Analog engineering, worth it?

[Enron]

I know there will always be IC fabrication, but what about analog circuits as a whole?

I'm not a big programming fan, but I like to play with audio circuits. I am just wondering
if there is anything really besides audio/radio circuits, that will still be valuable to learn, if
done using analog?

I ask because I want to move on to something else besides audio, but don't want to
program. I want to look into LED light fixtures, but we all know how many times we had to
make a LED light up in class...

[Baliszoft]

Do some research in analog filters. They also come handy when it is about doing AD conversion (before digital signal processing).

[EEVblog]

I'm not a big programming fan, but I like to play with audio circuits. I am just wondering
if there is anything really besides audio/radio circuits, that will still be valuable to learn, if
done using analog?

Tons of stuff.
Precision sensor amplifiers (wheatstone bridge for example), filters, PLL's, integrators, power supplies, to name but a few.

Dave.

[Enron]

I suppose I could work on a project that I always wanted to do. I wanted to do a frequency multiplier, with a guitar through a pedal. I was going to go about that with some kind of analog PLL setup, but maybe I can go ahead and try DSP.

[Enron]

Thanks for some ideas guys! Back to the lab.

[Rerouter]

there is also op amps covered in analog engineering, and they are quite a large chunk of most analog systems, and while they seem simple enough there are hundreds of little traps about them, from offset voltages to common mode ranges, and others,

i myself enjoy the feedback loops side of analog engineering, like using a dozen op amps, with 5 inputs and 3 outputs to make a stable and inter-related control system where all the corrections and actions are done in a very fast analog loop with no need for digital intervention other than feeding in set points and reading off values for operator convenience,

then there are tricks you otherwise might never learn without the variety, like why your 0.1uF capacitor across your logic chips supply rails was not a good choice and causes problems, or why your adc's 10K input resistance is skewing your readings,

[MetraCollector]

You do not have to worry.

I work for ON Semiconductor automotive division as a student in test engineering and there are many workplaces here.
Analog designers are the most respected and desirable engineers. They have usually highest salaries.
Despite it is about IC designing, they have to design filters, voltage converters or design test benches and many other activities.

Gigahertz bands are not usually suitable for digital processing, so it is essential not to forget about analog design 

[thb]

Analog: work is fun, result is boring.
Digital: work is boring, result is cool.

[Neilm]

If you think about most of the test equipment you can think of will use analogue engineering. The front end of an oscilloscope, the input on you multimeter, the input of logic analysers, Dave's ucurrent the list goes on.

I work for a company that makes very sensitive test equipment. It is impossible to measure femto-amps without some very good engineering. We keep finding that recruiting engineers that can do digital is easy. Getting someone who is very good at analogue electronics is hard. Last time we looked we went for 18 months and still didn't get what we wanted.

Yours

Neil

[AndyC_772]

Analog: work is fun, result is boring.
Digital: work is boring, result is cool.

At anything faster than trivial speeds, digital electronics *is* analogue 

[Simon]

At anything faster than trivial speeds, digital electronics *is* analogue 

Very true, digital is but an abstraction of the real world with rules that keeps you in its controlled boundaries which were determined by analog in the first place.

Digital is therefor easier and so often preceded as many can program too now.

I think soon finding people good in analog will be like trying to find a web developer to write in code versus one that just installs WordPress.

[Enron]

Thanks for all the replies!

You guys mentioned some great stuff, to keep me motivated in analog design.
I'm going to explore some things that where mentioned, build them, and keep
adding to my portfolio.

IC design may still be an option. I just never gave it much thought.
 

[AndyC_772]

Digital is therefor easier

Ever implemented a DDR2 or DDR3 memory interface, or anything else that relies on correctly matching a high speed transmission line? The days of digital being just ones and zeroes ended a long time ago...

[StubbornGreek]

Digital is therefor easier

Ever implemented a DDR2 or DDR3 memory interface, or anything else that relies on correctly matching a high speed transmission line? The days of digital being just ones and zeroes ended a long time ago...

Sorry but can't help being a bit of a smart-ass here; it still comes down to the 1's and 0's, just like the words 'yes' and 'thermodynamics' use the same alphabet - one is more complex than the other but they use the same building blocks.

Analogue circuitry is the backbone of all electronics (IMHO).

[saturation]

From an electronics level, all of the natural world speaks analog, so long as you interface with it, take data from or talk back to it, it will require the concepts of analog.   

[AndyC_772]

Sorry but can't help being a bit of a smart-ass here; it still comes down to the 1's and 0's, just like the words 'yes' and 'thermodynamics' use the same alphabet - one is more complex than the other but they use the same building blocks.

With due respect, I think you've just answered my question beautifully. The answer is clearly 'no', you've not implemented such an interface yourself.

If you had, then you'd know all about setup and hold timings (analogue), noise margin (analogue), slew rate (analogue), clock jitter (analogue), duty cycle asymmetry (analogue), termination voltage (analogue), monotonicity (analogue), supply voltage (analogue), termination impedance (analogue), length matching (analogue), crosstalk (analogue), propagation delay (analogue), and the way all those things vary with time, from device to device, and with temperature - all analogue.

The fact that your PC's memory *looks* like a digital system from the outside, and does in fact allow you to store and retrieve ones and zeroes reliably, is a remarkable feat of analogue design.

If you have any doubts at all that it's an analogue design problem, do dig out any CPU manufacturer's application notes on how to lay out a DDR2 or DDR3 PCB. It's quite the eye-opener.

[jahonen]

Yep, successful SI design requires understanding of analog stuff. And also let's not forget about one very important aspect of a product design today, EMC. Failing to meet EMC requirements can cause detrimental loss for the company. Sooner the requirements are met with minimal cost, the better. Good EMC design is all about understanding analog properties of the signals (and some RF concepts which "real" RF designers have had to live with from day zero).

I have used to think that digital is just non-linear high-bandwidth analog stuff, purely digital system exists only in daydreams of mathematician. One good example is to connect a series of gates to a high-frequency clock signal, and when enough gates have been added, the signal just ceases to exist due to the skew accumulation (analog property of logic gates), although from purely digital view, there should be no problem

Regards,
Janne

[Simon]

Digital is therefor easier

Ever implemented a DDR2 or DDR3 memory interface, or anything else that relies on correctly matching a high speed transmission line? The days of digital being just ones and zeroes ended a long time ago...

OK I over simplified and was referring to lower speed stuff like micro's, a previous post had already pointed out that high speed digital is effectively analog.

Example, if your running a motor with 1 KHz PWM most mosfets will be fast enough and you can practically assume the output as square. But go into 100's of MHz and rise and fall times are considered and the fact that the output is somewhat square is acknowledged.

[StubbornGreek]

With the same due respect, I don't understand your point (honestly). I have an understanding of skew, jitter, termination voltage, etc., etc. Also, I don't contest that implementing DDR2 (or 3) requires knowledge and talent. I was simply stating that digital will always amount to 1's and 0's, no matter how complex the overall design becomes.

As far as analogue electronics are concerned, the digital world doesn't even exist without it. 

Sorry but can't help being a bit of a smart-ass here; it still comes down to the 1's and 0's, just like the words 'yes' and 'thermodynamics' use the same alphabet - one is more complex than the other but they use the same building blocks.

With due respect, I think you've just answered my question beautifully. The answer is clearly 'no', you've not implemented such an interface yourself.

If you had, then you'd know all about setup and hold timings (analogue), noise margin (analogue), slew rate (analogue), clock jitter (analogue), duty cycle asymmetry (analogue), termination voltage (analogue), monotonicity (analogue), supply voltage (analogue), termination impedance (analogue), length matching (analogue), crosstalk (analogue), propagation delay (analogue), and the way all those things vary with time, from device to device, and with temperature - all analogue.

The fact that your PC's memory *looks* like a digital system from the outside, and does in fact allow you to store and retrieve ones and zeroes reliably, is a remarkable feat of analogue design.

If you have any doubts at all that it's an analogue design problem, do dig out any CPU manufacturer's application notes on how to lay out a DDR2 or DDR3 PCB. It's quite the eye-opener.

[EEVblog]

Ever implemented a DDR2 or DDR3 memory interface, or anything else that relies on correctly matching a high speed transmission line? The days of digital being just ones and zeroes ended a long time ago...

Digital is one and zero by definition, and always will be
There are two separate issues here. One is designing and implementing a system to allow it to work in a digital way. And then so long as you operate the system within the designed margins for digital operation, it then switches from being an "analog" system to a "digital" system. So in theory "digital designers" do not need to be concerned with the analog world, because someone with analog knowledge (who is not necessarily an analog design specialist) has designed that system to work for them.
But in practice, for that to be the case, you are no longer a "digital designer" but simply a "programmer". And programmers rightly should have no concern with the analog world.

There is no way anyone would call someone laying out a PCB, let alone high speed DDR, a "digital designer". You need at least rudimentary analog knowledge to lay out almost any PCB properly. A modern high speed DDR layout does not require a specialist analog design engineer though, but it does require a designer with enough basic knowledge of analog to understand things.

There have always been issues with designing digital systems, it's not just about speed.
For example, the early (non CMOS) digital systems had fanout current drive issues, even operating at 1Hz.
Similar issues still exist for example with FPGA "digital designers". You can't just use the one clock into your FPGA to drive unlimited numbers of internal gates in unlimited locations across the chip.

Basically, it's impossible to do any electronics design without a basic foundation of analog.
But one does not need to be an analog specialist for the majority of electronics design these days, but there are still a ton of niches were analog specialists are essential.

Dave.

[AndyC_772]

With the same due respect, I don't understand your point (honestly).

All I'm getting at is that most of the issues a designer faces when implementing a very fast interface like this aren't what you might associate with 'traditional' digital design. The memory interface in a product like a modern PC is switching at UHF frequencies, getting on towards microwave - and so the design process would be more familiar to a radio engineer than to someone brought up writing software or building slower digital circuits that just flip between 1 and 0.

The signals don't simply switch between high and low; they spend most of their time in a transitional period between the two. Nor are they simply driven from one device and received by another - they start at one device but then propagate across the PCB, changing shape, reflecting and interfering with themselves and each other as the wave makes its way to the receiver.

Read the data sheet for a modern memory IC and you might be led to believe it's a digital device. It's specified in terms of its logic 0 and logic 1 voltages, setup and hold times, access delays, minimum pulse widths and so on. But in order to actually meet those specifications, the design engineer can no longer just add up the delays and check there's still some margin at the receiver - it's not that easy any more!

I had a quick look around for images to show the actual signals on a DDR interface, but unfortunately most of Google's image results show working interfaces which have been implemented correctly, and the signals really do resemble 'traditional' 1-and-0 digital plots. But I did find this one:



...which sure looks like an analogue signal to me.

[AndyC_772]

You can't just use the one clock into your FPGA to drive unlimited numbers of internal gates in unlimited locations across the chip.

The Altera Cyclone series FPGAs that I've been using for years have dedicated global, high fan-out, low skew clock trees implemented on the silicon for just this purpose; the synthesis tool should spot global clocks and assign them to these clock trees automatically. I'm sure other vendors have the same, they're invaluable.

I agree that if you run out of these global nets, the clock has to go via regular logic instead, and the skew soon adds up to a design with a very slow Fmax.

[rolycat]

Ever implemented a DDR2 or DDR3 memory interface, or anything else that relies on correctly matching a high speed transmission line? The days of digital being just ones and zeroes ended a long time ago...

Digital is one and zero by definition, and always will be

Quite apart from any analogue design considerations, this is incorrect.

Binary is composed of ones and zeros by definition, but digital can mean any set of discrete values.

Even at the lowest level, balanced ternary (1, 0 and -1) computers have been built and may in the future replace binary systems. For example, ternary representations could prove more efficient in optical computers.

[StubbornGreek]

With the same due respect, I don't understand your point (honestly).

All I'm getting at is...

I pretty much agree and see what you're saying but again the end result is either a 1 or a 0 (intermediate states aside).

[StubbornGreek]

Ever implemented a DDR2 or DDR3 memory interface, or anything else that relies on correctly matching a high speed transmission line? The days of digital being just ones and zeroes ended a long time ago...

Digital is one and zero by definition, and always will be

Quite apart from any analogue design considerations, this is incorrect.

Binary is composed of ones and zeros by definition, but digital can mean any set of discrete values.

Even at the lowest level, balanced ternary (1, 0 and -1) computers have been built and may in the future replace binary systems. For example, ternary representations could prove more efficient in optical computers.

I'm pretty sure you mean this is incorrect when considering (possible applications of) quantum computing and some Russian tech from the 50's. Personally, I'm not aware of any other applications. Are there and I'm just in the dark here?