Weird and Wacky Engineering
"Hi, I’m your data; can you hear me now?"
Bill Schweber
5/16/2012 5:35 PM EDT
I just saw a fascinating story in a recent issue of Physics Today on “listening” to your data—literally, see “Shhhh. Listen to the data.” The article showed how making real-world data audible allows the ear and brain to sense patterns, extract features, and find occurrences which might otherwise not be found by conventional data-analysis packages.
This may seem a counterintuitive throwback to those quaint, ancient methods in our software-intensive world, but reality is that the brain can extract things that even our most impressive computers and algorithms can’t, or which require significant computing power to achieve. Also, the brain is good at dealing with the unexpected, while even the best data-analysis package can only find what it has been “programmed” to expect.
A few years ago, I spoke to some people doing software for the DARPA autonomous-vehicle road race, and asked them about the biggest challenges they faced. The answer was pretty quick and unambiguous: having the vehicle “see” where the actual road was, and not be misled by trees, signs, fences, obstacles, distractions, road irregularities, and the almost countless other realities of what the vehicle’s cameras could see. Many lines of code and corresponding MIPS were dedicated to image recognition and feature extraction, they added.
The irony is that seeing and then knowing where the road is turns out to be pretty easy for almost anyone, even those with poor actual driving skills. Yet the brain is not executing millions of lines of code, nor doing MFLOPS of processing to figure it out. Whether using audible, visual, or other senses, the brain is amazingly good at determining patterns and anomalies. And don’t kid yourself: we have almost no idea how the brain does this, despite what the neuroresearchers would like you to think.
Experienced engineers use all their senses when designing, assessing what’s going on, and finding out what’s not going as expected. Good design and debug is a combination of formal tools and also the human ones: sight, sound, feel, and yes, smell. The best debugging methods I have seen and used are also the oldest: look, listen, expect the unexpected, and then stop and think, before jumping to your next step.
Have you ever used the informal tools of human senses individually or in combination, to find the source of your problems or assess your designs? Did you do this intentionally or accidentally? ◊
|
|
|
|
|
David Ashton
5/17/2012 3:26 PM EDT
I remember in the days of modems we used to use a 511 bit pseudo-random error pattern to test data links. If you listened to it on the line, it was a surprisingly repetetive sound compared to real random data. So I can see how you'd "hear" patterns in data this way.
Sign in to Reply
selinz
5/17/2012 5:20 PM EDT
While I agree that we are able to see patterns and automatically expand, rescale, and focus on anomolies with our eyes, I don't beleive the same is true with our ears. They are much more 2d(or arguably, 1d) and are thus easily outdone by relatively unsophisticated algorithms.
Sign in to Reply
GordonScott
5/18/2012 4:24 AM EDT
Believe it.
I used to work in radiopaging sending digital data packets and during development would often say or hear "there was something wrong with that packet". We were rarely, if ever, wrong.
Most people who use(d) modems will have had a "that didn't sound right" moment.
Sign in to Reply
David Ashton
5/18/2012 8:27 PM EDT
Very much so, I used to deal with multisite modems on a polling system and got to know when there was something wrong on the line by the sounds. For instance, you could teell how many sites were responding, or if one was retransmitting a lot.
While Selinz may be sceptical, just as your eyes will register small changes in what you see, so the ear can distinguish small changes or patterns in modulated data. I can see how that could be used to find patterns or other irregularities in data - it would only give you a starting point for further investigation but you could get through a lot of data quickly.
Sign in to Reply
przemek
5/17/2012 6:07 PM EDT
old-time computer folks used to debug with an AM radio next to their system. Back in those days, clocks used to be in the hundreds of kilohertz, and you could hear the program flow (loops, jumps, interrupts) in the buzz on the corresponding AM frequency.
Sign in to Reply
krh
5/17/2012 9:19 PM EDT
Good comment przemek. And folks still use the AM radio technique. Put it near the AC service panel of a house and see (hear) if any x10, etc. devices are wired somewhere and perhaps therefore interfering with some other system. Turn off individual breakers to isolate and find.
Sign in to Reply
Charles.Desassure
5/21/2012 2:34 AM EDT
These concepts need to be implemented back within today’s classroom for better understanding. Good article.
Sign in to Reply
Earl54
5/22/2012 4:16 PM EDT
Power supply designers use touch all the time to tell if things are running hot. Low voltage circuits, of course. And sight, smell, and hearing give a quick indication that an electrolytic is in backwards. Or, what is left of it was in backwards. If something is hot, we can often tell by the smell if it's a semiconductor or a resistor.
Sign in to Reply
WKetel
5/23/2012 8:50 AM EDT
Evaluating data by sound , smell, or touch is indeed a very good method, particularly when the required results are qualitative, such as "good or bad". But it calls for skilled personal judgement and so it would not be found acceptable by that segment of managers who strongly want to eliminate the need for such skills in the name of uniformity. Think ISO9000 and similar policies.
But listening is still a very handy way to detect problems, beyond any doubt.
Sign in to Reply
Earl54
5/23/2012 9:54 AM EDT
Agreed, by ISO9000, and even good solid logic, measurements must be done by instumentation. But a first cut, fire it up, qualitative run can use all the senses. And when something goes wrong, a good nose or ear is often the first indication. On a power supply, the first indication of a failure is sometimes that loud popping sound, flash of light, and acrid smell. Of course, if that's the case, the fact that the product no longer works makes it obvious.
Sign in to Reply
Fabio007
5/24/2012 5:21 AM EDT
Agreed! Experienced power electronics engineers use all their senses to detect anomalous circuit operation, and hearing is very important. A circuit that is slightly unstable, say, only during a step change in either input or output, often has a distinctive change in noise, in many cases a "chirp", that indicates that some control loops are not quite happy. Back in the days before deep memory scopes I relied on this to detect stability problems. Even today a keen ear can often detect a problem long before someone relying on a scope image.
Off-peak power: here in Australia, one can hear the high-freq signals sent by the power supply authorities to turn off-peak relays on and off. Some times (older) equipment malfunctioned due to this signal, and the best test we had was to plug in a particular soldering iron and just listen for the distinctive "off-peak" sound....
Sign in to Reply
David Ashton
5/24/2012 6:12 AM EDT
Another example - switching power supplies - sometimes a a faulty SMPSU will emit audible "chirps" or a usually supersonic oscillation will become audible - many's the time I've used that as a diagnostic.
Sign in to Reply
Graham H
6/7/2012 6:59 AM EDT
I was called in once by a power supply company that had a conducted emissions EMC problem on a new PSU. I tested one and emissions were some dB's above the limit. I then asked to test the PSU they submitted for testing. It was below the limit. But I then took the lid off and I new straightaway why it had passed. I could hear that it was unstable and this was conveniently spreading the emissions spectrum. I cured the marginal stability problem but had to add an extra filter.
Sign in to Reply