Modems are harder to get right than CPUs. A CPU needs to do math and work with digital data lines. A modem needs to talk to the mobile network, using radios, and a bunch of standards that were built up since the 90s.

Tacking on: the devil in a modern baseband is in the firmware. The analog domain stuff is generally handled by increasingly commoditized RF frontends outside of the baseband processor, and then either clean analog IQ or more recently, even digital samples (DAC and ADC moved to frontend) are sent to the baseband processor.

A baseband is a really fancy specialized SDR. Most are based on arcane VLIW DSP architectures like Qualcomm’s Hexagon or Samsung Marconi - you’ll usually have several DSPs handling the different physical layer channels, and then some coordinating DSPs doing L1 channel mixing and timing (in 4G and 5G, various logical and transport channels are muxed into the same physical channels).

Then a set of higher level processor cores (usually referred to as CP, sometimes still a DSP but often a general purpose application processor like ARM) will handle the MAC and above. There will be occasional fixed function blocks for some common protocol functions, but generally it’s less “analog magic” in the way people think when they hear “radio” and more “DSP magic.”

If you look at the math it's not even that complicated. Basic statistical inference and information theory.

Seems like the difficult part is doing that effectively while avoiding IP issues -- patents on software and math have entrenched Qualcomm's dominance.

Imagine if all of the IP for ML or AI were held by a single company that got the regulating body (ITU/3GPP) to require their use. Makes a mockery of FRAND terms.

Patents are the main thing here.

Building a cellular modem, complete with working protocol stack, entirely in software has been done as a 1 person open source project.

Making a production -ready modem is clearly more complex, but far from hard for a company with the resources of Apple.

However, doing so whilst not violating any of qualcomms huge array of patents is the real challenge.

I'm willing to bet that the release date of this phone probably closely aligns with the expiry date of a patent they couldn't work around.

Patents are the dumbest thing on the planet.

Where do you draw the line for protection of innovative ideas?

There might be someone in government right now willing to flush obvious software patents

It is extremely hard to design a modem like this. Just the ADCs and DACs you need are like 2-3 year development and 1-2 year testing before going into production.

Also patents. Everything in the cell modem world is horribly encumbered by patents -- many of which I believe Qualcomm owns.

The original M-series chip team left years ago, or at least the major drivers. Also: radios are hard, super hard.

I wouldn't expect these initial baseband implementations to be fantastic software wise.

That's probably why it's in a weird release between 16 and 17.

The final round of beta-testing has been assigned to Apple's economy-class customers.

Someone has to be first.

CPUs can be easily developed in an ecosystem where annual updates and the ability to end-of-life hardware older than a decade are the norm. Cellular network opereators share none of those properties with Apple, and so any new modem chip designed to interoperate with carriers would necessarily be an order of magnitude more difficult to implement than an CPU that only needs to interoperate with Xcode would be.

Except making a decent modem is indeed harder than start of art CPU. People loves to shit on Qualcomm but dont appreciate that amount of work involved.

The hard part is not the modem design per se.

The hard part is designing a good modem while also unambiguously working around all the Qualcomm patents in all the jurisdictions that have iPhone, which is all of them.

Because if you don’t do that, you’re still paying Qualcomm which defeats an important purpose of making your own modem.

All those patents are standards-essential, which means Qualcomm (and anyone else involved in the standardization process for cellular networks) has to license them under "FRAND[0]" terms, which at least for this scenario means "for the same price we'd license them to Samsung or Intel[1]". It also means that design-arounds are not possible; if you design a different technology from the one Qualcomm owns, you're not speaking 5G anymore. That's why we have these legal rules on standards-essential patents.

[0] Fair, reasonable, and non-discriminatory. Each one of those words has a funny legal definition separate from whatever English you're thinking of.

[1] Who, incidentally, sold Apple the modem division that made the C1, because Intel is nothing but a bottomless pit of bad management decisions

I’ll admit with not being up to speed here, my information is actually fairly old, but Qualcomm is famous for skirting its FRAND obligations and as far as I know “no license, no chips” is still de rigor. https://www.sunsteinlaw.com/publications/no-license-no-chips...

Apple is most probably paying Qualcomm for these technologies. They think however that it is commercially beneficial to have their own implementation.

In 2025 it is time to give credit for Apple's PR spreading whole misinformation online during and after the trial. Macrumours and 9to5mac loves to repeat those narrative about Apple no longer has to pay any patent fees when they make their own modem.

As the reply below, Apple still has to pay its SEP. Given Apple has its current deal with Qualcomm until 2027 with time to extend further we likely won't know the full details. Previously it was 5% of Wholesale price for all Qualcomm patents whether they are SEP, wireless or not. With a cap or maximum $20 per smartphone meaning the Pro range don't have to pay a lot more. And rebate towards the modem Apple purchase. The reality after deducting rebate Apple was paying closer to 5%. For reference Ericsson ask for 3% on 5G SEP, previous Cory ruling suggest reality was closer to 2%.

I would buy that a decent modem is harder than many CPU designs, maybe even most. But harder than state-of-the-art? Surely not, have you seen the complexity?

And even CPUs (esp state of art) have to worry about radio effects, as in avoiding internally and across chipset.

I never worked on radio modems, but I've worked on wireline modems: the issue is not the math or the standard algorithms, it's making it interoperable with everything that's existing out there and all environment conditions while working around bugs of other implementations.

It is a ridiculous amount of work and if you're new to the business, it takes a long time just to be build the lab test suite. And you need to support not just the latest and greatest protocols but also legacy ones. The operators have their own say and test labs as well and they all have slightly different setups and requirements.

> And even CPUs (esp state of art) have to worry about radio effects, as in avoiding internally and across chipset.

Radio effect are rarely an issue with regular chips. Crosstalk within a chip only happens between wires that are within hundreds of nanometers separated from each other.

> Radio effect are rarely an issue with regular chips.

I'm an amateur, so welcome correction with this, but I'm also not convinced. For one example, ref this RFI mitigation with Intel's gen 12 processors: https://edc.intel.com/content/www/cn/zh/design/ipla/software...

I meant RF impact on the internal functioning of the chip, not the other way around. EMI has always been a thing, it's not specific to state of the art CPUs.

State of the art CPUs aren't fundamentally different to more basic CPUs. They just have fancier microarchitectures, better branch predictors, more cache, etc.

Easy to believe radios would be an order of magnitude harder, what with the ancient proprietary standards and actual physical radio stuff. (The closest CPUs get is serdes and in my experience those are bought in from Synopsys et al.)

s/work/patent blackmail/g

The CPU/SOC operates independently with limited protocols (such as PCIe, NVMe, etc.). In contrast, a baseband processor must communicate with various base stations and core networks, requiring precise timing, physical signal alignment, and correct message formats. Even though the 3GPP defines mobile network standards, achieving full compatibility remains extremely challenging.

When you're apple, you can just be mostly compliant, and the worlds mobile networks will have to work around your bugs.

Nobody will use a phone company that doesn't work with iPhones.

A lot of the mobile networks use off the shelf hardware. That was one of the reasons Huawei was painful to remove from global phone networks. If they don't work, they will not work over multiple networks.

If there are reports of iPhones failing to work reliably from Kansas City to Kuala Lumpur then it would be unlikely to be the operators being blamed here.

Not necessarily. People will notice if it doesn’t work as well as their last iPhone.

In fairness, Apple started by licensing the ARM core which gave them a base to work from. The modem is fantastically harder unless you can license an existing design that's not qualcomm.

They bought out Intel's modem business (giving them something like to work from), afaik.

One difficult I remember reading is: they can't always do things an obvious way someone else did because... patents.

Which is ironic, considering that patents are supposed to be non-obvious.

We're talking about a company that took another one to court because both of their products had rounded corners.

And similarly, copyright was supposed to "promote the Progress of Science and useful Arts", but here we are

Their first CPU was launched in 2010, and development likely began in 2006-2007. The M1 launched to much fanfare in 2020.

M chips are ARM so while they designed their particular architecture for it they were still building on some pretty solid foundations.

A modem, well nobody's going to help them build one of those; it'd put them out of business - ARM's business _is_ selling access to its body of work.

Their CPU is based on ARM IP, it’s not like they designed it from the ground up. The modem they might have had to? Also I guess a big challenge are all the patents you need to be aware of and find ways to license or circumvent, not sure if that’s worse for modems but it seems all wireless / cellular technology is riddled with patents (not saying CPUs aren’t but again ARM is probably taking care of some of that for Apple).

No, that's not how the Apple Silicon works - it was at the very start, but the processor cores haven't been based on Arm IP since the A6, since then the Arm designed cores were replaced with fully custom cores that are just compatible with the Arm instruction set.

Same with the GPU except that was later on - it was a licensed PowerVR IP until Apple started using their own custom GPU from the A11.

They bought intel’s entire modem unit, so it’s not a ground up design, but maybe that makes it worse… who knows what was causing intel to not be competitive in that space…

> Their CPU is based on ARM IP, it’s not like they designed it from the ground up.

It's the opposite, AArch64 was designed by ARM in order to produce their CPUs.