What is a good design?

[spec]

Good grief...

And to make matters worse it was setting the threshold for target detection on an airborne cloud and collision system.

The authorities had reported a variation of performance between units. It got a bit political and I was asked to do a design review, with help from other specialist engineers, of course.

The radar boys did all sorts of testing and analysis- radar return loss etc but the RF side came out OK. We didn't suspect the circuit design.

I had a bit of trouble understanding the circuit- it did not seem to make sense and the schematic was a spaghetti nightmare, with lines snaking all over the place and random voltage points peppering the whole thing. To cap it all the inputs and outputs were buried in the schematic too, rather than being along the edges.

In the end I redrew the whole circuit and then it became clear about the opamp input bias.

I contacted the engineer who did the design and asked him about the offset method and he confirmed what I thought. He said it was an elegant design with the minimum of components.

spec

 

[ericgibbs]

So what is a good design? What are the criteria? I think I know, but what are your views?

hi spec,
Practical engineers work in the world of the 'approximate', if it meets the clients specification it's 'perfectly' good enough.

All too often I have seen engineers who have the Christmas tree gene.
Adding more 'bell's and ball's' and finally a Fairy on top, at which point the tree becomes unstable and falls over.

Eric

 

[JonSea]

Sorry I offended you. I'll take my need to express myself somewhere else

 

[spec]

hi spec,
Practical engineers work in the world of the 'approximate', if it meets the clients specification it's 'perfectly' good enough.

All too often I have seen engineers who have the Christmas tree gene.
Adding more 'bell's and ball's' and finally a Fairy on top, at which point the tree becomes unstable and falls over.

Eric

Hi Eric,

I have seen exactly the same thing with the bells and whistles. Often the ornaments are hung on a fundamentally flawed design too.

Then you have requirements creep, especially from the customer: I know it is not in the requirements specification but could you ...

Also there is the desire to innovate rather than using a tried and tested approach.

One chap I knew delighted in using the reverse breakdown voltage of a BJT as a Zener diode.

He once had a task of measuring the frequency of a square wave. He converted the frequency into a voltage and then put that voltage into an ADC.

Yes, practical is the word, and a very important aspect of design is to make accurate approximations- nice oxymoron for you. That way you save hours of nugatory maths.

So often, especially with grads, they will leap into s parameters, simulations, etc etc, with out checking if particular component can handle the power, or is even available.

Rules of thumb are also essential: BJT VBE ~ 0.6V, -VBEmax ~ 5V and so on.

spec

 

[spec]

Sorry I offended you. I'll take my need to express myself somewhere else

You didn't offend me in the slightest. There is no place in design for emotions.

No need to go elsewhere- I for one would like to hear your views about design- it is quite clear from your posts on ETO that you have a lot of experience.

spec

 

[alec_t]

it was amazing that the total rack power supply load was close to the current corresponding to an average chip current consumption.

Perhaps not so surprising. The chip manufacturer's figures for 'typical' current were probably derived by using racks of chips and calculating the average current .

 

[spec]

Perhaps not so surprising. The chip manufacturer's figures for 'typical' current were probably derived by using racks of chips and calculating the average current .

Quite true.

But sometimes you get batches of chips that are all one way or the other, and the same component from different manufacturers can have quite different actual performance.

I was doing a 100MHz 50 Ohm video amplifier once. The video amp chips from one particular manufacturer sort of met the stated bandwidth, but Analog Devices video amplifiers far exceeded their specified bandwidth and at less current consumption too.

I often wondered what we would do if all the chips in the batch we bought were on the high end of their current consumption. You would not have a case against the manufacturer because the chips would be in specification.

spec

 

[Nigel Goodwin]

As a TV service engineer for many years I came across bad design on a frequent basis, I could often look at the circuit of a new TV chassis and predict where faults were likely to occur. It's not difficult, as they repeatedly made the same mistakes.

Typical problems were putting resistors in series or parallel, rather than using a correctly rated single resistor, this was a VERY common failure. I've never been able to find out if it was a mistake by the original designer, or a change introduced on the production line to reduce costs.

Another common failing was putting rectifiers in parallel with no series resistors, and this still goes on today in the Vestel LCD sets - with rectifier failure another VERY common fault.

 

[OBW0549]

Then you have requirements creep, especially from the customer: I know it is not in the requirements specification but could you ...

"Creeping featurism" was a chronic problem at one place I worked, and the perpetrator was usually Marketing. Some product line marketing managers were easy to work with because they maintained a rigorous distinction between "must have" and "would be nice to have." Others were a right pain in the ass because of their habit of over-inflating product performance requirements and/or imposing new requirements after the bulk of the design work had been completed. Some engineers were equally vexatious due to their relentless pursuit of the "perfect" at the expense of the "good enough." One senior engineer was a PITA of legendary proportions because he'd come up with a "bold, new, innovative idea" every couple of weeks that would force a hard reset on the design effort, and we'd have to start all over from scratch to incorporate his latest inspiration. Then Engineering Management would scratch their heads wondering why the project was running late.

So often, especially with grads, they will leap into s parameters, simulations, etc etc, with out checking if particular component can handle the power, or is even available.

...or without even reading the data sheet to check the Absolute Maximum Ratings.

 

[cowboybob]

"... Thus "design" may be a substantive referring to a categorical abstraction of a created thing or things (the design of something), or a verb for the process of creation, as is made clear by grammatical context ...". (Wikipedia)

Clear as mud.

But, within the context of this thread, to me "design" is more art than fact or, more colloquially, within the "eye of the beholder" paradigm. Thus, a universally definitive exposition of the term "good design" doesn't exist, other than that of the current viewer.

Thus, I can categorically state that ALL of my designs are "good" (if not awesomely brilliant).

Others might opine (and I would expect no less) that they are not.

To the point, however, I suggest the 555 timer: remarkably adaptable and still in use with the original design, since 1972.

 

[spec]

As a TV service engineer for many years I came across bad design on a frequent basis, I could often look at the circuit of a new TV chassis and predict where faults were likely to occur. It's not difficult, as they repeatedly made the same mistakes.

Typical problems were putting resistors in series or parallel, rather than using a correctly rated single resistor, this was a VERY common failure. I've never been able to find out if it was a mistake by the original designer, or a change introduced on the production line to reduce costs.

Another common failing was putting rectifiers in parallel with no series resistors, and this still goes on today in the Vestel LCD sets - with rectifier failure another VERY common fault.

So true- the same mistakes over and over.

I was expecting you to mention another tenant of good design: maintainability with all that implies- accessibility- modularity and so on.

spec

 

[spec]

"Creeping featurism" was a chronic problem at one place I worked, and the perpetrator was usually Marketing. Some product line marketing managers were easy to work with because they maintained a rigorous distinction between "must have" and "would be nice to have." Others were a right pain in the ass because of their habit of over-inflating product performance requirements and/or imposing new requirements after the bulk of the design work had been completed. Some engineers were equally vexatious due to their relentless pursuit of the "perfect" at the expense of the "good enough." One senior engineer was a PITA of legendary proportions because he'd come up with a "bold, new, innovative idea" every couple of weeks that would force a hard reset on the design effort, and we'd have to start all over from scratch to incorporate his latest inspiration. Then Engineering Management would scratch their heads wondering why the project was running late.

It almost seems like we worked for the same company. One of our engineering managers was a terror. At meetings he would suggest to the customer that he could having all sorts of new features while having no idea what that would imply. In the end we had to make sure he was busy or on holiday when we had customer meetings.

spec

 

[spec]

"... Thus "design" may be a substantive referring to a categorical abstraction of a created thing or things (the design of something), or a verb for the process of creation, as is made clear by grammatical context ...". (Wikipedia)

Clear as mud.

As you say- clear as mud. Who writes nonsense like that; not a designer that is for sure.

spec

 

[ronsimpson]

Good Design:
Parts have minimum/typical/maximum numbers. Often min or max is not on the data sheet.
If you build one of some thing then you change things until it works and done.....
If you build millions......you really need all the parts to work.
In the testing phase you probably can not get min or max parts.
A transistor's current gain has a large difference between min and max. (max might = ?)
Too many times we have received "max" parts for a year then suddenly you have 10,000 "min" parts.

For years I used high current high voltage capacitors. In some designs +/-5% is a little hard to deal with. The +/-2% parts are more money.
What I noticed; some times a batch of caps would all be -5% to -2%. Then the next batch is +2% to +5%. What I found is that some one grabbed all the +/-1% parts. Then most of the +/-2% parts are removed. Through some negotiation I found I could get the (-5%) caps for the same price as +/-5%. (which usually was -5% to -2%) I have bags of 1000V caps marked (-) and others marked (+).

Example: MPS2222 transistor, current gain, min and max. 1:3 range. (at 150mA only)

 

[spec]

Good Design:
Parts have minimum/typical/maximum numbers. Often min or max is not on the data sheet.
If you build one of some thing then you change things until it works and done.....
If you build millions......you really need all the parts to work.
In the testing phase you probably can not get min or max parts.
A transistor's current gain has a large difference between min and max. (max might = ?)
Too many times we have received "max" parts for a year then suddenly you have 10,000 "min" parts.

For years I used high current high voltage capacitors. In some designs +/-5% is a little hard to deal with. The +/-2% parts are more money.
What I noticed; some times a batch of caps would all be -5% to -2%. Then the next batch is +2% to +5%. What I found is that some one grabbed all the +/-1% parts. Then most of the +/-2% parts are removed. Through some negotiation I found I could get the (-5%) caps for the same price as +/-5%. (which usually was -5% to -2%) I have bags of 1000V caps marked (-) and others marked (+).

Example: MPS2222 transistor, current gain, min and max. 1:3 range. (at 150mA only)
View attachment 103720

Ah, I thought you would have an input on design, Ron.

I spent a bit of time in a department that analyzed components and component faults. There was some system where you would allocate numbers for different categories of faults and try to work out a trend. We had reams and reams of data going back ten years, but nobody ever did anything with it.

spec

 

[OBW0549]

It almost seems like we worked for the same company. One of our engineering managers was a terror. At meetings he would suggest to the customer that he could having all sorts of new features while having no idea what that would imply. In the end we had to make sure he was busy or on holiday when we had customer meetings.

We didn't have that problem quite so much; most of our engineering managers knew better than to make promises to customers (or to Marketing) without consulting us engineers first to make sure we could deliver without blowing schedules and/or cost targets. When they did transgress, we did everything possible to make sure they ended up regretting it.

 

[ronsimpson]

Good design:
Heat is one of the biggest killers. In the case of capacitors it might take a year to kill a cap with heat.
(Much of my work has been in the fields of power)
Some of my designs have had fans but people don't like the noise.
Some designs had metal boxes where I could dump the heat. (cost)
Most of the designs could not have fans, had plastic, and the vent holes must be small.

All my high voltage supplies were resonant back to 1980. But the low voltage were not.
In about 1990 (+/- some) I was having problems with 100 to 200 watt off line supplies. (this was when Power Factor Correction was just an experiment) With no air flow the heat sinks were too big. I could see this was not going to have a life of 10 years. I found a way of chopping my 400 V supply using resonant switches. The prototypes ran with out heat sinks. I believe I added small clip on heat sinks at production.

Heat kills. It kills in may ways. It is common for the end user to stack piles of paper on you box. Fans die in time. So watch for hot spots.

 

[JoeJester]

Typical problems were putting resistors in series or parallel, rather than using a correctly rated single resistor, this was a VERY common failure. I've never been able to find out if it was a mistake by the original designer, or a change introduced on the production line to reduce costs.

Sounds like they wanted one resistor to take the place of many. 2x series, 0.5x parallel, 0.67x series-parallel, and so forth. They must get great multiple item discounts.

 

[Nigel Goodwin]

Sounds like they wanted one resistor to take the place of many. 2x series, 0.5x parallel, 0.67x series-parallel, and so forth. They must get great multiple item discounts.

Obviously, but it doesn't work, it means it's very likely to be unreliable (and that is proved in practice).

It's not just a question of quantity (they already use resistors in such huge quantities a few more wouldn't make much difference), it's the fact that buying a 'small' number of bigger resistors probably costs more than three small ones, and it makes it more expensive to assemble with the component insertion machines as well.

 

[tcmtech]

To me good design is just intuitively known when you see and work with it.
The vast majority of the work I do is repair work and much of that comes with having to reverse engineer a component or system to try and figure out what the original designers were aiming for but missed for any number of reason.

My personal peeve with designs is the 95/5 problem as in 95% of the design and manufacturing was done very well but 5% of it is the constant source of failures because it was either under built, sloppily designed or cheaply made or a bit of all three when there was absolutely no justification to have done so given what can be seen of the rest of the overall 95% part of the design and manufacturing.

To me its like seeing 50 cent brass bushings in a $400 motor on a $10,000 unit that the manufacture knows is going to be in a 50,000+ our expected service life application where a set of $5 bearings that would have fit the same location and never failed.

Or that crappy 25 cent plastic toggle switch on a control panel that breaks off constantly where a $2 all metal one would never have issues.

Or having that outrageously expensive proprietary special limited production component that always fails doing a key function despite a far cheaper simpler robustly built and mass produced component of identical function could have been put in its place.

To me that either addressing or overlooking simple but critical low cost components or putting in high cost unjustified specialty ones that will cause an overall good well designed and built device or system to fail is what determines good or bad design.