How will small design companies deal with higher frequency systems?

[TheUnnamedNewbie]

So this is something I've wondered about a few times and discussed at work a lot with my colleagues.

I work on systems at 150 GHz. Everything we do at these frequencies requires expensive tools. PCBs, antennas, packages, traces, are all simulated in $$$ software packages, we need people whose only job and expertise is how to get a 150 GHz signal off a chip and into a PCB trace, and they spend weeks tweaking these interfaces. Other people spend weeks designing antennas and complaining to the mechanical guys that no, you cannot, in fact, just slap some plastic there in that mold because it absolutely destroys the antenna performance.
With 60 GHz consumer applications also becoming a thing, I wonder if this will make it harder for small companies to design these products. You can't get away with rules of thumb anymore. I find it hard to see 1-man design houses design products with these higher frequencies integrated.

What is your perspective on this? Perhaps from someone who is not in this industry, as I'm clearly not unbiased or anything.

[hans]

Don't have a clear answer... All I can think of, with your mentioned example of even getting a signal to interface nicely onto a PCB or antenna... put everything on a module and CE/FCC certify it. End users integrate said module onto their design with keepout areas or necessary ground planes as to not affect the modelling work/assumptions for sensitive circuits or radiating elements.

If you want a tighter integration, then you need to do more work. A module may have the impedance matching etc. all figured out for the right stackup, antenna, etc. Or you grab the RF chip itself with unmatched I/O and S-parameter files, and do the impedance matching yourself with e.g ADS.

[T3sl4co1l]

Whaddya mean, we don't even design 433MHz connections, let alone GHz!  (THz maybe )

Maybe a few like routes and antennas, preferably following the appnote to have some confidence of success.  Even just having a VNA is extra (but, much more affordable nowadays), let alone how to use it properly (remember, RF theory/practice in general is not that widespread below say masters levels, and most EEs are BS).

Just as well, since certifying those modules is a big deal, and being able to plop one in and exempt that bit of spectrum from the EMC test is big value.

Tim

[jonpaul]

AFIK, Above 5 GHz the links are for network infrastructure eg undersea cables and long haul microwave links.

The attenuation of the atmosphere and precipitation makes mm wave systems >>20 GHz short haul.

The limited applications require the specialized components and tools.

j

[2N3055]

There are few RFID chips that have integrated antennas. TI has 77GHz radar transceiver with antenna on package.
Rising frequencies allow that for short range devices.
We are going to buy those things as a ready made modules like GPS..

I already buy 24.125 GHz radar modules for vehicle detection as ready made modules. Not only design but regulatory compliance, certifications etc. are nightmare for small teams. Cost is high and recouped only in volume.

[nctnico]

Likely modules are the way to go for now but I can see this kind of technology trickling down into cheaper or even free software. When I started out 16MHz was very fast for a microprocessor on a board. Nowadays I design boards that have 5Gbit/s connections running over them and the software tools to design & simulate such boards are affordable.

[T3sl4co1l]

AFIK, Above 5 GHz the links are for network infrastructure eg undersea cables and long haul microwave links.

The attenuation of the atmosphere and precipitation makes mm wave systems >>20 GHz short haul.

The limited applications require the specialized components and tools.

These are things like short range radar (as mentioned above), and extremely high bandwidth, short range communication; think Bluetooth but able to stream Blu-rays in minutes, maybe even seconds.  Extremely high bandwidths with reasonable signal quality and synthetic aperture (diversity to the Nth degree, shall we say) all on chip are feasible on these scales.  Scary crazy fancy stuff.

Likely modules are the way to go for now but I can see this kind of technology trickling down into cheaper or even free software. When I started out 16MHz was very fast for a microprocessor on a board. Nowadays I design boards that have 5Gbit/s connections running over them and the software tools to design & simulate such boards are affordable.

A ton of work has gone into making these links feasible for the non RF expert to lay out, and on economical materials (FR-4).  Stuff like transmitter preemphasis, error correcting coding, aligning multiple bitstreams, etc.  The SNR / eye opening possible on these systems is, of course, excellent.  (I suppose they don't even use ECC very much, since the BER can be so low; although I don't recall if newer PCIe versions do.  Radio obviously benefits, optimizing power levels.)

I suppose the final frontier is not to bother with dirty wires at all; as long as there's some kind of path or waveguide out of the box, let the transceiver echo cancellation figure it out.  When there's bandwidth there's a way.

We're already seeing something like that with 5G+, smaller cells with ever more bandwidth.

Tim

[TheUnnamedNewbie]

I suppose the final frontier is not to bother with dirty wires at all; as long as there's some kind of path or waveguide out of the box, let the transceiver echo cancellation figure it out.  When there's bandwidth there's a way.

We're already seeing something like that with 5G+, smaller cells with ever more bandwidth.

Tim

It is a problem still with computational power. You see this already with DSPs in highspeed ethernet. Doing that stuff at 25 Mbaud is nice and all, but doing it at 25 Gbaud is a whole other kettle of fish. Your tranciever for 112 Gbit/s consumes 250 mW. The DSP to deal with all of the non-ideal crap uses 5 W.

I guess you are right though, let the MIMO algorithms figure a bunch of this stuff out. I'm working at the limits of what is possible, so everything has to be optimized to the nth degree, and you can't leave a few dB's hanging somewhere because you can be sure your competitor wont.

But I support getting rid of wires. Wires suck.

Likely modules are the way to go for now but I can see this kind of technology trickling down into cheaper or even free software. When I started out 16MHz was very fast for a microprocessor on a board. Nowadays I design boards that have 5Gbit/s connections running over them and the software tools to design & simulate such boards are affordable.

While I'm sure that the technology trickels down to cheaper software (though they have been saying that about IC design software too and there is still no viable analog IC design flow out there that doesn't cost >tens or hundreds of k$). However, I am more worried about the complexity of design and the knowledge required from the designers, there is just too much crap (I think) for one person to be fluent in designing the antenna and the package and the housing and the digital FPGA design and....

AFIK, Above 5 GHz the links are for network infrastructure eg undersea cables and long haul microwave links.

The attenuation of the atmosphere and precipitation makes mm wave systems >>20 GHz short haul.

The limited applications require the specialized components and tools.

j

Atmospheric attenuation is actually not that big (except some dips in the spectrum due to molecular resonances, such as around 60 GHz). Sure, link loss is higher due to the smaller wavelength, but once you throw some arrays at it (which you can easily do if your antenna is wwaaay smaller) and you get the same range.

I think it is a chicken and egg problem - the expertise required makes it hard to penetrate a general market. It's not that there are limited applications, it's just a pain in the ass to design for them.

[rstofer]

Capitalism will again provide a solution!  As the technology finds its way to the mainstream, more people will start to design for it.  Equipment will improve and cost of entry into the market will decline and, eventually, the technology will be quite pedestrian.  Or, the technology doesn't provide a mass benefit and it will forever be used by niche players.

In my undergrad years, TTL was the big new thing.  We had RTL and DTL but just barely.  Had it not been for the space program of the '60s, these technologies might have taken much longer to reach the market.  Still, I don't think a sober mind would have envisioned multi-core, multi-GHz CPUs.  These advances were also expensive during the startup years.  Now anybody can use them.  Somebody knows how to design the motherboards.

[T3sl4co1l]

It is a problem still with computational power. You see this already with DSPs in highspeed ethernet. Doing that stuff at 25 Mbaud is nice and all, but doing it at 25 Gbaud is a whole other kettle of fish. Your tranciever for 112 Gbit/s consumes 250 mW. The DSP to deal with all of the non-ideal crap uses 5 W.

I guess you are right though, let the MIMO algorithms figure a bunch of this stuff out. I'm working at the limits of what is possible, so everything has to be optimized to the nth degree, and you can't leave a few dB's hanging somewhere because you can be sure your competitor wont.

Yeah, will need a few generations of DSP optimization to make that combination feasible.

I wonder, at what point do you just throw everything into neural networks -- can they be made to run fast enough?  (Are memristors at all useful up here (or at least at whatever IF might be)?)

Stupid I'm sure, to throw the "we don't know how it works" box at an otherwise well-understood problem we just need a certain well-defined amount of computation to handle -- but eh, if there's a way to do it fast, and cheap enough (in terms of power), eh... who cares if it spits out a few more bits in error if the radiation pattern saves 10s of dB and watts of P?

How not-impressed would the FCC be if you marry a "we don't know"-box to an antenna, anyway?......

...Also, sustained rates likely don't matter, at least for a lot of applications?  You'll max out an SSD (maybe not NVMe) pretty easily at these rates, and, there's not too much you'd use on a phone that needs more than streaming video?  If the thing can shut down quickly between bursts (and, Idunno, maybe it's TDMA, probably it's something much cleverer), and you've got well over 5W of peak capacity from your next-generation battery, well... that might turn out fine anyway.

Atmospheric attenuation is actually not that big (except some dips in the spectrum due to molecular resonances, such as around 60 GHz). Sure, link loss is higher due to the smaller wavelength, but once you throw some arrays at it (which you can easily do if your antenna is wwaaay smaller) and you get the same range.

I think it is a chicken and egg problem - the expertise required makes it hard to penetrate a general market. It's not that there are limited applications, it's just a pain in the ass to design for them.

"Build it and they will come".  Just a little trouble with that tiny word "build"...

Tim

[rstofer]

"Build it and they will come".  Just a little trouble with that tiny word "build"...

Tim

As I was told many years ago: "Never underestimate technology!"

It's old news but we walked on the Moon on July 20, 1969.  How hard can this new stuff be?

[T3sl4co1l]

It's considerably more engineering effort, but I have no doubt a conveniently-packaged solution can exist.

Tim

[Kasper]

[...]put everything on a module and CE/FCC certify it. [...]

I agree. 

[Marco]

Line of sight RF has limited applications in consumer electronics and competes with light, which in many ways is easier to handle. Though neither LiFi nor mmWave have had any real success for in house broadband.

Imaging mmWave radar is going to be huge I guess, but it will be a whole for that to be cheap enough for anything but high end cars. To me it seems it will be very niche till it's cheap and then just modularized, look at the cheap 24 GHz modules coming out.

[tszaboo]

The same way we were dealing with RF circuits and such. Build several version of the circuit with small changes and pick the one that works.
And you see how a hacked TV receiver chip is providing very cheap VNAs and spectrum-o-meters. Same will happen eventually to that.

[nvmR]

The smaller companies will hire consultants instead of have inhouse engineers with expertise, and rent equipment instead of buying. This lowers the bar, but requires excellence to succeed.