Betting against Wi-Fi: The odds of Private 5G for digital transformation
Since the advent of the ETHER Network in 1973, Ethernet has been on a seemingly unstoppable tear – first in the Local Area Network (LAN) and more recently in the Wide Area Network (WAN). As (arguably superior) alternative technologies came and went, an adage emerged that would become jokingly known as Metcalfe’s second law: “Never bet against Ethernet.”(1)
While it has remained unsaid until now, a similar phrase certainly comes to mind when we think about the evolution of wireless technologies in the enterprise. Wi-Fi not only shares a standardization body work group (IEEE 802 LAN) and a common MAC-layer lineage,(2) it has also enjoyed an almost exclusive dominance in the enterprise over the last two decades – to the point of replacing it’s wired alternative, in the 10 years or so, as its speed and reliability have improved. Indeed, Wi-Fi is perhaps the only technology I would readily double-down on. Maybe that’s why the thought of staring down its 6th generation (Wi-Fi 6) in preparation for a fight for the next generation of private wireless enterprise networks has always made me more than a little concerned.
Industrial IoT becomes table stakes.
As seemingly mysterious, at first glance, the acceleration of digital transformation initiatives in the enterprise has much to do with the availability of toilet paper in 2020: The tenuous nature of manufacturing and supply chains in today’s increasingly personalized consumer marketplace. Consequently, between 2020 and 2021 we witnessed an acceleration in the ongoing evolution of workplace autonomous automation that we recognize as Industry 4.0. This in turn is triggering a societal evolution (Society 5.0) (3) where people - rather than industries - are once again taking center stage. Society 5.0 is an advancement that is best described as being enabled by “the interdependence of man and machine using cognitive computing and human intelligence [for delivering] mass customization and personalization for humans.”(4)
When the ITU laid out its vision for international mobile telecommunications beyond 2020 (IMT-2020) in 2015,(5) it included a now infamous pyramid diagram highlighting several business and consumer usage scenarios along with the three new service categories that would support them: Ultra-reliable and low-latency communications (URLLC), massive machine type communications (mMTC) and enhanced mobile broadband (eMBB). The ITU expected the fifth generation of (public) mobile network technologies (5G) to deliver on these requirements and handed the keys over to the 3GPP to make it happen. The result was a new radio access technology (RAT) first defined under release 15(6) as the 38 series of specifications and (rather unimaginatively) called New Radio (NR).
I’ve written a few pieces, in the past,(7) detailing the how New Radio achieves its reliability and low latency goals, thereby providing theoretical proof of why 5G trumps Wi-Fi. However, as these techniques often apply only to (or are only beneficial in) the high band frequency ranges and generally assumed licensed or lightly licensed spectrum is employed, comparing NR to Wi-Fi as if on an equal footing had always concerned me a little. ?I’m in marketing, so I obviously didn’t lose any sleep about it. Indeed, I am generally more than happy to take nuanced technical theory and promote it as a broad fact. That’s what pays the bills, after all. But in this case, it was always lurking deep (deep) in the back of my mind. Fortunately, it has apparently been front and center in the minds of some technologists who recently published their findings in an IEEE paper entitled “Comparing Wi-Fi 6 and 5G Downlink Performance for Industrial IoT.”(8)
Network simulation lays all cards on the table.
That report documents a comparative study of NR in the lightly licensed 3.5 GHz Citizens Broadband Radio Service (CBRS) (9) band with two variations of New Radio Unlicensed (NR-U) (10) and Wi-Fi 6. All variations were deployed in the 5 GHz range using a 40 MHz channel size, which provided a reasonably level playing field across these radio access technologies. The fact that the 6 GHz spectrum was not employed for this study is of little consequence as the simulations assumed a highly controlled RF environment at the deployment site, with minimal interference from other wireless networks. Employing a blend of the indoor factory (InF) and office (InH) scenarios outlined within 3GPP TR 38.901, (11) the underlying environment, base station infrastructure and traffic models are also grounded by a common reference document.
The blended InF/InH scenario floorplan with 2 infrastructure options employed in the simulation.
To keep the maximum number of base stations required for full coverage to 12, the authors opted for a (L)120m x (W)60m x (H)5m space (~InH). They then applied InF environmental options where just 20% of the surface area is occupied with ‘clutter’, averaging 2m in height. The complete opposite of my house - in all aspects. The model calls for either one centrally located, ceiling mounted, 5G gNodeB’s (gNB) or Wi-Fi access points (APs) or 2, 4, 8 or 12 evenly distributed up to 20m apart. In these simulations, the authors assumed 60 static user (IoT) endpoints uniformly distributed across the area and evenly connected between however many base stations were active during the test. This represents the sort of careful planning and design needed for industrial wireless networks. The arrival of 50-byte packets with varying quality of service classes to an endpoint is modeled using a distribution probability theory, frequently employed in queueing, called the Poisson process.(12)
As an Industrial IoT simulation, the tests focused on the ability of these various radio access technologies to deliver on the promise of reliability and low latency. While the authors set a fixed limit of 99.999% (five nines) reliability,(13) their latency cap was 100ms. This is particularly welcome as, while URLLC with it 1ms latency target gets a lot of the attention, most of the deployment action will ultimately be considered mMTC – a broader category of which URLLC is a subset.(14) According to McKinsey, of the two, URLLC will be first out of the gate but by 2025, mMTC growth will start to accelerate. This is primarily an artifact of wide-scale chipset availability and doesn’t negate early trials and testing.(15)
Forecasted 5G IoT unit sales (Source: McKinsey & Company)
These broader mMTC service classes are a little more forgiving to latency across the air (i.e. in the 2ms, 2-10ms and 10-50ms ranges) (16) because the end-to-end delay tolerances for the applications is higher. Indeed, survival times - the time that an application consuming a communication service may continue without an anticipated message - for some applications can be as high as 100ms.(17) Given that one of the early conclusions from the simulation results is that achieving sub-1ms latency with five nines data reliability is particularly tricky, that news is quite welcome. To achieve even modest data rates of just a few Mbps, it was found that traffic with these constraints must operate in environments with only one to two gNB’s or APs. This was due to the effects of RF or cross-channel interference and contention or simply queuing delays as the 60 users vied for base station resources.
Increasing the spread improves the odds
With just a little latency leeway, we can expand the coverage by adding significantly more gNB’s or AP’s. We can also dramatically increase the overall load on the network. By how much? Well, it’s probably finally time to look at the simulation results. As noted, the radio access technologies employed in the test were NR, NR-U and Wi-Fi 6.?While the authors tested NR Frequency Division Duplex (FDD) because the dedicated transmit and receive frequencies would inherently deliver lower latencies, it was also noted that frequency ranges above ~2600 MHz (including CBRS) are not paired, making Time Division Duplex (TDD) the only option. I pretty much ignored FDD, for that reason. TDD transmissions can be delayed as packets wait for their transmission timeslot, but it offers far better performance as loads increase. The introduction of the mini-slot as part of the 5G release specifications was designed to address this by reducing the wait time but is only really applicable in the high band frequency ranges.
The simulation did take advantage of another NR feature, though: The ability to select two channel access mechanisms. Wi-Fi exclusively employs the Load Based Equipment (LBE) method as its Listen Before Talk (LBT) procedure. Although any other radio equipment operating in the same spectrum range must also use LBE (this would be category 4 (CAT4) LBT in NR-U/LTE-LAA parlance), if there is no Wi-Fi present, NR-U can make use of Frame Base Equipment (FBE) LBT instead. Working on a fixed frame period, FBE was designed specifically to support industrial IoT (IIoT) applications demanding lower latencies.(18) This was developed and defined as part of the 3GPP release 17 specifications that kicked off in 2019 with eye towards enhancing IIoT support in NR-U. To determine exactly how advantageous it is to completely kick Wi-Fi out of the room and run NR-U FBE, the simulation compared 2 x gNB’s using LBE with 2 x gNB employing FBE. All other NR tests, for the record, assumed FBE.
Maximum load achievable in the 3GPP factory scenario with five nines reliability and various packet arrival delays employing various radio access technologies and 2 or 12 AP’s or gNB’s.(19)
5G is Industrial IoT’s ace in the hole
To simplify the findings without compromising the integrity of the results, I focused on two of the base station deployment scenarios, comprising 2 and 12 AP’s or gNB’s. The report concluded that 5G NR does indeed deliver a discernable advantage over Wi-Fi 6 in mMTC and URLLC environments. Whereas relative Wi-Fi capacity dramatically declines, with the addition of access points, the inherent scalability of 5G is clear. This can be primarily attributed to superior LBT options and far better performance when multiple antennas are employed. Generally, NR-U results were on-par with those of NR in the lightly licensed spectrum range. This is likely the result of simulation assumptions which defined a highly controlled unlicensed RF environment with no other active antennas. As a footnote, the anomalous disparity between NR and NR-U with 12 gNB’s in the 2ms test results could be a little misleading as up to 280 Mbps was actually achieved with 2.5ms latency.
Obviously, Wi-Fi ultimately has a far broader portfolio of use cases for which it is the ideal technological and business case solution, but there is no doubt it will struggle in Industrial IoT applications. For sure, Wi-Fi can achieve the latency and reliability metrics demanded of these applications, but completely fails to scale. And we have not even covered other areas where Wi-Fi may also lag 5G NR, such as UE density, mobility, and security. Regardless of whether Wi-Fi 6 or NR are ultimately employed, it is apparent that an incredibly high degree of frequency planning and interference control is required when building out wireless infrastructures for IIoT. However, the simulation conclusively proved that 5G new radio deployed using either licensed, lightly licensed or completely unlicensed spectrum can scale its coverage and capacity while still achieving the strict latency and reliability goals demanded by URLLC and mMTC. With that knowledge, I know where I’m now placing my bets.
---
领英推荐
?Disclaimer:
None of the ideas expressed in this blog post are shared, supported, or endorsed in any manner by my employer. The bit about my house is not supported or endorsed by my wife.
?
?Footnotes:
(1)??????Bob Metcalfe – the father of Ethernet – is also known for Metcalfe’s (n2) Law which states that “the value of a?telecommunications network?is?proportional to the square?of the number of connected users of the system.” While Metcalfe himself is not directly credited with the phrase, he quoted it in a 2002 article which he bylined.?
(2)??????Carrier Sense Multiple Access (CSMA) with Collision Detection (CD) in wired Ethernet and Collision Avoidance (CA) in Wi-Fi
(3)??????While an entirely new industrial revolution (Industry 5.0) is often claimed this is not entirely true. The confusion is likely exacerbated by the fact that an evolution (not revolution) in Industry 4.0 is powering the societal revolution, which are on different evolutionary tracks that do not necessarily align. In the same way the German government can claim Industry 4.0, the Japanese Government first talked about Society 5.0. This is not entirely unexpected, given Japan’s well documented decline in critical demographics.
(4)??????Is COVID-19 pushing us to the Fifth Industrial Revolution (Society 5.0)? (nih.gov). Ignore the title. Or ignore my ramblings above.(3) Move on, or just stop reading now. I can’t say I’d blame you, at this point.
(5)??????Aka IMT Vision 2020 (See what they did, there?!) IMT Vision – Framework and overall objectives of the future development of IMT for 2020 and beyond (itu.int)
(6)??????Work on New Radio (NR) continued in release 16 and beyond.
(7)??????Like this What is 5G ultra reliable low latency communications (URLLC)? (metaswitch.com) and this Supporting Low Latency Enterprise Applications with 5G on Public Clouds (metaswitch.com)
(8)??????Open access: Comparing Wi-Fi 6 and 5G Downlink Performance for Industrial IoT | IEEE Journals & Magazine | IEEE Xplore
(11)????Specification # 38.901 (3gpp.org) “Study on channel model for frequencies from 0.5 to 100 GHz”
(12)????The Poisson Distribution and Poisson Process Explained | by Will Koehrsen | Towards Data Science
(13)????Meaning no more that 0.001% of a 20 byte packet can be lost, for anyone who’s counting.
(14)????There have been subsets of both URLLC (Scalable URLLC) and mMTC (Critical mMTC) defined but let’s just park that fact, for the sake of this post and my sanity.
(15)????the-5g-era-new-horizons-for-advanced-electronics-and-industrial-companies.pdf (mckinsey.com)
(19)????Data derived from larger results, comparing 1, 2, 4, 8 and 12 AP’s or gNB’s in Comparing Wi-Fi 6 and 5G Downlink Performance for Industrial IoT | IEEE Journals & Magazine | IEEE Xplore.?