Invited Article: Channel performance for indoor and outdoor terahertz wireless links

One of the most exciting future applications of terahertz technology is in the area of wireless communications. As 5G systems incorporating a standard for millimeter-wave wireless links approach commercial roll-out, it is becoming clear that even this new infrastructure will not be sufﬁcient to keep pace with the rapidly increasing globaldemandforbandwidth.Onefavorablesolutionthatisattractingincreasingatten-tionforsubsequentgenerationsofwirelesstechnologyistousehigherfrequencies, above 100 GHz. The implementation of such links will require signiﬁcant advances in hardware, algorithms, and architecture. Although numerous research groups are exploring aspects of this challenging problem, many basic questions remain unad-dressed. Here, we present an experimental effort to characterize THz wireless links in both indoor and outdoor environments. We report measurements at 100, 200, 300, and 400 GHz, using a link with a data rate of 1 Gbit/s. We demonstrate both line-of-sight and non-line-of-sight (specular reﬂection) links off of interior building walls. This work represents a ﬁrst step to establish the feasibility of using THz carrier waves for data transmission in diverse situations and environments. © 2018 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/). https://doi.org/10.1063/1.5014037


INTRODUCTION
Global mobile data traffic has seen explosive growth over the past few years and is predicted to reach to 49 exabytes per month by 2021, 1 with growth driven largely by the rise of the Internet-of-Things. 2 To accommodate this trend, wireless network infrastructure will need substantial upgrades beyond the capabilities of existing systems.Even the anticipated 5G roll-out will only provide a peak data throughput of a few Gbit/s, far short of the anticipated needs.As a result, many researchers are considering the possibility of employing carrier frequencies beyond 95 GHz, beyond the range where the US Federal Communications Commission (FCC) has well-established service rules.][5][6] Exploiting this spectral range is not easy; there are many hurdles to overcome, in almost every aspect of wireless technology.Of particular interest here are the channel characteristics, which in many situations are not well established.Clearly, at these higher frequencies, THz signals suffer significant power loss and degradation due to atmospheric attenuation and free space path loss.In predicting the performance of a channel which employs transmitter (Tx), receiver (Rx), and an intervening medium, the Friis formula is usually used, 3 with P t and P r as the transmitted and received power.G r and G t are the antenna gains for Rx and Tx, respectively.F is the normalized intensity pattern function due to scintillation effects with angles θ and φ referred to spherical coordinates, and e αd corresponds to the power attenuation factor due to both absorption and scattering in the atmosphere.ε p is the polarization coupling efficiency.λ and d are the signal wavelength and the link distance between Tx and Rx.The emergence of high power transmitters and high gain antennas will be critical for extending the link distance.In addition, the strong variation of the atmospheric attenuation with frequency dictates the use of certain frequency windows, in order to avoid the relatively narrow absorption peaks that are mostly due to water vapor.Figure 1 illustrates a survey of the existing literature on studies of THz links from the past 10 yr, indicating that nearly all of these studies have avoided frequencies near water vapor absorption lines (Ref.7).
The Friis formula provides important guidance about the feasibility of various link configurations in the THz range.According to Eq. ( 1), the received power is proportional to the gain of the transmitter and receiver antenna subsystems, which are each proportional to the ratio of the effective antenna area to λ 2 . 3As a result, for a given set of antennas, P r varies as (λd) 2 .This wavelength dependence reflects the decrease in the diffractive spreading of the propagating wave front, with increasing frequency.THz beams are much more directional than the signals in today's 4G and cellular systems, which operate at ∼100 times larger wavelength.This factor of 10 4 increase in received power due to directionality can compensate to some extent for atmospheric and scattering losses.However, these highly directional beams are much more susceptible to blockage by obstructions.5][56][57][58] Although numerous LOS link demonstrations have been described in the literature (see Fig. 1), the possibility of specular non-line-of-sight (NLOS) link pathways has been less well investigated.The idea of engineering specialized reflectors to facilitate this sort of specular NLOS link has been discussed, [59][60][61] and a few examples of NLOS links have been modeled 62 and demonstrated. 63The impact of surface roughness has also been considered, 54,57 as it could play an important role in degrading the energy transfer through a specular reflection.Yet, in the few cases where experimental measurements have been reported, most of the results have been discussed in terms of THz spectroscopy, not using modulated data to evaluate the data transfer performance.In many cases, a simple spectroscopic characterization is not enough; the evaluation of link performance using realistic data streams can provide new information about the capabilities and limitations of a given configuration. 64n this work, we describe characterization of both LOS and specular NLOS links, with direct comparisons of link performance for both indoor and outdoor environments.We performed our measurements using a continuous wave (CW) signal at four discrete frequencies: 100, 200, 300, and 400 GHz.Our THz source is based on a frequency multiplier chain (Virginia Diodes), which up-converts a modulated baseband signal to the desired output frequency, and radiates using a horn antenna to produce a highly directional vertically polarized signal.We used a pulse pattern generator to produce a pseudo-random bit pattern for modulating the carrier wave at 1 Gbit/s via amplitude shift keying (ASK) modulation.We detected the signal using a zero-bias Schottky diode, also coupled to a horn antenna.The Schottky signal is amplified to drive a bit-error-rate (BER) tester (Anritsu MP1764A) for real-time signal analysis.Both the Tx and Rx antennas are mounted at a height of 1.35 m above the ground, for all measurements.Their gains and half-power-beam-widths (HPBW) are shown in Table I.To improve the overall gain of the antenna subsystems, we used a transmissive (PTFE Teflon) lens in front of the horn antennas ( f = 7.5 cm) for both the Tx and Rx horns.We note that, when a directive antenna with HPBW smaller than 15 • is used, unintended multipath components are generally suppressed because any multipath coming from outside of the maximum gain direction of the antenna is attenuated by the antenna radiation pattern. 65Yet, as discussed below (Fig. 7), multipath effects can have an impact even in cases where the HPBW is small.

TH INDOOR WIRELESS LINK MEASUREMENTS
Figure 2 shows our experimental configuration used to explore the possibility of a specular NLOS link from a typical indoor wall, composed of painted cinderblock.The photo illustrates the Tx and Rx subassemblies, with PTFE lenses, directed toward a spot on the wall.The distance from the Tx/Rx antenna to the wall is a constant 1 m (d 1 = d 2 = 1 m).The antennas were adjusted to point to the wall at the same angle of incidence ( 50• ) for all frequencies, and a direct LOS free-space calibration was performed at each measured frequency at a Tx-Rx separation distance of 2 m (d 1 + d 2 = 2 m).To assure a linear behavior of the Schottky barrier diode (SBD), we used calibrated attenuators after the Tx antenna.We compare the results for three cases, in which the signal was reflected by the bare painted cinderblock wall, by a conformal metal foil attached on the wall, and by a smooth metal plate.In the first case, we expect the signal to be degraded by absorption in the paint and cinderblock material, as well as by diffuse random scattering from the rough surface. 54,66In the second case, the metal foil (which is thicker than the skin depth and is assumed to act essentially as a perfect metal) eliminates absorption losses in the underlying cinderblock surface.However, the scattering losses remain since the thin foil conforms to the rough shape of the cinderblock surface.In the third case, both scattering and absorption losses are essentially eliminated since the smooth metal plate is nearly a perfect reflector.Here, the size of foil and metal plate (around 26 cm × 35 cm) is much larger than the radius of first Fresnel zone (R), which is 3.9, 2.7, 2.2, and 1.9 cm for frequencies of 100, 200, 300, and 400 GHz.To measure the performance of the communication link, and ensure robustness against possible alignment error, we rotated and aligned the detector antenna to optimize the BER.
These optimal BER results are compared in Figs.2(b)-2(e), for four different carrier frequencies.The results are plotted as the log of the BER as a function of the output power from the transmitter, which is characterized using the attenuators in front of the Tx-module and the calibration data for the Schottky diode.For each carrier frequency, the curves display nearly identical slope on these log-log plots, as expected for a simple scattering process.The shifts in the curves indicate the extra transmitter power that is required to reach, e.g., error-free performance (BER = 10 9 ) due to the scattering and absorption losses.We note that these data are not corrected for the frequency-dependent detector responsivity (shown in Table I), which contributes significantly to the variation in the minimum power level required for error-free performance.Nevertheless, we can draw some interesting conclusions from comparisons among the various results.Most interestingly, at all frequencies, the effect of scattering from the rough surface (i.e., the difference between the black and red curves) is significantly smaller than the effect of absorption (the difference between the red and blue curves).
We also observe that the absorption losses increase moderately with frequency, as expected, rising from about 8 dB at 100 GHz to nearly 11 dB at 400 GHz.These data provide strong evidence that, contrary to most conventional wisdom, specular NLOS paths can realistically be used in indoor THz links even up to 400 GHz, with only moderate and manageable losses.In the case of a conventional painted cinderblock wall, these losses are dominated by absorption, not surface roughness scattering.
To further characterize the signal loss due to reflection, we varied the angle of incidence by mounting the Tx and Rx antennas on movable rails that pivot with respect to the reflection point on the wall, as shown in Fig. 3(a).We measured the BER performance at 5 different incident angles, from 20 • to 60 • .Here, the carrier is at 400 GHz, where we anticipate the maximum effect of random scattering due to surface roughness.As shown in Fig. 3(b), the curves shift to the left with increasing angle because of the smaller loss at larger angles.As earlier, the slope is invariant, as expected for a linear scattering process.
In order to elaborate upon these results, we used a Fresnel analysis.Scattering effects due to rough surfaces can be calculated by numerical simulation based on the Maxwell boundary value problem; 67 however, it has been shown that a simpler analytic approximation is generally suitable in THz scattering problems. 54In this model, a Rayleigh roughness factor (ρ) was added to the reflection coefficient (r) 68 as where ρ = e g/2 with g = (4π• σ • cos θ i /λ) 2 as a parameter to represent the effect of roughness.Here, σ represents the root mean square value of the roughness (deviation from perfectly flat).A rougher surface has a larger σ and therefore exhibits a larger value of g.This model assumes a Gaussian height distribution of the surface, as well as local smoothness (i.e., a long surface correlation length) so that the σ parameter is well defined; nevertheless, it is still an adequate approximation for our purposes.We used a value of σ = 0.06 mm in our model, which is close to values used in previous studies on similar surfaces. 54The Fresnel reflection coefficient r for a smooth surface can be calculated using the usual approach for vertical (p) polarization, with refraction angle θ t = arcsin(sin(θ i ) • Z/Z 0 ).Z 0 = 337 Ω is the wave impedance in free space and Z is the wave impedance for the material, where ε 0 and µ 0 represent the free space permittivity and permeability, f is the incident frequency, and c is the free-space velocity of light.We used α = 500 m 1 as the absorption coefficient of the wall material, which is an approximation since the wall is a multilayer (paint on top of cinderblock). 69The refractive index n is 2.09 as measured in Ref. 55.Using the same experimental setup, we obtained a reference measurement by placing a smooth metal reflector on the wall.If we assume that this smooth reflector is a perfect reflector as above, this allows us to deduce the power loss which results from the reflection off of the wall.We then compare this measured power loss to that calculated using the formalism described earlier.This result is shown as the inset in Fig. 3(b).The dashed curve in this inset shows the result when scattering is neglected (i.e., by setting σ = 0 in the calculation).We observe that neglecting scattering shifts the curve by just a few dB, which emphasizes again the dominant role of absorption in this NLOS link.The fact that scattering losses are relatively small in these measurements is somewhat counterintuitive, given that the surface roughness of the reflector is comparable to the wavelength.To clarify this result, we characterized the beam pattern in two different configurations: a LOS path and a specular NLOS path.In the two cases, illustrated in Figs.4(a) and 4(b), we mounted the receiver on a movable rail which permits it to translate along an axis perpendicular to the incoming beam's propagation direction.By scanning the detector along this line, we mapped the spatial distribution of the beam arriving at the receiver horn.The link path is 10 m in both cases, to facilitate comparison.As earlier, the detector angle was optimized by rotation to ensure accurate pointing, prior to each scan.In the NLOS case, the angle of incidence on the wall was 15 • .The measurement was performed for 100, 200, and 400 GHz, as shown in Figs.4(c)-4(h) with Fresnel zone radii of 8.6, 6.1, and 4.3 cm, respectively.We observe that, in all cases, the specular reflection from the rough surface introduces perturbations to the beam pattern which resemble interference fringes due to diffraction.However, in all cases, the beam power is sufficient to maintain a data link with very low BER, if the angle and position of the receiver are optimized.
In addition to the above results, we have explored a number of other arrangements for indoor THz links.Figure 5 illustrates a couple of examples, which demonstrate the versatility of the channel in realistic configurations.Figures 5(a) and 5(b) show a 30 m link at 200 GHz, with a single reflection at near-normal incidence (∼5 • ).As earlier, we compare reflection from a bare smooth wall to that from a conformal metal foil, which eliminates absorption but preserves the random scattering due to surface roughness effects.We observe that absorption imposes a ∼8 dB penalty in this configuration.Again, error-free performance can be obtained even over this range.Figures 5(c) and 5(d) show a similar configuration, except that this time the beam reflects twice-once off of a (painted metal) door and a second time off of the painted cinderblock wall as in Fig. 2(a).The total link path is ∼35 m in this example.In Fig. 5(d), we compare the BER performance of this link with a single reflection (off of the painted wall) to the double reflection (door + wall).We see little change in the link performance vs. received power, emphasizing again the surprising robustness of these data links.
Finally, we demonstrate that an NLOS link can be established even when the LOS path is fully blocked, as illustrated in Fig. 6(a).Here, the signal undergoes two specular NLOS reflections from the same sort of painted cinderblock wall as described in Fig. 2(a), with a total path length of 5.5 m (d 1 + d 2 + d 3 = 5.5 m).The Tx and Rx antennas cannot see each other, so no LOS path can be measured.The Tx and Rx antennas were directed so that both reflections have the same 45 • angle of incidence.Figure 6(b) shows the BER performance for this link, at 100 and 200 GHz carrier frequencies.Here, as earlier, the signals are not corrected for the different detector responsivities (which differ by about 4 dB in this case), so the shift between these two curves cannot be attributed entirely to the attenuation of the beam due to the two reflections.Even so, this result demonstrates the feasibility of robust data links employing higher-order scattering.

TH OUTDOOR WIRELESS LINK MEASUREMENTS
For indoor wireless communications, multipath effects are often inevitable over long distances in confined spaces.But for outdoor measurements in open environments, the LOS path can be more predominant.1][72][73][74][75] For higher frequencies with more directional antennas, this phenomenon would be expected to be even more pronounced.Even so, specular-type NLOS scattering can interfere with LOS links even at higher frequencies, especially in the case of grazing incidence from the ground. 76e performed preliminary outdoor measurements by transferring the experimental apparatus outside. 9,36These measurements have been performed under the authority of an experimental license issued by the Federal Communications Commission (FCC), which covers outdoor operation of a transmitter at any of the frequency bands described here.Both the transmitter and receiver were mounted on tripods for easy repositioning, with the same beam height of 1.35 m.To extend the link range, we used a double-pass configuration as shown in Fig. 7, where the beam was reflected from a smooth metal plate (90 cm × 120 cm) situated at a distance.Calculation predicts a much smaller radius of the first Fresnel zone (27.4,19.4,15.8, and 13.7 cm for carrier at 100, 200, 300, and 400 GHz) compared to the plate size for a link distance of 100 m.That is, the reflecting surface is sufficiently large to cover the first Fresnel zone.Measured BER performance is shown in Fig. 7(c) for propagation over a grass surface and in Fig. 7(d) for operation over a concrete sidewalk.The environmental conditions for these measurements were a temperature of 16-19 • C, humidity of 50%-70%, and a variable wind velocity of 5-14 MPH.These plots display the BER as a function of Tx-to-Rx path distance and are not corrected for either the transmitter power or the receiver sensitivity, which is why the results do not vary in a monotonic fashion with frequency.Using the known transmitter power and receiver sensitivity (see Table I), we can compute the maximum transmission distance at each frequency, from Eq. ( 1).The results are comparable to our measured distances but somewhat larger; specifically, predicted distances are 140, 210, 35, and 35 m for 100, 200, 300, and 400 GHz, respectively.The discrepancies with the measurements arise from several factors, including the effect of the movement of the metal reflecting plate due to wind gusts, and to multipath effects.This latter effect can be seen more clearly through a comparison of the experimental results from the two different surfaces.Significant differences result from interference of the direct LOS path with a specular NLOS path associated with reflection from the ground.This interference signal is sensitive to the nature of the ground; for example, it has a greater impact on the BER when the ground is a solid concrete surface and less effect on a grassy surface where the absorption is likely to be much higher.As a result, the distances at which error-free performance is observed are larger for the grassy surface.
From these data, we can extract the path loss at different distances.According to Eq. ( 1), and neglecting multipath effects, the square root of the path loss (i.e., √ P r /P t ) should be proportional to fd, where f is the carrier frequency and d is the path length.In Figs.7(c) and 7(d), we plot this square root vs. fd.We find that the data fall approximately on a common line, for all frequencies, as anticipated.The slope of this line incorporates all sources of loss, including interference from multiple paths, atmospheric absorption, and scintillation due to atmospheric turbulence.As anticipated, the slope in the case of propagation over pavement is slightly larger than that for the case of propagation over a grassy surface, by about 26%, due to the extra interference from the ground reflection.The effect of multipath interference, due to ground reflections, can lead to a change in the slope, especially at longer propagation distances. 77A more quantitative analysis of the effects of fading on these signals will be forthcoming in future work.

CONCLUSIONS
We present a series of link measurements at several discrete frequencies over a wide spectral range, under a number of different realistic conditions.We observe that indoor links are quite robust, even when one or two specular NLOS reflections are included.Scattering losses due to surface roughness are relatively low since their effect is primarily to introduce diffractive effects.As a result, a good BER performance can still be achieved if the receiver direction can be tuned.We demonstrate a link in which there is no line-of-sight path from the transmitter to the receiver, which suggests the possibility that high quality links can be established even with an obstruction in the beam path.For outdoor measurements, we show that interference from unintentional NLOS paths can limit the BER performance and that this effect is more pronounced on a dry relatively flat surface such as pavement, when compared with a rougher and wetter grassy surface.Although our results have been acquired using a fairly low data rate (1 Gbit/s), we believe that the general conclusions remain applicable even for the much higher data rates anticipated for wireless systems operating in the THz range.These results suggest many optimistic possibilities for future system architectures.

FIG. 1 . 3 Ma
FIG. 1. (Top) Distance versus carrier frequency obtained for indoor and outdoor wireless communication systems.The numbers in brackets indicate the reference from which the data point was taken; 8-50 the data points circled in red indicate the results from this paper.(Bottom) Impact of atmospheric attenuation of THz waves under different humidity with humidity from 60% to 100%.[51][52][53]

FIG. 2 .
FIG. 2. (a) Photo of the 2 m distance link and of the modified wall conditions used in these measurements.[(b)-(e)] Log(BER) vs. transmitter output power when the signal is reflected by a bare painted cinderblock wall (blue curves), a conformal metal foil attached to the wall (red curves), and a smooth metal plate (black curves), at the frequencies shown.

FIG. 3 . 6 Ma
FIG. 3. (a) Configuration of the 2 m distance link with movable rails for changing the angle of incidence; (b) log(BER) vs. transmitter output power for different incident angles, using a carrier frequency of 400 GHz.The inset shows the measured (stars) and computed (solid curve) power loss (relative to a smooth metal mirror reflector).The dashed curve in the inset shows the predicted result when scattering losses are neglected.

FIG. 4 .
FIG. 4. Configurations for comparison of (a) LOS and (b) specular NLOS links.Both links have the same 10 m distance, to facilitate comparison.[(c)-(h)] Received power and BER pattern measurements at different carrier frequencies for LOS (black data curves) and specular NLOS (blue curves) links.

051601- 8 Ma
FIG. 5. (a) Photo of the 30 m link at 200 GHz (Fresnel zone radius r = 11 cm) with a single near-normal-incidence reflection.(b) Log(BER) vs. transmitter output power when the signal is reflected by the bare wall (black curve) and by a conformal metal foil (26 cm × 26 cm) attached to the wall (blue curve); (c) photo of a 35 m link at 200 GHz (R = 11.5 cm) with two specular reflections: one from the (painted metal) door and a second from the wall.(d) A comparison of the BER performance for single and double reflection as a function of received power, showing that they are nearly identical.

FIG. 7 .
FIG. 7. Photo of measurement site on (a) lawn and (b) sidewalk; the BER performance with respect to link distance on (c) lawn and (d) sidewalk, for 100 (black), 200 (blue), 300 (red), and 400 (green) GHz carrier frequencies.Inset: square root of path loss vs. the product of distance and frequency, which shows a linear relation, as anticipated.

TABLE I .
Terahertz wireless communication system specifications.