Cellular Wireless Networks in the Upper Mid-Band

The upper mid-band - roughly from 7 to 24 GHz - has attracted considerable recent interest for new cellular services. This frequency range has vastly more spectrum than the highly congested bands below 7 GHz while offering more favorable propagation and coverage than the millimeter wave (mmWave) frequencies. The upper mid-band can thus provide a powerful and complementary frequency range to balance coverage and capacity. Realizing the full potential of these bands, however, will require fundamental changes to the design of cellular systems. Most importantly, spectrum will likely need to be shared with incumbents including communication satellites, military RADAR, and radio astronomy. Also, the upper mid-band is simply a vast frequency range. Due to this wide bandwidth, combined with the directional nature of transmission and intermittent occupancy of incumbents, cellular systems will need to be agile to sense and intelligently use large spatial and frequency degrees of freedom. This paper attempts to provide an initial assessment of the feasibility and potential gains of wideband cellular systems operating in the upper mid-band. The study includes: (1) a system study to assess potential gains of multi-band systems in a representative dense urban environment and illustrate the value of wide band system with dynamic frequency selectivity; (2) an evaluation of potential cross interference between satellites and terrestrial cellular services and interference nulling to reduce that interference; and (3) design and evaluation of a compact multi-band antenna array structure. Leveraging these preliminary results, we identify potential future research directions to realize next-generation systems in these frequencies.


I. Introduction
Cellular wireless systems up to the fourth generation (4G) had largely operated in a range of microwave frequencies below 6 GHz.Given the severe spectral shortage in these bands, 5G systems [1], [2] introduced new capabilities in the millimeter wave (mmWave) frequencies above 24 GHz [3]- [5].The wide bandwidths available in the mmWave range, combined with spatial multiplexing capabilities, have enabled massive multi-Gbps peak rates.However, extensive measurements also now show that practical performance is often intermittent [6], [7] with limited penetration inside [8], [9].At root, this poor coverage is due to the inherently limited range of mmWave signals and their high susceptibility to blockage [10]- [14].
Against this backdrop, the upper mid-band spectrum, roughly from 7-24 GHz, is being considered to provide a good balance of coverage and bandwidth, overcoming the spectral shortage of the sub-6 bands while having favorable propagation and penetration relative to the mmWave frequencies.For these reasons, the band has recently attracted considerable attention for commercial cellular communications and has been cited by industry as a leading candidate spectrum for next generation (NextG, 5G and beyond) wireless networks [15]- [18].The Federal Communications Commission (FCC) Technical Advisory Committee (TAC) has recently identified the upper mid-band as vital for meeting the growing data rate demands [19], [20].The 3rd Generation Partnership Project (3GPP), the organization that sets cellular standards, has also begun study of these bands [21] -see Section II for more details.
In 3GPP, the upper mid-band is referred to as Frequency Range 3 (FR3).FR1 initially referred to the traditional spectrum below 6 GHz, which was later expanded to 7.125 GHz [22].FR2 refers to the mmWave spectrum that was introduced in the fifth generation (5G) standard.

A. Challenges
Development of cellular services in the full range of these bands faces significant challenges: • Need for sensing and spectrum sharing with incumbents: Most importantly, these bands are already used by several vital incumbent services.This situation stands in contrast to the mmWave deployments where the bands were significantly less occupied.As we review in Section C, commercial satellite services already widely use these bands and have been interested in increased bandwidth allocations, particularly for broadband rural access.The upper mid-band is also vital for military and commercial RADAR and contains scientifically important spectral lines that are fundamental for radio astronomy.
• Large spectral range: The upper mid-band is simply a vast range of spectrum -well beyond the span of most commercial cellular front-ends that typically operate over a small percentage of the carrier frequency.This large range, combined with the need to dynamically share spectrum and directionally transmit, demands new front-ends that adaptively and intelligently sense and exploit large bandwidth and spatial degrees of freedom.• Spectral resiliency: Sensing capabilities, combined with the ability to adapt in space and frequency, can also yield improved defenses to external attacks from jammers and signal disruption.Such features can provide robust protection for vital future cellular infrastructure.

B. Contributions
The broad goal of this paper is to assess the feasibility and potential of agile wideband cellular systems operating in these bands.We will also identify key challenges and areas for research.Towards this end, our contributions are: • Channel modeling and estimation of multi-band coverage: We briefly review channel measurements in the upper mid-band range and discuss open areas of research (Section A).Also, based on current propagation models, we conduct a ray-tracing simulation in a representative dense urban area (New York City) to assess the coverage and capacity of a potential cellular system in the upper mid-band (Section B).Our results demonstrate that mobiles that can adaptively select one of multiple bands across the upper mid-band can improve coverage and data rates over mobiles restricted to a single band.This finding argues for multi-band, adaptive systems to gain the full benefits of the frequency range.• Interference with incumbents: We perform an additional ray tracing simulation (Section B) to assess the potential interference of terrestrial cellular services on the satellite uplink.The analysis shows that interference from both user equipment (UEs) and base stations (BSs) can be significant, but directional nulling can be effective in mitigating the interference with some loss in the terrestrial network capacity.We also discuss some potential interference issues with passive radio astronomy in the upper mid-band.• Wideband antenna systems: Finally, we present (Section V) a compact, multi-band antenna array to demonstrate the feasibility of front-ends that operate over the entire upper mid-band in a small form factor.

A. Prior Developments in the Sub-6 and mmWave Bands
The interest in the upper mid-band has to be seen in the context of developments of commercial cellular systems in both the sub-7 and mmWave frequency ranges over the last decade.Early mmWave experiments and capacity analyses such as [3], [4], [23], [24] suggested the possibility of massive data rates in micro-cellular deployments owing to the wide bandwidths and spatial multiplexing gains available in the mmWave frequencies.Based partially on these and other results, the mmWave bands emerged as an integral component of the 5G New Radio (NR) specification [1], [2].Commercial mmWave deployments appeared shortly after the release of the specification, particularly in the US [5].
Coverage at reasonable cell densities in these deployments, however, has been an enormous challenge.Several recent measurements in commercial mmWave networks now clearly demonstrate that outdoor coverage can be highly intermittent [6], [7] with limited penetration indoors [8], [9].MmWave signals are simply blocked by many common building materials such as concrete [10], [14] as well as the human body and other obstacles [11]- [13].See also Section D below.
Parallel to the developments in the mmWave, significant spectrum was also released in portions of the mid-band frequencies above 3 GHz, including the Citizens Broadband Radio Service (CBRS) band (3.55 to 3.7 GHz) [25] and the C-band [26] (3.7 to 4.2 GHz).These bands were instrumental for both so-called private 5G networks [27] as well as spectrum expansion in many wide area public networks.More recently, there has also been significant discussion of unlicensed bands from approximately 6 to 7 GHz, extending the mid-band further [22].In particular, both wireless LAN and cellular services have been considering enhancements to operate in these bands [28], [29].
Measurements from commercial network monitoring companies, now offer an opportunity to compare the practical performance of these networks.On the one hand, Ookla reported in 2022 [30] that the tested mmWave networks offer an incredible 1.6 Gbps median downlink throughput, up to seven times higher than systems in C-band.Similarly, in 2021, Open Signal demonstrated [31] that Verizon's mmWave downlink capacity was at least three times as high as networks relying mostly on mid-band spectrum.Nevertheless, networks with mid-band alone often provided downlink speeds in the hundreds of Mbps.Moreover, the coverage for mid-band networks was much more uniform than mmWave.For example, Open Signal's study [31] showed that users connect to mmWave less than 1% of the time.

B. Recent Standardization Efforts in the Upper Mid-band
The relative success of commercial cellular systems in the mid-band, combined with the high data rates, but intermittent coverage, of mmWave has set the stage for interest in the upper mid-band.The hope is to provide the high data rates close to those available in the mmWave range, but with much more uniform coverage.As mentioned in the Introduction, the bands have been identified by industry [15], [16] and 3GPP has recently formally started study of services in the 7 to 24 GHz band [17], [21], where they are called frequency range 3 or FR3.Note that, in 3GPP terminology, frequency range 1, or FR1, has been extended from sub-6 GHz to sub-7 GHz to include the frequencies in the 6 to 7 GHz range in FR1 [32].
In 2022, the FCC also began discussion on two 500 MHz bands from 12.2-12.7GHz [33], and 12.7-13.2GHz [34] for possible cellular use.While the first band has now been rejected for cellular services, there is wide recognition of the need for opening the upper mid-band for cellular services.For example, the FCC Technical Advisory Committee (TAC) has recently published a comprehensive survey of spectrum from 7.125 to 24 GHz [19] as well as a 6G working paper [20].These analyses emphasize the need for considering larger portions of the upper mid-band for terrestrial cellular services in order to meet the growing data demand.

C. Incumbency and the Need for Spectrum Sharing
The FCC TAC analyses [19], [20] also emphasize that a key issue in allocating the upper mid-band is incumbents, particularly commercial satellite services that also need bandwidth.In fact, the rejection of the 12.2-12.7GHz band proposed in [33] for cellular services was largely due to the interference onto ground satellite units.We will perform some simple cellular-satellite interference calculations in Section B.
The FCC analysis [19] has thus considered several potential models for spectrum sharing in the upper mid-band.Interestingly, spectrum sharing was critical in CBRS and C-Band allocations [35], [36].Spectrum sharing mechanisms in these bands included spectrum access systems SAS and more general licensed spectrum access (LSA) schemes are now being considered in the upper mid-band as well.More dynamic methods for spectrum sharing, as is being considered by 3GPP [37], [38] could also be used.
In short, the upper mid-band is a vast and valuable frequency range for numerous services.How to allocate and share the spectrum between various users will be one of the fundamental design and policy challenges going forward.

A. Prior Channel Measurements and Studies
Aspects of channel propagation in the upper mid-band have been studied for over two decades.For example, [40], [41] studied performed indoor radio measurements up to 11.5 GHz in the early and mid-1990s.Significant research has continued.For example, additional indoor measurements can be found in [42] (from 2.4 to 61,GHz) and in corridors from 9 to 11,GHz in [43].Attenuation measurements for various building materials at 800 MHz to 18 GHz were conducted in [44].Indoor and outdoor 5G diffraction measurements at 10, 20 and 26 GHz can be found in [45].The work in [46] performed wideband outdoor channel measurements at 3-18 GHz and a more recent comprehensive set of outdoor and indoor measurements at 3.3, 6.5, 15 and 28 GHz was conducted in [47].Satellite propagation has also been extensively studied -see, for example, [48] on earth satellite measurements at 12 GHz.A key and consistent finding of these channel measurement studies is that the factors that influence large-scale propagation, such as transmission losses, reflectivity, and diffraction, vary considerably over the upper mid-band.As expected, the lower frequencies in the upper mid-band provide more favorable coverage in most indoor and outdoor scenarios.At the same time, higher frequencies offer better bandwidth since bandwidth allocations generally scale with the carrier frequency.

B. Multi-frequency Outdoor Capacity Gains
One implication of the variability of propagation across the upper mid-band is that cellular systems would ideally have access to bands across the spectrum with realtime band selection.Wideband mobiles could use the higher frequencies with higher bandwidth when coverage is available, and switch to lower frequencies when the higher frequencies are blocked.To estimate the potential gain of such a multi-band, frequency adaptive system in the upper mid-band, we consider a dense urban area, i.e., Herald Square in New York City, as shown in Fig. 2a and downlink scenario (gNB→UE).In dense urban scenarios, capacity a key requirement.At the same time, providing satisfactory coverage is challenging at high frequencies due to blockage, as have been experienced in the mmWave bands [11], [12], [51].Within this area, we consider a hypothetical systems operating at up to four potential frequencies: 6, 12, 18, and 24GHz.
The simulation parameters of the potential cellular downlink systems at these frequencies are shown in Table 1.Note that the bandwidth in each frequency scales with the carrier, as is typical in deployments today.As shown in Fig. 2b, the size of the gNB antenna array also scales so that the aperture is approximately constant.Furthermore, in the simulation area, we manually selected the locations of 18 BSs (gNB) on rooftops, corresponding to an inter-site distance (ISD) of approximately 200 m, typical for urban microcellular evaluations.Terrestrial UEs are randomly placed outside buildings, and ray tracing was performed using Wireless Insite [52], which has recently been proven fairly accurate by conducting real-world measurements in the mmWave bands, as reported in [53], [54].
As mentioned in the Introduction, a critical factor in the performance of mmWave systems is the susceptibility of signals to blockage [10]- [14].To model the effect of blockage, we consider two scenarios: • No blockage; and • Blockage modeled with the 3GPP Blockage Model B [55] with K = 4 random human blockers.
Fig. 2c shows a coverage map of the unblocked wideband SNR for a single BS deployment as an example.As expected, we see that coverage is greatly reduced as frequency increases.
For both the unblocked and blocked cases, we compute the wideband SNR for each UE and BS and assume the UE is served by the BS with the strongest unblocked SNR.Then, given an SNR, we assume the achieved rate R (goodput) follows a standard realistic model [56]: where B is the bandwidth, α is a system bandwidth loss factor to account for overhead and receiver imperfections, and ρ max is a maximum spectral efficiency.Following [56], we adopt 0.57 for α and 4.8 for ρ max .Fig. 3 shows the resulting SNR distribution for frequencies 6, 12, 18 and 24 GHz at all outdoor UE locations in the study area without blockage.As can be seen in Fig. 3, UEs at lower frequencies experience uniformly better SNRs than higher frequencies due to favorable propagation and reduced noise power from using a smaller bandwidth compared to higher frequencies.Fig. 4a shows the corresponding rate distribution.We observe that, even though the SNRs are lower, as expected, the use of higher frequencies (18 and 24 GHz) ensures superior data rate due to the wider bandwidth.Note that from (1), there is a maximum rate of R = Bρ max which corresponds to 0.48, 0.96, 1.44 and 1.92 Gbps at frequencies 6, 12, 18 and 24 GHz, respectively.Hence, higher peak rates can be achieved at higher frequencies, where more bandwidth is available.However, UEs at the cell edge (for example, UEs up to the bottom 10% percentile of the CDF) suffer a significantly worse rate in the higher frequency bands (18 and 24 GHz) compared to UEs at lower frequencies (6 and 12 GHz).This property is also expected, as propagation is much less favorable for UEs at the cell edge at higher frequencies.
Fig. 4b shows the rate distribution with blockage.We observe that the cell edge rates at the high frequencies drop sharply.Indeed, UEs in 18 and 24 GHz show significantly worse performance approximately 35% of the time, compared to UEs in 6 and 12 GHz.Thus, when blockers, such as humans or vehicles, surround an UE, as would occur commonly in an urban scenario, the highfrequency coverage suffers significantly, while the lowfrequency coverage is significantly more uniform.The curves labeled "best choice" in Fig. 4a and Fig. 4b correspond to the rates for the UEs that select the best BS and frequency.As expected, the best choice is uniformly better than any individual frequency.Indeed, the best choice achieves the same peak rates as the UEs in the highfrequency bands, along with the improved cell edge rates of the low frequencies.These results motivate wideband cellular systems that adaptively select across a range of bands.Such systems can obtain the bandwidth benefits at high frequencies while providing robustness and resistance to blockage at low frequencies.

C. Penetration Loss
In the previous section, we focused on outdoor UEs.Coverage for indoor from outdoor cell sites is called outdoor to indoor (O2I) penetration.To understand the potential for O2I coverage in the upper mid-band, Fig. 5 plots the 3GPP model [55,  for the loss of the O2I path for several common exterior building materials.For each material, the 3GPP model for pathloss is given by where L is the path loss in dB, f is the frequency in GHz, and a and b are linear constants that depend on the carrier frequency and are given in [55, .Note that the curve in Fig. 5 does not appear linear since the pathloss is plotted on a logarithmic scale.We see that standard glass and wood are relatively permeable throughout the frequency range (<8 dB for both materials), while infrared reflecting glass (IRR) is relatively impermeable across all RF frequencies (>20 dB).Concrete, however, makes a sharp transition from relatively permeable to inpenetrable, precisely in the upper mid-band.

D. Multi-frequency Indoor Capacity Gains
To understand the effect of penetration loss on capacity, we re-run the simulation with users randomly placed indoors.Note that the 3D models used in the ray tracing have approximate material classifications for the facades of each building.Using the penetration loss models from Eq. 2 combined with the exterior wall classification, Fig. 6 plots the estimated CDFs of the SNR and data rate for indoor UEs.Fig. 6a shows that the SNR for indoor UEs is considerably reduced at higher frequencies due to high penetration loss, as mentioned in the previous section.In this setting, concrete is a dominant exterior wall material and significantly reduces the signal penetration -see Fig. 5. Indeed, Fig. 6b shows that, for roughly 65% of the users, the data rates at lower frequencies are higher than at higher frequencies.These results indicate a further potential gain of wideband cellular systems in FR3: the lower frequencies can provide valuable indoor coverage behind materials such as concrete while the higher frequencies can opportunistically offer high capacity for outdoor users and selected indoor users with minimal blockage (e.g., indoor users next to non-IRR glass windows).

E. Effects of Interference
As the last discussion, we consider a full buffer interference scenario where BSs transmit data to their associated outdoor UEs simultaneously, and each UE receives interference signals from other BSs.Note that this is the worst-case scenario to show the lower bound of the system capacity gains.As before, considering 200 m ISD, 18 BSs are manually selected and concurrently transmit data to 18 UEs.Fig. 7 shows the CDF of the signal-to-interference-noise ratio (SINR) and the corresponding data rate calculated using (1).As shown in Fig. 7a, SINRs are degraded at lower frequencies (6 and 12 GHz) because the small number of antenna arrays produces wider beamwidths.In contrast, due to directional transmissions with narrow beamwidths, UEs at higher frequencies experience less interference.
As a result, we observe in Fig. 7b that data rates at higher frequency are almost uniformly better than those at lower frequency.Thus, the "best choice" always selects the highfrequency band.This result suggests yet another benefit of adaptive systems: in interference-limited scenarios, BSs or UEs can automatically choose bands where directionality can reduce interference.
Note that to clearly see the effect of interference, we have not added blocking in this simulation as we did earlier.If blocking were added, then the high frequency performance would degrade and not be uniformly better.The point is that adaptive systems can naturally select the optimal bands with multiple factors, including path loss, penetration, and directionality.

IV. Interference with Incumbent Services A. Commercial Communication Satellites
One of the most vital and growing incumbents in the upper mid-band are commercial communication satellites [57].Fig. 8 shows the standard satellite bands.We see immediately that the X, Ku, and K bands -all widely used by satellites -fall entirely in the upper mid-band.In fact, recent FCC reports [19], [20] indicate extensive use of the bands by various commercial satellite services.Moreover, with the rapid growth of satellite Internet services, there is enormous demand for increased bandwidth, particularly in the 7-15 GHz bands.As an example, Fig. 8 shows the bands granted to the commercial satellite provider Starlink in the recent FCC grant [58].As depicted, the bands in the grant are in both the uplink and downlink, as well as gateway-satellite and terminal-satellite links.3GPP has also begun considering 5G NR services from satellites to mobile devices [59]- [61], including communication in the S, K, and Ka bands, which fall partly within the upper mid-band.More generally, satellite services are growing rapidly [62] and if terrestrial cellular services are to be deployed in significant fractions of the upper mid-band, co-existence with satellite services will be crucial.

B. Estimating Satellite Interference
Several recent works have analyzed interference between satellite and terrestrial networks.For example, [63] considers the selection of the network between satellite and terrestrial services.The works [64], [65] analyze interference of terrestrial networks onto the satellite downlink, i.e., satellite ground BSs, in the C-band.Similarly, [66] considers interference on the geostationary (GEO) and medium Earth orbit (MEO) satellite downlink in the Ka band.The work [67] uses stochastic geometry to estimate interference in the satellite uplink, also in the Ka band.In this work, we estimate the potential interference between terrestrial cellular networks and satellite services in the upper mid-band.As illustrated in Fig. 9, there exist possible interference channels to the low Earth orbit (LEO) satellite uplink by terrestrial networks.We will examine both the interference from the terrestrial downlink (DL, when the gNB is transmitting) and the terrestrial uplink (UL, when the UE is transmitting).When terrestrial UE or gNB transmits data, some portion of that signal energy is received as interference at a victim satellite.To obtain multipath channels for the DL and UL cases, we conducted an extensive ray tracing simulation.Based on the ray-tracing data, we build and run a simple system-level simulation to estimate the interference distribution.
The system-level simulation parameters are listed in Table 2. Satellite altitude, h, and antenna gain-to-noisetemperature for satellites, G/T , are taken from [60, Section 6.1.1.1]for the Set-1 LEO-600 case in the Ka band.We presume that the peak gain is achievable in the entire simulation area, corresponding to the case where we are looking at a satellite beam focused in the simulation region.
Interference in a victim satellite is calculated under the assumption that there is no additional attenuation L a , except for pathloss provided by the ray-tracing simulation, to make the analysis conservative (including this attenuation will reduce interference).The antenna pattern for the gNB and UE are taken [50] with half-power beamwidths of 65 • in both azimuth and elevation, maximum element gain of 8 dBi, and front-to-back gain of 30 dB.
We selected a typical rural area shown in Fig. 10 and examined a rural setting, as such areas are essential for satellite coverage.A satellite is placed at an altitude h = 600 km located at a random azimuth angle and an elevation angle θ uniformly distributed in [10 • , 90 • ].Note that the line-of-sight (LOS) distance to the satellite is the so-called slant distance given as where R E is the earth radius.A gNB and UE are randomly selected in the simulation area with the constraint that the UE is within 1 km of the BS.
We use ray-tracing data to estimate the multi-input multiple-output (MIMO) channel matrix, H ter , between the gNB and UE and the channel vector, h sat , from the gNB to the satellite for the DL case.In the UL case, h sat is the channel vector from the UE to the satellite.In both cases, as further discussed in the next section, h sat can be approximately tracked using ephemeris data.For simplicity, we treat the satellite as a single antenna receiver, since the beamforming gain of the satellite is already incorporated into the G/T value.Furthermore, the channel matrices H ter and h sat , include the gain of the antenna element for gNB or UE and the multipath components.
In our first simulation, we take into consideration the case where the terrestrial TX and RX select TX and RX beamforming vectors w t and w r to maximize the beamforming gain on the terrestrial link, where the optimization is solved over unit vectors.Assuming TX and RX perform channel estimation using pilot signals, the solution to ( 4) is given by the maximum singular vectors of H ter [68].Importantly, the selection of the beamforming vectors from (4) do not take into account the interference to the satellite uplink -it only maximizes the SNR on the terrestrial link.After taking the TX beamforming vector w t , the resulting channel from terrestrial TX to satellite RX will be w H t h sat .Hence, referencing the definition of carrier-to-noise ratio [59], the interference-to-noise ratio (INR) in dB at the satellite will be where P tx is the total TX power of the UE or gNB, G/T is the satellite RX gain to thermal noise ratio in dB, L a are Typical satellite systems require an INR < −6 dB corresponding to an SNR degradation of ∆ ≈ 1 dB.Plotted in Fig. 11 are the CDFs of the INR for both the uplink and downlink at two upper mid-band frequencies: f c = 6 and 18 GHz.There are three important conclusions: • Possibility of high interference to the satellite uplink: We see that the INR can be high.For example, at 6 GHz carrier frequency we see that approximately 27% of the downlink transmissions result in an INR ≥ −6 dB, the level at which the satellite SNR degrades by more than ∆ = 1 dB.• Both uplink and downlink terrestrial interference sources can be significant: We see that the INRs can be large for transmissions by both the UE and the gNB.Although the UE transmits with a lower power, its antenna arrays can be in an arbitrary orientation.Therefore, with some probability, it can be oriented directly to the satellite.In the left panel of Fig. 11, we see that, even with terrestrial uplink transmissions, the INR > −6 dB about 25% of the time when f c = 6 GHz.
In contrast, the gNB is equipped with fixed downtilted antenna arrays.In such a case, the interference to the satellite uplink is caused by sidelobes occurring due to transmit beamforming of gNB.
• Reduced interference at higher frequencies: By Friis' law, the INRs are lower at higher frequencies.This property is considered to be useful for frequency adaptation to reduce interference to satellite services.

C. Reducing Satellite Interference with Nulling
To mitigate the interference, the terrestrial transmitter can employ interference nulling as follows: Assume, for the time being, that the terrestrial transmitter knows the channel vector h sat to the satellite along with the MIMO channel matrix H ter between the gNB and the UE.We will discuss the estimate of the satellite channel vector in the next subsection.As before, the receiver selects the beamforming vector w r from (4).However, to eliminate interference, the transmitter selects a vector w t via the regularized cost: where λ > 0 is a regularization parameter.The regularization term, |w H t h sat | 2 , penalizes the interference in the satellite receiver and attempts to create a null along the satellite channels.The solution w λ t to ( 7) is given by maximum eigenvector of To assess the effectiveness of the nulling, we re-ran the identical simulation as in the previous subsection where the TX beamforming vector is computed from (8).We have considered the 6 GHz carrier frequency since the interference is higher in the lower band.The left panel of Fig. 12 plots the CDF of INR caused by terrestrial downlink transmissions to the satellite uplink at different values of λ.Regarding the curves labeled "with errors", we will discuss them in the next sub-section.From the left panel of Fig. 12, we see that with sufficiently high λ, the interference can be well mitigated, assuming ideal tracking.For example, when λ = 10 8 , INR ≥ −6 dB less than 3% of the time.
Plotted on the right panel of Fig. 12 is the degradation in the SNR on the terrestrial link defined by which is the ratio of the beamforming gain with the optimal beamforming vector w t from ( 4) and the beamforming gain with the regularized beamforming vector w λ t from (7).We see that with a value of λ = 10 8 , the degradation of the terrestrial SNR can be kept at ρ <0.1 dB for 97% of the time.
Fig. 13 similarly plots the CDF of the INR for terrestrial uplink transmission under different values λ along with the terrestrial degradation of the SNR ρ.We see here that the UEs can also reduce the interference, but the degradation is higher on the terrestrial link.We observe that the degradation of the SNR is less than 3 dB for 78% of time when λ = 10 7 .This arises from the fact that the UE has a lower number of antennas, and hence directional nulling is more costly.

D. Tracking Interference Channels to Satellites
A practical issue with the above interference nulling method is that the UE and gNB must track the wideband channels to the satellite.Specifically, the cellular transmitter (UE or gNB) must estimate the channel h sat at all frequencies.In this sub-section, we assess the possibility of using satellite ephemeris data to track azimuth and zenith angles of satellites.Ephemeris data are typically publicly availablesee, for example, [69].The use of these data has been considered by 3GPP [60].
To realistically assess the tracking with the ephemeris data, we need to take into account several practical challenges.First, knowing the satellite locations, one can only compute the LOS interference channels for victim satellites.Therefore, we can use the ephemeris data to estimate the LOS component of the interference channel, ignoring the NLOS components.Since we are considering a rural scenario, the channels are dominated by LOS components, so considering only LOS interference channels should not significantly degrade performance.We will verify this assumption in the simulation below.
A second issue is the orientation calibration for ground antenna arrays.To use the ephemeris data, the orientation of the gNB or UE relative to the global frame must be known.For the gNB (base station), the orientation can be measured, for example, with a high-precision compass [70].For the UE, the orientation can be measured with a magnetometer along with an inertial motion unit (IMU) and filtering techniques for tracking.systems will introduce some angular error.Even if small, performance could be impacted, since spatial nulling tends to create very sharp nulls.To see this effect, we simulate the angular errors shown in Table 3 taken from the published measurements in [70], [71].Specifically, we applied a random angular error given by the values in Table 3 to determine the estimated interference channels h sat towards the victim satellites.
To better account for the angular errors, we modified the transmit beamforming procedure as the following where h sat includes only LOS component of interference channel to a satellite.The optimization is identical to (7) except for the regularization term E|w H t h sat | 2 .The expectation is taken over the distribution of channel vectors h sat given the measured angular errors.This expectation encourages the creation of nulls, not just for the estimated angles of the satellites but rather for small regions around those angles to account for angular errors.The solution to (10) is described in a way similar to (8), but the expectation for h sat h H sat should be taken over the distribution of the channel vectors h sat .
The final potential issues are variation of the tracked satellite location over the transmission time interval (TTI) and time synchronization.These time-related errors are, however, negligible.For example, since the TTI in 5G is generally less than 1 ms, and the velocity of a LEO satellite at an altitude of h =600 km is approximately v = 7.56 km/s, the maximum angular variation is at most ∆θ = tan −1 (vT /h) ≈ 7.2 × 10 −4 degrees.This error is considerably less than the angular errors from the orientation calibration and can thus be neglected.The variation in tracked information due to synchronization errors in gNBs is also negligible, as network synchronization protocols such as the Precision Time Protocol (PTP) in the open radio access network (O-RAN) require nanosecond-scale timing synchronization errors [72].
Under these assumptions, we evaluate the modified interference nulling scheme given in (10) by running the same simulation as in the previous section.Fig. 12, shows the INR and gain loss CDFs with angular errors for the DL on the curve labeled "λ = 10 9 with errors".We see that the performance is almost identical to the curve without errors.Similarly, as shown in the curve labeled "λ = 10 7 with errors" of Fig. 13, we see that there is no significant change of INR due to angular errors for the UL case.Note that while tracking errors for UEs are assumed to be larger, as shown in Table 3, the effect of angular error is not as strong, as UEs have fewer antennas and produce less narrow nulls.Furthermore, since the nulls are created only on the LOS component, in the right panel of Fig. 12, we see the lower loss in SNR compared to the case without errors.For both gNB and UE, these results suggest that the modified interference nulling technique can work with reasonable tracking errors, even for high-speed LEO satellites.

E. Radio Astronomy
Radio astronomy studies the electromagnetic emissions from distant astronomical sources and high energy events [73], [74].These emissions are wideband in nature, and reach the surface of the Earth strongly attenuated by their long-distance propagation throughout the interstellar or intergalactic media, but also due to atmospheric absorption.
Astronomical emissions can be characterized by their spectral shapes.They feature persistent continuum emissions depending on their nature [75], [76]: blackbody radiations for the cosmic background, free-free emissions, e.g. for starforming regions, or synchrotron emissions e.g., for neutron stars.
Gaseous or ionized sources also embed narrow features known as emission or absorption spectral lines [77], [78].These lines are characteristics of the chemical elements present in the astronomical source, and are used to trace their composition, structure, and density.They are the only probes available to the interstellar medium and to external galaxies.They also reveal complementary information, such as gas temperatures, ionization, and fluid dynamics.More importantly, their shift in frequency from a given rest frequency, known as redshift, provides information on the age of the source and its distance to an observer.The International Astronomical Union (IAU) defined a list of the most important spectral lines, in which more than 25% fall in and below the mid-band (< 24 GHz).Notably, the complex prebiotic molecules, essential to the understanding of life processes in the Universe, have spectral signatures concentrated between 10 and 15 GHz [79], [80].
The spectral flux density of an astronomical sources is expressed in jansky (Jy), which is defined as 1 Jy = 10 −26 Wm −2 Hz −1 .To appreciate how small this flux density is, recall that an isotropic antenna at f c = 6 GHz would have : an aperture of A = λ 2 /(4π) ≈ 2(10) −4 m 2 .A signal of 1 Jy with this antenna would thus be received at −267 dBm, more than 90 dB below the noise floor.Observing milli-to micro-Jy sources are not uncommon with modern radio telescopes.
Detecting these weak emissions requires high sensitivity, which is achieved with radio telescopes using large collecting areas from wide reflectors or combined across multiple dish antennas, receivers with low system temperatures and wide bandwidths, and long integration times spanning seconds to hours of data integration.
Radio telescopes are also sensitive to Radio Frequency Interference (RFI), which can impact astronomical observations at various levels.Weak sources of RFI are detected after data integration in either or both time and frequency domains, and the associated corrupted time and frequency resource blocks are then discarded before further astronomical information extraction processes [81], [82].The loss of data associated with this procedure not only reduces the sensitivity of an astronomical observation, which can possibly be recovered by longer observations, but may also prevent the observation of transient sources, such as Fast Radio Bursts [83] or counterparts of sources of gravitational waves, such as super massive black holes mergers [84], which are individual and non-repeatable events.Similarly, redshifted spectral lines can fall outside protected frequency bands where the astronomical information may be fully lost.This is for instance the case with the redshifted galactic Hydrogen line with rest frequency at 1400 MHz falling into the lower Global Navigation Satellite System (GNSS) bands (1145-1310 MHz), and the observation of the edges of the local Universe [85].Stronger sources of RFI can drive the electronics components of a telescope receiver into a nonlinear regime, leading to the complete loss of data loss.
The impact of RFI is minimized by locating radio observatories in remote areas with low population and terrestrial transmitter densities, exploiting the propagation losses of these transmissions due to their large distance from a radio observatory [86].Further protection can be sought through coordination with active services to prevent the deployment of future radio frequency infrastructure, as is the case in the National Radio Quiet Zone (NRQZ) [87], [88].Emerging work involving artificial intelligence (AI) and machine learning (ML) based nonlinear signal processing may help partially offset the effects of RFI from LEO constellations on radio astronomy [89].

A. Challenges
The antennas and RF circuits of upper mid-band transceivers need to support wide bandwidths, high degree of tunability, and large numbers of antenna elements for directionality.Existing technologies have a number of bottlenecks in meeting these requirements including: a) antenna sizebandwidth-gain tradeoffs, 2) high frequency losses, 3) degraded SNR while scanning, 4) poor interference tolerance, and 5) inefficient spectrum management.To address the above shortfalls, here we combine a mix of wideband elements that are tunable and reconfigurable to mitigate interference and improve SNR across dynamically large spectrum swaths in the upper mid-bands.
Even though an antenna element can be designed to operate across the full range of interest, there are trade-offs that limit performance/ sensitivity and directional gain as a strong function of frequency if a single element is used in an array configuration for sensing applications.We want to avoid such compromises and trade-offs while covering the full band of interest, which requires several innovations in the antenna and microwave circuit areas for a costeffective and scalable high-performance solution.Another key challenge in developing wideband systems for cellular applications is the antenna form factor.While there has been extensive work in wideband antennas (see, for example, [90]), most designs require physically thick profiles that are not suitable for portable devices.

B. Compact Wideband Aperture in Aperture Antenna Arrays
To overcome the aforementioned challenge, a lowprofile aperture-in-aperture (AiA) realization is investigated, considering 3 classes of antennas: a) ultra-wideband (UWB) tightly coupled arrays (TCDAs) [91], [92] [93]- [96], b) low profile planar circular monopole [97], [98] [99]- [101], and c) UWB patches [102], [103].[104], [105].Specifically, the unit cell of this array (see Fig. 14 was designed based on existing work in literature, which was later optimized to meet the requirements of the FR3 bands.Notably, the UWB performance offered by the TCDA is based on Wheeler's current sheet principle.In addition, the ground plane inductance is canceled using the capacitive overlaps of the dipoles, thereby eliminating unwanted resonances at certain frequency spots [93], [94].With the monopole, the ground plane, which serves as the counterpoise for the monopole, is designed to enhance the antenna's performance [99]- [101].The size and shape of the ground plane play a role in determining the bandwidth.Similarly, by expanding the above approach, the bandwidth of the patch antenna was also increased.That is, the edges of the partial ground plane of the patch antenna were modified to allow for radiation, resulting in improving the impedance bandwidth of the patch antenna.Of course, this list is not exhaustive, and there are other types of wideband antennas that can also be considered.For instance, the holographic antennas can give the required bandwidth, as the holographic technique allows the use of sub-wavelength unit cells, and these reflectarrays and transmitarrays, such as [106]- [108], can be used in cellular network towers. Notably, this AiA brings forward several unique features with game-changing impact in antenna array design and performance features.Specifically, they are: 1) scalable across all frequencies and geometries; 2) highly compact in terms of element size and thickness; 3) wide and continuous bandwidth of more than; 4) low-cost and easy to deploy in a highly conformal manner to mounted and deployed easily on any platform.Indeed, realizing low-cost fabrication and beamforming across wide bandwidths is one of the foundational challenges in our footsteps.The AiAs overcome this challenge quite effectively and with all the required characteristics rather naturally.
Unlike existing UWB arrays which use a single aperture for the entire band, here we present three separate antenna designs for each band.Doing so, provides the desired beam scanning across all the bands which is not the case with the former.Due to finite array size, using a single aperture significantly limits the beamscanning performance of the array due to the finite electrical array size at the lower bands.For instance, a typical 8 × 8 UWB array operating across a 4:1 impedance bandwidth, will have an antenna element spacing of λ high /2.This corresponds to an interelement spacing of λ low /8 at the lowest band, implying that the effective number of elements at the lower band (6 GHz) is only 2. This significantly impacts the array scanning performance.
To address the above shortfall, a single aperture-inaperture (AIA) antenna array comprising of 3 separate antenna arrays is designed.The lower band from 6−12 GHz will comprise of closely spaced dipole arrays employing a co-axial feed, the mid-band from 12 − 18 GHz will employ circular monopole, while the high-band from 18 − 24 GHz will be realized using patches.The top panel of Fig. 14 depicts a simple illustration of the AiA geometry.The unitcell dimensions are provided in Table   The constant spacing between elements has been set to be 9 mm.Thus, the entire 8-element array measures 74 mm × 25 mm.
As depicted, a single aperture houses antennas operating across the entire upper mid-band band.The design optimization and simulation were performed using commercially available Ansys HFSS.The unit cell comprises of one antenna of each type with a semi-infinite boundary condition set-up.Hence, we used only 3 ports for our simulation depicted in yellow in Fig. 15.Active unit-element voltage standing wave ratio (VSWR), gain, and efficiency are provided in the bottom panel Fig. 14.As shown, all three antennas have a very good return loss or VWSR < 3. Further, the unit cell gain ranges from -2 to 4 dB across the operational FR3 bands.Finally, the average efficiency of this aperture is 82% with the upper bands experiencing a minimum efficiency of 67% at the upper ends.Fig. 16 provides the single element port-port isolation of the AIA.In Fig. 16, P1 corresponds to port-1 of the lower band antenna antenna radiating across 6 -12 GHz.P2 corresponds to the midband antenna operating between 12 -18 GHz, and P3 represents the ports of the patch antenna designed to operate between 18 -24 GHz.As expected, the isolation between monopole (P2) and the patches(P3) is the maximum due to their increased physical separation, in comparison to the placement of the dipoles.Nevertheless, the overall isolation or mutual coupling between the bands of AiA is still better than −18 dB.Finally, Fig. 17 shows the radiation pattern at the center frequencies of each band, specifically at 7.5 GHz, 15 GHz, and 21 GHz.It should be noted that the radiation pattern corresponds to a single antenna element within each band.

VI. Open Research Problems
The upper mid-band presents an enormous potential for cellular systems to deliver high data rates with consistent coverage and uniformity.Nevertheless, significant technical challenges remain to realize such systems.We summarize some open research problems indicated by the preliminary studies in this paper.

A. Channel measurements and capacity analyses
While there has been extensive channel measurements in the upper mid-band, certain aspects need further study.Most importantly, most of the measurement campaigns described in Section A captured the omni-directional path loss.Phased array systems, similar to those now widely-used in mmWave bands, could provide insights into the spatial structure of the channel which will be necessary for modeling MIMO and beamforming.The dynamics of blockage, across the band, will also need to be investigated, given how challenging blockage has been for mmWave systems.Also, our study in Section B suggested the possibility of significant gains with wideband adaptive systems operating across the upper mid-band.However, the study was limited to a single urban area with only outdoor users.To analyze how general these results are, statistical models and more data, validated through measurements, will be needed.Current statistical models, such as those used by 3GPP [55], will also need to be extended.Large-scale statistical dependencies between multiple bands are not well modeled, and datadriven techniques, such as those recently proposed in [109] may be valuable.

B. Interference with incumbents
Similarly, our analysis in Section B suggested that high density LEO constellations can be susceptible to interference from terrestrial cellular services, but spatial nulling may be able to mitigate these effects.Further research, however, is required.Interference nulling will require tracking, including tracking of NLOS components, such as ground reflections, which can be significant when one wants to suppress interference by more than 20 dB.Protocols that can also selectively avoid the time and frequency of transmissions that cannot be mitigated will also be needed.Further work will be needed for other constellations as well as interference with radio astronomy and other passive sources, which we have only briefly mentioned.

C. Antennas and circuits
The design presented in Section V indicates the possibility of a compact multiband antenna structure that can cover the entire wideband with appropriate RF switching.One limitation in the current design is that the elements are assumed to be probe fed and further work will be needed to build microstrip fed structures and packaging to realize such antennas in practical devices.In addition, tightly coupled arrays require signal processing to account for the mutual coupling at lower bands.In addition, we have not addressed the design of the RF circuits and switches that will need to operate across a wideband with a large number of antenna elements.

D. Security and resiliency
Due to space considerations, this article has not touched on the vital topic of security, a key area mentioned in the FCC study [20].Spectrally agile systems in the upper mid-band could provide new resiliency to hostile attacks by sensing signals and frequency hopping.There is broad literature on such systems, but little has been researched specifically in the context of cellular multiband systems.Furthermore, we have seen that satellite signals can be significantly impacted by low-power random terrestrial signals at these frequencies.This fact suggests that an adversarial attacker could significantly disrupt vital LEO satellite services.Methods to detect and mitigate such attacks, possibly leveraging terrestrial measurements, will be an area of paramount importance.

VII. Conclusion
We have provided a detailed assessment of both the potential benefits and challenges that may arise for cellular use in the upper-mid band.This analysis yields valuable insights: Capacity gains: The potential capacity gains of a wideband FR3 was assessed in an urban scenario.As expected, we showed that the overall system capacity was maximized when lower frequencies are dedicated to cell edge users, due to the more favorable propagation features.Similarly, such frequencies are indispensable for indoor UEs, owing to the substantial penetration losses in higher frequencies introduced by materials such as walls and glass.On the other hand, the key benefits of the higher frequencies stem from (1) the availability of high bandwidths and (2) the intrinsic directionality, and hence interference isolation, which translate in higher data rates.These results demonstrate the value of systems that can dynamically select frequency across the upper mid-band.
Coexistence: Then, we analyze the coexistence between cellular services and satellite incumbents, and conclude that interference may lead to substantial degradation in the performance of satellite networks.For this reason, we propose an interference nulling scheme that enables terrestrial networks to significantly reduce that interference.
Antenna design: A compact wide-band antenna array is presented based on a low-profile aperture-in-aperture (AiA) realization and three classes of antennas (coupled dipole array, circular monopole, and UWB patches).The presented aperture provides contiguous coverage across the entire FR3 bands.

FIGURE 1 .
FIGURE 1.The upper mid-band (called frequency range 3 or FR3 in 3GPP) is a potential new band for the cellular services that offers a balance of coverage and spectrum.To utilize the full band, cellular services will likely require sharing spectrum between satellite, and radio astronomy along with resliency to malicious jammers.

FIGURE 2 .
FIGURE 2. Simulation results obtained via ray-tracing a dense urban area of NYC at four frequencies in the upper mid-band for one example BS site.Left: 3D model of the NYC area (Herald Square) that was used for raytracing.The model was downloaded from [39], which enabled a fairly accurate mapping of foliage and building materials.Center: gNodeB antenna array architecture and bandwidth at each carrier frequency.Right: Coverage map obtained from raytracing data from an example single BS.The reduced coverage with higher frequencies is readily visible.

FIGURE 3 .
FIGURE 3. Aggregate multi-band SNR distribution for all outdoor UEs without blockage.
(a) Data rate CDF without blockages (b) Data rate CDF with blockages

FIGURE 4 .
FIGURE 4. Aggregate multi-band rate distribution for all outdoor UEs.

FIGURE 6 .
FIGURE 6. Aggregate multi-band SNR and rate distribution for all indoor UEs.

FIGURE 7 .
FIGURE 7. Aggregate multi-band SINR and rate distribution for all outdoor UEs in a multi-cell setting with interference.

FIGURE 8 .
FIGURE 8. Top: Satellite bands.Bottom: Recent requests for the Starlink frequency band to the FCC [58].Additional requests in the E-Band are not shown.

FIGURE 9 .
FIGURE 9. Terrestrial to satellite interference where terrestrial transmissions between a terrestrial gNB and UE interfere with the satellite uplink (UL).

FIGURE 11 .
FIGURE 11.INR distribution from UL and DL terrestrial cellular sources.

FIGURE 12 .
FIGURE 12. INR distribution from DL terrestrial cellular sources for different λ values at a carrier of 6 GHz.

FIGURE 13 .
FIGURE 13.INR distribution from UL terrestrial cellular sources for different λ values at a carrier of 6 GHz.

FIGURE 14 .FIGURE 15 .
FIGURE 14. (Top) HFSS illustration of tri-band array, (Bottom) Simulated plots showing active VSWR, gain and efficiency of the array when all ports are excited. .