31
\$\begingroup\$

Do seasoned designers tend to do a fair amount of calculations or are large parts of circuits designed intuitively? I'm asking because it seems like design engineers tend to have a sense of what value cap you want to have here, resistor there, for common parts of circuits. If that's the case is it because they're just recycling designs? To the novice this is mind blowing. Though, books like the Art of Electronics seem to encourage the approach of making approximate calculations on the fly.

\$\endgroup\$
6
  • \$\begingroup\$ And when you do have to do them do you just use Matlab or do folks do it the old fashioned way? \$\endgroup\$
    – inbinder
    Apr 5, 2016 at 16:19
  • 7
    \$\begingroup\$ When a greybeard pulls a component value "from the sky" it generally isn't by calculation. Pure intuition, maybe, or a hunch, or a very rough empirical estimate. Not often by doing mental integrals. Past experience can be a cruel teacher after all... \$\endgroup\$ Apr 5, 2016 at 16:27
  • 5
    \$\begingroup\$ It is a mixture of experience (remembering values that worked in the past in similar situation) and cognitive shortcuts (simplifying calculations so that approximate answers can be generated very quickly). A cognitive shortcut would be pretending Pi = 3 to help quickly compute the value of a filter component. An example of experience might be choosing the value for a DC blocking capacitor in a headphone application. You are probably going to use the same value in all your designs. You just need to remember what it is. \$\endgroup\$
    – mkeith
    Apr 5, 2016 at 16:39
  • 2
    \$\begingroup\$ The other thing is that there are only so many choices. If you put a series resistor in a digital clock line, it is probably going to be 0, 10, 22 or 33 Ohms. You could do calculations based on load capacitance and rise time and setup and hold, or you could just pick a value, then look at the signal on a scope. I am more of a pick-a-value then look at the signal type of person, mainly because you have to do that anyway (you cannot release a design to production without verifying your choices). \$\endgroup\$
    – mkeith
    Apr 5, 2016 at 16:44
  • 7
    \$\begingroup\$ @inbinder, you are probably learning more than you realize. At some point you will help someone fix a problem and they will look at you with gratitude and amazement that you were able to help so easily. \$\endgroup\$
    – mkeith
    Apr 5, 2016 at 16:45

5 Answers 5

46
\$\begingroup\$

I'm a professional electrical engineer who routinely designs new circuits for volume production, and have been for over 35 years.

Yes, I frequently do calculations to determine the exact part specs. There are also many cases where experience and intuition are good enough and the requirements loose enough that I just pick a value. Don't confuse that with a random value, though.

For example, for a pulldown resistor on the MISO line of a SPI bus, I'll just spec 100 kΩ and be done with it. 10 kΩ would work fine too, and someone else picking that wouldn't be wrong either. If I'm using a 20 kΩ resistor elsewhere, then I might spec another one on the MISO line to avoid adding another part to the BOM. The point is sometimes you have a lot of leeway, and intuition and experience are good enough.

On the other hand, looking at the schematic of my latest design, which I'm in the middle of bringing up first boards of now, I see a case where I spent some time not only specifying the part value but calculating the result of variance on the rest of the system. There were three cases of two resistors used in the feedback to a switching power supply. Here is the problem worded like homework:

A powersupply chip feedback input threshold is 800 mV ±2%. You are using three instances of this chip, to make the 12 V, 5 V, and 3.3 V power supplies. You have previously decided to use around 10 kΩ for the bottom resistor of each voltage divider. Determine the full resistor specs in each case, and determine the min/max resulting nominal supply voltage. Stick to readily available resistor values. Use 1% if suitable and spec accordingly.

That's a genuine real world problem that took a few minutes with a calculator. By the way, I determined that 1% resistors were good enough. That's actually what I expected, but did the calculations anyway to make sure. I also noted the full nominal range for each supply right on the schematic. Not only might this be useful to refer to later, but it also shows that this issue was considered and the calculations done. I or someone else won't have to wonder a year later what the tolerance of the 3.3 V supply is, for example, and re-do the calculations.

Here is a snippet from the schematic showing the case described above:

I just picked R2, R4, and R6, but did the calculations to determine R1, R3, and R5, and the resulting power supply nominal ranges.

Added about the SHx parts (response to comment)

The SH parts are what I call "shorts". These are just copper on the board. Their purpose is to allow a single physical net to be broken into two logical nets in the software, which is Eagle in this case. In all three cases above, the SH parts connect the local ground of a switching power supply to the board-wide ground plane.

Switching power supplies can have significant currents running across their grounds, and these currents can have high frequency components.

Much of this current just circulates locally. By making the local ground a separate net connected to the main ground in only one place, these circulating currents stay in a small local net and do not cross the main ground plane. The small local ground net radiates far less, and the currents don't cause offsets in the main ground.

Eventually power has to flow out of a power supply and return via the ground. However, that current can be filtered much more than the high frequency internal currents of a switching power supply. If done right, only the well behaved output current of the switcher makes it out of the immediate vicinity to other parts of the overall circuit.

You really want to keep local high frequency currents off the main ground plane. Not only does that avoid the ground voltage offsets those currents can cause, but it prevents the main ground from becoming a patch antenna. Fortunately, many of the nasty ground currents are also local. That means they can be kept local by connecting the local ground net to the main ground in only one spot.

Good examples of this include the path between the ground side of a bypass cap and the ground pin of the IC it is bypassing. That's exactly what you don't want running across the main ground. Don't just connect the ground side of a bypass cap to the main ground thru a via. Connect it back to the IC ground via its own track or local ground, then connect that to the main ground in one place.

\$\endgroup\$
3
  • 1
    \$\begingroup\$ Good thinking in having the nominal range of voltages on the schematic, BTW. \$\endgroup\$ Apr 5, 2016 at 22:46
  • \$\begingroup\$ Can you show us snippet of the layout? I am asking because of SH1/SH2/SH3. Is the ground of dc-dc connected with main ground plane in one point only? \$\endgroup\$
    – Bip
    Apr 8, 2016 at 19:04
  • 1
    \$\begingroup\$ @Bip: See addition to answer. \$\endgroup\$ Apr 8, 2016 at 21:13
15
\$\begingroup\$

I do mainly low volume commercial & industrial market stuff, so this may be different elsewhere.

At least 75% of a typical schematic is usually building block sort of engineering, "I need a 5V rail at 3A, 5% tol, I have 15V", there is nearly no point in designing that, when Ti/Linear/Micrel have all got perfectly good designs in their datasheets, it is just a case of picking one (And the choice usually does not much matter). I can of course design from first principles, but that is not what I get paid for.

Same thing applies for many other subsystems.

Then there are the "It just needs to have the right order of magnitude" cases, pull up and pull down for cmos, series resistors for indicator LEDs, stuff like that. My usual practise here is to leave figuring these out until I see what values I need in those few places it really matters, then pick something from those values if at all possible. "Power on LED, green, 12V rail? Ok, led will drop a couple of volts more or less, and I probably want somewhere in the 1 - 10mA or so range, so anywhere in the couple of K region will be fine, oh look I needed a 3k9 resistor for that filter, one of those will get it done".

The real trick is knowing when that 'finger in the air' guess is NOT going to cut it, usually things like filters, matching networks and timing circuits, pll and other feedback things involving significant phase shifts are likely bad places for guessing. Places like that where you actually do need to get your math on (Usually matlab/scilab/ads will get the job done, no need to actually remember much of the standard tables of integrals beyond the very basic trig).

It is actually quite rare (and very nice when it happens) to end up in that place where the electronics meets the physics meets the maths, sure it happens, path loss calcs, noise calcs when doing analogue, that sort of thing, but that is maybe 10% of a design, the rest is usually cookie cutter stuff.

\$\endgroup\$
1
  • \$\begingroup\$ matlab/scilab/ads - what is ads here ? \$\endgroup\$
    – Grebu
    Apr 8, 2016 at 7:45
9
\$\begingroup\$

Particularly when using an analog IC, there will typically be one or more suggested application circuits in the datasheet. For example, I am currently designing a Qi receiver for a project. The capacitors in the inductive loop are dependent on a number of variables, and the datasheet provides some equations for determining their values:

enter image description here

So it is just a matter of plugging in the numbers, breadboarding the circuit and trying it out.

\$\endgroup\$
1
  • 2
    \$\begingroup\$ And if the datasheet doesn't help you, don't forget to take a look at the application notes (AN-documents). \$\endgroup\$
    – Mast
    Apr 6, 2016 at 6:25
7
\$\begingroup\$

For analog design, we do the calculations for most part. Some things, like coupling and bypass/filter capacitors we might just pick a "typical" value, knowing that it's going to work for the application. But note that "typical" will be different for DC, audio, and radio circuits -- this is something we have to be familiar with.

For bias and gain resistors we usually do the calculations. I do them by hand, since the equations are simple. Often we want a "gain of about 10" circuit, so the ratios are simple enough to do in your head, and the values (1K vs 1Meg) are chosen for the type of circuit.

\$\endgroup\$
1
\$\begingroup\$

The accuracy required by your application, is what dictates the amount of reuse, intuitive design, and/or formal design that one would use. One example of each is: audio amplifier, low noise amplifier for TV, and ultra low noise amplifier for a radio telescope, respectively. It should be clear that how "formal/accurate" your design should be, depends on how "critical" the application is (as well as how much time and money is available for the design).

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.