What's Design Margin?

Chuck D. Bones

Circuit Wizard
Design margin, simply stated, is the difference between a design working and a design not working. So why do we want design margin and how much is enough?

We can build a circuit, have it work fine and think that we're done. Then we build 10 more and surprise, surprise, half of them don't work. This actually happened to me early in my engineering career and I have seen it happen to others, many many times. The root cause was insufficient design margin. A circuit can have 1% design margin and work fine. Take that margin down to 0% and all Hell breaks loose. Component tolerances, temperature drift, aging and variations in the power supply voltage are some of the things that can erode the design margin.

So what do we do about it? The pros perform worst-case analysis (WCA) to see what the circuit will do under the worst-case combination of component and environmental variations. They perform worst-case testing where all of the external variables, including power supply voltage and temperature, are taken beyond the expected operating limits. Needless to say, all this is time-consuming and expensive, so it should come as no surprise that most pedal designs never see any worst-case analysis or testing. The customers end up doing the worst-case testing at gigs. If the manufacturer gets enough warranty returns, they might actually do something about it because it starts cutting into their profits. Things manufacturers do about it:
  • Build the cost of warranty returns into the sale price
  • Discontinue the product
  • Revise the design
What I do about it is look for design weaknesses and try to correct them. I'll perform simulations and a partial WCA based on experience. I'll breadboard the circuit and vary some key component values to see what happens. My methods are not super rigorous, but they are better than nothing.

Many years ago, I was negotiating requirements of a product with a customer and the topic of design margin came up. I told him that the amount of margin he wanted was going to increase the size and weight. That's a big deal when the product is an electronics package that goes on a satellite. Archie's succinct response was "margin doesn't save size or weight, the only thing it saves is your ass!"
 
Last edited:
That's a lotta wisdom is such a short post. Perfect quote, too.

Short of a more sophisticated analysis method, I do always try think about fault potential in designs. For pedals, you can pretty much guarantee someone will plug anything into the DC socket that will fit - and sometimes even when it won't. One of the cringiest things I still see are designs that use a parallel diode for polarity 'protection.'
 
I guess the biggest trouble with this comes up with designs that use highly variable components like our 'beloved' jfets or Germ transistors...?
Great wisdom there indeed.
 
That's a lotta wisdom is such a short post. Perfect quote, too.

Short of a more sophisticated analysis method, I do always try think about fault potential in designs. For pedals, you can pretty much guarantee someone will plug anything into the DC socket that will fit - and sometimes even when it won't. One of the cringiest things I still see are designs that use a parallel diode for polarity 'protection.'
Keen to learn more...

I've only seen diodes in line, such as 1N5817, or diode to ground such as a 1N4001 — could you give an example of a parallel diode in a schematic masquerading as polarity protection, please?
 
I've only seen diodes in line, such as 1N5817, or diode to ground such as a 1N4001 — could you give an example of a parallel diode in a schematic masquerading as polarity protection, please?

Diode to ground is parallel.

I despise that method of polarity "protection".... When the diode shorts the pedal is dead. To the average end-user the pedal is effectively broken, the diode protected nothing.

And this is assuming it doesn't take out your power supply with it.

I suppose it was passable when 9V batteries were the norm, but these days power supplies have huge amounts of output current capability. Enough to blow traces clean off the board when all goes wrong.


Many years ago a friend of mine brought over one of those Danelectro pedalboards, the molded plastic case with five pedals all daisy-chained together. After he plugged everything up he mentioned that none of the pedals were working... I happened to notice that he was using an AC adapter for a Nintendo. I suppose he just grabbed the first AC adapter that said 9V and ran with it.

All five pedals were dead in one instant.

This was before my days as a repair tech... Now I know it was a simple fix, but to two idiots armed with nothing more than a screwdriver and a joint it was just a pile of broken junk.
 
Last edited by a moderator:
Thanks for the explanation.
I figured parallel might be the "to ground" option, but my mind said "parallel to what?"



My friend had a brand new Electric Mistress back in the day, a couple days old; he auditioned for a band and for some reason I don't recall, he had to leave his gear and when he went back to retrieve it a couple days later, the Mistress was fried and denied.

The mint-exterior Mistress is now in my possession, but the price of a SAD1024 if you can even find one (and I did and it was $300 and sold before I could finish reading the ad for it)... ouch.
 
Parallel to the power supply (or circuit / load).
 
Last edited by a moderator:
Yup, well, pretty sure. He took the pedal to a highly-touted repair guy (after a few decades in shoe-box storage, and only at my pushy insistence). The guy said everything was fried and that it would cost too much in repair fees for him to fix and be worthwhile. I suggested to my friend that maybe the repair guy is shady, but buddy said pricey repair-dude has too much rep on the line to pull a fast one and get caught out on it. Still, people do the strangest things... and get caught. Cosby, Ryder, DeLorean — or closer to home: Adams, Marangella, Ken...

I've seen SAD chips recently fetching a "more reasonable" $100ish. Do I want to spend the $$$ and then find the pedal still doesn't work? (I could use the chip in a Moosapotamus ADA clone or something, but $till...). I've been busy with building a 5F2a (finished), repairing my Trend 5w (just finished). Busy preparing an overseas move, now, but I'll try to spelunk the Mistress' depths and assess the extent of the damage before I have to pack her up. Even though my friend bequeathed the Mistress to me, it's been my dream to get it going again for him. If I can get it going, I will insist he take it back.

Based on my "parallel" question here, and other questions I've asked in the forum you'll surmise that I've plenty of room left on the solder bench to add quite a few notches.


Sorry to derail and prattle on. I shall pray there's enough Design Margin in the Mistress that it can be fixed.

Thanks again for your wisdom and edification, Gentlemen.
 
I'm thinking a little triage is in order. Is the PCB itself salvageable? Or are there burned traces?

Everything except the SAD1024 is still in production.
 
I didn't see anything burnt, but I've not pulled it completely apart yet. I'll start up its own thread when I do.
 
I sold a wonderful saffron squeeze clone (including metal can ota mojo) only to have the guy message me that he hooked it up to the nintendo power supply ... told them that whatever they do, don't open it up and let the mojo smoke out
 
Yes it does.

The (expen$ive) commercial versions of SPICE contain two specific functions for performing Worst Case Analysis (WCA): Sensitivity and Monte Carlo. Sensitivity calculates how much the circuit performance changes when a specific component parameter (ex: resistor value) changes. Monte Carlo varies all component parameters randomly within their specified tolerance ranges, runs a simulation, and then repeats with different random parameters. This is repeated 1,000 times or more and then the range of results is evaluated by the engineer. The person setting up the simulation must specify all of the component tolerances that they wish to vary.

LTSpice does not contain either of those specific WCA functions, but it is possible to do them anyway using the .step param function. This article explains how.

I was never a fan of the Monte Carlo method. It has its uses in certain cases, but is not worth the hassle with simple circuits like pedals.

I do use the extreme value method when I'm looking for how much component variation effects a circuit parameter like gain, freq response, bias, headroom, distortion, etc. Varying a transistor parameter, like hFE, is possible, but is more labor intensive than varying a capacitor value. When I want to investigate the effect of hFE or Vp variation using LTSPICE, I do multiple simulations with different part numbers. It's not a true WCA because all of the transistor models that come with LTSPICE have nominal parameters, but it's good enough to indicate trends.

Get comfortable with LTSPICE, then try out the .step param function.
 
I can see a few reasons why you're not keen on the Monte Carlo method, one being too much data produced and then having to sift through it all — setting up the parameters alone would be excessively time-consuming.

I'm still trying to learn Eagle, but I can foresee needing to learn Spice, too.


Mistress Mystery thread is up, by the by.
 
Back
Top