different branches (the parallel case), timing, magnitude, and initial phase impairments in recorded sub-band data may exist. The key element in recovering the original signal is the precise estimation of the sub-band impairments. For more general input signals, the timing correction is mostly resolved through either circular cross correlation or trigger signals among different sampling branches [65], [67], [68], [69]. However, these methods do not ensure an accurate estimate for the timing impairment. Therefore, estimation methods with higher accuracy are needed. Afterwards, both the sequential and the parallel setup with vector signal analyzers (VSAs) are discussed. The advantages and disadvantages of both setups are analyzed. For the timing correction, a two-step delay estimation method, which combines the common circular cross correlation and an additional refining step is proposed. 3.1.2.2 Setups and planning Figure 69) shows both setups. 80 (a) (b) Figure 69: Diagrams of a) sequential setup and b) parallel setup. The center frequency of the VSA of the nth trigger event or the nth sampling branch is denoted as fn Figure69a)presentsthe sequentialsetup. Asitisvisibleit requiresonlyoneVSAin themeasurement. Acommon trigger signal is used to control the transmission of the signal source and the reception of the VSA. Every time when a trigger event occurs, the signal source replays the input signal from the beginning. Let’s suppose that the spectrum of the input signal is divided to N sub-bands. During the nth trigger event, the center frequency of the VSA, fn, is set to the center frequency of the nth sub-band. Then, the in-phase and quadrature (I/Q) data of that sub-band are recorded. The parallel setup is presented in Figure 69a), and it is composed of multiple VSAs conducting parallel recording in a simultaneous manner. The number of VSAs is equal to the number of sub-bands. Likewise in the previous setup, the reception of all VSAs is controlled by a common trigger signal. The center frequency of the VSA at the nth sampling branch, fn, is set to that of the nth sub-band. One trigger event is required to capture the I/Q data of all sub-bands, because different sub-bands are recorded at their respective sampling branches in the same time. It is recommended that the VSAs and signal source are locked to the same reference frequency, e.g., through a 10-MHz signal, to minimize any carrier frequency offset for both setups. If this is not the case, then the carrier recovery may be needed for each sub-band. In comparison to the parallel setup, the sequential setup has the advantage of a much lower cost of the system cost, because only one VSA is used in the measurement. However, the measurement with the sequential setup takes N times longer than that for the parallel setup. 81 Figure 70: Diagram of the relations among sub-bands. The nth sub-band is centered at fnwith the analysis bandwidth ofBanalysis. Thespacingbetween thecenterfrequenciesofadjacentsub-bandsis∆f. The overlapbandwidthbetween adjacent sub-bands is Boverlap. The total stitched bandwidth is Bstitch Figure 70) shows a diagram of the relations of the individual sub-bands and the total stitched band. It is visible that all sub-bands have a bandwidth equal to the analysis bandwidth Banalysis. The spacing between the center frequencies of the adjacent sub-bands is ∆ffor all sub-bands. The input signal is first filtered by the analysis filter of the VSA [74]. The analysis filter can be seen as an effective filter that includes the total filtering, e.g., at the IF analog filter, the anti-aliasing filter, and so on. The bandwidth of the analysis filter is Banalysis and it is often referred to as the analysis bandwidth in instrument manuals [74]. Banalysis determines the bandwidth within which the spectrum of the raw input signal can be recorded approximately without distortion, i.e., Hanalysis(f)≈ 1 within the analysis bandwidth. The value of the analysis bandwidth is usually scaled to the sampling rate of the VSA by a factor of 0.8 [74]. For instance, if the sampling rate is 100 MHz, then the corresponding analysis bandwidth is 80 MHz. The reader can refer to equations (1)-(2) in [71], showing the relations between the different frequency bands. 3.1.2.3 Sub-band impairment correction The timeline of the recording of two sub-band signals is presented in Figure 71. 82 Figure 71: Timing impairments between sub-bands with respect to the original signal for a two-band-stitching scenario. The input signal arrives at each VSA after a sampling path delay. Each VSA starts recording after a triggering path delay, indicated with the red dashed line. The recording duration is T for both VSAs. The green dashed line indicates a virtual start of recording after delay ∆τcorrection. The recoverable duration of the data is ˜ Twith a maximum value of max˜ T= T−|∆τ| The three types of impairments are estimated with the recorded sub-band I/Q signals in the overlap band. We assume that when there is no impairment, the signals of adjacent sub-bands in the overlap band shall be about the same if noise is not taken into an account. Recall that Hanalysis(f)≈ 1 in the overlap band. We will consider the parallel setup in Figure 69b) and let’s suppose a common trigger signal is sent to both VSAs. Owing to the potential different lengths of the sampling paths, triggering paths, or jitters, the starting of recording of the two VSAs (see Figure 71) may not necessarily coincide at the same position relative to the beginning of their respective sub-band signals. This timing impairment from various sources can be identified as a delay offset, ∆τ(between the recorded data of the two sub-bands). Equations (3)-(6) in [71] give the relations between the ∆τand the spectrum of the signals. 83 Figure 72: Photos of the measurement setup 3.1.2.4 Validation measurements Table I in [71] represents the general details about the conducted measurements, while the settings of the sub-band measurements for the individual band-stitching cases are given in Table II in [71]. A photos of the measurement setup is presented in Figure 72. An example of the measured I/Q converted to the frequency domain is presented in Figure 73 for the B1 and B3 cases of Series 1, i.e., the conventional full-bandwidth measurement case and the three-band-stitching cases. 84 Figure 73: Measured I/Q converted to the frequency domain. Each sub-band I/Q of the B3 case has been interpolated to 122.88-MHz sampling rate and mixed to their respective IF frequencies, i.e., −30 MHz, 0 Hz, and 30 MHz Each sub-band I/Q of the B3 case has been interpolated from its recording sampling rate, 62.5 MHz, to the target sampling rate 122.88 MHz, and further mixed to their respective IF center frequencies, i.e., −30, 0, and 30 MHz. The small frequency components on the very left-hand and right-hand sides of the “sub-band 1” and “sub-band 2” are caused by the anti-aliasing filter and decimation of the VSA while recording the I/Q data. We can see the envelopes of the three sub-bands, if superimposed, follow that of the B1 case within the channel bandwidth. The same example (the B3 case of Series 1) showed in Figure 73 is taken for the illustration. The impairment estimation and correction are done over the 20-MHz overlap frequency band between adjacent sub-band I/Q. The initial and refined delays between “sub-band 1” and “sub-band 2” are estimated to be ∆τ′= 1 ·δτand ∆τ′= 0.03·δτwith ∆τ′= 1/122.88 μs. Between “sub-band 2” and “sub-band 3,” the delay estimates are ∆τ′= 1·δτand ∆τ′= 0.18·δτ. The phase alignment in the two overlap frequency bands is shown in Figure 74 for the cases without delay correction, with the initial delay correction, and with the refined delay correction. 85 Figure 74: Phase alignment in the overlap frequency band between the adjacent sub-band I/Q, i.e., (top) between “subband 1” and “sub-band 2,” and (bottom) between “sub-band 2” and “sub-band 3.” Three cases are shown, i.e., the phase alignmentwithout anydelay correction (raw), thatwith the initialdelay correction (∆τ′), andthat with therefined delay correction (∆τ′′ ) We can see a clear improvement in the alignment with the initial delay correction, and that of the refined delay correction seems marginal. The initial phase offset is estimated from the phase alignment with the refined delay correction, and it is ∆ϕ= 39.92°between “sub-band 1” and “sub-band 2” and ∆ϕ= −74.83°between “sub-band 2” and “sub-band 3.” Similarly, the magnitude alignment in the two overlap frequency bands is shown in Figure 73. Figure 75: Magnitude alignment in the overlap frequency band between the adjacent sub-band I/Q, i.e., (top) between “sub-band 1” and “sub-band 2,” and (bottom) between “sub-band 2” and “sub-band 3.” Three cases are shown, i.e., the magnitude alignment without any delay correction (raw), that with the initial delay correction (∆τ′), and that with the refined delay correction (∆τ′′) 86 No significant difference among the three cases is seen, which is reasonable, since the estimated delay offsets between adjacent sub-bands are relatively small. The estimated magnitude offsets in the linear scale are ∆α= 0.9986 between “sub-band 1” and “subband 2” and ∆α= 0.9995 between “sub-band 2” and “sub-band 3,” which indicates that the magnitude offsets between the adjacent sub-bands are almost negligible. A summary of the estimated delay, initial phase, and magnitude offsets for all three repetitions of the B3 case of Series1 isgivenin TableIII in[71]. Wecansee thatthe estimateddelayoffsets (∆τ′and∆τ′′ )forallthreerepetitionsremain relatively small, i.e., within two delay resolutions. Besides, an interesting observation is that the estimated initial phase offset ∆ϕchanges over the repetitions under the condition that the signal generator and the VSA were locked to the same 10-MHz reference frequency during the measurement, which indicates that the impairment correction is probably needed every time a new band-stitching measurement is conducted. From the EVM evaluation point of view, it is specified in the standard [75] that the EVM is evaluated over a complete one-frame signal from its frame head. Therefore, we stitch the I/Q data for a two-frame duration to ensure that there exists at least one complete frame of the signal within the stitched duration. Moreover, we generate a frame of the golden I/Q data for the considered “NR-FR1-TM3.1a-FDD-100 MHz–30 kHz” signal according to the standard [75], [76]. Both the I/Q data of the B1 and B3 cases are time and power aligned to the golden I/Q data, and a segment of 1-μs duration of the aligned I/Q data is shown in Figure 76 in terms of the real and the imaginary part, respectively. No significant differences among the three datasets are seen. Figure 76: Real part (top) and the imaginary part (bottom) of the golden I/Q data and that of the B1 and B3 cases, respectively. A 1-μs segment is shown 87 Figure 77: Frequency-domain counterpart ofthe golden I/Qdata and that ofthe B1 and B3cases, respectively, (top) over the full 122.88-MHz band and (bottom) that zoomed to the frequency range around 40 MHz where a slight deviation is observed The I/Qdata arefurther comparedin thefrequency domain, whichis moreimportant forOFDMsignals, as shown in Figure 77. Within the center 100-MHz channel bandwidth, both the B1 and B3 cases have a good match to the golden I/Q data, except for the frequency range around 40 MHz. Since the deviation happened in both the B1 and the B3 cases in the same way, it is probably caused by the signal generator that the original RF signal itself is slightly distorted in that frequency range. The normalized root-mean-square error (NRMSE) is further calculated to quantitatively measure the deviation from the B1 and B3 cases to the golden data, respectively, within the center 100-MHz channel bandwidth. The NRMSE is defined by eq. (10) in [71]. The resultant NRMSE for the B1 case and the B3 case is 3.32 % and 3.78 %, respectively. A summary of the resultant NRMSE for all measurements is given in Table IV [71]. We can see that the NRMSEs are stable across the three repetitions. Taking the values of the B1 case, i.e., the conventional full-bandwidth measurement, as the reference, the B4 case results in the largest difference of about 1 % for Series 1, and the B3 case results in the largest difference of about 0.5 % for Series 2. An in-house demodulation and EVM evaluation algorithm is implemented according to the standard [75]. Given the known golden data, the frame head is found, and the carrier recovery is conducted. Given the known frame structure, the symbols are further demodulated. Moreover, equalization is performed with respect to the demodulation reference signal symbols. Finally, the EVM is calculated for each slot within the frame over the symbols in the physical downlink shared channel (PDSCH). The constellation of the demodulated symbols of the first slot in the PDSCH channel is shown in Fig. 10 in in [71] for the golden I/Q data and that of the B1 and B3 cases, respectively. The root-mean-square averaged EVM is calculated over the individual EVMs of the 20 slots within the frame to represent the EVM of the whole frame. The resultant 88 Figure 84: AiP elements stability measurement results Excluding the result of the seventh measurement, all the rest of the measurements have a good agreement with the tenth one in terms of both the amplitude and phase excitation of array elements. Additionally, the average AiP chips temperature is constant throughout the measurements, except for the seventh measurement. Obviously, the offset in the result for the amplitude and the phase in the seventh measurements can be explained by the fact that this measurement is done after the AiP being off for some time and again then turned on, the average temperature of the AiP chips is a bit high during the early part of the measurement procedure until it again stabilizes. Therefore, the operating temperature of the AiP has a significant impact on the stability of the elements. It is recommended that the AiP must be used when it is in its stable state (the temperature is stabilized). The beamforming pattern measurements of the AiP are also performed in the CATR. The block diagram of the AiP beamforming pattern measurement setup is presented in Figure 85. 95 Figure 85: Block diagram of the AiP beamforming pattern measurement setup The AiP beamforming patterns at different angles are presented in Figure 86. The AiP beamforming performance is validated by showing that the peak directions of the patterns varies with the beam-steering angles. Figure 86: AiP beamforming patterns at different angles The 8 × 8 AiP is used also as a validation platform to investigate the calibration accuracy of the calibration methods on large arrays and the performance of AiP-based channel sounding, respectively. The 8 × 8 AiP array with 64 elements can be defined as a large array. The effectiveness of the ‘inverse’, ‘on-off’ and ‘least squares’ calibration methods on large phased array is investigated. Details on the obtained results can be seen in Figure 9 - Figure 11 in [87]. The reader can find more details about the calibration methods and most particularly how the amplitude and the phase of some elements of the AiP are changed intentionally. The ‘inverse’ method is not able to effectively calibrate the AiPwith a largearray. Thecalibrationexcitation matrixis large, whichmakes the calibration influencedsignificantly by the measurement noise. Therefore, the ’inverse’ method is not suitable for calibration on large phased arrays. The AiP array calibration excitation matrix of the ‘least squares’ method is based on the Hadamard matrix [89], and the condition number is 1. Therefore, the phased array calibration is less affected by the measurement noise. The errors between the calibration results of the other two methods (‘least squares’ and ‘on-off’) for the four intended changed elements and the corresponding reference values are shown in Table VII. 96 TABLE VII VARIATION BETWEEN THE RESULTS OF AIP ELEMENTS CALIBRATION MEASUREMENTS Element ’On-off’ Amp. Err. [dB] ’LS’ Amp. Err. [dB] ’On-off’ Pha. Err. [ ◦]’LS’ Pha. Err. [ ◦] 1 -0.2 -0.2 1.6 2.8 2 -0.3 0.2 5.0 -1.9 3 0.7 -0.4 2.6 1.7 4 1.5 -0.2 2.9 -3.6 The results show that both methods can perform effective measurements on the AiP array. As expected, some error are also observed. The errors are mainly introduced by the limited control accuracy of the element attenuator and phase shifter and the measurement noise. Most particularly, when the element complex filed amplitude is small, the measurement noise has a large impact on the ‘on-off’ calibration accuracy. The environment and system setup of the AiPs channel sounding measurement can be seen in Figure 87. (a) (b) Figure 87: a) Photo and b) block diagram of the AiPs channel sounder setup 97 Figure 88: Placement of the Tx, Rx and metal plate The 4 × 4 AiP described in [88] is employed as the Tx, while the 8 × 8 AiP configuration is used as Rx. A metal plate is used to act as a reflector in the channel environment. The height of the Tx and the Rx is 1.2 m. The placement of the Tx, the Rx and metal plate is shown in Figure 88. Port 2 of the VNA is connected to the Rx, while port 1 to the Tx. By measuring the S-parameter with the VNA, the channel frequncy response between the Tx and the Rx is obtained. When the AiPs channel sounder measures the channel, laptop 1 controls the Tx AiP beam-steering, which enables the AiP to beam-steer in the horizontal plane from −53°to 53°. For each beam steered by the Tx AiP, labtop 2 controls the Rx AiP steering beams from −72°to 72°in the horizontal plane. 98 (a) (b) Figure 89: Measurement results for a) without the metal plate and b) with the metal plate The channel sounding measurements results are shown in Figure 89 for scenarios without and with the metal plate. The AOD and AOA of the LOS path between Tx and Rx are the same and is approximately 18.4°. It is found that the highest peak appears when both the Tx AiP and Rx AiP steering the beam in approximately 18°, demonstrating the ability of the AiPs based channel sounder to detect accurately the LOS propagation path in the measured channel. Figure 15 and Figure 16 in [87] present more measurement results. In this sub-chapter, the hardware structure and control program of an 8 × 8 AiP array experimental platform are proposed and discussed. Further, the stability, element weighting control accuracy and beamforming feasibility of the AiP are measured and verified. It is demonstrated that temperature has a high impact on the AiP performance. The calibration accuracy of three different calibration methods for large phased array is investigated with the AiP platform. Due to the measurement noise, the results show that the calibration accuracy of ’inverse’ method for large phased array is low. In the channel sounding measurements, the dominant propagation components were analysed in terms of AoA, AoD and delay. The results validate the effective work of the channel sounder. It would save measurement time and make the measurement setup simplified due to the capability for swift beam control of the phased array. 99 4. Terahertz (THz) Channel Modeling in 6G Deployment Scenarios via Ray Tracing (RT) 4.1 Introduction Terahertz (THz) communication, which is planned to use (0.1-10 THz) frequncy band, is considered to be the key enabling technology to meet the requirement for the high data rate demand in 6G. The main advantage of the THz communications is the abundant available spectrum resources. One precondition for implementing the THz communication systems is to understand and model the THz radio channel for 6G in realistic deployment scenarios. However, channel sparsity, NF propagation and large-scale antenna configuration are among the new radio characteristics in THz bands. They bring new opportunities, but also challenges to channel modeling in terms of accuracy and complexity of the modeling. It was proved experimentally in [90] that the THz channels show directionality and can be characterized by a few dominant paths. Therefore, the propagation channel in the THz bands will be highly sparse and specular. For long-range applications, the THz communication systems will rely mainly on the LOS path, while for short-range applications, a few dominant paths (LOS and reflected paths) can be used. Owing to the use of large-scale antenna configurations (to limit the severe propagation loss), combined with the short-range deployment scenario and short wavelength in the THz bands, channel non-stationary is another aspect that one must take into an account in channel modeling [91]. Considering the channel characteristics and applications of the THz communication systems, a RT method may be a promising approach to model the propagation characteristics and parameters in THz bands for 6G. RT is deterministic modeling approach based on electromagnetic field theory and geometrical optics. Therefore, it enables precise modeling of the dominant paths of THz waves. Moreover, the quasi-optical property of THz waves secures the precision of the RT approach. The requirement for precise modeling of dominant paths can efficiently reduce the complexity of RT (in terms of database precision and interaction orders of propagation mechanisms). For future data-driven research, the RT simulations can provide the essential massive input data, which may not feasible and manageable with measurement data. This chapter based on the work in [92] presents that deterministic RT is a promising solution to channel modeling in THz bands. Two sets of indoor measurement campaigns in THz bands are carried out to validate the obtained RT simulation results. Good agreement is achieved between the RT simulation and measurement results, which demonstrate the effectiveness of RT approach in THz bands for 6G. In this work, the commercial RT tool Wireless InSite is used to perform the RT simulations. 4.2 Channel characteristics of THz communications systems 1. High propagation loss: The high propagation loss of the THz channel is defined by FSPL and non-negligible atmosphere molecular absorption. The THz channel is known to have high FSPL according to the theory (the Friis formula). Therefore, to compensate for the high propagation loss and to consider the power limitations of the current THz transceivers, high-gain directional antennas must be used at both Tx and Rx ends. Moreover, the absorption of oxygen and water vapor is very common in atmospheric absorption in the THz channel, as the absorption of water vapor is dominant. Unlike at lower frequency bands, the molecular absorption is not a negligible factor and should be also considered for long-range communication scenarios. 2. Sparsity: As it was experimentally demonstrated, the propagation channel becomes sparser and more specular with the increase of the frequency. As it was discussed, the THz propagation channels will be characterized only 100 by a few dominant paths, i.e. LOS and a few low-order reflection paths, due to high diffraction [93], penetration, and reflection losses of THz waves. 3. Near field and spatial non-stationary: When analyzing the massive Multiple-input-multiple-output (MIMO) systems, the FF and spatial stationarity assumptions applied for traditional MIMO systems may be violated [94]. Typically, due to the high frequency (HF) of THz waves and the small size of the antennas, a large number of antennas can be embedded in limited space. Generally, the assumption of plane waves is not valid, when the propagation distance is less than the Rayleigh distance. Due to the high propagation loss, the communication range will be rather short, making the NF (i.e., plane-wave model assumption is not valid) and spatial non-stationary (i.e., the spatial stationary assumption is violated) effects a significant problem in the THz bands. Therefore, it may be important and beneficial to model these effects in the THz channel modeling. 4.3 Measurement-calibrated ray tracing Channel modeling can be generally classified into two groups - based on using deterministic or statistical methods. RT is a typical representative of the deterministic modeling approach. The RT-based channel modeling has been extensively used in sub-6 GHz and mmWave bands. However, the channel modeling using the RT approach in the THz band is different from that in mmWave and sub-6 GHz bands. Due to the inherent characteristics of RT, i.e. it is based on geometric optics and it can be explained by the high-frequency approximation of Maxwell’s equations, the quasi-optical characteristics make the RT results in the THz band more reliable and suitable than those in the sub-6 GHz band. Moreover, for the THz channel characteristics, the high propagation loss and sparsity characteristics of the THz channel define the requirement for site-specific analysis, which can be provided by the RT simulation. Compared to the sub-6 GHz, the propagation range of THz waves is shorter. Therefore, the RT simulation needs only to be performed over a limited distance and in a limited space. Moreover, considering the sparsity and specularity of the THz channel, RT simulation of the dominant paths, taking into an account the limited propagation mechanism and orders of reflections, can be sufficiently precise. However, some challenges to RT-based channel modeling in THz bands must be outlined. A fundamental difficulty is that it does not exist a complete set of EM properties for the materials at THz bands. Moreover, there is a lack of adequate measurements for validating and calibrating RT. In this work, the performance of the deterministic RT in the THz band is explored by using measurements to calibrate the material EM properties in the corresponding scenario. The measurement-calibrated RT flow is presented in Figure 90. 101 Figure 90: Flowchart of measurement-calibrated RT channel simulation The knowledge of the measurement scenario (including the Tx and Rx coordinates) is used in the RT simulation to model the environment simulation scenario. The EM calculations are performed and the results such as power gain, delay, and angle for each path (i.e. the RT channel parameters) are obtained. The RT channel parameters are then processed to obtain the required channel information, which is shown as RT simulation data in Figure 90. The calibration process is shown in Figure 90. Afterwards, the material EM properties in the RT simulations are tuned to obtain the best agreement in terms of power and delay for the dominant propagation paths. In the beginning, the RT simulation results, corresponding to the initially used EM parameters for the materials in the RT simulation, are compared with the measurement data. If the results of power and delay parameters for the dominant paths match (a minimization of the root mean square error is targeted), the simulation results are considered as the expected (desired) output. If the results do not match, the relative permittivity and conductivity of the materials are updated, simulation is performed again and a comparison between the results is made again. This operation repeats until the best agreement (match) between the simulation and measurement results is achieved. The performance of the measurement-calibrated RT is analyzed based on two sets of indoor measurement campaigns at the THz bands. Channel measurements were carried out by using the VNA based channel sounder (utilizing the RoF technique to extend the measurement distance) at Aalborg University, Denmark [1]. The measurement setting parameters are listed in Table VIII. The measurements were conducted in two indoor typical scenarios, i.e., the empty room and the spacious hall. The investigated scenarios were chosen such that the channel characteristics are different and therefore the accuracy of RT simulation in different deployment scenarios can be demonstrated. The obtained parameters of several dominant paths are presented in Table IX for comparison of the measurements and RT simulations. The gain in the empty room scenario is calculated with the antenna gain included. TABLE VIII 102 MEASUREMENT SETUP Parameter Empty room scenario Spacious hall scenario Frequency (GHz)100 300 Bandwidth (GHz)6 2 Transmitted power (dBm)0 5 Measurement description Virtual array (Aperture = 1 m)Horn antenna rotation TX antenna type (gain, Omnidirection (4.5dBi,−)Horn antenna (26dBi,8◦) HPBW) Fixed [−90 : 4 : 90] TX rotation Omnidirection (4.5dBi,−)Horn antenna (26dBi,8◦) RX antenna type (gain, [−180 : 0.5 : 180]( virtual array ) [−180 : 4 : 180] HPBW )6.5 4.2 RX rotation 1.25 1.25 TABLE IX PARAMETERS OF MEASUREMENT AND RT PATHS *: Note that the path parameters of this scenario are obtained with one element of the virtual large-scale array as the reference. 103 (a) (b) (c) (d) Figure 91: Empty room scenario a) RX view in the empty room scenario b) Ray trajectories c) Power-delay-element measurement result d) Power-delay-element simulation result Figure 91 presents the empty room scenario and the obtained results. This measurement campaign was carried outat100GHzwithabandwidthof6GHz. TheTxusesaverticallypolarizedomnidirectionalantenna,whileavirtualUCA antenna with a radius of 0.5 m and 2400 elements was created and used at the Rx end. With the used UCA antenna, accurate parameters of MPCs (path gain, delay, and angle) can be extracted from measurement data via parametric estimation [95]. The gain of the Tx and Rx antennas is 4.5 dBi. The distance between the Tx and the center of the UCA is 6.5 m. The height of both Tx and Rx is 1.25 m. A metal plate is positioned near the Rx during the measurements (as shown in Figure 91a). Figure 91b presents the diagram of the ray trajectories of the dominant path. The trajectory color changes from red to yellow, indicating the decrease in the received power. It was found that the dominant paths come from first-order reflection and second-order reflection paths. As expected, the received power of the paths generally decreases as the reflection order increases. The measured power delay profile across elements of UCA is presented in Figure 91c. The power and delay of MPCs vary with the elements, identifying clearly the existence of non-stationary characteristic in the spatial domain. The non-stationary characteristics of the path with high delay are more obvious, definedbytheextendedrangeofpowervaluesandtheincreaseofthecurvatureofthedelay-elementcurve. Therefore, itcanbeconcludedthat thespatialnon-stationarycharacteristicsof higher-orderreflectionpathsmaybemorevisible. Figure 91d presents the RT results with performed measurement-based calibration. Comparing the RT results and the measurement results, the trajectory of MPCs across the array elements is almost the same. The three strong MPCs are denoted by numbers. A good agreement is obtained between the measurement and RT simulated paths in terms of power, angle, and delay. 104 (a) (b) (c) (d) (e) (f) Figure 99: Simulated Rays plotted within the model of the Atrium scenario for the LOS (a, b) and NLOS (c, d) cases. The simulated CIR with the paths overlaid for the (e) LOS and (f) NLOS scenarios. 111 measurements and simulations in two indoor typical scenarios, i.e., the empty room and the spacious hall. A good agreement is achieved between the measurements and RT simulations for both investigated scenarios. Future work may include more accurate measurement-calibrated RT channel model, more measurements to calibrate the EM properties of materials more precisely in RT simulation and more various measurement (simulation) scenarios. 112 5. Conclusion The first key contributions in Chapter 2 are the development of novel channel sounder designs for sub-THz frequency bands, which are critical for B5G and toward 6G wireless communication systems. Traditional channel sounders face significant limitations at these high frequencies, necessitating innovative designs that offer improved resolution, dynamic range, and real-time processing capabilities. Firstly, channel sounder architectures covering respectively the (75 and 110) GHz and (220-330) GHz are proposed. Their analysis includes link budget estimations and information on system calibration. Further, a RoF technique and a novel phase compensation scheme for a channel sounder at (220-330) GHz have been proposed and their effectivenesses were analyzed. The second important aspect in Chapter 2 are the conducted measurement campaigns across different deployment scenarios at 100 GHz and 300 GHz by deploying VAA, DSS and DDSS schemes and the presented respective measurement results. By providing key insights into the propagation characteristics at sub-THz frequencies, these measurements are crucial for understanding the feasibility of high-frequency wireless communication and optimizing antenna and beamforming techniques for B5G and toward 6G networks. The results indicate that Omni-VAA provides a broad, but lessdirective channelcharacterization, making ituseful for initiallarge-scale propagation studies. However, its lack of directional selectivity results in higher PL and increased multipath interference, which may limit its effectiveness in highly dynamic environments. On the other hand, Dir-VAA enhances measurement precision by focusing on specific propagation paths, reducing interference and improving signal clarity. This method is proved particularly to be effective in urban and indoor environments, where LOS and NLOS conditions vary significantly. Lastly, Chapter 2 contains DDSS sub-THz channel measurements at 300 GHz in a large hall scenario providing valuable insights into the channel characteristics in both LoS and NLoS conditions. The measurements reveal that the LoS path exhibits a strong concentration of power around the LoS direction, with a decrease in power observed as the measurement distance increases. This decrease is associated with an increase in the number of MPCs, which become more prominent in the NLoS scenarios due to the blocking of the LoS path. These results show that the multipaths are sparse and specular in nature, with the angular spreads in the NLoS cases being significantly larger than in the LoS cases. The empirical PL data, which shows a good agreement with the theoretical FSPL model, highlights the accuracy of themeasurementsetup. The delayspreads inthe LoS cases arefound to increasewith theTx-Rx distance, suggesting that the decrease in LoS power leads to the closer power levels of the MPCs, making them more comparable. In the NLoS cases, the angular spreadand delay spreadaremorepronounced, as theLoS path isblockedand thecontribution of the MPCs becomes dominant. Additionally, the PAS comparisons across different Rx locations show that the MPCs contribute significantly to the angular spread in NLoS scenarios, while in LoS scenarios, the power is more focused, leading to smaller angular spreads. The measurements also confirm the sparse and specular nature of the channel, particularly at the high-frequency 300 GHz band. These findings underscore the importance of conducting further long-range 300 GHz channel measurements to enhance the understanding of propagation characteristics at sub-THz frequencies, especially in scenarios that involve both LoS and NLoS conditions. The measurement results provide a solid foundation for future research in the field, with implications for the design and optimization of communication systems operating in the 300 GHz band. Chapter 3 explores the challenges and advancements in enabling high-fidelity UWB radio channel emulation, particularly focusing on band-stitching and DP techniques. The need for effective radio CEs is highlighted, especially as wireless communication technologies evolve to support wider bandwidths, such as the 400 MHz bandwidth used 113 in 5G NR systems. Existing commercial CEs, which can typically support emulation bandwidths of up to 160 MHz, face limitations when attempting to emulate broader bandwidths. To address this, the band-stitching technique is proposed, wherein multiple digital channels are combined to simulate a wider bandwidth. This process requires careful calibration to ensure phase and amplitude coherence between the adjacent channels, a task traditionally done through time-consuming trial-and-error methods. However, these conventional calibration techniques have their limitations, including reduced accuracy and time inefficiency. To improve this process, a novel DP scheme is introduced, based on MMSE equalization. This approach enhanced the accuracy of the channel emulation by compensating for the intrinsic distortions of the CE, ensuring that the emulated channel closely matches the target channel’s frequency response. The pre-distortion method is found to be particularly beneficial in band-stitching, as it corrects the intrinsic frequency response of the digital channels, making it closer to the ideal raised-cosine shape, which is crucial for achieving high-fidelity emulation. Through examples and numerical simulations, it is demonstrated that the DP improves the performance of band-stitching, leading to more accurate and reliable channel emulation. The results show that pre-distortion significantly reduces deviations in both magnitude and phase compared to traditional calibration methods. Overall, the DP approach offers a promising solution to achieve high-fidelity UWB channel emulation, which is critical for the testing and validation of next-generation wireless devices. Moreover, Chapter 3 delves into the challenges of wideband vector measurements for the 5G NR signals, especially given the significant increase in bandwidths, i. e. from 20 MHz in 4G LTE to 400 MHz in 5G FR2. As wider bandwidths demand higher analysis bandwidth from measurement instrumentation, many existing commercial instruments become insufficient for these tasks. To address this, methods for extending the measurement bandwidth without hardware modifications have been proposed, particularly focusing on time-interleaved and frequency-interleaved approaches. The concept of band stitching is central for solving this problem. This technique is utilized in various applications like channel sounding, emulation, and signal generation, where it allows multiple sub-bands of a signal to be measured and then stitched together to form a wider effective bandwidth. Band stitching involves several critical steps: spectrum division, sub-band sampling, interpolation and mixing, sub-band impairments calibration, and finally, sub-band stitching. These steps can be organized into two main configurations: sequential and parallel structures, each with its own advantages and challenges, particularly in dealing with impairments like timing, magnitude, and initial phase. Both configurations are explored for VSAs, comparing their effectiveness. It also emphasizes the importance of precise estimation and correction of sub-band impairments, especially timing errors. To improve timing accuracy, a two-step delay estimation method is proposed, which refines the common circular cross-correlation technique. By addressing these challenges, the band stitching method allows wideband signals to be effectively measured and analyzed, making it a crucial tool for the development and testing of modern 5G NR systems. Lastly, Chapter 3 focuses on the experimental validation and applications of an 8 × 8 AiP array platform designed for mmWave communications. As 5G communications expand into the mmWave frequency bands (28–73 GHz), they promise significant increases in wireless data transmission speed. However, challenges such as high PL and low SNR at these frequencies limit the effectiveness of these systems. Directional high-gain antennas can be used to overcome these limitations, but they come with the trade-off of reduced coverage area, necessitating beam-tracking for reliable communication in dynamic environments. Phased array systems can be a solution to this challenge, allowing for real-time beam steering to track dominant propagation paths. The AiP technology integrates antennas and control circuitsintoasinglepackage,offeringbenefitslikesmallsize,lowcost, lowpowerconsumption,andfastbeam-steering capabilities. The presented 8 × 8 AiP array operates between 26.5 and 29.5 GHz, with various features such as 6114 bit phase shifters, attenuators, and power amplifiers for both Tx and Rx modes. The experimental setup includes calibration and beamforming measurements in a CATR. The results show that the AiP array is stable, with temperature playing a significant role in the stability of the array’s elements. Calibration methods, including ”inverse,” ”on-off,” and ”least squares” methods, were evaluated for large arrays. The ”inverse” method proved ineffective for large arrays due to the impact of measurement noise, while the ”least squares” method was more reliable. Additionally, the AiP was used in mmWave channel sounding experiments to evaluate its ability to detect the LOS propagation path. The channel frequency response was measured using a VNA, and beam-steering control enabled precise channel sounding. The results demonstrated the effectiveness of the AiP-based channel sounder in detecting the dominant propagation components, such as AoA, AoD, and delay. Chapter 4 discusses the use of RT simulations for modeling THz communication channels, considering special features as high propagation loss, sparsity, NF and spatial non-stationarity. Special attention has been put on the calibration of RT simulations using measurement data to achieve higher accuracy in channel characteristics, such as power, delay, AoA and AoD. The calibration process involves adjusting the EM properties of the materials in the simulation until the results from the RT simulation closely matches the measurement data, minimizing errors in key parameters. The effectiveness of the measurement-calibrated RT simulations is demonstrated for two scenarios - empty room and spacious hall. In the case of an empty room scenario, the measurements are carried out at 100 GHz, with a 6 GHz bandwidth. The Tx uses a vertically polarized omnidirectional antenna, and the Rx uses a virtual UCA with 2400 elements. The distance between Tx and Rx is 6.5 meters. Ray trajectories in the empty room show that the dominant paths come from first-order and second-order reflections. The received power generally decreases as the reflection order increases. A good match between the measured and RT-simulated results is achieved, especially for the dominant propagation paths in terms of power, delay, and angle. The delay differences between the three dominant paths are within 0.1 ns, and the power differences are within ±0.3 dB. In the case of a spacious hall scenario, the measurements are conducted at 300 GHz, with a 2 GHz bandwidth. Horn antennas are used at both Tx and Rx ends, with Tx and Rx rotations to capture various angle-dependent propagation effects. The dominant paths come from the LOS path and reflections off adjacent columns. The RT simulation results show good agreement with the measurements for both spatial and delay domains. The delay differences for the five dominant paths are within ±2 ns, and the maximum differences for AoA, AoD, delay, and power gain were all within reasonable bounds (e.g., -1.8°, 0.7 ns, -0.7 dB). Future work will focus on improving the calibration process, obtaining more precise measurements of the EM properties of materials, and applying this method to a wider variety of deployment scenarios. This research paves the way for more accurate THz channel modeling that can aid in the development of advanced communication systems. 115 6. Appendix Chapter 3, containing different measurements for channel characterization, uses lots of common equations. Therefore, it is useful to include all used fundamental equations in this section. When a specific equation is used to calculate a channel parameter (or characteristic), then the reader is referred to this section. The measurements are conducted within the [fmin, fmax] frequency band, where fmin and fmax can be specified respectively as the minimum and maximum frequencies. Then, the bandwidth (BW) of the measured signal is: BW =fmax −fmin (6.1) The delay resolution ∆τ( i.e. the interval between the two adjacent delay samples) is: ∆τ=1 BW (6.2) The frequency step is related to the chosen number of frequency points by: ∆f=BW Nf(6.3) Then, the maximum excess delay is: τmax =1 ∆f(6.4) When using directional antennas for both the Tx and Rx on a rotational table, the CIR can be characterized by the delay τ, the AOA θ, and the AOD ϕ. Therefore, the CIR can be written in the form: h(τ, θ, ϕ) = L ∑ l=1 αlδ(τ−τl)δ(ϕ−ϕl)δ(θ−θl)(6.5) A PDP is calculated by maximazing over all angles for each delay: P DP (τ) = max ϕmax θ|h(τ, ϕ, θ)|2(6.6) The PADP is obtained by the maximum the power for each θRx rotations over all ϕTx rotation: P ADP (τ, θ) = max ϕ|h(τ, ϕ, θ)|2(6.7) PAS is calculated by summing the delay for each Tx and Rx angles: PAS(ϕ, θ) = ∑ Nτ |h(τ, ϕ, θ)|2(6.8) 116 The average delay µDS of the PDP can be calculated as: µDS =∑ττP DP (τ) ∑τP DP (τ)(6.9) The root mean squared (rms) delay spread σDS of the PDP is calculated via: σDS =√∑τ(τ−µDS )2P DP (τ) ∑τP DP (τ)(6.10) The AoA angular spread is was calculated as the second central-moment of the azimuth-power: σθ=√∑θ(θ−µθ)2P(θ) ∑θP(θ)(6.11) where the mean azimuth µθis calculated as, µθ=∑θP(θ) ∑θP(θ)(6.12) The Friis equation defines the FSPL as: FSPL = 20 ∗log10 (4πfcd c)(6.13) where fcis the center frequency of the transmitted signal, cis the speed of light and drepresents the link distance between the Tx and Rx. The PL can be calculated as the power summation of the estimated path power. The calculated PL results can be fitted by the log-distance dependent PL model, i.e., alpha-beta-gamma (ABG) model: PLABG(d)[dB] = 10 ·nABG ·log10 (d d0)+PL (d0) + χABG σ(6.14) where PLABG denotes the ABG path PL over distance. As in the previous equation, d is the Tx-Rx distance with d ≥ 1 m and d0= 1 m; nABG is the PL coefficient, P L(d0)is the optimized offset value for PL. χABG σis the large-scale signal fluctuations. Note that the Fraunhofer distance to distinguish the NF or FF is defined as: dfar = 2D2/λ(6.15) whereDis thelargestdimension ofthesingleantenna. Inthecaseof VAAandmost particularUCA, D=2Rdenotes the array aperture of the VAA. R is the radius of the UCA. λ is the wavelength at the carrier frequency. 117 The estimated PL from the measurements can be also matched with the floating intercept model, also referred to as the alpha-beta (AB) PL model: PLAB(d) = 10 ·α·log10 (d/d0) + β+χAB σ(6.16) where PLAB(d)[dB] denotes the AB PL model as a function of distance. The Tx-Rx distance is denoted as d [m], where dis constrained to d≥1,m, and a reference distance of d0= 1 m is also used. αand βcorrespond respectively to the PL exponent and the optimized intercept for the PL curve. χAB σis the large-scale signal fluctuations (i.e., shadowing), which is commonly modeled as a zero-mean normal distribution as χAB σ∼ N (0, σχ2), where σχis the standard deviation of the shadowing. 118 References [1] Yejian Lyu, Pekka Kyösti, and Wei Fan. “Sub-THz VNA-based Channel Sounder Structure and Channel Measurements at 100 and 300 GHz”. In: IEEE 32nd Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) (2021). [2] Yejian Lyu et al. “Design and Validation of the Phase-Compensated Long-Range Sub-THz VNA-based Channel Sounder”. In: IEEE Antennas and Wireless Propagation Letters 20.12 (2021), pp. 2461–2465. [3] Allan Wainaina Mbugua et al. “Phase-compensated optical fiber-based ultrawideband channel sounder”. In: IEEE Transactions on Microwave Theory and Techniques 68.2 (2020), pp. 636–647. [4] Wei Fan, Allan Wainaina Mbugua, and Kim Olesen. “Accurate channel sounding with a phase stabilizing scheme”. In: XXXIIIrd General Assembly and Scientific Symposium of the International Union of Radio Science (2020). [5] M Mechaik. “Signal attenuation in transmission lines”. In: 2nd International Symposium on Quality Electronic Design (2001), pp. 191–196. [6] M Calhoun, S Huang, and R L Tjoelker. “Stable photonic Links for Frequency and Time Transfer in the Deep-Space Network and Antenna Arrays”. In: IEEE Transactions on Microwave Theory and Techniques 95.10 (2007), pp. 1931– 1946. [7] Lawrence Carslake, James Skinner, and Tian Hong Loh. “Design and Preliminary Indoor Assessment of a LongRange sub-THz VNA-based Channel Sounder between 500 GHz and 750 GHz”. In: 18th European Conference on Antennas and Propagation (EuCAP 2024) (2024). [8] Yejian Lyu et al. “Virtual Antenna Array Based Channel Sounding at 300 GHz: Implementation and Field Measurements”. In: IEEE Antennas and Wireless Propagation Letters 23.12 (2024), pp. 4174–4178. [9] Mengting Li et al. “Antenna Deembedding in Directional Channel Measurements with Virtual Array Concept and Experimental Validation”. In: IEEE Transactions on Antennas and Propagation 72.1 (2024), pp. 1063–1068. [10] Yejian Lyu et al. “Virtual Antenna Array for W-Band Channel Sounding: Design, Implementation, and Experimental Validation”. In: IEEE Journal of Selected Topics in Signal Processing 17.4 (2023), pp. 729–744. [11] Yejian Lyu et al. “Enabling Long-Range Large-Scale Channel Sounding at Sub-THz Bands: Virtual Array and RadioOver-Fiber Concepts”. In: IEEE Communications Magazine 62.2 (2024), pp. 16–22. [12] JNoda,KOkamoto,andYSasaki.“Polarization-MaintainingFibersandTheirApplications”.In:JournalofLightwave Technology 4.8 (1986), pp. 1071–1089. [13] V Budinski and D Donlagic. “Fiber-Optic Sensors for Measurements of Torsion, Twist and Rotation: A Review”. In: Sensors 17.3 (2017). [14] Fengchun Zhang and Wei Fan. “Near-Field Ultra-Wideband mmwave Channel Characterization Using Successive CancellationBeamspaceUCA Algorithm”.In: IEEETransactionson VehicularTechnology 68.8 (2019),pp.7248–7259. [15] S L H Nguyen et al. “Comparing Radio Propagation Channels between 28 and 140 GHz Bands in a Shopping Mall”. In: 12th European Conference on Antennas and Propagation (2018). [16] T Niiho et al. “Multi-Channel Wireless LAN Distributed Antenna System Based on Radio-Over-Fiber Techniques”. In: The 17th Annual Meeting of the IEEE Lasers and Electro-Optics Society (2004), pp. 57–58. [17] R Sambaraju et al. “Impact of Channel Characteristics on the Performance of a 60 GHz Radio Over Fiber (RoF) System”. In: 2008 International Topical Meeting on Microwave Photonics (2011), pp. 157–160. [18] I Oppermann, J Talvitie, and D Hunter. “Wide-Band Wireless Local Loop Channel for Urban and Sub-Urban Environments at 2 GHz”. In: IEEE International Conference on Communications (1997), pp. 61–65. 119 [19] L Tarlazzi et al. “Characterization of an Interleaved F-DAS MIMO Indoor Propagation Channel”. In: Loughborough Antennas and Propagation Conference (LAPC) (2010), pp. 505–508. [20] Antenna. “W-band omnidirectional antenna”. In: Available: https://sftp.eravant.com/content/datasheets/SAO7531140230-10-S1.pdf (2025). [21] Antenna.“Flannstandardgainhornantenna27240-20”.In:Available:https://flann.com/products/antennas/standardgain-horns-series-240/ (2025). [22] N A Abasi et al. “Double Directional Channel Measurements for THz Communications in an Urban Environment”. In: IEEE International Conference on Communications (2020), pp. 1–6. [23] Yejian Lyu et al. “Large Virtual Antenna Array-Based Empirical Channel Characterization for Sub-THz Indoor Hall Scenarios”. In: IEEE Transactions on Antennas and Propagation (2023). [24] N Czink and C Mecklenbrauker. “A Novel Automatic Cluster Tracking Algorithm”. In: IEEE 17th International Symposium on Personal, Indoor and Mobile Radio Communications (2006), pp. 1–5. [25] L Greenstein, D Michelson, and V Erceg. “Moment-Method Estimation of the Ricean k-factor”. In: IEEE Communications Letters 3.6 (1999), pp. 175–176. [26] N A Abbasi et al. “THz Band Channel Measurements and Statistical Modeling for Urban D2D Environments”. In: IEEE Transactions on Wireless Communications 22.3 (2023), pp. 1466–1479. [27] 3GPP. “Study on Channel Model for Frequencies from 0.5 to 100 GHz”. In: 3GPP, Technical Report V17.0.0 (2022). [28] T S Rappaport and Y Xing et al. “Wireless Communications and Applications Above 100 GHz: Opportunities and Challenges for 6G and Beyond”. In: IEEE Access 7 (2019), pp. 78 729–78 757. [29] Z Ma and B Ai et al. “A Non-Stationary Geometry-Based MIMO Channel Model for Millimeter-Wave UAV Networks”. In: IEEE Journal on Selected Areas in Communications 39.10 (2021), pp. 2960–2974. [30] Z Huang and X Cheng. “A General 3D Space-Time Frequency Non-Stationary Model for 6G Channels”. In: IEEE Transactions on Wireless Communications 20.1 (2021), pp. 535–548. [31] Y Yuan and R He et al. “A 3D Geometry-Based THz Channel Model for 6G Ultra Massive MIMO Systems”. In: IEEE Transactions on Vehicular Technology 71.3 (2022), pp. 2251–2266. [32] AParssinenandMSAlouinietal.“WhitePaperOnRFEnabling6G–OpportunitiesandChallengesfromTechnology to Spectrum”. In: 6G Flagship Ecosystem https://www.6gchannel.com/items/6g-white-paper-rf-spectrum/ (2021). [33] J Zhang and P Tang et al. “Channel Measurements and Models for 6G: Current Status and Future Outlook”. In: Frontiers of Information Technology and Electronic Engineering 21.1 (2020), pp. 39–61. [34] Mikkel Bengtson, Yejian Lyu, and Wei Fan. “Long-range VNA-based channel sounder: Design and measurement validation at MmWave and sub-THz frequency bands”. In: China Communications 19.11 (2022), pp. 47–59. [35] S Priebe and M Kannicht et al. “Ultra Broadband Indoor Channel Measurements and Calibrated Ray Tracing Propagation Modeling at THz Frequencies”. In: Journal of Communications and Networks 15.6 (2023), pp. 547–558. [36] E M Vitucci and M Zoli et al. “Tri-Band mm-wave Directional Channel Measurements in Indoor Environment”. In: 29th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) (2018), pp. 205–209. [37] K Guan and B Peng et al. “Channel Characterization for Intrawagon Communication at 60 and 300 GHz Bands”. In: IEEE Transactions on Vehicular Technology 68.6 (2019), pp. 5193–5207. [38] Heng and S Sangodoyin et al. “THz Cluster-Based Modeling and Propagation Characterization in a Data Center Environment”. In: IEEE Access 8 (2020), pp. 56 544 –56 558. 120