System design of the physical layer for Loon’s high-altitude platform

This paper describes several aspects of the physical layer and over the air interface of Loon. Loon utilizes stratospheric balloon-based high-altitude platforms (HAPs) that use Long-Term Evolution (LTE) to connect people with standard User Equipment (UEs) to the Internet. In particular, topics covered include the Loon prototype eNodeB (eNB) antenna pattern, the observed channel, UE battery life, and coexistence with terrestrial networks using the same spectrum. While channel models from a HAP to the ground have been well studied in the past, the use of polarization diversity to establish Multi-Input Multi-Output (MIMO) communication to real UEs below 1 GHz has not. In addition, a theoretical analysis of terrestrial coexistence and an analysis of the estimated impact on UE battery life when communicating with HAPs are presented. Finally, results from several measurement campaigns and from experiments with polarization diversity are presented as a spot check of theory.


Introduction
The use of HAPs to deliver Internet connectivity on the ground has been discussed for almost two decades, with many research articles on various aspects of such a system, including the wireless link. Loon has developed the only stratospheric HAP that has provided connectivity to hundreds of thousands of people. To accomplish this, the Loon HAP is comprised of high-altitude balloons, floating between 17 and 21 km above the Earth, which beam standard compliant LTE signals to the ground. This is accomplished via an eNB that resides on the balloon which facilitates communication with standard UEs. For backhaul, the balloons are able to communicate over multiple balloon-to-balloon hops before landing the traffic on the ground with a balloon-to-ground hop, all using a proprietary high speed link. In this sense, one can consider the balloons as having formed a mesh network in the sky.
The design of such a system needs to address some very interesting challenges: -Interoperability with regular UEs, including those that are 3GPP Release-8 compliant. Modifications to *Correspondence: sananth@loon.com Loon LLC, 100 Mayfield Ave, Mountain View, CA 94043 USA the standard are not desired, as the goal of Loon is to serve users in emerging markets. -UE antennas are optimized for communication with terrestrial cellular towers and not HAPs. -Power budget is constrained due to the use of solar power, and mass budget is constrained by the physical limits of the prototype Loon HAP. -Transmit power on the UE is limited and cannot be increased. -The spectrum used by the Loon HAP may be the same as that used on the ground by terrestrial cellular operators. Therefore, the Loon HAP needs to coexist with these transmissions.
In prior work, channel models for HAP communications have been researched extensively (e.g., [1][2][3][4][5][6][7][8][9][10]). In addition, chapter 10 in [11] goes over recent work in stratospheric channel models. However, only a few papers (e.g., [12][13][14]) discuss HAP-MIMO channel models and only a few ( [15,16]) discuss the use of polarization to achieve diversity. For instance, [13] discusses the advantages of having MIMO for HAP communication by evaluating the performance of downlink (DL) HAP channels. Through the use of simulation, this work shows that using 2 × 1 MIMO improves LTE performance by 1.4 to 12.3 dB and that using 2 × 2 MIMO can improve the performance by 7.7 to 15.7 dB. This research uses data from [17] to model the channel as an elevationdependent Ricean. The authors assume two independent, Rician faded channels in their simulation but it is unclear how this independence is achieved. Michailidis and Kanatas [12] derive a three-dimensional, geometrybased, single bounce channel model for MIMO channels in Ricean fading environments. Using derived theoretical expressions, an evaluation of the HAP antenna interelement spacing requirement for achieving uncorrelated responses in HAP MIMO channels is derived. However, using polarization to achieve the uncorrelated channels is not considered. In addition, [9] measures building penetration loss (BPL) as a function of elevation angle and polarization. In order to simulate receive antennas with directional patterns, the authors place antennas in an orthogonal configuration (i.e., one antenna pointing vertically and another pointing horizontally). This configuration is used to measure the impact of polarization on BPL as a function of elevation. Figure 5 from [9] shows a peak differential in BPL of 5 dB between vertical and horizontal polarization at 2 GHz with test receivers. However, typical UEs do not have nicely orthogonal antenna patterns below 1 GHz. Oestges [10] discusses the effect of rain and ice depolarization on a HAP at 47 GHz, using dual-polarized antenna arrays. Dong et al. [15] analyzes diversity performance from multiple HAP networks while also considering the single HAP use case. The authors show that due to the very close distance between the antennas in a single HAP, the use of traditional MIMO techniques cannot overcome large-scale fading. Due to the predominant line of sight (LOS) channel conditions in a HAP operating environment, propagation channels are highly correlated and most diversity techniques are not applicable. However, the authors also show that there may be exceptions such as using spatial diversity on the ground or using multiple HAPs. Also, using polarization to achieve diversity from HAPs is not explored. Michailidis et al. [18] provides a mathematical model for polarization-based diversity from HAPs by calculating the XPD between orthogonal polarizations. Figure 2 from this paper is particularly interesting, as the computations show that XPD is expected to be low for an urban region (e.g., London) even when the HAP is directly overhead. This is somewhat counter intuitive, as a high XPD would be expected due to the strong LOS conditions. However, the analysis uses a Ricean K factor of 0 since it is for a dense urban area. This figure is not applicable to more rural areas, where the Ricean K factor is high. In addition, the authors assume an isotropic antenna pattern for the UE, which does not hold for typical UEs in real use. Nikolaidis et al. [16] provide measurement data for XPD in LOS channels from airships using dual polarized antennas. Table 1 in this paper shows that an XPD of greater than 15 dB is expected for all conditions (e.g., LOS, non-LOS). Section III B, and the discussion around the Demmel condition number, leads to the conclusion that at high elevation angles, a large amount of multiplexing (MIMO) based transmission should be expected. However, using co-located HAP antennas at less than 1 GHz with real UEs, given their antenna limitations, has yet to be explored. Using polarization to provide diversity gain in terrestrial base stations has also been researched extensively (e.g., [19][20][21][22]). For instance, [19] demonstrates that polarization provides a means of realizing two independently fading signals with co-located antennas by relying on the ability of scatterers in the channel to depolarize and decorrelate the signals. In addition, [21] demonstrates that polarization diversity is mostly preserved in LOS conditions for terrestrial applications. Finally, coexistence of transmissions from HAPs and terrestrial deployments using the same frequency has been researched extensively (e.g., [23][24][25][26][27][28]). For instance, [23] discusses coexistence of 3G in disaster scenarios, where some terrestrial towers are disabled due to an emergency. The terrestrial network is then overlapped by a HAP-based 3G network, and the impact of the HAP network to the terrestrial network is analyzed. Based on the parameters chosen, the authors are able to demonstrate that the simultaneous application of HAP and terrestrial networks impacts the terrestrial signal, particularly in suburban and urban macro cellular areas. Likitthanasate et al. [24] discuss coexistence of WiMax at 5 GHz. The authors consider a single HAP with a single terrestrial base station located 10 km away from, but still within, the HAP coverage area. Based on the parameters chosen, the authors conclude that the HAP and terrestrial base station can coexist with low data rate modulation schemes. To get higher data rates, the authors expect that the UE antenna beamwidth would have to be narrow (e.g., less than 30 • ). A similar idea of exploiting antenna directionality in the UE is explored in [25]. Here, the authors discuss coexistence among a constellation of HAPs, where interference from multiple HAPs is reduced by using a narrow antenna beamwidth at the  [26,27] discuss coexistence of Code Division Multiple Access (CDMA) in terrestrial and HAP deployments. In this research, the minimum distance between terrestrial CDMA coverage regions and HAP CDMA coverage regions is computed to be between 2.5 and 9 km. However, coexistence for LTE at less than 1 GHz, with large HAP coverage, and omni-directional UE antennas has not been explored. While existing research has been done in many areas applicable to the Loon use case, many design challenges require more research. This paper intends to build on the existing research by measuring the impact of polarization diversity using co-located HAP antenna bands below 1 GHz, demonstrating why the HAP channel model in this use case is not sufficient to support MIMO communications with standard UEs and discussing LTE coexistence between HAPs and terrestrial cellular deployments for real UEs.

Frequency band
Loon aims to provide network expansion for telecommunication partners by utilizing their existing spectrum allocations and fully integrating into their existing network. Although any of the existing LTE bands available from a partner could be chosen for use on a Loon system, LTE bands below 1 GHz are preferred as they allow for the widest possible coverage. Due to its wide availability, LTE band 28 (703 to 748 MHz for uplink (UL) and 758 to 803 MHz for DL) has been made available by some of Loon's terrestrial partners for use in actual Loon deployments and will be used for the analysis in this paper.

eNB antenna pattern
The prototype antenna pattern used by Loon's eNB was designed to facilitate outdoor communication over large geographic areas. To this end, a single Loon HAP is expected to cover a region~40 km in radius. Beyond this range, even though a UE can receive Loon's DL signal, the system becomes UL limited. In addition, the prototype antenna pattern was designed to not have peak gain directly below the balloon where path loss is the lowest. Tables 5, 6, 7, and 8 in [29] point out that a coupling loss of no more than 132 dB is required to achieve 384 kbps on the UL and a coupling loss of no more than 140 dB is required to achieve 14.4 kbps. Assuming a UE antenna gain, including body loss, of negative 10 dB on average, the antenna gain required at the eNB can be derived as: AntennaGain >= PathLoss + 10 dB − CouplingLoss (1) Where PathLoss is due to free space path loss (FSPL) and CouplingLoss is either 132 dB or 140 dB. The actual gain of the prototype antenna for Loon's eNB is plotted against these two coupling losses in Fig. 1. As can be seen, 384 kbps on the UL is expected up to~65 • (~43 km radius) for an outdoor UE with LOS conditions.
In addition, Loon's eNB uses a two transmit and two receive antenna configuration with co-located antenna elements using two different linear polarizations (horizontal and vertical). Experimentation has been performed with other antenna configurations, including spatial separation.

UE antenna pattern
Research has shown that achieving diversity at the UE for frequencies below 1 GHz is difficult. For example, [30] say that, in general, two linearly polarized antennas, located orthogonal to each other, provide polarization diversity by reducing the mutual coupling. They go on to point out that this technique does not work well for UE antennas at lower frequencies (e.g., LTE at 700 MHz) due to ground plane sizes being much smaller than wavelength (λ is 429 mm at 700 MHz). Derneryd et al. [31] show that there is a real challenge when designing multiple antennas with low antenna correlation and high efficiency in small hand-held devices, especially at low frequencies. Hagerman et al. [32] describe the result of a field study, conducted by Ericsson and Verizon, which shows that with careful placement of UE antennas it is possible to achieve good MIMO rank even for LTE at 700 MHz. This study used mock UEs with various antenna placements and sizes corresponding to smart phones and feature phones. These UEs were then tested for MIMO rank and throughput in a precommercial LTE network. Given that the feature phones have a smaller size, the observed rank-2 MIMO performance in this case was lower than was observed in smart phones. However, the study did not differentiate between polarization and spatial diversity when assessing MIMO performance. This is key, since spatial diversity is difficult to achieve from HAPs, as will be shown.
A typical metric used in LTE handset antenna design is the Envelope Correlation Coefficient (ECC) as described in [30]. In order to achieve good MIMO performance, an ECC value of less than 0.5 is recommended. Typically, UEs exhibit high ECC for the lower bands and lower ECC for the higher bands. For example, a typical popular smart phone will have an ECC around 0.4 to 0.5 at 700 MHz and an ECC of lower than 0.1 at higher frequencies. However, the ECC metric treats spatial and polarization diversity equally (i.e., we can have low ECC by having two antennas point in the same direction with orthogonal polarization or we can have low ECC by having two antennas pointing in different directions with the same polarization). From inspecting the antenna radiation patterns of popular UEs, low ECC in the 700 MHz range is achieved by the antenna pointing in different directions rather than polarization diversity. An impact analysis of antenna pointing direction on ECC can be performed by computing the ECC using the total antenna gain, as opposed to the usual computation using the two orthogonal polarizations. If the ECC obtained by these two methods are roughly the same one can conclude that the primary driver for ECC is antenna pointing rather than polarization. If the ECC obtained by the total gain method is high but the ECC obtained by the normal computation method is low, one can conclude that the low ECC is obtained predominantly achieved by polarization diversity. For typical UEs, we observe that modified ECC is only 20 to 30% higher than the regular ECC computation (at low frequencies). This further corroborates that the low ECC at 700 MHz is due primarily to the antennas pointing in different directions. Figure 2 shows the total directivity, phi-polarized directivity, and theta-polarized directivity of an a typical popular smart phone. From these plots, it can be seen that the main and diversity antenna patterns point in different directions and that the theta-polarized directivity of both antennas is quite poor. This leads to the UE having predominantly one polarization requiring very strong signal strengths to overcome the weak gain for the second polarization to allow MIMO communication. However, in this scenario, there would be an imbalance in the channel capacity between the two MIMO streams due to the large gain difference.

eNB transmit power
The Loon eNB is currently configured for 37 dBm of conducted power per transmit port, giving a total conducted power of 40 dBm across both ports. The peak conducted power is only achieved when the eNB has 100% Physical Resource Block (PRB) utilization and is reduced as the number of utilized PRBs is reduced. This property is shown in Table 1 for a 5-MHz channel bandwidth.
In addition, signal strength from a terrestrial network will dominate at the UE as a Loon HAP approaches a more dense deployment. When accounting for proper reselection, UEs will prefer camping on a terrestrial tower compared to the Loon HAP. Due to the lowered demand, the transmitted peak power from the Loon HAP is also reduced.

Channel model
Due to the coverage pattern from a single Loon HAP being 40 km and the relatively low rate of drift, the channel for most outdoor UEs is expected to be LOS Additive White Gaussian Noise (AWGN). Shimamoto et al. [17] describe the standard deviation in mean received power versus elevation angle from a HAP. From this, a standard deviation of 0.5 dB (a Ricean K factor of around 18) is expected at high elevation angles (i.e., when a HAP is directly overhead), a standard deviation of 3.9 dB is expected at an elevation angle of 40 • , and a standard deviation of 5 dB (a Ricean K of 1) is expected at low elevation angles of 10 • . Although future Loon HAPs will support indoor usage, the prototype Loon system that was used for the measurement campaign in this paper is designed to support outdoor use cases. Considering the indoor use case, the model described in [17] changes and a higher standard deviation is expected.

Terrestrial transmit power and antenna pattern
This is certainly operator dependent; however, for macro sites, a range of 46 to 49 dBm per antenna port with a peak antenna gain of 18 dBi seems to include most use cases.
For the rest of the analysis, the worst case assumption of 46 and 18 dBi will be used.

Spacing of terrestrial towers
This again is operator dependent and is also geography dependent. At 700 MHz, spacing will be used for rural and 3.76 km (2 × 1.88 km) spacing will be used for suburban.

Location of users
Although the prototype Loon HAP was designed for outdoor usage and future Loon HAPs are being designed to additionally support indoor usage, both indoor and outdoor users will be considered in the analysis. It is expected that a higher percentage of users will be indoors.

Loon propagation model
As was already discussed, a LOS-AGWN channel is expected. This channel is dominated by FSPL which will be used for this analysis. Using FSPL for interference modeling is optimistic for propagation prediction and presents a worst case for the coexistence analysis. In addition, FSPL is used in many HAP coexistence studies (e.g., [26-28, 34, 35]).

Terrestrial propagation model
For terrestrial propagation, the impact of terrain must be taken into account. A simplified version of COST231 Hata as given in [33] is used for this analysis. In this model, the eNB height is assumed to be 35 m and the UE height is assumed to be 1 m. Path loss is then calculated as a function of distance (d) between the UE and the tower in km: Log normal fading margin is not applied here, as this is intended to be a worst case analysis. In typical terrestrial deployments, an additional margin of 10 dB is expected for log normal fading as given in Table 11 of [36].

Building penetration loss
BPL is very much dependent on the type of construction used for building roofs. From observation, developing countries have a wide variety of roofs. This ranges from roofs which pass radio frequency (RF) signals with minimal attenuation (e.g., thatch) to roofs which add substantial attenuation (e.g., reinforced concrete). For this analysis, an average of 13 dB BPL is used when considering the indoor scenario. In the outdoor scenario, BPL is 0 dB.

UE characteristics
As was discussed previously, UE antenna gain and body loss is expected to be around negative 10 dB. However, most studies (e.g., [23,36]) model the UE antenna gain to be 0 dB. Considering that there may be customer-premises equipments (CPEs) with higher antenna gain, coexistence analysis in this paper will use 0 dB. Table 6.11 from [33] defines − 4 dB as the minimum required SINR for LTE service, and [37] requires that the error vector magnitude (EVM) for a transmitted signal be less than 8%. This 8% EVM requirement translates to a peak SINR of 25 dB. For this analysis, the maximum possible SINR is limited to 30 dB to allow for implementations that outperform the minimum requirement. In addition, [24] defines two metrics called "percentage coverage area served" and "percentage coverage area not served. " Using these definitions, this analysis defines a coverage area and measures the SINR distribution. This SINR distribution is measured by sampling the coverage area uniformly and plotting a cumulative distribution function (CDF).

Simple interference without Loon HAPs
Considering first a single terrestrial tower without a Loon HAP, Fig. 3 shows the expected SINR on the ground. When a second terrestrial tower is added 13.16 km from the original tower at the same frequency (i.e. a rural, single frequency network), Fig. 4 shows the expected SINR from both towers combined. Although the plots show that the cell radius extends out to 20 km, this is only due to the optimistic assumptions on UE antenna gain (0 dB) and lack of log normal fading margin. As described earlier, this is to present a worst case scenario. In all cases, the SINR is limited to 30 dB. As can be seen, the presence of additional terrestrial towers reduces the peak SINR at the UE from greater than 30 dB to slightly less than 15 dB.

Simple interference with a Loon HAP
From this base analysis, a Loon HAP with 100% eNB load is added at a distance of 30 km from the terrestrial tower. This distance aligns the peak antenna gain from the Loon HAP with the terrestrial tower. In order to represent the worst case, all UEs are considered to be outdoors and the Loon HAP is at maximum transmit power. Figure 5 shows the expected SINR for this case. As expected, there is a further drop in the SINR near the terrestrial towers. However, there is also a large region where coverage has improved substantially.
Having considered the worst case scenario for coexistence analysis with all UEs outdoors, it is interesting to consider the inverse condition with all UEs indoors. The SINR from this scenario is shown in Fig. 6. In this case, there is minimal impact due to the Loon HAP while the outdoor coverage gain is still realized.
Considering now a suburban case, Fig. 7 shows the two terrestrial towers spaced 3.76 km apart for outdoor UEs, still with 100% load on the Loon eNB.
In this case, there is a degradation in SINR due to the Loon HAP. However, it is substantially less than the degradation in SINR due to the terrestrial towers interfering with each other.
In all cases and in all areas, the combined curves are above the minimum SINR threshold of − 4 dB.

Fig. 5 Expected outdoor SINR with two rural terrestrial towers and a Loon HAP
This means that even though the interference causes cell shrinkage, with proper re-selection, UEs do not lose coverage in any locations and there is a substantial increase in coverage due to the presence of the Loon HAP.

Interference with terrestrial deployment
Moving beyond the simplified terrestrial case of two towers, a terrestrial deployment with 37 hexagonal sectors is considered. The radius of each cell is 6.58 km to simulate a rural deployment. All users are considered to be outdoors to simulate the worst case. This deployment scenario results in a coverage area with 46 km radius, the SINR on the ground for this deployment can be seen in Fig. 8. For the analysis, a single Loon HAP is placed randomly within the 46 km. In addition, the Loon eNB is assumed to have 100% load for a worst case analysis and toroidal wrap is used to model a wider area. Figure 9 shows the resulting CDF of the SINR for the entire region, computed before and after placement of the Loon HAP. As can be seen, a negligible delta is observed when a Loon HAP is directly over a fully built out terrestrial network.
In order to model situations where the terrestrial network is not fully built out, the same basic model can be used with just 3 terrestrial towers instead of 37. Figure 10 shows three different CDFs of SINR from this scenario.
The first CDF shows the SINR of the terrestrial network without the Loon HAP. The second CDF shows the SINR of the network when the Loon HAP is added. As can be seen, a large delta is observed in SINR and the degradation of the terrestrial network is non-negligible when the Loon HAP is present. The third CDF shows the SINR of the terrestrial with a Loon HAP added assuming that reselection is happening at the UE without any hysteresis. In this case, the UE will always choose the strongest signal, while all other signals contribute as interference. Under these assumptions, there is no degradation of SINR on average, since a decrease in SINR is seen above~6 dB due to interference and an increase is seen below due to the large coverage improvement from the Loon HAP. This is intuitive as there are large coverage holes that are covered by the Loon HAP when there are only 3 terrestrial towers.

UL interference
For the uplink, Loon's eNB will naturally observe signals from UEs within~40 km that are communicating with terrestrial towers. In fact, there is a direct link between the number of terrestrial towers the loading experienced in these terrestrial towers, the roof material in these locations and the amount of UL interference that is to be expected. However, in general and with proper reselection, UEs will prefer a terrestrial network if one is available. As a Loon HAP approaches a large terrestrial deployment, this means that fewer UEs will stay camped on the Loon HAP. This certainly reduces the amount of DL interference which the Loon HAP generates, but it also allows for better frequency selective scheduling in the UL for the remaining UEs.

UE transmit power and battery life
Due to the larger distance between a Loon HAP and UEs, as compared to a terrestrial use case, a common assumption is that the UE transmits at higher power more often, leading to a reduction in battery life. This can be analyzed by first understanding the distribution of UE transmit powers in terrestrial networks. Joshi et al. [38] shows that in current terrestrial networks a UE is transmitting at maximum power 50% of the time. In rural areas, this paper also shows that the 95% case has a UE transmitting~7.9 s out of 15 min, while the average is 3.8 s out of 15 min. Given this information, the following worst case assumptions will be used for the rest of the analysis: • Eight seconds of UL transmission out of 15 min • UL transmissions will be at maximum power 50% of the time and 12 dB lower for the remaining time In addition, Fig. 1 from [39] shows that 2.6 W are consumed when a UE transmits at maximum power. The same figure shows that 1.7 W are consumed when transmitting 12 dB below maximum. Finally, Fig. 3 from [40] shows that the average power consumption during light sleep is 11 mW per 1 ms and the average power consumption while receiving is 500 mW per 1 ms.
Considering a typical paging duty cycle of 640 ms as shown in Fig. 11, the total energy consumption in idle mode is: 11 mW × 637 ms + 500 mW × 3 ms = 8507 μJ   TSME ultra compact drive test equipment [41] was used to measure the channel characteristics while Key Performance Indicators (KPIs) gathered from the Loon network and the terrestrial network were used to assess overall system performance.

Received signal power at the UE
In Puerto Rico, data was captured using the TSME in both indoor and outdoor environments. The TSME includes an omni directional external antenna which has an antenna pattern that is designed to mimic that of a real UE. During the measurement, two Loon HAPs were present at 40 km and~25 km from the test location and data was collected from both simultaneously. Channel impulse response data was captured for 1 h while keeping the TSME antenna stationary using the LTE pilot signals from both Loon HAPs. Measurements were taken on the ground floor of a single story building for the indoor results. Figure 12 shows the signal power distribution from the HAP that is~40 km (i.e., elevation angle of~25 • from the test location towards the HAP). Both indoor and outdoor measurements are shown. The difference in mean between outdoor and indoor is 7 dB with a standard deviation of 4.6 dB indoors and 4.7 dB outdoors. Shimamoto et al. [17] measures a standard deviation of 2.65 dB at 20 • , 1.75 dB at 30 • , and 3.9 dB at 40 • . Measurements from the Loon HAP are slightly higher than this at this location.
However, a higher standard deviation is expected indoors, indicating that there is not a significant amount of multi-path. More testing in indoor locations is required to further understand this observation. Figure 13 shows the same plot from the HAP that is 25 km away from the test location (i.e., elevation angle of~45 • from the test location towards the HAP). In this case, the difference in mean is 6.7 dB with a standard deviation of 2.5 dB indoors and 3.5 dB outdoors. This roughly matches the measurement of 3.9 dB at 40 • from [17].

Diversity/MIMO communication
Using KPI data collected from the Loon network in Peru and Puerto Rico, on average, rank-2 communication is realized less than 10% of the time. This is expected due to the prior discussion on UE antenna construction in bands lower than 1 GHz. Figure 14 shows the CDF of polarization diversity by taking the ratio of rank-2 requests to the sum of rank-1 and rank-2 requests from KPI data. An experimental configuration where the Loon eNB employed spatial diversity instead of polarization diversity was also tested. In this experiment, the transmit antennas were placed~1 m apart with the same polarization. From the theory presented in [12], the antennas should have been spaced further apart. However, the prototype Loon platform limitations restricted the spacing that could be achieved. Figure 14 also shows the CDF of this experiment. It can be seen that polarization diversity is significantly better than this particular configuration of spatial diversity. In addition, neither diversity scheme is sufficient for MIMO communication to a large number of UEs. It is expected that improving antenna gain and tweaking the ECC at the eNB will improve this situation. However, even with this improvement, we expect the MIMO performance from a HAP platform at lower frequencies to be worse than what is seen in a terrestrial network.

Terrestrial coexistence 4.2.1 DL interference-terrestrial only
An analysis of DL terrestrial self interference was performed against a live terrestrial network. For this analysis, the TSME equipment was used to take measurements in a vehicle as part of a limited drive test of the terrestrial network. The SINR from this measurement is presented in Fig. 15 and the Received Signal Strength Indicator (RSSI) from the same measurement is presented in Fig. 16.
As can be seen, the RSSI from both towers is quite high for the duration of the test and especially around 4000 s. However, since the two towers have similar RSSI, the interference resulted in an SINR that is much lower than expected. In this case, the average SINR is on the order of 10 dB with substantial dips around 4000 s when both terrestrial towers have elevated RSSI.

DL interference-terrestrial with a Loon HAP
Several measurement campaigns were conducted in Piura, Peru against a live, terrestrial partner network. During the campaigns, SINR was systematically measured on the ground using the TSME equipment at various distances from a terrestrial tower. A tower close to the edge of a terrestrial network's coverage area was chosen for this campaign. A Loon eNB was then turned on and off at known time intervals to assess the impact on SINR. During all measurements, no degradation of the terrestrial tower was observed.
Assessment of KPIs from the partner network in the Piura, Peru region were also observed. For this Fig. 17 UL Interference observed from Loon near Lima assessment, DL and UL throughput and PRB utilization KPIs were collected over the course of a month. During this time, the Loon network was not operating at all times and the times of operation were recorded. An analysis was then conducted over all the KPIs to determine the delta between when the Loon network was present and not present. This analysis showed that no degradation was observable in the partner network's KPIs.

UL interference
KPIs from the Loon network were collected over the course of several months over Lima, Peru to analyze the effect of UL interference. For this analysis, UL noise KPIs were averaged over all Loon HAPs using a geographic grid with 11 km spacing. These noise measurements are not calibrated and can therefore only be used for a relative assessment and not for an absolute assessment. Figure 17 shows the collected UL noise data as a heatmap. As can be seen, the delta in UL noise is 20.1 dB (a range of − 4.8 to 15.3 dB is observed) with the highest interference being closest to Lima, where the terrestrial network was the most dense.

Conclusion
This paper describes physical layer aspects involved in the Loon system design. Specifically, the following has been shown: • Coexistance of terrestrial and HAP LTE networks: With careful system design it is possible for a HAP LTE network to coexist with terrestrial LTE networks and that the addition of a Loon HAP has a minor effect on DL interference, which is offset by a substantial increase in coverage. • MIMO challenges below 1 GHz: At frequencies below 1 GHz utilizing polarization diversity to achieve MIMO links has challenges due to UE antenna pattern limitations. These limitations are more pronounced as the UE becomes smaller. • UE energy consumption when used with HAP LTE networks: Increase in energy consumption in the UE while communicating with HAPs is not significant in normal use cases (i.e., less than 1% of UE battery life).