I Introduction
With growing demands on data rates, reduced endtoend latencies, and connectivity across a diversity of new applications such as the industrial Internet of Things (iIoT), automotive, massive machinetype communications (mMTC), etc., the Fifth Generation New Radio (5GNR) standard specifications for wireless systems build on prior standard releases and try to address these complex design objectives seamlessly. The Release 15 specifications for nonstandalone and standalone deployments have been completed and approved in December 2017 and June 2018, respectively, with continued evolution expected over the next few months and years. Given this state of wireless evolution, the scope for followup work in terms of future releases and enhancements, and more specifically, broader questions on relevant physicallayer research problems for study by both academia and industry become pertinent. Such questions are important and existential, and have been repeatedly asked in the coding theory [1], information theory [2], digital signal processing [3] and communications theory [4] communities over many decades.
To address this question in the context of 5GNR, we first study the historical evolution of WiFi and cellular modems’ capabilities in terms of peak download data rates and spectral efficiencies. We show that the peak data rates have been consistently growing exponentially over the last twenty five years for both systems, with cellular on the brink of catching up with WiFi very soon. More importantly, the projection of these growth rates into the nearterm (next five years) can be met with a modest increase in bandwidth acquisition and signal processing complexity of the modem. On the other hand, a meaningful stretching of this roadmap into the far future (next ten years) would require significant bandwidth accretion at higher carrier frequencies (than currently commercially available), commensurate signal processing complexity scaling, and further network densification allowing the use of higherorder modulation and coding schemes. While the business usecases for such increased peak data rates in future modems are unclear as of now, applications such as pervasive health monitoring, advanced driver assistance systems (ADAS), cellularWiFi coexistence, etc., appear to be reasonable drivers for sustained rate increases and latency reductions.
From a physicallayer research perspective, the evolution of 5GNR and WiFi systems into the millimeter wave regime signal a key transition point into the increased focus of communications systems at higher carrier frequencies. We show by the way of two simple examples (on delay spreads and hand blockage losses) that the traditional abstraction of using simplified modeling techniques, sometimes inherited from sub GHz^{1}^{1}1Recent 3GPP standardization work has extended the upper point of the frequency regime of interest in traditional systems from GHz to GHz. Thus, the above usage should be seen with the technical caveat of sub GHz systems. systems, for systems studies at higher carrier frequencies is not sufficient. This is because simplified models often have no capability to emulate unknowns that remain unknown till a deeper systemic understanding of the impact of different components of the system on the concerned figuresofmerit.
Thus, these illustrative examples hint at a paradigm shift in the evolution of systems research with time. In particular, systems research has to incorporate far more of circuits and device level abstractions and capabilities in generating relevant models, the space of possible solutions, and in closing the feedback loop on the efficacy of the generated solutions for the intended original problem. Both science and technology policy as well as telecommunications departments need to evolve with these emerging trends in terms of the scope and shape of systems research.
Ii Cellular System Roadmaps
Iia Historic Trends
We start with a brief glimpse into the WiFi and cellular systems’ roadmaps in terms of the log^{2}^{2}2All logarithms are to base in this work. of the peak download data rates and spectral efficiencies^{3}^{3}3Spectral efficiency in bps/Hz is simplistically computed as the ratio of peak data rate and the maximum occupied bandwidth necessary to achieve this rate. over the past (approximate) twenty five years. In particular, the timeperiod of interest is from January 1997 through December 2020 (including future forecasts) over a period of months.
For WiFi, we use opensource data from [5, Table 1] on peak data rates, maximum occupied bandwidths, and release dates (approximated to the month level) of different standard specifications. This data is also presented in the Appendix for the sake of selfcontainment of this work. For cellular systems, we use opensource data from Qualcomm (see Appendix) on the approximate commercial sampling dates at the month level of different (in both standalone and integrated forms) cellular modems as well as their corresponding capabilities such as supported bandwidth, peak modulation scheme, number of antennas, and signal processing complexity [6]. For the signal processing complexity, we consider the number of spreading codes in a codedivision multiple access (CDMA) system, or the number of antennas in a digital beamforming system such as Long Term Evolution (LTE), or the number of radio frequency (RF) chains in a hybrid beamforming system such as in 5GNR. The modems studied in this work encompass Mobile Station Modem (MSM) 3000/3100 addressing the IS95 (2G) specifications through the X50 modem addressing 5GNR with intermediate stopping points in the CDMA2000 (2.5G), WCDMA (3G) and LTE (4G) families.
The raw data for WiFi and cellular systems lead to scatter plots of or spectral efficiency, as illustrated in Fig. 1
. For this data, linear regression models of the form
(1) 
and a confidence interval around the regression fit are generated. The methodology behind the linear regression modeling (including confidence interval estimation) is described in
[7, Chap. 2.4.2]. In (1), denotes the metric of interest, denotes the month index, and denotes the random error term. The best linear regression fits are obtained for WiFi data (peak rates) with andcorresponding to a standard deviation of error term of
. Similarly, for cellular, we have , with the standard deviation of error term being . Further, most of the data points (especially for cellular in the recent past) fall within the twosided confidence interval around the regression fits, as seen in Fig. 1. These observations suggest a reasonable fit with the regression model in (1) for the two sets of data.(a)  (b) 
From this study, we observe that the peak data rates have grown almost exponentially (with time) for both WiFi and cellular systems. In particular, the peak rates with WiFi and cellular systems have doubled over a timeperiod of and months, respectively. These observations suggest that both the WiFi and cellular industries have been successful in developing progressive roadmaps with increasing capabilities for their respective modems over time. This relentless growth has been possible due to a number of enhancements over multiple generations of wireless standardization efforts such as:

Increase in bandwidth of signaling transmissions corresponding to higher levels of carrier aggregation [6], as well as increasing bandwidth accretion by mobile network operators over time.

Increase in the number of antennas and the number of RF chains/layers, corresponding to a commensurate increase in cost, complexity, power consumption and realestate at the user equipment (UE) end [6].

Increase in total and effective isotropically radiated powers (TRPs and EIRPs).

Better coding schemes that can achieve higher reliabilities with lower overheads [10] (or higher rates).

More efficient coordinated transmissions and multiple access strategies that manage and mitigate interference, etc. [11].
While such observations have been made in the past [17], this work presents evidence over a significantly longer timeframe and evolution across multiple generations of standardization efforts. From Table II, a more careful compartmentalization of some of the specific factors leading to this exponential growth (with time) in cellular systems is provided in Fig. 2(a). From this plot, we note that a simple linear regression fits the evolution of the peak modulation scheme (with bits/symbol as the metric), log of the peak bandwidth (in MHz) and number of antennas (as a proxy for complexity) with time. Thus, the peak bandwidth appears to be the only exponentially scaling factor with time from the three factors studied here.
More interestingly, we also observe that while WiFi has dominated in peak rates for almost all the time, cellular systems have been catching up rather quickly. This trend is further clear from the spectral efficiency behavior in Fig. 1(b), where cellular has dominated WiFi for a long time. A number of explanations can be offered for these observations. As carrier aggregation efforts have speeded up in cellular systems along with coexistence in unlicensed bands, the main differentiator in performance has been in terms of the number of RF chains/layers and power levels. Since WiFi primarily targets indoor scenarios, the TRP/EIRP is limited to ensure regulatory compliance, which is compensated with wider bandwidths for higher rates, but resulting in poorer spectral efficiencies. Further, the lower cost factor associated with the WiFi modem (relative to the cellular modem) limits the hardware features/capabilities and peak rates, leading to the observed trends.
(a)  (b) 
(c)  (d) 
IiB Future Prospects for Cellular
We now consider the implications of these trends in terms of the future trajectory of cellular evolution. Projecting the historic trends over the next ten years, the peak data rates should evolve as presented in the second column of Table I (see more details later) for certain key milestone points in time. Due to the exponential growth rate in the past, it is not surprising to see projections for peak data rates on the order of a few Tbps in 2030.
Bandwidth necessary ( in GHz)  

Time  Peak data rates (in Gbps)  256QAM ()  1024QAM ()  
Dec. 2020  10.1  1.07  0.54  0.27  0.86  0.43  0.22 
Dec. 2022  29.8  3.17  1.59  0.79  2.54  1.27  0.63 
Dec. 2025  151.1  16.06  8.03  4.02  12.85  6.43  3.21 
Dec. 2030  2259.9  240.22  120.11  60.05  192.17  96.09  48.04 
We assume a similar subframe structure for the airlink specifications as used in 5GNR [18, Sec. 5.3] (namely, a kHz subcarrier spacing with that is possibly expanded/extended to higher carrier frequencies), as illustrated in Fig. 2(b). With a modest bandwidth occupancy, a simplistic calculation shows that the number of modulated symbols per second (denoted as ) that is theoretically feasible with a bandwidth allocation of Hz is given as
(2) 
The peak data rate (in Gbps) assuming an layer transmission, a transmission efficiency^{4}^{4}4Transmission efficiency is a simplistic metric to account for overheads such as control signaling, reference and synchronization signals, and coding. While more involved calculations as in LTE and/or 5GNR can be performed, the simplistic calculations done here shall suffice to extract the main trends on cellular evolution. of , and a ary modulation scheme is given as
(3) 
In particular, with a practical choice such as and different values for (from to ) and (from for QAM to for QAM), the bandwidth necessary to meet the projected peak data rates (at different points in time over the 202030 period) are presented in Fig. 2(c). This information is also presented in Table I for certain milestone points in time over this period. From this data, we observe that the nearterm projections and demands (ca. 2023) can be met with a spectral efficiency improvement of the cellular modem. In particular, a small bandwidth expansion (from MHz to GHz), a hardware complexity expansion (from to RF chains), and a commensurate signal processing overhead increase are sufficient to meet these demands.
However, as we stretch out the roadmap far into the future (ca. 2030), a substantial portion of the bandwidth necessary to meet these peak data rates (at least GHz) is not realizable except at the millimeter wave, submillimeter wave and THz regimes. Further, such a rate scaling depends on licensingrelated complexities across multiple disparate geographies to be resolved before the commercialization of these future modems. In a more realistic setting of an average yearonyear bandwidth accretion (denoted as ) of MHz, GHz, or GHz, Fig. 2(d) plots the achievable peak data rates relative to the current trends projected into the future. This plot also reinforces the ability to meet nearterm trends, but not the trends far into the future.
IiC Potential UseCases for Future Cellular Modems
The above studies showed that even partially meeting the historic trends on cellular data rate growth could only be possible with increased bandwidths of signaling transmissions. Such an increase would take us into higher carrier frequencies than those currently envisioned or used today (e.g., , , , , , or GHz, etc.). Meeting these high data rate and low latency requirements can be extremely challenging, if not impossible, beyond certain key milestones at price points of commercial interest. Nevertheless, even reaching these milestones would require the design of robust, lowcost and energyefficient hardware (antennas, RF frontends, etc.) that work across multiple wide frequency bands and at higher carrier frequencies than possible commercially today.
While the necessity/usecases for the sustained high peak data rates as projected above are unclear as yet, three possible applications are listed below.

Applications on the cellular phone can coordinate with other devices/sensors and monitor human health nearconstantly and noninvasively producing large amounts of data. Transmitting such data from the phone to other inferencing nodes in an edge computing framework could necessitate high data rate bursts as well as addressing securityrelated concerns [19, 20].

Advanced driver assistance systems (ADAS) that help with cognitive distraction detection, collision avoidance and accident prevention, semiautonomous driving, etc., are expected to be a cornerstone of post5G systems as the cellular industry attempts to address the needs and demands of other horizontal industry segments. Such systems are expected to consist of a number of devices/sensors performing realtime monitoring tasks in highly dynamic environments. Coordinating such systems with other vehicles, processing/inferencing nodes on busy streets or downtown settings, or even other pedestrians via either the Cellular VehicletoEverything (CV2X) or the Dedicated Short Range Communications (DSRC) protocols requires high data rate links with ultralow latencies.

Coexistence of Bluetooth, WiFi and cellular systems to offer a single universal ecosystem providing universal mobile coverage in an always on, always connected framework has been long overdue and a possible solution could require higher rates and lower latencies than possible today. In this context, from Fig. 1(a), we observe that while WiFi has dominated in peak rates for almost all the time, cellular systems have been catching up rather quickly. This trend is further clear from the spectral efficiency behavior in Fig. 1(b), where cellular has dominated WiFi for a long time. Such a crossover is bound to have significant impact on viable coexistence solutions.
Iii Challenges at Higher Carrier Frequencies: Illustrative Examples
At this point, it is important to take a segue and to note that the key progresses in communications and information theories [2, 4] (as well as much of technology and engineering) have been built on simple models that reflect real systems, that are mathematically elegant, and lead to a deep intuition on system design and practice, aptly summarized by the famous maxim of George Box [21]:
“Since all models are wrong the scientist cannot obtain a ‘correct’ one by excessive elaboration. On the contrary following William of Occam he should seek an economical description of natural phenomena. Just as the ability to devise simple but evocative models is the signature of the great scientist so overelaboration and overparameterization is often the mark of mediocrity.”
As we march into the post5G era of higher carrier frequencies, more care is necessary especially since much of system design intuition relies on simplistic models, primarily inherited from our understanding of sub GHz systems. We illustrate how such legacydriven understanding can fail with two studies on GHz systems.
Iiia Discrepancies in Delay Spread
The delay spread is an important metric characterizing a wireless channel and is used to understand its frequency coherence properties. The first step in estimating the delay spread is to estimate the gains and delays of all the propagation paths from the transmitter to the receiver. The excess and rootmean squared (RMS) delay spreads of the channel are then computed as given in [22, (4) and (5), p. 6526]:
(4)  
(5) 
where and denote the delay and power corresponding to the th path in an omnidirectional antenna scan.
(a)  (b) 
In the first study, an indoor office environment (the third floor of the Qualcomm building, Bridgewater, NJ) described in [22] and [23] is studied at and GHz with five transmitter locations offering coverage for the whole area. A number of receiver locations in the building are considered and two scenarios are studied: the transmitter that provides the best link margin is chosen for each receiver location, and only one a priori chosen transmitter is made active for all the receiver locations. All the transmitter and receiver locations are deployed with omnidirectional antennas at either or GHz. In either scenario, the gains and delays from the transmitter to the receiver are estimated using an electromagnetic raytracing software suite such as WinProp^{5}^{5}5See more details at https://altairhyperworks.com/product/FEKO/ WinPropPropagationModeling.. More details on the experiments conducted are described in [24]. Fig. 3
(a) illustrates the cumulative distribution function (CDF) of the RMS delay spread for either scenario at the two frequencies. From this study, we observe that transmitter diversity reduces the delay spread as expected. More importantly, these studies show that the RMS delay spreads are comparable across
and GHz, and the medians are less than in both cases.In the second study, a channel sounder (described in detail in [22]) that allows omnidirectional antenna scans at , and GHz is used to study the RMS delay spreads at the same indoor office location. The transmitreceive location pairs used for CDF generation here are similar to those used in the raytracing study described previously. Fig. 3(b) illustrates the CDF of the RMS delay spreads for lineofsight (LOS) and nonlineofsight (NLOS) links at these three frequencies. Unlike the earlier study, we observe that the RMS delay spreads of NLOS links generally decrease with carrier frequency, whereas the LOS behavior is inconsistent with frequency. Further, the raytracing study appears to significantly underestimate the true delay spreads estimated from measurements.
Two plausible explanations are put forward in [22] to explain the discrepancies seen with raytracing: i) Waveguide effect where long enclosures such as walkways/corridors, dropped/false ceilings, etc., tend to capture more electromagnetic energy than a simplistic LOS scenario in raytracing and also increase observed delay spreads with frequency, and ii) Radar crosssection effect where small objects of sizes commensurate with the roughness of surfaces such as walls, light poles, metallic objects, etc., take part in propagation at higher frequencies (by increasing the number of channel taps) and distort the delay spreads. In general, a raytracing software primarily captures scattering due to buildings and large objects/macroscopic features in the environment, and only those features that are explicitly modeled. Thus, raytracing misses out on many potential (small) reflectors and scatterers and cannot be relied on to accurately capture the delay spread in a wireless channel at higher carrier frequencies.
IiiB Discrepancies in Hand Blockage Loss
Another important feature of transmissions at millimeter wave, submillimeter wave and THz carrier frequencies is blockage of the transmitted signal by obstructions in the environment. In particular, electrically small objects at microwave carrier frequencies become electrically large at higher frequencies affecting the antenna’s radiation performance. Specifically, the blockage loss associated with the hand holding a formfactor UE has become an important metric to understand at these carrier frequencies.
(a)  (b)  (c) 
To understand the effect of the human hand, in the first study, a simplified model of the UE corresponding to a typical size of and designed for transmissions at GHz is studied. As is common with sub GHz frequencies, a simplified model of the UE is studied in an electromagnetic simulation framework (see details in [25] and [26]). In particular, several layers of materials emulating a realistic formfactor design such as glass with a thickness of , LCD shielding beneath the glass, FR4 board, etc., are incorporated in the simulation framework. In addition, a battery and few shielding boxes of random sizes are placed over the printed circuit board (PCB) and are modeled. All the metallic objects are connected to the ground plane of the PCB which covers its bottom plane. For the antennas, multiple subarrays are placed on the longleft and topshort edges of the UE as illustrated in Figs. 4(a)(b) for the Portrait and Landscape modes, respectively. The antenna modules are designed on a relatively low loss dielectric substrate (Rogers 4003) and are placed on the FR4 substrate. The antenna elements are either dipole elements or dualpolarized patch elements with the size of each subarray being . The antennas are designed to radiate at GHz and are simulated in Freespace (with no hand) and with a hand phantom model, as also illustrated in Fig. 4.
In terms of electromagnetic properties, the hand is modeled as a homogeneous dielectric with the dielectric properties of skin tissue. These dielectric properties determine the penetration depth of signals into the hand and the reflection of electromagnetic waves from the hand. At GHz, a relative dielectric constant and conductivity S/m are used in the studies [27]. The UE is then simulated with and without hand using a commercial electromagnetics simulation software suite such as CST Microwave Studio^{6}^{6}6See https://www.cst.com/products/cstmws for details.. Fig. 4(c) illustrates the CDF of hand blockage loss using simulated data captured as the differential in beamforming array gains between Freespace and Portrait/Landscape modes over a sphere around the UE.
In the second study, a GHz experimental prototype described in [23] and capturing the attributes of a 5G basestation as well as a formfactor UE design is used to study hand blockage loss based on measurements. The UE design used in these studies corresponds to the same setup studied with simulations earlier. In these studies (see details in [25] and [26]), the UE is grabbed by the hand and the hand completely covers/envelops the active antenna arrays on the longleft edge. All the subarrays at the UE side except the enveloped subarray are disabled in terms of beam switching thus allowing us to capture the hand blockage loss in terms of received signal strength differentials between the pre and posthand blocked scenarios. Multiple experiments are performed with different hand grabbing styles, speeds, with different air gaps between fingers, and with different people. For each experiment, ten received signal strength indicator (RSSI) minimas spanning the entire event from signal degradation to recovery upon removing the hand are recorded. Link degradation is computed as the RSSI difference between the steadystate RSSI value and the ten minimas. The empirical CDF of hand blockage loss corresponding to such experiments is plotted in Fig. 4(c) along with a simple Gaussian fit (specifically, of the form ) to the data.
This study illustrates the wide discrepancy between simulationbased studies and true measurements of blockage loss. Underestimating blockage losses can lead to a poorly designed UE with less antenna module diversity than necessary to effectuate its seamless functioning. A number of plausible explanations can be provided for these discrepancies. These include a poor understanding of the wide variations in material properties (such as the human hand) at higher carrier frequencies as well as the dynamics of hand blocking, impact of materials in the formfactor UE on signal distortion and deterioration [28], capability of simulation studies to only capture those features that can be deterministically modeled, etc. This example illustrates the need for great care in extrapolating established techniques for systems studies, often based on sub GHz systems, to higher carrier frequencies.
Iv Implications on Broader Research Aspects
The arguments put forward in Sections II and III focus on the importance of higher carrier frequencies and the difficulty of simplistic simulation studies in capturing the true impact of these systems. These observations have a number of broad implications for future directions in physicallayer research.

In terms of the specific blockage study of Sec. III, overcoming blockage losses at higher frequencies requires mechanisms that endow path diversity at far higher levels than sub GHz systems. One such mechanism is the use of modular UE designs with multiple antenna arrays. In contrast to sub GHz systems (such as LTE), such modular designs would require a careful optimization of the antenna modules to tradeoff power consumption, diversity/spherical coverage, cost and implementation constraints such as realestate issues. The divergence from sub GHz systems in terms of UE design would require further careful studies of antenna module placement tradeoffs [28]. Another mechanism could be the use of densified networks with multiple transmission points, which would also naturally allow the use of higherorder modulation and coding schemes.

At a general level, the studies described in Sec. III clearly demonstrate the gap between traditional simulation studies with simplistic models (often, but not always, inherited from legacy systems) from real observations in the field with measurements. Thus, in terms of philosophy, without closing the gap between theory and practice of higher carrier frequency systems, the results produced from simplified models can become meaningless in terms of the big picture in the post5G era.

This closing of the loop requires multiple steps:

A careful understanding of the different components of the system and how they interact with each other.

Accurate models that capture these interactions and the contours of the solution space along with the objective function(s) for optimization.

A proposed solution which can then be applied to the real scenario and studied in terms of its efficacy in solving the original problem of interest.

Refinement of the model, the solution space, the proposed solution(s), and its/their fit to the original problem.


At an algorithmic level, the studies described in Sec. III
suggest a possible role for nonparametric or even machine learninginspired approaches
[29, 30, 31, 32] in supplanting traditional statistical signal processing and inferencing solutions in a number of applications in the cellular phone at the sensing, processing and communications levels. However, their success would rely on engineers’ ability to extract intuition into the structure of these solutions. 
In terms of physicallayer transmissions, directional hybrid beamforming approaches over sparse channels are of importance at higher carrier frequencies than traditional digital beamforming approaches [33, 34, 35, 36, 28, 37, 38]. The cost and complexity tradeoffs in implementing such approaches need further study as 5G standardization efforts mature and branch off to even higher carrier frequencies.

While higher carrier frequency systems are important for post5G evolution, advances of sub GHz systems cannot be ignored (or deemphasized) in future research and development efforts. In particular, advances in terms of formfactor UE designs with realestate constraints targeting lower power consumption and acceptable thermal stability, advanced physicallayer capabilities and feature sets for different/emerging usecases as well as highlymobile applications, robust coverage with carrier aggregation over multiple contiguous/noncontiguous spectral bands, and meeting various regulatory compliance requirements (all at similar or lower cost and complexity in implementation [39]) are of importance at both sub GHz and higher carrier frequencies. For example, while 4G systems primarily targeted the enhanced mobile broadband (eMBB) usecase, 5G (and beyond) systems already target other important usecases such as ultrareliable low latency communications (URLLC), iIoT and mMTC with enhancements to nonterrestrial networks (e.g., drones), integrated access and backhaul, coexistence in unlicensed bands, positioning systems and vehicular coverage, etc., expected shortly. With such a diverse set of applications, both sub GHz and higher carrier frequency systems are expected to play a prominent role in future efforts.

While fulfilling all these objectives in a reasonable manner takes a significant amount of time and energy, such endeavors should be actively encouraged and rewarded in terms of research funding and policy initiatives. Some recent examples in this direction include the U.S. National Science Foundation’s Platforms for Advanced Wireless Research (PAWR) program (
https://www.nsf.gov/funding/
pgm_summ.jsp?pims_id=505316
) and Millimeter Wave Research Coordination Network program (http://mmwrcn.ece.wisc.edu) for fostering academiaindustry interactions in the post5G era. 
All this said, the fundamental dilemmas confronting a theoretician in physicallayer research will continue to grow manifold as these systems will continue to breach the boundaries of circuit theory, electromagnetics, communications, optimization, statistics, signal processing, and economics. Thus, it is imperative that a modern telecommunications department develop a core curriculum that spans these hitherto distinct focus areas. Furthermore, it is important that such a department equip itself with at least one advanced wireless testbed and offer handson exposure to the theory and practice of stateoftheart telecommunications technologies to its students and researchers.
V Concluding Remarks
The last twenty five years have been witness to a remarkable exponential scaling in cellular modem capabilities with time. A number of technological innovations such as carrier aggregation, higherlayer multiantenna transmissions with more device and circuit complexities, higherorder modulation and reliable coding schemes, network densification, coordinated transmissions and interference management, etc., have played a key role in this relentless growth. Sustaining these growth rates at historic levels into the far future is both difficult as well as possibly needless due to lack of strong business usecases (at least as of now). Nevertheless, there are enough opportunities in terms of both technological innovations and emergent usecases to sustain slower growth rates in modem capabilities with time. A central component in the evolutionary roadmap of the cellular modem would be operation over a significantly wider bandwidth across a number of higher carrier frequencies (than currently possible today).
While such a reality is already visible today given that 5GNR addresses millimeter wave systems (e.g., Qualcomm’s X50 modem), the focus of this work has been on the more dramatic implications of such trends for physicallayer research problems that would be of relevance in the next few years. This paper philosophically argues that systems research in its own cocoon and isolated from other areas such as circuits/device design, electromagnetics, economics, etc., would be futile, especially as we march inexorably to communications at higher carrier frequencies. Syncretic systems research needs to be both encouraged and advanced from a policy standpoint, and in the nature and scope of curriculum development and departmental structure across universities.
Acknowledgment
The authors would like to thank Jung Ryu and Andrzej Partyka for studies on delay spread, and M. Ali Tassoudji, Lida AkhoondzadehAsl, Joakim Hulten and Vladimir Podshivalov for studies on hand blockage reported in this paper. The authors would also like to acknowledge the critical feedback and encouragement of Thomas J. Richardson on the evolution of this paper. The authors would like to thank the feedback from M. Ali Tassoudji, Kobi Ravid, Jung Ryu, Tianyang Bai, Ashwin Sampath, Ozge H. Koymen, YuChin Ou, Wei Yu, Erik G. Larsson, Emil Björnson, David J. Love, Srikrishna Bhashyam, Akbar M. Sayeed, Wei Zhang, and Durga Malladi on earlier drafts of this article.
References
 [1] R. W. Lucky, “Coding is dead,” Lucky Strikes Again: (Feats and Foibles of Engineers), pp. 243–245, 1993.
 [2] R. Blahut, I. Csiszár, D. Forney, P. Narayan, M. Pinsker, and S. Verdú, “Shannon theory: Present and future,” IEEE Inform. Theory Newsletter, vol. 44, no. 4, pp. 1–10, Dec. 1994.
 [3] G. Frantz and A. Gatherer, “The death and possible rebirth of DSP,” IEEE ComSoc Tech. News, Apr. 2017, Available: [Online]. https://www.comsoc.org/ctn/deathandpossiblerebirth dsp, Accessed on June 19, 2018.
 [4] M. Dohler, R. W. Heath, Jr., A. Lozano, C. B. Papadias, and R. A. Valenzuela, “Is the PHY layer dead?,” IEEE Commun. Magaz., vol. 49, no. 4, pp. 159–165, Apr. 2011.
 [5] C. Links, “White paper on ‘WiFi data rates, channels and capacity’,” Dec. 2017, Available: [Online]. https:// www.qorvo.com/resources/d/qorvowifidatarates channelscapacitywhitepaper, Accessed on June 6, 2018.

[6]
Qualcomm,
“Snapdragon modem comparison,”
Aug. 2017,
Available: [Online]. https://www.qualcomm.com/
snapdragon/modems/comparison
, Accessed on June 6, 2018. 
[7]
D. C. Montgomery, G. G. Vining, and E. A. Peck,
Introduction to Linear Regression Analysis
, Wiley, 5th edition, 2012.  [8] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner, A. Damnjanovic, R. T. Sukhasvi, C. Patel, and S. Geirhofer, “Network densification: The dominant theme for wireless evolution into 5G,” IEEE Commun. Magaz., vol. 52, no. 2, pp. 82–89, Feb. 2014.
 [9] J. An, K. Yang, J. Wu, N. Ye, S. Guo, and Z. Liao, “Achieve sustainable ultradense heterogeneous networks for 5G,” IEEE Commun. Magaz., vol. 55, no. 12, pp. 84–90, Dec. 2017.
 [10] T. J. Richardson and S. Kudekar, “Design of lowdensity parity check codes for 5G new radio,” IEEE Commun. Magaz., vol. 56, no. 3, pp. 28–34, Mar. 2018.
 [11] D. Gesbert, S. V. Hanly, H. Huang, S. S. Shitz, O. Simeone, and W. Yu, “Multicell MIMO cooperative networks: A new look at interference,” IEEE Journ. Sel. Areas in Commun., vol. 28, no. 9, pp. 1380–1408, Dec. 2010.
 [12] S. Verdú, “Spectral efficiency in the wideband regime,” IEEE Trans. Inform. Theory, vol. 48, no. 6, pp. 1319–1343, June 2002.
 [13] K. Wang, Y. Wang, Y. Sun, S. Guo, and J. Wu, “Green industrial Internet of Things architecture: An energyefficient perspective,” IEEE Commun. Magaz., vol. 54, no. 12, pp. 48–54, Dec. 2016.
 [14] F. Han, S. Zhao, L. Zhang, and J. Wu, “Survey of strategies for switching off base stations in heterogeneous networks for greener 5G system,” IEEE Access, vol. 4, pp. 4959–4973, 2016.
 [15] J. Wu, S. Guo, J. Li, and D. Zeng, “Big data meet green challenges: Big data toward green applications,” IEEE Systems Journ., vol. 10, no. 3, pp. 888–900, Sept. 2016.
 [16] J. Wu, “Green wireless communications: From concept to reality,” IEEE Wireless Commun., vol. 19, no. 4, pp. 4–5, Aug. 2012.
 [17] S. Cherry, “Edholm’s law of bandwidth,” IEEE Spectrum, vol. 41, no. 7, pp. 58–59, July 2004.
 [18] 3GPP TR 38.802 V14.2.0 (201709), “Technical Specification Group Radio Access Network; Study on New Radio access technology physical layer aspects (Rel. 14),” Sept. 2017.
 [19] R. Atat, L. Liu, J. Wu, G. Li, C. Ye, and Y. Yi, “Big data meet cyberphysical systems: A panoramic survey,” IEEE Access, vol. 6, 2018, Available: [Online]. https://ieeexplore.ieee.org/document/8533338.
 [20] R. Atat, L. Liu, H. Chen, J. Wu, H. Li, and Y. Yi, “Enabling cyberphysical communication in 5G cellular networks: Challenges, spatial spectrum sensing, and cybersecurity,” IET CyberPhysical Systems: Theory & Applications, vol. 2, no. 1, pp. 49–54, 2017.
 [21] G. E. P. Box, “Science and statistics,” Journal of the American Statistical Association, vol. 71, no. 356, pp. 791–799, Dec. 1976.
 [22] V. Raghavan, A. Partyka, L. AkhoondzadehAsl, M. A. Tassoudji, O. H. Koymen, and J. Sanelli, “Millimeter wave channel measurements and implications for PHY layer design,” IEEE Trans. Ant. Propagat., vol. 65, no. 12, pp. 6521–6533, Dec. 2017.
 [23] V. Raghavan, A. Partyka, S. Subramanian, A. Sampath, O. H. Koymen, K. Ravid, J. Cezanne, K. K. Mukkavilli, and J. Li, “Millimeter wave MIMO prototype: Measurements and experimental results,” IEEE Commun. Magaz., vol. 56, no. 1, pp. 202–209, Jan. 2018.
 [24] J. H. Ryu, A. Partyka, S. Subramanian, and A. Sampath, “Study of the indoor millimeter wavelength channel,” Proc. IEEE Global Telecommun. Conf., San Diego, CA, pp. 1–6, Dec. 2015.
 [25] V. Raghavan, L. AkhoondzadehAsl, V. Podshivalov, J. Hulten, M. A. Tassoudji, O. H. Koymen, A. Sampath, and J. Li, “Statistical blockage modeling and robustness of beamforming in millimeter wave systems,” Submitted to IEEE Trans. Microwave Theory Tech., 2018, Available: [Online]. https://arxiv.org/abs/1801.03346.
 [26] V. Raghavan, V. Podshivalov, J. Hulten, M. A. Tassoudji, A. Sampath, O. H. Koymen, and J. Li, “Spatiotemporal impact of hand and body blockage for millimeterwave user equipment design,” IEEE Commun. Magaz., vol. 56, no. 12, pp. 46–52, Dec. 2018.
 [27] N. Chahat, M. Zhadobov, R. Augustine, and R. Sauleau, “Human skin permittivity models for millimeterwave range,” Electronics Lett., vol. 47, no. 7, pp. 427–428, Mar. 2011.
 [28] V. Raghavan, ML. Chi, M. A. Tassoudji, O. H. Koymen, and J. Li, “Antenna placement and performance tradeoffs with hand blockage in millimeter wave systems,” To be published in IEEE Trans. Commun., 2019.

[29]
S. Dörner, S. Cammerer, J. Hoydis, and S. ten Brink,
“Deep learning based communication over the air,”
IEEE Journ. Sel. Topics in Sig. Proc., vol. 12, no. 1, pp. 132–143, Feb. 2018.  [30] A. Alkhateeb, S. Alex, P. Varkey, Y. Li, Q. Qu, and D. Tujkovic, “Deep learning coordinated beamforming for highlymobile millimeter wave systems,” IEEE Access, vol. 6, pp. 37328–37348, June 2018.
 [31] X. Chen, J. Wu, Y. Cai, H. Zhang, and T. Chen, “Energyefficiency oriented traffic offloading for wireless networks: A brief survey and a learning approach for heterogeneous cellular networks,” IEEE Journ. Sel. Areas in Commun., vol. 33, no. 4, pp. 627–640, Apr. 2015.
 [32] RF. Liao, H. Wen, J. Wu, H. Song, F. Pan, and L. Dong, “The Rayleigh fading channel prediction via deep learning,” Wireless Commun. and Mob. Comput., vol. 2018, no. Article ID 6497340, 2018.
 [33] V. Raghavan, S. Subramanian, J. Cezanne, and A. Sampath, “Directional beamforming for millimeterwave MIMO systems,” Proc. IEEE Global Telecommun. Conf., San Diego, CA, pp. 1–7, Dec. 2015.
 [34] V. Raghavan, J. Cezanne, S. Subramanian, A. Sampath, and O. H. Koymen, “Beamforming tradeoffs for initial UE discovery in millimeterwave MIMO systems,” IEEE Journ. Sel. Topics in Sig. Proc., vol. 10, no. 3, pp. 543–559, Apr. 2016.
 [35] V. Raghavan, S. Subramanian, J. Cezanne, A. Sampath, O. H. Koymen, and J. Li, “Singleuser vs. multiuser precoding for millimeter wave MIMO systems,” IEEE Journ. Sel. Areas in Commun., vol. 35, no. 6, pp. 1387–1401, June 2017.
 [36] M. R. Castellanos, V. Raghavan, J. H. Ryu, O. H. Koymen, J. Li, D. J. Love, and B. Peleato, “Channel reconstructionbased hybrid precoding for millimeter wave multiuser MIMO systems,” IEEE Journ. Sel. Topics in Sig. Proc., vol. 12, no. 2, pp. 383–398, May 2018.
 [37] V. Raghavan, J. J. Choi, and D. J. Love, “Design guidelines for limited feedback in the spatially correlated broadcast channel,” IEEE Trans. Commun., vol. 63, no. 7, pp. 2524–2540, July 2015.
 [38] G. Li, J. Wu, Z. Chen, X. Luo, T. Tang, and Z. Xu, “Performance analysis and evaluation for active antenna arrays under threedimensional wireless channel model,” IEEE Access, vol. 6, pp. 19131–19139, 2018.
 [39] L. Chiaraviglio, N. BlefariMelazzi, W. Liu, J. A. Gutierrez, J. van de Beek, R. Birke, L. Chen, F. Idzikowski, D. Kilper, P. Monti, and J. Wu, “Bringing 5G into rural and lowincome areas: Is it feasible?,” IEEE Commun. Standards Magaz., vol. 1, no. 3, pp. 50–57, Oct. 2017.
Appendix A Explanation of Data Used in Our Study
For the time evolution axis of WiFi and cellular data rates, we begin with Jan. 1997 as Month 1 and Dec. 2020 as Month 288 for the data. Table II provides the peak WiFi data rates and standard specification release dates. This table also provides the peak cellular downlink data rates, release dates and modem capabilities used in our study.
WiFi 
Standard  Specification release date  Peak downlink data rate (in Mbps)  Bandwidth (in MHz) 

802.111997  June 1997  2  22  
11a  Sept. 1999  54  20  
11b  Sept. 1999  11  22  
11g  June 2003  54  20  
11n  Oct. 2009  150  40  
11ac  Dec. 2013  866.7  160  
11ad  Dec. 2012  6757  2160  
11ax  Dec. 2019  1134  160  
11ay  Dec. 2019  20000  8000 
Modem  Approx. commercial  Peak downlink  Bandwidth  Modulation  Antennas  Processing  
identifier  sampling date  data rate (in Mbps)  (in MHz)  complexity  
IS95 
MSM 3000  July 1998  0.0144  1.25  BPSK  1  1 code 
MSM 3100  July 1999  0.064  1.25  BPSK  1  1 code  
CDMA2000 
MSM 5000  Sept. 2000  0.1536  1.25  QPSK  1  1 code 
MSM 5100  May 2001  0.3072  1.25  QPSK  1  1 code  
MSM 5500  July 2001  2.4576  1.25  16QAM  1  1 code  
WCDMA 
MSM 5200  Apr. 2002  0.384  5  QPSK  1  1 code 
MSM 6275  Sept. 2005  1.8  5  QPSK  1  5 codes  
MSM 6260  Jan. 2007  3.6  5  16QAM  1  5 codes  
QPSK  1  10 codes  
MSM 6280  July 2006  7.2  5  16QAM  1  10 codes  
QSC 7230  Mar. 2009  10.8  5  16QAM  1  15 codes  
MSM 7x30  Sept. 2009  14.4  5  16QAM  1  15 codes  
MDM 8200  Jan. 2010  21.6  5  64QAM  1  15 codes  
MDM 8200  Jan. 2010  28.8  5  16QAM  2  15 codes  
MDM 8220  July 2010  42.2  5  64QAM  2  15 codes  
10  64QAM  1  15 codes  
LTE 
MDM 9x00  Mar. 2010  100  20  16QAM  2  2 layers 
MDM 9x25/X5  Apr. 2013  150  20  64QAM  2  2 layers  
MDM 9x35/X7  June 2014  300  40  64QAM  2  2 layers  
X10  Apr. 2015  450  60  64QAM  2  2 layers  
X12  Oct. 2015  600  60  256QAM  4  6 layers  
X16  June 2016  1000  80  256QAM  4  10 layers  
X20  June 2017  1200  100  256QAM  4  12 layers  
X24  June 2018  2000  140  256QAM  4  20 layers  
5G 
X50  June 2019  5000  800  64QAM  UEspecific  2 layers 
Comments
There are no comments yet.