I've always accepted that technology advances. Being born in the 90s, everything just becomes faster, smaller, cheaper and generally better if you wait a few years. This was most obvious with consumer electronics such as TVs, PCs and cellphones.

However, it occurs to me now that I know what drives most of this changes, except for one. Computers and cellphones get better and faster mainly because we are able to build smaller and more efficient transistors (I hear about twice the transistor count per unit of silicon area every two years).

The Internet got faster first with DSL which pushed the bandwidth of landline copper twisted pair to its maximum. When we ran out of usable spectrum inside the copper wire we turned to optic fiber, and it was a whole new game.

TL;DR: But, what is it that makes it possible for cellular networks to keep getting faster? I've had 2G, 3G and now LTE cellphones and the speed differences are astronomical, akin to the differences observed in household internet in the last decade.

Yet, LTE channels don't necessarily have a bigger bandwidth (in fact, I believe LTE uses less: 3G uses 5 MHz channels, whereas LTE can have smaller channels, from 1.4 to 20 MHz). Moreover, I've heard many times that LTE is more efficient in terms of bps per channel Hz (I would add 'citation needed' here, I'll be the first to admit that it at least sounds dubious).

So what is it? Just more spectrum? Better and smaller electronics? Or are we getting better at this in other ways? How so?

  • 11
    \$\begingroup\$ Greed is probably the ultimate driver : businesses are after Profit and lower costs... \$\endgroup\$ – Solar Mike Feb 11 '18 at 14:56
  • 4
    \$\begingroup\$ @SolarMike well I guess you're right, but I didn't mean that. I meant from a technical perspective, what makes it possible. \$\endgroup\$ – freejuices Feb 11 '18 at 15:25
  • \$\begingroup\$ So, what is your question - what will we invent tomorrow or next week that will be faster? \$\endgroup\$ – Solar Mike Feb 11 '18 at 15:28
  • \$\begingroup\$ @SolarMike No, the question would be how are they going to make it faster. PCs will be faster next year because Intel will build CPUs with smaller and more efficient transistors, so they can cram more hardware, for less money, clock it faster and use a lower VDD. But why will the successor of LTE be faster? From a technological stand point, what makes it possible? Digiproc mentioned something along the lines of better algorithms to exploit the channel capacity, that's what I'm looking for. \$\endgroup\$ – freejuices Feb 11 '18 at 15:37
  • \$\begingroup\$ So, you want to know tomorrow's theories today? ie the "how" ?... \$\endgroup\$ – Solar Mike Feb 11 '18 at 15:38

what is it that makes it possible for cellular networks to keep getting faster

Basically, good old Moore's law.

The handset is only half the equation. More modern and powerful silicon does help in getting better channel quality, less noise, etc. However this can't go above the channel bandwidth as per Mr. Shannon.

A simple way to boost the bandwidth available to each user is therefore to slice the landscape into smaller cells. Directional antennas on top of towers slice the "round" cell into quarters, like an orange.

Installing lots of micro/picocells everywhere in densely populated areas means each base station only handles a smaller number of users. Less users per cell means more bandwidth per user. This is enabled by reducing the price of base station hardware (ie, cheap silicon, Moore's Law, and MMICs which integrate the RF bits on-chip).

A smarter system also helps. For example, in GSM, even when you don't talk, your bandwidth time slot is reserved for you, which is wasteful.

An important thing is also the availability of these at a reasonable price:

  • Big FPGAs with truly insane computation power
  • Fast ADCs/DACs
  • Microwave ICs

These enable digital radio, and this is where the juicy bits are, like MIMO and adaptive antenna arrays with real-time beamforming and channel equalization, advanced (and adaptive) modulations, plus strong error-correction codes which require lots of computing power, etc.

  • \$\begingroup\$ Where do FPGAs come into play with respect to cellular networks? I would've thought everything is an ASIC? \$\endgroup\$ – user541686 Feb 11 '18 at 20:31
  • 2
    \$\begingroup\$ ASIC has cheaper unit cost, but FPGA is field-upgradeable... \$\endgroup\$ – bobflux Feb 11 '18 at 20:37
  • 3
    \$\begingroup\$ FPGAs can be economical in lower volumes, or when reconfigurability is necessary. The higher per-unit cost of FPGAs can be preferable to the huge expense of developing an ASIC. FPGAs can be used some very high performance network hardware, cell base stations, etc. which are relatively low volume. For picocells, ASICs are a stronger possibility because there will be a lot more of them. \$\endgroup\$ – alex.forencich Feb 11 '18 at 20:43
  • 2
    \$\begingroup\$ OK! FPGAs are in the base stations. Phones sell in enough quantity to justify ASICs, and they get "upgraded" quite often when people buy new ones anyway. \$\endgroup\$ – bobflux Feb 11 '18 at 21:01
  • 1
    \$\begingroup\$ The gradual proliferation of better ground technologies is a big part of it too; handsets only form a portion of a cellular network. e.g. packet-based timing synchronisation which we've had for years but these things take a long while to seep into industry at scale \$\endgroup\$ – Lightness Races in Orbit Feb 12 '18 at 1:14

I think following are some of the key technologies/techniques driving up cellular data rates.

  1. Move to higher carrier frequencies where wider bandwidths are available. Soon we will have millimeter wave technology being used in cellular.

  2. Multi Input Multi Output (MIMO) Antenna systems allowing parallel transmission of data streams.

  3. Advance modulation schemes such as OFDM and QAM.

  4. Stronger forward error correction codes not requiring re-transmissions and bringing us ever closer to Shannon Capacity.

  5. Shrinking cell sizes. Now we have the same frequency divided among a smaller number of users.

  • 1
    \$\begingroup\$ short and to the point. +1 \$\endgroup\$ – Sredni Vashtar Feb 12 '18 at 1:16

Assuming the same bandwidth, the only way to boost datarates is better coding: QAM versus GSM's MSK, 16QAM versus QAM, 256QAM versus 16QAM,

And in all this, multipathing and fading must be handled.

With more bits per Hertz, the SignalNoiseRatio (SNR) needs to improve, tho coding provides a one-time 5 or 10 dB assist here. To improve SNR, the link needs more ERP (focused TX antennas), higher-gain receiver antennas (more elements, phased arrays, etc giving more area to gather more energy) and shorter paths to reduce pathloss.

  • 2
    \$\begingroup\$ Still, eventually, Shannon's limit will be reached. Once this happens the only possibility for increased speed would be more bandwidth per user, meaning smaller cells. Eventually one could end up with a system that looks like low-powered WiFi where only a few users are on a cell, and at that point standard RF design would be at maximum possible throughput... \$\endgroup\$ – madscientist159 Feb 12 '18 at 8:42

Or are we getting better at this in other ways? How so?

There will possibly come a day when our handsets (or the system) will be able to store the mathematical nuances of our individual voices and manipulate it to form other words algorithmically. Then all that needs to be transmitted in a voice call is "text" and the receiving phone can reconstruct our voices and sound like the actual person.

So to say "have a nice day" would take 15 ascii characters or 120 bits for two seconds of speech.

  • 1
    \$\begingroup\$ Don't forget a few bytes for a smiley at the end, unless you expect everyone to sound very serious on the phone in the future. \$\endgroup\$ – Dmitry Grigoryev Feb 11 '18 at 20:49
  • 1
    \$\begingroup\$ I sure hope it doesn't ever go that far, lest we have something like this happen, but for voice communications instead of scanned documents. I'll call it "autocorrect for audio". \$\endgroup\$ – Aleksi Torhamo Feb 11 '18 at 22:27
  • \$\begingroup\$ So when this becomes a reality, can we no longer trust our friends' voices for the same reason that we can't trust their e-mails or caller ID today? (spoofing) \$\endgroup\$ – AaronD Feb 11 '18 at 23:15
  • \$\begingroup\$ @AaronD actually it will be that you can't trust phone calls. The friend (and their voice) themself remains as trustworthy as ever. \$\endgroup\$ – user253751 Feb 11 '18 at 23:17
  • \$\begingroup\$ @immibis Yeah, that's what I meant. Guess I left some ambiguity in there. Thanks for clarifying. \$\endgroup\$ – AaronD Feb 11 '18 at 23:19

Another critical advancement that hasn't been mentioned is improved utilization of optical fibre networks. An optical fiber can carry an entire spectrum of wavelengths. They haven't always done so, however. Optical filters of increasing precision now allow dozens (or more) "channels" to now be crammed into single fibers where previously they would have only been using two. This lets existing infrastructure (fiber in the ground) carry increasing amounts of data with only the need to upgrade the endpoint equipment. Cellular networks basically sit on top of fiber backbones, so better and faster fiber is a critical part of broader, faster cellular.

This is similar, in some ways, to how the POTS copper went from 2400bps to 50MBps in the span of a few decades.


Not only are designers still coming up with better algorithms to do dynamic audio compression, dynamic channel coding (i.e getting closer to Shannon's limit), and dynamic adaptation to multipath, clutter, and interferers; but as transistors get smaller, we can use more elaborate algorithms for the same amount of battery energy.

  • 1
    \$\begingroup\$ How much impact does channel coding has given that most communication is, or should be, encrypted and hence should not be possible to distinguished from white noise? \$\endgroup\$ – Maciej Piechotka Feb 12 '18 at 1:50
  • \$\begingroup\$ @MaciejPiechotka Not coding as in compression, coding as in modulation. And coding as in error correction (strange as it sounds, adding error correction can increase the data rate as now your "actual" connection can be faster and less accurate to make up for it). \$\endgroup\$ – user253751 Feb 12 '18 at 4:51
  • \$\begingroup\$ @immibis Oh so things like 10b/8b. Makes sense \$\endgroup\$ – Maciej Piechotka Feb 12 '18 at 14:52
  • \$\begingroup\$ @MaciejPiechotka I assume you mean 8b/10b encoding? That coding scheme is mainly for clock recovery and DC balance and only transmits 0.8 bits per symbol. Transmissions with 16-QAM have 4 bits per symbol and transmissions with 64-QAM have 6 bits per symbol. \$\endgroup\$ – Toothbrush Feb 12 '18 at 17:31
  • \$\begingroup\$ @Toothbrush sorry. Last time I dealt with this type of material was at uni and I didn't remember the notation (and before coffee I haven't noticed that google put correct order when I was checking it). \$\endgroup\$ – Maciej Piechotka Feb 12 '18 at 17:48

Not the answer you're looking for? Browse other questions tagged or ask your own question.