While this article is about frequencies you typically use in Wifi communication, it touches another point.
I am a ham radio operator and with many radio devices delivered these days I witness the following. Mostly high quality components are used but they are assembled in a way that lacks knowledge of harmonics, interference, cross-talk, parasitic effects in general. PC board design makes it obvious and results in a sub-quality end product.
It seems like a basic engineering craft which got lost in the last 30 years.
I don't consider RF PCB layout basic engineering and I don't think it's lost. Plenty of us EEs have read Pozar and could make a competent RF layout given time and resources. I'd first guess that bad layouts are a management decision.
Yes. But if you really want to get into it you need to do it. You could make some cheap PCBs and measure them with a cheap network analyzer. RF design has never been more accessible. The books will get you started and you will start to think about things like de-embedding circuit elements which is very important and not covered in books. You will also start to see why people consider RF design challenging; e.g. curves in the books will depict a "trend" in the circuit and in real-life, everything is a capacitor and/or inductor which causes the real life measurements to deviate substantially from the ideal for reasons that are not always obvious.
An even more cheaper accessible intro to RF is just plain old analog electronics. Building amplifiers ,etc.
I had the good fortune of being exposed to both Pozar, an excellent lecturer and a wizard of an RF lab postdoc when studying RF engineering, and the utility of simply spending lots and lots of time in the lab, trying all sorts of implementations of filters, transmission lines, antennas and whatever caught our fancy, prototyping with wild abandon, verifying (or getting stymied by) the results, refining our measurement skills along the way...
Priceless.
Then all RF design jobs disappeared overnight and I ended up doing PLC coding instead of working with radio - but, hey, University was great fun, and RF design definitely is one of the fields where there's still room for intuition and a bit of artistry in addition to just knowing the hard theory of the subject.
As I said, nature is analog, it will never go away. Interesting areas of analog research are in things such as switched capacitor architectures where you can filter as close to the antenna as possible, reducing the impact of an interferer on the rest of your signal chain. Also similar is continuous time signal processing where you can perform your signal processing on an analog signal without discretizing it in time, effectively giving you the ability to make sense out of your analog data as close to the sensor (or antenna) as possible.
The problem is you can't tweak these things with a software update. Their relative inflexibility makes them unpopular outside of a defense or research setting but analog solutions are typically higher in performance than digital solutions but that can depend on definition as well.
Where are all the jobs going? There are a lot of RF things going on at an ever increasing pace but I don't see a lot of people hiring in the field. Do all the defense primes just have a silo of gray beards they keep in a dungeon somewhere?
Yes. Many analog design groups in college went from 10's of students 10-15 years ago to just a few now, not due to lack of funding but due to a lack of interest. Our interface to nature is through analog signals but due to the heavy reliance on HW intuition and its relative inflexibility (but more 'elegant solutions' if you ask me) analog electronics continues to grow in unpopularity in industry and academia follows. Yes, there are just some older guys updating old designs in many instances. The push for RFSOCs and similar products is partially to reduce reliance on the greybeards. Its a shame because there are very interesting areas of analog electronics such as continuous time signal processing.
TLDR: To answer your question, to reduce the reliance on a relative few engineers, the market solved it with a few products that are 'just good enough'. This has become a self perpetuating cycle, race to the bottom. There will still be many analog/RF EEs but relegated to a few niche sub-industries serving mainly the DoD and RFIC industries. Even the telecom industry is planning to pipe back raw, downconverted baseband data to a central location to do SDR DSP rather than rely on a modem at the basestation. All for flexibility in lieu of power/efficiency, NRE etc.
Would you recommend making the transition if it was available? I've been eyeing a lateral move from optical control/processing software into a sort of full stack rf role. Nobody is hiring from what I've seen but everywhere I've been I end up moonlighting with the RF team helping with DSP and high throughput ADC on down stuff. They're always understaffed and I have my amateur extra and know just enough to be dangerous so I get pulled on to help.
This is the right idea, in many cases, the "HW" team might focus on RF but does "antennas to bits" (i hate that term) because of the need to get a prototype up and running to test it. If you have baseband circuit and DSP experience you could be poised for a lateral move especially since the operational frequency will be less critical for your skills. For example, without years of mmWave experience a move into doing mmWave RF wouldn't make sense but that doesn't mean you can't work on other parts of the mmWave system.
RF HW design is very much a systems engineering field and folks that have experience in each block are more desirable than ones that focus on only one piece in most cases.
You aren't painting a fair picture. RFSoCs cost more but they also have much more dynamic range (both amplitude and temporal). It's just just about configurability, but also performance.
You really are proving my point. Based on your misuse of nomenclature its obvious you're not a HW guy. The ADC is the last in an RF chain and typically sets the systems dynamic range if you've done things right. There is nothing special about the ADC in the RFSOC other than the designer will not need to worry about creating a physical, digital interface to the programmable logic.
I don't even know where to start on your "temporal dynamic range" comment....
I'd love some lab space to really get into this. All the old high performance stuff used to be designed basically by hand — that takes money and true craft.
I spend most of my time in theoretical work of some kind, but get a real urge to build...
If you want to dig in just have a look at the RX-888 and it's ilk. Pretty much a case study in an effort to hook a cypress usb3 chip up to an ADC with the lowest BOM cost possible. Forget the matched predriver, forget buffers, forget analog, forget the clock, forget the reference designs and application notes altogether really. But oh yeah make sure you spec the 16 bit part-- we'll have none of that 14 bit shit thank you.
Here's a super crude ELI5, feel free to correct me.
Signals crash course: Radio comms have a channel they need to fit into (let's say 1.1GHz - 1.2GHz as an example). This means that what you send over the air is basically a high frequency (~1.15GHz) sine wave. You encode data by slowly varying this sine wave (amplitude and/or frequency). The variations are going to be super slow, as fast changes in either amplitude or frequency will make your signal go outside the frequency limits. It's only the sine wave (aka the "carrier frequency") that's high frequency (1.15GHz in our case).
There are several ways to digitize this signal. You can do "direct RF sampling" - you need a very high sampling rate (2.4GHz+ due to Nyquist in our case) and just sample the high frequency signal.
You can multiply (in analog domain) the RF signal from the antenna by another frequency to keep all the slow changes but change the carrier frequency. This means that you don't need such a high lower sampling rate (you can get away with much less than GHz). If your new carrier frequency is just some low but non-zero frequency, that's super-heterodyne (IF). If it's zero, that's Zero-IF.
I was surprised they didn't use the word homodyne.
One issue they didn't mention, but probably shows up in the noise figures is that homodyne receivers suffer from 1/f noise in the relatively wide band "direct conversion" amplifiers. Of course all of the heterodyne receivers suffer from 1/f up-conversion at the mixer.
It is rather a soup of terminologies namely homodyne, zero-IF, and direct conversion architecture are all referring to the same thing.
It seems they are now settling on RF sampling for the newer transciever architecture technology but sometime direct RF is also being used for example this latter terminology is referred by Intel for their new line of FPGA and RF transceivers product integration.
This is the first time I've seen it explained like this, and having struggled with these concepts in the past, I found your explanation really easy to understand. Thanks!
For both RF sampling and Zero-IF the two of the most popular analog manufacturers are TI and Analog Devices.
It's rather intriguing that for RF sampling technology, Intel has recently join the bandwagon and called it a game changing technology when the RF sampling is integrated with high speed FPGA [1].
Intel Delivers a Game-Changing, Analog-Enabled Direct RF FPGA Portfolio:
Not push back to me. As an "analog", I love SDR-SOC but what this is saying is that people may lack a fundamental knowledge when designing hardware devices that is contributing to these challenges.
Id also push back against the article in the sense that I think the industry shot itself in the foot trying to provide too many options that all overlap. But, I do agree with the sense that minimizing the radio architecture to only what it needs to operate is smart.
Not only that, but it also stems from "management" or "customers" unable to contemplate tradeoffs in the design. The reason FPGAs and RFSOCs are pushed so much in certain industries (we all know which one) is because the customer wants everything reconfigurable because they can't make decisions now, or in most cases, not competent enough to make decisions now. However, they are OK with buying a $25K+ FPGA or RFSOC which is why the market is there and also why few understand the actual art of RF design and radio architecture anymore. RF budgeting is almost non-existant. Most Jr. engineers have never seen a level diagram and the only places you will find them is in old watkins johnson app notes. SDR design is ultimately a race to the bottom in most cases in terms of good design but not necessarily cheap design.
I am a ham radio operator and with many radio devices delivered these days I witness the following. Mostly high quality components are used but they are assembled in a way that lacks knowledge of harmonics, interference, cross-talk, parasitic effects in general. PC board design makes it obvious and results in a sub-quality end product.
It seems like a basic engineering craft which got lost in the last 30 years.