Qualcomm, Nvidia ready for 'AI-native' 6G, if only the world knew what it was

It seems like just yesterday that the 5G rollout started. Now, at Mobile World Congress, major companies are already talking about commercializing 6G. Never mind that binding 6G standards haven't been nailed down yet. The world has been joking about when telecoms would push the sixth generation of mobile networking ever since …

[Rich 2]

Utter bollocks

Firstly, “… ever since 5G came to dominate in the earlier part of the decade”

Maybe I’ve been living in a cave, but this is news to me. I thought 4G was still much more widely used than 5G

As for the “AI” shite, I would love to know how sprinkling a turd shower of “AI” on to the mobile network it’s going to magically change fundamental radio modulation schemes - the number one distinguishing feature between each ‘G’ from 1 to 5 so far. Saying that “AI” Will allow for updates (say to 7G) without reinventing the hardware is total bollocks. It’s why “software defined radio” in general is such a half-arsed concept because it too completely ignores the fundamental RF side of things

[elsergiovolador]

Re: Utter bollocks

Ignores as in software defines it. *wink*

[ABugNamedJune]

Re: Utter bollocks

Truly. Whenever I hear news like this I have to wonder if it's time we pump the brakes a little bit. I only got a 5g enabled phone last year, and the bandwidth is fine. My partner was running her whole house off a little 5g puck, and while it wouldn't suit my needs, it worked for her just fine to play games and stream shows.

If no one's asking for hardware updates, might as well shoe in "AI" and rake in billions from dull-eyed "investors"

[that one in the corner]

Re: Utter bollocks

At the risk of being a kill-joy over the enjoyable "what you on about, ain't no useful 5G here, fool"

> ever since 5G came to dominate in the earlier part of the decade

is probably meant to refer to dominating the industry media machine's self-pleasuring exercises. Now that prattling on about 5G phones has moved from the Mobile World Congress to the Carphone Warehouse telly ads.

And, sorry to be disagree, but

> “software defined radio” in general is such a half-arsed concept

SDR lets (some) radio applications be performed more cheaply by allowing more generic hardware be applied, only changing the software/configuration side of things. That is all it need do to be a useful concept: let a few (suitable) things be done a bit more cheaply.

Of course, like absolutely everything else, you'll then get it hyped as far as it possibly can be, to the point of total charlatanry - although you gotta love the way that the UFO and ghost hunters have taken to running GNU Radio on their laptops: that scrolling frequency map just looks so sciencey!

[Rich 2]

Re: Utter bollocks

But it’s the RF bit that’s the most complex and expensive. If you’re not changing that bit in your new version of your SDR then you’re just twiddling with the upper layers. Ok - that might be all you need to do for your application but you probably don’t need all the baggage that the typical SDR framework brings with it to achieve that - in fact you’re almost certainly better off without it

This classic article is also an amusing read…

https://arstechnica.com/information-technology/2012/06/how-to-blow-6-billion-on-a-tech-project/

[Swiss Anton]

Multiplexing

I believe it was the 1930/40s film actress Hedy Lamarr who played a pivotal role in her other career as inventor in laying the foundations for multiplexing that allows for more efficient use of the radio spectrum. The point being that you don't need to update the RF hardware to make it better, you just need to us it more effectively, and I can certainly see a case for AI doing that.

[Rich 2]

Re: Multiplexing

If you take a radio system with the intention of improving its spectrum efficiency, then you either need to change the modulation scheme, or the media access protocol or something similar. These things must be done at the RF level - in this day and age, often via some FPGA signal processing magic. Regardless, it’s right at the coal face - exactly the bit that all SDR models ignore because “it’s too hard”

Shoving an LLM (because that’s what “AI” means these days) into your upper software layers isn’t going to change this. All it will do it eat up power and processor bandwidth and memory for no benefit other than to make some AI outfit richer.

[Swiss Anton]

Re: Multiplexing

Except that we already know how to do that at the RF level. What I'm thinking of is that Ai can reduce the data that needs to be sent by sending a vectors that are generated when it initially analysis an input. If the sender and receiver are using the same LLM, then the sender only need to send the token that represent a bit of input, not the whole thing.

[Rich 2]

Re: Multiplexing

You can tokenise your data regardless of the medium. And you certainly don’t need an LLM to do that - we’ve doing it for decades

[DS999]

No one wants 6G

People eagerly awaited 3G, then LTE and then 5G because they needed more speed and/or more capacity so they didn't have to live with plans limited by MB or GB.

With 5G that's a thing of the past. All but the absolutely bargain basement plans are unlimited (for reasonable values of "unlimited") and in areas with decent infrastructure, fast enough that there isn't much need for more speed.

As infrastructure continues getting upgraded, and 5G Advanced appears later this decade, there will be zero demand for 6G. No one will upgrade their phone to get one that supports 6G, or switch carriers to get one that promises 6G connectivity. The mobile industry has yet to wake up to this fact.

[Kevin McMurtrie]

A sequel of the characters you don't like

I thought 5G has all of that AI edge support, but nobody wants it. The spectrum efficiency boost is great and some telcos are even phasing out LTE. The rest of the 5G features are silly.

I guess every patent troll on Earth wants a piece of 6G licensing. Over here in America, we doomed our broadcast TV upgrade by cramming it full of DRM, interactive ads, and patent trolling.

[Bebu sa Ware]

"AI-native, open, secure and trustworthy platforms"

Pick one and hope for the least worst outcome.

"Open" possibly has the fewest death traps but open as in "open wound" also comes to mind.

[Anonymous Coward]

6G networks, built on AI-RAN architecture

Too late, they blew that up over the weekend. Try Baghdad? Although they only have Nvidia's existing kit built in its AI-RACK architecture.

[Swiss Anton]

Fire resistant?

Will the 6G transmitting equipment be less flammable than the current 5G tech? (Asking for a friend).