Modeling and Analyzing Millimeter Wave Cellular Systems

I. INTRODUCTION

• Sub-6GHz spectrum scarcity and growing broadband and wireless-data demand have driven intense interest in mmWave cellular, despite challenging propagation and handset size, cost, and power constraints.Source: §I · p. 1

A. Millimeter Wave: What’s New?

• Large antenna arrays provide required directionality, the key distinguishing feature of mmWave cellular, with far-reaching implications for modeling, analysis, design, and implementation.Source: A. Millimeter Wave: What’s New? · p. 2
• Blocking is more severe in mmWave than in Sub-6GHz, as higher penetration losses and weak diffraction make otherwise effectively LOS Sub-6GHz environments NLOS for mmWave.Source: A. Millimeter Wave: What’s New? · p. 2
• Directionality and blocking require changes to cellular architecture, deployment, modeling, and analysis, the theoretical adaptations that this paper principally surveys.Source: A. Millimeter Wave: What’s New? · p. 3

B. Scope and Organization

• The tutorial covers physical-layer modeling and analysis, including a baseline system model, SINR distribution derivation, and approximation of per-link rate distributions.Source: B. Scope and Organization · p. 3
• The tutorial addresses initial access, noise-versus-interference behavior with densification, self-backhauling, spectrum licensing, and extensions for uplink, joint Sub-6GHz coverage, and beyond-analog MIMO.Source: B. Scope and Organization · p. 3

II. A BRIEF HISTORY ON MILLIMETER WAVE SYSTEMS

• Although mmWave frequencies have long served other applications, serious consideration for commercial cellular, notably 5G, is recent; this section traces how and why they became viable.Source: §II · p. 4

A. Pre-cellular mmWave

• Early mmWave communication applications focused on medium-to-long-range LOS links, including backhaul and satellite communication, and vehicular communication –.Cellular-like mmWave systems had been considered as early as 1985, but received little attention for roughly the next 25 years.Source: A. Pre-cellular mmWave · p. 4
• The 60 GHz unlicensed band, enabled short-range WPAN/WLAN standards – delivering several Gbps and advancing small adaptive arrays and low-power RF circuitry.These products targeted very short-range cable replacement, typically within a room.Source: A. Pre-cellular mmWave · p. 4

B. Understanding the mmWave channel

• Outdoor mmWave links achieved acceptable SNR up to approximately 200 m; 38 GHz outdoor urban cellular channels were measured across the UT Austin campus.Source: B. Understanding the mmWave channel · p. 5
• Directional antennas alter the effective channel: they reduce delay and Doppler spread but can cause beam misalignment, while lower delay spread may still mean greater equalization demands.The classic two-ray ground-reflection model can imply an LOS path-loss exponent of 4 rather than 2, but focused beams make such reflections less likely.Source: B. Understanding the mmWave channel · pp. 5–6
• mmWave channels differ sharply between LOS and NLOS: poor diffraction makes NLOS depend on reflections and scattering, with greater attenuation and indoor–outdoor penetration loss than at Sub-6GHz.Indoor users usually cannot be served by outdoor base stations; channels are often angularly sparse, with a few scattering clusters and a dominant LOS path.Source: B. Understanding the mmWave channel · pp. 5–6

C. The recent push for mmWave cellular

• About 1 Gbps was reported without mobility from Samsung’s prototype, over up to 1.7 km in LOS or 200 meters; an earlier link budget projected Gbps rates with 15–30 dB antenna gains.Samsung used hybrid beamforming with four subarrays of eight antennas, each served by one RF unit.Source: C. The recent push for mmWave cellular · p. 6
• The mobility study in claims users moving at about 8 km/hr can achieve 500 Mbps with 1% block error rates using steerable antennas, although this is much less mobility than LTE can support.Source: C. The recent push for mmWave cellular · p. 6
• About 3 mm was the skin penetration depth at 28 GHz, compared with 45 mm at 1.9 GHz; the study found absorbed energy was more localized and largely limited to the epidermis.The study used a prototype phone with two 16-element antenna arrays.Source: C. The recent push for mmWave cellular · pp. 6–7

D. Performance analysis

• 3 to 5.8 Gbps mean throughput and 25 to 1400 Mbps cell-edge rates were reported as base stations increased from 36 to 96 in a 0.72 km2 region.The density change corresponds to 50/km2 to 133/km2, and cell-edge throughput is strongly noise-limited.Source: D. Performance analysis · p. 7
• Stochastic-geometry analysis in indicated that, with large arrays satisfying the link budget, mmWave could offer comparable SINR coverage and much higher rates than conventional cellular networks.The framework models base-station locations as a stochastic point process to derive downlink SINR distributions.Source: D. Performance analysis · p. 8

III. NOVEL MODELING ASPECTS: BLOCKING

• Shadowing alone does not accurately capture blocking in dense networks: blockage can dramatically change the effective path-loss exponent, and blockage-aware trends can differ from a single-exponent power-law model,, –.The section motivates accurate yet tractable blockage models and introduces empirical 3GPP and analytical approaches.Source: §III · pp. 8, 10
• Building penetration losses of 40−80 dB can isolate indoor and outdoor mmWave systems, while buildings produce distinct LOS and NLOS path-loss laws, and much higher NLOS path loss,.Diffraction is weak, and more signal energy is scattered at mmWave frequencies.Source: §III · pp. 8–9
• Blocking also attenuates interfering signals, which are typically farther from the receiver and thus more likely to be blocked.The section notes that buildings and other static objects are not the only blockers: human-body penetration loss is 20-40 dB,, making it a severe, intermittent effect that changes on a smaller timescale.Source: §III · pp. 9–10
• The section presents the empirical 3GPP blockage model, random-shape, LOS-ball, and Poisson-line analytical models, body and foliage blocking, and comparisons using real geographic data.Source: §III · p. 10

A. 3GPP model for incorporating blockages

• The 3GPP model, classifies links as LOS or NLOS for different path-loss laws and treats P_LOS(d) as a non-increasing distance-to-probability mapping, not a conventional probability distribution.The mapping takes values in [0, 1] and is specified differently for urban, suburban, and rural environments.Source: A. 3GPP model for incorporating… · pp. 10–11
• For large distances, the urban 3GPP LOS probability has a heavier tail than the suburban function.In a regular urban street grid, this corresponds to the possibility of LOS signals from distant base stations on the same street.Source: A. 3GPP model for incorporating… · p. 10
• The 3GPP blockage parameters [65] draw on a relatively limited set of measurements from the pre-LTE WINNER II 2007 document.Later work modified parameters to incorporate building heights and refitted urban LOS probabilities using New York City measurements,.Source: A. 3GPP model for incorporating… · p. 10
• For mmWave bands, the narrow Fresnel zone above 15 GHz supports frequency-independent building-blockage models, while Sub-6GHz may require a different model.Frequency-dependent blockage-model research remains nascent.Source: A. 3GPP model for incorporating… · p. 11

B. Random shape theory model

• The Boolean germ-grain model represents buildings with Poisson-distributed centers and independently distributed shapes, sizes, and orientations; uniform orientation is one analyzed assumption.The example uses a Boolean model of rectangles.Source: B. Random shape theory model · pp. 11–12
• The random shape analysis derives a LOS probability function with the same negative-exponential form as the 3GPP suburban model and analytically computes its parameter C from building statistics.This enables parameter estimation from area statistics without relying solely on extensive simulations or measurements.Source: B. Random shape theory model · pp. 11–12
• Perimeter- and area-based building statistics provide quick approximations of C without extensive simulations or measurements.The area-based estimate also uses the fraction of area covered by buildings.Source: B. Random shape theory model · p. 12
• For the UT Austin topology, the two estimates of C differ: 100 m using the perimeter-based estimate and 85 m using the area-based estimate.The estimates can differ because real building deployments are not necessarily rectangular.Source: B. Random shape theory model · p. 12

C. LOS ball model

• The LOS ball model approximates LOS probability with a step function and chooses RB to match average LOS area from another model or geographic data; at high base-station density, its SINR distributions differ only slightly from random shape theory.The LOS area is represented by a ball of radius RB around a typical user.Source: C. LOS ball model · p. 12

D. Poisson line model

• The Poisson line model represents streets as intersecting Poisson lines and makes LOS links depend on shared street lines, tractably capturing correlations that can change SINR-distribution tails.Users and base stations on streets are outdoor, while locations inside blocks are indoor.Source: D. Poisson line model · p. 13

E. Human body and foliage blockage models

• Smaller-obstruction models include foliage loss in dB linear in path length through tree canopies, ray-tracing-fitted LOS probability linear in link length, and cone-based self-body blocking in outdoor mmWave cellular networks.For, LOS probability has the form min(ax + b, c), where a, b, and c depend on deployment.Source: E. Human body and foliage blockage… · p. 13
• Most existing analyses treat blockages and users as static; future work could model moving obstacles and study penetration-loss effects on coverage and handover, including self-body cones shifting over time.The passages identify dynamics of mobile obstacles as an avenue for analysis.Source: E. Human body and foliage blockage… · p. 13

F. Comparison and conclusions on blockage models

• The 3GPP-like model fits LOS probability with RMSEs of 2.18% in Austin and 1.45% in Los Angeles, but that fit does not ensure accurate SINR or rate coverage.The passage notes that fitting empirical LOS probability often gives poorer coverage estimates than the LOS ball or random shape models.Source: Figure 3 · p. 14, F. Comparison and conclusions on… · p. 15
• With RB = 200 m, the generalized LOS ball model accurately fits SINR coverage in both Austin and Los Angeles, and radii of 150–300 m have worked well for dense deployments.The model uses pl = 0.3027 for Austin and pl = 0.2419 for Los Angeles.Source: Figure 4 · p. 15, F. Comparison and conclusions on… · pp. 15–16
• The random shape theory model gives a reasonably tight upper bound on SINR coverage near Austin, but underestimates coverage in Los Angeles.Its parameterization uses C = 85 m for Austin and 42 m for Los Angeles, with κ = 0.27 and 0.42, respectively.Source: Figure 4 · p. 15, F. Comparison and conclusions on… · p. 15
• The analytical framework accepts a generic P_LOS(d), while LOS-ball assumptions permit simpler analytical forms; comparison models neglect cross-link blockage correlation, which the more geometry-specific Poisson line model captures.LOS-ball and random shape models are simple to incorporate analytically.Source: F. Comparison and conclusions on… · p. 16

A. Hardware constraints and the need for different transceiver architectures

• Early mmWave research and prototypes suggest arrays of 32−256 antennas at base stations and 4−16 at mobile users,,,.Source: A. Hardware constraints and the need… · p. 16
• With wide bandwidths and large arrays, fully digital processing is difficult because per-antenna RF chains and high-resolution ADCs impose high cost and power consumption,.The passage says it is presently infeasible for mmWave receivers to have 32–256 full-resolution ADCs.Source: A. Hardware constraints and the need… · p. 17
• Analog beamforming uses RF-domain phase shifters to steer transmit and receive beams along dominant propagation directions, but quantization constraints limit it to single-stream use and hinder multi-user MIMO,.Its phase-shifter weights can be tuned to shape and steer beams.Source: A. Hardware constraints and the need… · pp. 17–18
• Analog beamforming was already available in Wireless HD and IEEE 802.11ad products, making it commercially viable in the near term; Section V analysis assumes this approach.Source: A. Hardware constraints and the need… · p. 18
• Hybrid precoding splits processing between analog and digital domains, providing a compromise between hardware complexity and performance while using NRF ≪ Ntx RF chains,, –.Source: A. Hardware constraints and the need… · p. 18

RFHFRFFBBs + W∗

• Hybrid architectures use far fewer RF chains yet support independent data streams for spatial multiplexing and achieve near-optimal performance versus fully digital transceivers,, –.The cited work reports near-optimal performance compared with fully digital transceivers despite the reduced RF-chain count.Source: A. Hardware constraints and the need… · p. 18, RFHFRFFBBs + W∗ · p. 18
• Proposed alternatives include switch networks to reduce hybrid-architecture power, lens antennas for RF precoding, and few-bit ADC receivers,,,,, [112].Source: RFHFRFFBBs + W∗ · p. 18
• Extending system analysis to capture all facets of hybrid precoding and other transceiver architectures remains future work.Source: RFHFRFFBBs + W∗ · p. 18

B. Spatial channel modeling

• Outdoor mmWave channels typically have few dominant scattering clusters, motivating geometric few-cluster models for capacity analysis and precoder design,, [113],,.Source: B. Spatial channel modeling · p. 19
• Quantizing uniform-array responses onto spatial angles yields a sparse virtual channel representation that produces ON/OFF interference, [116],.Source: B. Spatial channel modeling · p. 19

C. Single stream analog beamforming

• When a channel has a dominant LOS path or few scatterers, single-stream beamsteering can maximize gain in a desired direction by matching weights to the array response.Source: C. Single stream analog beamforming · p. 20
• For tractable coverage and rate analysis, antenna patterns are often approximated by a step function with constant main-lobe and side-lobe gains,,, [119].Source: C. Single stream analog beamforming · p. 20
• Reusing the same time-frequency resources across all access-point sectors can be feasible at mmWave, offering a low-cost, potentially suboptimal densification option.Source: C. Single stream analog beamforming · p. 21

D. SU-MIMO

• Because mmWave hardware constrains precoder dimensions and entries, approximating fully digital SVD is more dubious and motivates new SU-MIMO precoding designs.Source: D. SU-MIMO · p. 21
• Hybrid precoding methods include matching-pursuit variants and matrix-decomposition or alternating-minimization techniques, [125]–.Source: D. SU-MIMO · p. 21
• Hybrid precoding can generally achieve very good spectral efficiency relative to fully digital SVD, especially when RF-chain count is close to the number of dominant paths, [107],,, [125]–.Source: D. SU-MIMO · pp. 21–22

E. MU-MIMO

• A two-stage hybrid precoder assigns each user an analog beam, then uses baseband processing to cancel inter-user interference.The approach achieved results close to unconstrained digital solutions while requiring low training and feedback overhead.Source: E. MU-MIMO · p. 22

A. Performance metrics

• MmWave systems can be noise-limited, making SNR a useful SINR approximation, unlike the often interference-limited Sub-6GHz case.Coverage probability is the fraction of users exceeding an SINR threshold; rate coverage measures the probability that an active user’s rate exceeds a threshold, with rate depending on SINR and resource share.Source: A. Performance metrics · p. 23

B. Baseline mmWave cellular system model

• The baseline model extends the traditional Sub-6GHz analytical framework with mmWave blockage and directional beamforming, [131].It is used to analyze downlink SINR and rate distributions; user-load effects are addressed as an extension for rate distributions.Source: B. Baseline mmWave cellular system model · pp. 23–24, A. Performance metrics · p. 23
• Base stations and users are modeled as independent Poisson point processes, with users associating to the base station offering the smallest path loss.The analysis considers a typical outdoor user and assumes all base stations transmit constantly, a pessimistic assumption for SINR.Source: B. Baseline mmWave cellular system model · p. 24
• Independent distance-dependent LOS probabilities split base stations into independent, non-homogeneous LOS and NLOS processes with different path-loss laws.The model leaves the LOS-probability function general, provided blocking events are independent across links.Source: B. Baseline mmWave cellular system model · p. 24
• Analog beamforming is assumed at both base stations and mobiles, with perfect channel knowledge and maximum-gain steering on the desired link.Interfering base-station steering angles are uniformly distributed, and the array pattern is approximated by a sectored model.Source: B. Baseline mmWave cellular system model · p. 24
• The model uses distinct LOS and NLOS path-loss exponents and independent Nakagami fading, while omitting shadowing and frequency selectivity.Nakagami parameters differ by link type; shadowing can be incorporated at some cost in tractability.Source: B. Baseline mmWave cellular system model · p. 25
• PPP modeling captures important SINR trends across actual base-station layouts and is typically within a fixed 2–3 dB SINR gap, pessimistic versus real locations,,.The passage notes that the PPP assumes independent base-station locations.Source: B. Baseline mmWave cellular system model · pp. 25–26
• Because no actual mmWave cellular deployments existed at the time of writing, any model was speculative, but PPP results across many base-station layouts made it a reasonable baseline.The passage describes the PPP SINR curves as similar to those for a very large class of base-station locations, including grids.Source: B. Baseline mmWave cellular system model · p. 26
• The authors extend established Sub-6GHz stochastic-geometry results to mmWave rather than repeat the existing analysis [131].The cited reference covers downlink, uplink, and multi-tier downlink SINR distributions.Source: B. Baseline mmWave cellular system model · p. 26

C. SINR downlink coverage probability

• Alzer’s Lemma [134] converts gamma-fading coverage calculations to a weighted sum of exponential-variable CDFs, an approximation shown generally tight in simulations.This addresses the Nakagami fading model used instead of Rayleigh fading.Source: C. SINR downlink coverage probability · p. 26
• LOS and NLOS base stations are modeled as independent inhomogeneous PPP tiers with distance-dependent densities, while correlations between nearby LOS probabilities are ignored.The LOS density is λPLOS(d); the NLOS process is also modeled as a PPP with distance-dependent density.Source: C. SINR downlink coverage probability · p. 27
• Random interference-channel directivity gains are represented by a discrete random variable under the sectored antenna approximation, simplifying the interference calculation.The gain randomness is incorporated into the interference model; the discrete gain distribution is specified in Table II.Source: C. SINR downlink coverage probability · p. 28
• Theorem 1 computes overall SINR coverage by combining conditional LOS/NLOS coverage probabilities and averaging over serving-link state and distance.The conditional coverage calculation uses the interference Laplace functional and the serving-link state and distance distributions.Source: C. SINR downlink coverage probability · p. 28
• In noise-limited networks, SNR coverage replaces SINR coverage with a simpler expression; setting noise power to zero yields the SIR distribution for interference-limited networks.When interference is ignored, the terms Wn(·), Qn(·), Vn(·), and Zn(·) in Theorem 1 become zero.Source: C. SINR downlink coverage probability · p. 29

D. Rate coverage probability

• Under the common round-robin assumption, equivalent to random user selection per time-frequency resource block, per-user rate depends on perceived SINR and available resources.Rate coverage is the probability that user rate exceeds threshold τ in bps.Source: D. Rate coverage probability · pp. 29–30
• Serving-cell load and SINR are correlated: larger association cells mean more users and longer links, while blockage creates irregular mmWave association cells.Finding the load distribution therefore requires the joint distribution of association area and SINR; even the marginal distribution of Poisson-Voronoi cell areas remains an open problem.Source: D. Rate coverage probability · p. 30
• For tractability, the model approximates mmWave association-area distribution with that of a typical Poisson-Voronoi cell of the same mean, assumes PPP-distributed users, and treats load as independent across base stations and of SINR.The independence assumption applies to the load at the serving base station and the loads of other base stations.Source: D. Rate coverage probability · p. 30
• The mean load is 1 + 1.28λu/λ for the base station serving the typical user, versus λu/λ for a typical base station.The serving-base-station load distribution assigns zero probability to no associated users.Source: D. Rate coverage probability · p. 31
• Rate coverage for threshold τ is derived from SINR coverage and the distribution of the serving base station’s load.The expression is an infinite sum that can be truncated to a finite number of terms with little loss in accuracy.Source: D. Rate coverage probability · p. 31

A. When will mmWave systems be noise/power-limited?

• Beamforming can improve SNR without increasing average interference, while blockage and misaligned beams suppress interferers, making mmWave systems more likely to be noise-limited than Sub-6GHz systems.The average transmit power is unchanged when SNR is improved through beamforming.Source: A. When will mmWave systems be… · pp. 31–32
• Baseline analysis finds SINR coverage varies non-monotonically with base-station density; enough interferers eventually shift mmWave networks from noise- to interference-limited behavior, without a crisp threshold.The transition occurs at much lower densities in Sub-6GHz systems, and depends on multiple system parameters.Source: A. When will mmWave systems be… · p. 32
• At 73 GHz, systems tend to remain noise-limited even at 100 m inter-site distance (λ = 200 BSs/km²); denser Los Angeles blockages further reduce interference than in Austin.The 28 GHz system shows greater interference impact with its smaller 200 MHz bandwidth.Source: A. When will mmWave systems be… · p. 32
• Interference increases with smaller bandwidth, wider beams from fewer antenna elements, and more users in MU-MIMO.The simulations compare 28 GHz system parameters using P(I/N > 1).Source: A. When will mmWave systems be… · pp. 32–33
• At 73 GHz, about half the inter-site distance—and thus four times the base-station density—is needed to reach I ≈ N with probability 0.1 compared with 28 GHz.The comparison holds for both Austin and Los Angeles; higher frequency tends to be more noise-limited through greater bandwidth, directionality, and possible blockage sensitivity.Source: A. When will mmWave systems be… · p. 33
• More blockages make systems more noise-limited because they can block stronger interfering signals as well as the desired signal.Denser buildings and obstacles therefore increase noise-limited behavior.Source: A. When will mmWave systems be… · p. 33
• Initial deployments should prioritize sufficient SNR; interference becomes significant in successful, extremely dense, heavily loaded networks, but may matter earlier when coverage requires inter-site distances below roughly 100 m.Smaller arrays or high blockage can create coverage holes that require such dense deployment.Source: A. When will mmWave systems be… · pp. 33–34

• Low pre-beamforming SNR and the need for large antenna gains make mmWave initial access challenging, while beam alignment requires searching at both the BS and UE –, [151].Source: B. Initial access in mmWave systems is… · p. 34
• Initial access has two stages: downlink cell search lets the UE acquire the BS signal, and uplink random access lets the BS acquire the UE signal and learn its receive beam; communication requires completing both.Source: B. Initial access in mmWave systems is… · p. 34
• Because analog beamforming prevents searching in another direction on a separate frequency band while transmitting data, initial access must use the entire band, raising overhead [151].Source: B. Initial access in mmWave systems is… · p. 34
• Testing more beams improves alignment but consumes slots, while performing access more often helps new or moving users connect quickly at the cost of data-transmission time.Source: B. Initial access in mmWave systems is… · p. 34
• Unless beams are very wide or coherence blocks very small, exhaustive search with full pilot reuse is nearly as good as perfect beam alignment [119].The effective reliable rate captures SINR penalties from possible misalignment and beam-training resource overhead; the analysis covers near-orthogonal control pilots and full pilot reuse. Further exploration and optimization of initial access is identified as a research need.Source: B. Initial access in mmWave systems is… · p. 34

C. The promise of self-backhauling

• Directional backhaul generally avoids access-link interference, letting wired BSs provide in-band wireless backhaul to other BSs in dense mmWave deployments.Gbps-burst backhaul can be difficult to provide over wired networks.Source: C. The promise of self-backhauling · p. 35
• Increasing the fraction of wired-backhaul BSs improves peak rates, but per-user rate saturates as BS density rises with wired-BS density fixed; saturation density scales with wired-BS density.Source: C. The promise of self-backhauling · p. 35
• Subdued interference makes it feasible for access and in-band backhaul links to use the same time resources,, motivating investigation of dynamic time-division duplexing.Source: C. The promise of self-backhauling · p. 35

D. Spectrum license sharing among cellular operators is possible

• Uncoordinated spectrum sharing achieves higher rates than exclusive licensing due to low inter-operator interference, while static coordination can further improve network performance and differentiate operators’ access [157], [158].Source: D. Spectrum license sharing among… · p. 35
• The cited work [157] reports that uncoordinated sharing of spectrum licenses is possible in both cases, provided antenna beams are relatively narrow, about 30 degrees or less.Source: D. Spectrum license sharing among… · p. 36

E. Ultra-densification in mmWave networks

• At low BS density, densification shortens serving distance and raises LOS serving probability and power without significantly increasing interference, generally improving SINR and rate coverage.After a threshold, enough LOS interferers accumulate to degrade SINR.Source: E. Ultra-densification in mmWave… · p. 36
• As BS density tends to infinity, SINR coverage reaches zero when αL < 2 but converges to a nonzero value when αL > 2; beyond a critical density, performance degrades.The critical density is normally on the order of the LOS range and decreases with narrower beams [119].Source: E. Ultra-densification in mmWave… · p. 37
• For 28 GHz, critical density tracks average LOS range—43 meters in Los Angeles and 85 meters in Austin—and decreases with narrower beams.Source: Figure 10 · p. 37, E. Ultra-densification in mmWave… · p. 37

VII. EXTENSIONS TO THE BASELINE MODEL

• The extensions target uplink analysis, heterogeneous deployments with Sub-6GHz fallback and control signaling, indoor coverage, and additional multi-antenna techniques.The overview identifies these as important extensions to the baseline SINR and rate coverage framework.Source: §VII · p. 38

A. Uplink

• With base stations modeled as a PPP, scheduled users across cells do not form a PPP because of Voronoi structure,; an approach models uplink interferers as a non-homogeneous PPP in Sub-6GHz heterogeneous networks.Source: A. Uplink · p. 38
• Because mmWave uplink transmit power is expected to be below Sub-6GHz levels, the uplink is even more noise-limited than the downlink, so SINR ≈ SNR can support coverage estimates using downlink formulas with mobile transmit power.The paper suggests power control may be negligible because mobiles will typically transmit near full power in power-limited mmWave networks.Source: A. Uplink · p. 38

B. Joint coverage with sub-6GHz systems

• Legacy Sub-6GHz spectrum can support control signaling for mmWave; users can be simultaneously connected to both systems, while mmWave base stations are used whenever possible to offload from the more congested Sub-6GHz network.Source: B. Joint coverage with sub-6GHz systems · p. 39
• The cited analyses model macrocell and mmWave BS locations as independent PPPs with different densities, and represent their coexistence by superposition of the two PPPs,.Source: B. Joint coverage with sub-6GHz systems · p. 39
• Beamforming-aware association increases the probability of connecting to high-bandwidth mmWave BSs, but excessive bias toward mmWave can reduce rate when link SNR is weak [165],.The paper notes an optimum mmWave bias and advises offloading to UHF only when mmWave communication is infeasible.Source: B. Joint coverage with sub-6GHz systems · pp. 39–40

C. Outdoor-to-indoor coverage

• Because outer building walls cause huge penetration losses, indoor users not near windows are unlikely to be served by outdoor mmWave base stations and need indoor base stations or lower-carrier-frequency systems.Over 80% of all current cellular communications are with indoor users, excluding WiFi offloading,.Source: C. Outdoor-to-indoor coverage · p. 40
• Indoor propagation models must account for first-order wall reflections, sometimes comparable in strength to direct paths, and inner-wall partition losses, which contribute a large fraction of indoor path loss,.In,, walls were treated as mirrors and transmitter images were modeled approximately.Source: C. Outdoor-to-indoor coverage · p. 40

D. MIMO techniques beyond analog beamforming

• The analysis finds that optimum base-station density for SINR coverage decreases with more MU-MIMO transmission, while denser SU beamforming yields higher cell-edge rates than equal-power-density MU-MIMO.Round-robin scheduling harms cell-edge rates, highlighting the importance of user selection for MU-MIMO.Source: D. MIMO techniques beyond analog… · p. 41
• A channel rank above one may change interference statistics, while the feasibility of multi-stream transmission depends strongly on channel rank.The paper identifies incorporating higher-rank channels as an extension of the baseline model.Source: D. MIMO techniques beyond analog… · p. 41
• Network-level comparisons of analog and hybrid beamforming, low-resolution ADCs, and CAP-MIMO could identify suitable architectures under deployment, antenna, RF-chain, and power constraints.The virtual channel representation offers a means to incorporate different precoders and combiners.Source: D. MIMO techniques beyond analog… · p. 41
• Imperfect channel information and channel aging merit analysis because errors in arrival or departure angles can significantly reduce post-processing SINR with narrow beams.Most stochastic-geometry analyses assume perfect transmitter channel-state information, with a few exceptions such as and [119].Source: D. MIMO techniques beyond analog… · p. 41
• For mmWave massive MIMO, an analysis incorporating blockage, channel sparsity, and mobile-station beamforming finds that good SINR coverage requires high base-station density.In certain dense mmWave networks, the asymptotic SINR distribution approximates cases with more than 256 antennas.Source: D. MIMO techniques beyond analog… · pp. 41–42
• Future analysis could incorporate mmWave-compatible channel training, such as compressed sensing in the angular domain, which could potentially reduce training overhead, –.Source: D. MIMO techniques beyond analog… · p. 42

VIII. CONCLUSIONS

• The baseline model and analysis account for mmWave’s susceptibility to blocking and need for strong directionality.The conclusion presents these as the two most important physical challenges addressed by the paper.Source: §VIII · p. 42
• Outdoor-to-indoor coverage is largely neglected, load balancing and offloading remain poorly understood, and mobility complicates beam alignment and handoff prediction when blocking dominates distance.Maintaining aligned beams in both downlink and uplink requires considerable effort and system overhead; blocking can include a user’s own body.Source: §VIII · p. 42