I'm not an expert in UMTS-style cellular networks, but I think the numbers in this article are misguided or incorrect. :-( E.g., the article seems to conflate

I'm not an expert in UMTS-style cellular networks, but I think the numbers in this article are misguided or incorrect. :-( E.g., the article seems to conflate "Internet-style" measurements of practical round-trip latency over a network path, with "telecom-style" measurements of round-trip latency across a single unloaded radio hop with immediate schedule availability. E.g. it gives a figure of "50ms" for the typical latency of 4G, and "1ms" for the typical latency of 5G. I don't think this is an acceptable comparison. "50ms" is a believable ICMP ping time from a commercial LTE service, including delay to get a resource block, to sojourn queues inside the network, etc. (I just tried on T-Mobile LTE and am getting 21-69 milliseconds RTT, mean 36 ms, to the next hop.) By contrast, "1ms" for 5G sounds like a 3GPP "user plane latency" figure (and even then, only for the "uRLLC" version of 5G that is probably only going to be used for more-exotic purposes). This is a lower-level latency measurement made on a single unloaded radio hop. You cannot compare the two. I think the more-accurate 3GPP user-plane latency figures would be: - LTE: about 5-6 ms (see https://communities.theiet.org/blogs/426/444 https://communities.theiet.org/blogs/426/444 or https://www.techplayon.com/5g-nr-user-plane-latency/ https://www.techplayon.com/5g-nr-user-plane-latency/ or https://www.artizanetworks.com/resources/tutorials/req_lte.html https://www.artizanetworks.com/resources/tutorials/req_lte.h...) - 5G eMBB ("mobile broadband" service ): about 3-4 ms (https://www.techplayon.com/5g-nr-user-plane-latency/ https://www.techplayon.com/5g-nr-user-plane-latency/ , https://www.itu.int/en/ITU-R/Documents/ITU-R-FAQ-IMT.pdf https://www.itu.int/en/ITU-R/Documents/ITU-R-FAQ-IMT.pdf) - 5G uRLLC ("Ultra-reliable low-latency communication"): about 0.2-0.7 ms (see above) The practical IP round-trip time with all of these networks is much larger than the above; e.g. ~21-69 ms for LTE, and 22-76 ms for 5G eMBB services (https://www.lightreading.com/the-edge/how-5g-is-pushing-envelope-on-latency-/d/d-id/766273 https://www.lightreading.com/the-edge/how-5g-is-pushing-enve...). And even those figures are in the absence of load to or from the UE; if traffic has filled up the queue of the uplink (on the baseband chip) or downlink (in the UE-specific queue on the base station/eNodeB/gNodeB), users can experience latencies of 500+ milliseconds. This is called "bufferbloat" and it comes from a refusal by the baseband and eNodeB/gNodeB vendors to deploy queue disciplines that queue on a per-flow basis, or mark or drop packets when queues are large. (I have talked to many vendors about this, and the prevailing attitude is somewhere between "we are the phone company; we're legally obligated to transmit 99.999% of all packets to the endpoint, no matter how delayed everything gets" and "if there were an actual problem with latencies over our networks, companies would be coming to us like Facebook did and paying us a bajillion dollars to deploy custom traffic rules to let their traffic skip the line; nobody else has offered us this kind of money for a custom traffic rule, ergo there is no problem.") Similarly, the article seems to somewhat conflate improvements in theoretical throughputs with improvements in the "typical download speed" a user might experience. The latter is not that strongly related to the air interface standard (e.g. LTE or 5G-NR) because any flow's experience depends heavily on how many other flows are sharing the same constrained resource, which in turn depends on (a) how much spectrum has the carrier licensed in your area, (b) how many base stations has the carrier decided to build/lease in your area, (c) how many other customers does this carrier have, trying to download right now through this base station, on this slice of spectrum, at this moment, and of course (d) is the base station actually close to you [do you have good SNR?] or is it spending all its airtime on slower modulation and coding? Carriers sometimes use improvements in the air interface (which improve throughput-per-Hz) to save money by deploying less spectrum or larger cells. So even if a newer/fancier air interface is x% faster on a per-Hz basis, the benefits may not fully accrue to the users, if the carrier takes advantage of the improvement to deploy bigger cells, or less spectrum per cell, etc. (I once sat in a meeting with AT&T around the time of the 3G-to-4G transition, and in at least this particular setting, they were planning for roughly 1/3 of the improved spectrum efficiency to accrue towards better service for their users, and the remaining 2/3 towards saving resources for themselves. So, yes, improvements in technology do somewhat improve service for users, but... in some cases that may be a minority of the overall benefit.) For more rants on this topic: http://blog.keithw.org/2013/07/3g-and-me.html http://blog.keithw.org/2013/07/3g-and-me.html https://www.marketplace.org/2010/11/05/tech/4g-networks-dont-exist-or-do-they/ https://www.marketplace.org/2010/11/05/tech/4g-networks-dont... And, obligatory plug! In September, the IAB is holding a workshop on Measuring Network Quality for End-Users (https://www.iab.org/activities/workshops/network-quality/ https://www.iab.org/activities/workshops/network-quality/), with submissions due August 2. If you care about this stuff, and maybe you know something or have a perspective that could be helpful to the Internet community, please consider submitting to and participating in the workshop.