95 GHz Sub- THz Multipath Propagation Measurement for Indoor Conference Room Desktop

This study conducts a wideband multi-path propagation measurement at the 95 GHz sub-terahertz band for short-range communication in a conference room desktop scenario. Regardless of the fact that the current 3rd generation partnership (3GPP) stochastic channel model (SCM) targets the frequency up to 100 GHz for various scenarios, neither detailed measurements at the 95 GHz band nor a compatible channel modeling/generation framework for indoor short-range communication scenarios have been conducted. To fill these voids, based on a real-world measurement at 95 GHz with a bandwidth of 4 GHz, this study analyzes the multi-path propagation characteristics and yields the following insights for developing a 3GPP SCM-compatible channel generation framework at this band. First, the exponential power decay with delay time and quasi-uniform azimuth angles of arrival (AAoAs) are observed, which should be revisited to develop a channel generation framework. Secondly, distribution models for root mean squared (RMS) delay/AAoA spreads and omnidirectional path loss model are derived, which serves as a foundation for developing a channel generation framework at this band. Moreover, these established models are compared with the recently conducted measurement results at the 60 GHz band in the same scenario, shedding light on the hypothesis that the models for these parameters at the 60 GHz can be generalized for the 95 GHz band.


I. INTRODUCTION
As the beyond fifth-generation (B5G) or sixth-generation (6G) mobile networks are envisioned as a platform involving not only humans but also autonomous systems, a huge demand for extremely high-rate and low-latency communications will be inevitable [1].For this background, exploring the subterahertz (THz) spectrum has attracted huge attention in both academic and industrial communities.The sub-THz band is generally referred to as a spectrum ranging from 100 GHz to 300 GHz [2] and offers an unprecedented amount of bandwidth relative to the sub-6 band ranging from approximately 3 GHz to 6 GHz and even millimeter (mmWave) band ranging from 30 GHz to 100 GHz.
To draw a guideline for designing wireless communication systems at sub-THz bands, many propagation measurements have been conducted to model multi-path arrival characteristics.For example, the measurement campaigns within the frequency range from 110 GHz to 220 GHz are exemplified by [3][4][5][6][7][8][9], where the path loss, large-scale parameters (e.g., root-meansquared (RMS) delay and angular spreads, k-factor) were measured, and several channel generative models were established.The measurements for the frequency range above 220 GHz up to 300 GHz and more are exemplified by [9][10][11][12][13][14][15], where the measurements particularly focused on indoor scenarios.Therein, the path-loss model, large-scale parameter model, and other multi-path characteristics have been derived based on real-world measurements.
However, regardless of the abundant measurement campaigns, to the best of our knowledge, few studies investigated the multi-path characteristics at the boundary between the mmWave and sub-THz spectrums, that is, the frequency range from 90 GHz to 100 GHz.The multipath characterization on this frequency band is also of importance in the sense of bridging the gap between mmWave and sub-THz bands.Moreover, as the dense deployment of terminal nodes is envisioned in the B5G and 6G, the usage of short-range internode communication will grow significance.For these joint backgrounds, this study serves as the first work that investigates the multi-path propagation characteristics in a short-range communication use case.It should be noted that exceptionally, the work in [7] conducted several measurements at 97 GHz.However, this measurement analyzes electromagnetic characteristics such as diffuse scattering and diffraction and lacks the overall multi-path characterization, particularly in terms of multi-path delay and angular metrics.
Indeed, the 3rd generation partner project (3GPP) technical report 38.901 [16] states that they developed a stochastic channel model (SCM) for the frequency range up to 100 GHz; hence, one can argue that the channel modeling of 90 GHz bands has already been completed.However, the 3GPP 38.901 SCM is not based on the real-world measurement at the 90-100 GHz band, where only a ray-tracing simulation was conducted for this band [17].Moreover, a past European channel measurement/modeling activity termed mmMAGIC also established a 3GPP 38.901 SCM-compatible channel model [18]; however, there are no measurement campaigns reported in this activity.These facts naturally lead us to the hypothesis that there are no 3GPP 38.901 SCM-compatible multi-path channel models verified by real-world measurements; hence, multi-path propagation measurements at this band are required to smoothly guide the 3GPP SCM to the sub-THz bands.
Motivated by this background, we conducted an indoor wideband channel measurement at a 95 GHz sub-THz1 band focusing on short-range communications.The measurement has been conducted with a bandwidth of 4 GHz in a small fact that the 28 GHz band is generally termed "mmWave" even though the mmWaves start from 30 GHz technically.conference room as in the first standardized mmWave wireless personal area network [19], and power angular delay profiles (PADPs) were measured at several RX locations.In this paper, we report the PADPs, omnidirectional power delay profiles (PDPs), and power angular profiles (PAPs); thereby drawing a guideline to develop a 3GPP 38.901 SCM-like multi-path channel model.Moreover, we derive the RMS delay spread model, RMS angular spread model, and path loss model, and we compare these results with a recently developed 3GPP 38.901 SCM-compatible model [20,21] in the same scenario at the 60 GHz band.Our results suggest that these parameter models at the 95 GHz band possess the affinity for those at the 60 GHz band; shedding light on the generalizability of the 60 GHz band for the 95 GHz sub-THz bands.

A. Setup of Channel Sounding System
We developed a channel-sounding system that can measure PDPs in the 95 GHz band.In this system, a sounding signal is transmitted with an omnidirectional antenna, and this transmitted sounding signal is scanned by a rotated receiver (RX) antenna as shown in Fig. 1.The RX was equipped with a 25 dBi horn antenna rotatable in the azimuth plane.The transmitter (TX) was equipped with an omnidirectional antenna in the azimuth plane.As a sounding technique to obtain the PDPs, we applied a Keysight channel-sounding system [22], which is one of the time correlation-based channel-sounding methods.It should be noted that considering the real-world use cases of a conference room desktop scenario, we simplified the measurement system in the sense of 1) limiting the measurement in the azimuth plane and 2) rotating only RX antennas while applying the omnidirectional antenna in the TX side.In the real-world use case of this conference desktop, the communication is performed in the azimuth place, and applying the double-directional antennas may not be required owing to its short communication range.Hence, this measurement suffices to acquire insight into the real-world use cases for the conference room desktop environment.
The overall procedure to deal with the sounding signals is performed similarly to the well-known super-heterodyne TX and RX, where a sounding signal possesses intermediate frequency (IF) in the TX and RX circuits.Firstly, the signal generator, which is an arbitrary waveform generator with 64 GS/s, directly generates a sounding signal with a center frequency of 6 GHz and a bandwidth of 4 GHz.At this point, this sounding signal exhibits the IF of 6.0 GHz, which modulates a pseudo-noise sequence to be used to calculate a PDP at the RX.Afterward, the sounding signal is up-converted to the radio frequency of 95.8 GHz and is received in the RX.The received signal is down-converted to the IF of 3.8 GHz and is sampled in an oscilloscope.At the same time, the oscilloscope calculates the PDPs based on the time-correlationbased sounding technique.The detailed parameters of this measurement are shown in Table I.
Before the field measurement, we conducted a channel sounding in an anechoic room to investigate the quantitative performance for resolving arrived multi-paths.In the anechoic room, we measured the PDP at the TX-RX distance of 1m while rotating the RX horn antenna with the angular step of 5°.Fig. 2 shows the picture of the anechoic measurement and obtained a PDP.In Fig. 2(b), the delay time is aligned so that the power peak of the line-of-sight (LoS) ray is at 0 ns, and in Fig. 2(c), the angle of the RX horn antenna is aligned so that the power peak of the LoS ray is at 0°.From Fig. 2(b), we find no spurious power peaks above −25 dB relative to the LoS peak in the delay time longer than 1.5 ns.It should be noted that the value of −25  dB will used for the acceptance threshold for multi-path components in the subsequent analysis.This PDP characteristic implies that arrived multi-paths separated for over 1.5 ns in delay time (i.e., 45 cm difference in path length) can be resolved in this measurement system.Moreover, from Fig. 2(c), we cannot find any spurious power peaks above −25 dB relative to the LoS peak in the RX horn antenna rotation angle larger than 20°, implying that arrived multi-paths separated for over 20° can be completely resolvable.

B. Measurement Environments for Conference Room Desktop
As shown in Fig. 3, the measurement is subsequently conducted in a typical conference room desktop environment, which was one of the real use case scenarios considered in the standardization activity of mmWave communications (e.g., the IEEE 802.15.3c [19] and IEEE 802.11ad [23]).The desks were placed in a rectangular shape, and the TX was located on one of the desks close to the glass window.As shown in Fig. 3(b), the measurement was conducted at 8 RX locations named RX1-RX8 where the TX-RX distances were 1.00 m, 1.53 m, 2.00 m, 2.63m, 3.00 m, 3.40 m, 3.64 m, and 4.00 m, respectively.The TX and RX were located at the same height of 0.19 m above the surface of the desk.
At each RX location, the PDPs are recorded while rotating the RX horn antenna.The RX horn antenna was rotated from −180° to 175° with the angular step of 5° where the angle of 0° was set to be the direction of the TX.Hence, for each RX location, 72 PDP records were obtained.This set of PDPs is referred to as PADP, which is written as: where  RX and  (& RX ) denote the RX rotation angle and delay time at the RX rotation angle  RX , respectively.As we rotated the RX horn antenna from −180° to 175° with the angular step of 5°,  RX ∈ {−180°+ 5 |  = 0,1, . ..71}.Moreover, in our measurement system,  (" RX ) is set such that for each RX rotation angle,  (" RX ) = 0 ns when the maximum power is observed in the recorded PDP; hence, (0 ns,  RX ) ≥ $ (" RX ) ,  RX ) for all  (" RX ) and  RX .In this measurement, the PDP was obtained in the time step of 0.164 ns with 4098 samples, where  (" RX ) = 0 ns is located at the center of the PDPs; hence,  (" RX ) ∈ { × 0.164 ns |  = −2048, −2047, . . ., 2046, 2047}.

C. Time Alignment of PDPs in Different RX Antenna Rotation
Angles and Omnidirectional PDP Estimation The aforementioned measurement system was conducted with the lack of information on the time of arrival (ToA) of each multi-path component observed by the antenna rotations.This sometimes occurs in the time correlation-based channel sounding system [24], which gives rise to the problem that we can acquire neither omnidirectional PDPs nor the RMS delay spread.
A motivating example to identify the problem is shown in Fig. 4. Let us consider a simple scenario where two PDP records are obtained in the measurement in different two RX antenna rotation angles.Therein, one of the RX rotation angles is 0°, capturing the LoS ray whereas the other RX rotation angle of − 60 ° captures another multi-path component.In our measurement system, as shown in Fig. 4(a), the maximum power peak is forced to 0 ns regardless of the fact that at least two strong components (i.e., LoS ray and the multi-path component) arrive in the different absolute time if they were received with omnidirectional RX antenna.This problem prohibits us from calculating the RMS delay spread.Hence, as shown in Fig. 4(b), an appropriate ToA of the multi-path relative to the LoS ray is estimated, and thereby, the recorded PADP should be aligned so that the omnidirectional PDP can be obtained from the recorded PDPs.
To this end, we applied the time alignment algorithm of the PDPs that has been recently proposed in [25] and well validated in the same desktop scenario.Briefly, this algorithm leverages the correlation of the PDPs obtained in the adjacent RX rotation angles to estimate the ToA of the multi-path component that gave rise to the maximum power peak in each PDP.In this paper, we do not detail this time alignment algorithm because our focus is on the 95 GHz channel propagation measurement, and for more details of the algorithm, readers are encouraged to consult [25].Hence, we limit the discussion to how the omnidirectional PDP can be obtained given the estimated ToAs.Let us consider that the ToA of the multi-path component that caused the maximum power peak in the PDP for each RX antenna rotation angle  RX is estimated as  B '() (" RX ) .Then, the time-aligned PADP for each  RX , denoted by  *+,-.(,  RX ), is mathematically expressed as follows: Notably, in the time-aligned PADP, the dependency of the delay time on the angle  RX does not appear, which exactly means that the delay time is aligned so that the 0 ns implies the  Omnidirectional PDP Analysis.Fig. 6 shows the omnidirectional PDPs that were calculated by eq. ( 1) for RX1, RX3, RX5, and RX7, where the power peaks above the threshold of −25 dB are manually marked.Several effective multi-path components are observed, which suggests the necessity to model the power and delay-domain characteristics in these multi-path components.In addition, from Fig. 6, the overall characteristics of the power decay with respect to the delay time can be found, suggesting that the exponential power decay model that is already applied in the 3GPP 38.901 [16] to generate cluster power is also applicable to this scenario.PAP Analysis.Fig. 7 shows the polar plot of the power angular profiles (PAPs) that are calculated by: E  *+,-.(,  RX ), 0 where the summation is taken over all observations in the delay domain.As was performed in the omnidirectional PDPs, the power peaks except for the peak of the LoS ray above the threshold of −25 dB are manually marked.From Fig. 7, we can find effective peaks above the threshold.Revisiting the standardized 60 GHz indoor systems, these multi-path components are valuable sources to maintain connectivity against the LoS blockage events; hence, modeling the AAoAs of these components is required to design the communication systems operated in this 95 GHz band.In addition, Fig. 7 leads to the hypothesis that these multi-path components can arrive within the full range of the angles, that is, within   Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
[ − 180 °, + 180 °].As was recently shown in [20], this characteristic cannot be modeled with the 3GPP 38.901 model.Hence, to model the AAoA characteristics, modification of the 3GPP 38.901 SCM is required.

B. RMS Delay and Angular Spreads RX Location-Wise Spread Values.
Table II summarizes the RMS delay and AAoA spreads that are calculated by the omnidirectional PDPs and PAPs respectively.For both calculations, the threshold of −25 dB relative to the LoS ray peak was applied, implying that only power samples above the threshold were used for the calculation.As for AAoA spread, we report these values as RMS AAoA spread because the antenna rotation was in the azimuth plane and was done on the RX sides.From Table II, a dependency of these spread values on the TX-RX distance is partially found, where the RMS delay spread tends to be larger for an increase in the TX-RX distance up to 3 m.Moreover, the RMS AAoA spread tends to increase as the TX-RX distance increases.This increase in the RMS delay and AAoA spread for the TX-RX distance can be attributed to the power of reflective multi-path components becoming more comparable with the LoS ray as the TX-RX distance increases.
Notably, the tendency of increasing RMS delay spread cannot be found for the TX-RX distance over 3 m (i.e., RX locations RX6-RX8), and rather, the RMS delay spreads at these locations are lower than that at RX5.This observation can be attributed to the RX locations being close to the corner of the conference room as shown in Fig. 3.This fact implies that the delay of the multi-path component can arrive in a shorter time relative to the LoS ray, resulting in a lower RMS delay spread relative to that at RX5.These types of relationships between the RMS spread values and TX-RX distances are not modeled in the current 3GPP 38.901 SCM; hence, this point should be included for accurate channel modeling and generation.

Distribution of RMS Spread Values.
Fig. 8 compares the distributions of these RMS delay/AAoA spread values with those in the 3GPP 38.901 technical report at the center frequency of 95.8 GHz in an indoor office hotspot scenario and those in the recently proposed model for the same desktop scenario at 60 GHz [20].In our measurement, the smaller RMS delay spread can be found relative to that in the 3GPP 38.901 report, which is attributed to a smaller room size in the measurement.In more concrete, the room size in our measurement was approximately 8 m × 6 m while in 3GPP, the office was considered to be 120 m × 50 m.As for the RMS AAoA spread, a smaller standard deviation was observed relative to that in the 3GPP report, which may be attributed to the shorter TX-RX distance than those considered in the 3GPP 38.901 report.Hence, a new distribution model for the delay and AAoA spread is necessary for this conference room desktop scenario.
Notably, the distributions of RMS delay/AAoA spread values found in our 95 GHz measurement results are closer to the recently proposed 60 GHz channel model for a similar conference room desktop environment [20].Hence, we can conclude that the room size and usage scenarios are the dominant factors for RMS delay/AAoA spread values.Moreover, from these comparison analyses, we can at least hypothesize that the 60 GHz channel model can be generalized for the 95 GHz band because both RMS delay/AAoA spread values distribute similarly.This hypothesis should be carefully validated, which is deferred to our future work.

C. Omnidirectional Path Loss Model
Finally, with the measurement results, we derived a path loss model based on a well-accepted close-in free-space reference distance path loss model [21,26].The path loss model is formulated as follows: where PL(, ) and FSPL( 1 , ) denote the path loss in the decibel scale at distance  and at center frequency  in gigahertz and the free-space path loss in the decibel scale at reference distance  1 = 1 m, respectively.The terms  and χ 3 ∼ (0, σ 4 ) denote the path loss exponent and the shadow fading term, respectively, which are found by a least-squared error fitting.Noticeably, the 3GPP 38.301 also uses this model for indoor scenarios with  = 1.7 and σ = 3 dB as the LoS indoor hotspot (InH)-office model; hence, we can compare the path loss model found by this measurement with the 3GPP model.The detailed algorithm to extract these parameters is found in [21].Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
Fig. 9 shows the path loss model extracted from the measurement data.From Fig. 9, we can find a larger path loss exponent than the free-space path loss model, that is,  = 1.8.This path loss exponent is rather closer to the 3GPP LoS InHoffice model.Moreover, a similar path loss exponent was observed in the 60 GHz and 105 GHz [21], which strongly validates that the path loss exponent ranging from 1.7 to 1.8 can be reusable for the 95 GHz band.Meanwhile, a smaller power dispersion is observed relative to the 3GPP LoS InH-office model, and this characteristic is observed also in the 60 GHz band.Hence, this smaller power dispersion should be newly added for the conference room desktop scenario.

IV. CONCLUSION
This paper reported an indoor wideband multi-path measurement at 95 GHz band.The measurement was conducted in a small conference room envisioning a desktop scenario.The PADPs, omnidirectional PDPs, and PAPs were obtained, and several guidelines to statistically generate a similar channel characteristic were found.Moreover, we modeled a distribution of RMS delay/angular spreads and path loss and compared them with the recently established model in the same scenario at the 60 GHz band, suggesting the generalizability of the 60 GHz band model to the 95 GHz sub-THz band.

Fig. 3 .
Fig. 3. Floor plan and picture of the measurement environment.The measurement was conducted at eight RX locations named RX1-RX8, where the TX-RX locations are 1.00 m, 1.53 m, 2.00 m, 2.63m, 3.00 m, 3.40 m, 3.64 m, and 4.00 m, respectively.

Fig. 4 .
Fig. 4. Motivating example for time alignment.(a) PADP obtained in our measurement (b) Time-aligned PADP, which should be obtained for analysis.

Fig. 7 .
Fig. 7. PAP obtained in the measurement with RX horn antenna rotations.

TABLE I .
PARAMETERS IN 95 GHZ BAND MEASUREMENT Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

TABLE II .
MEASURED RMS DELAY AND ANGULAR SPREADS