At The Forefront of Noise Evaluation: Real-Time Analysis

Control: management of fans, temp/rpm monitoring via soft/hardware

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
Rory Buszka
Posts: 175
Joined: Tue Oct 04, 2005 4:58 pm

At The Forefront of Noise Evaluation: Real-Time Analysis

Post by Rory Buszka » Tue Jul 25, 2006 4:08 pm

Please read, and give me your feedback. Much of the "Background" section can be skipped; I'm mainly looking for comments on my testing methodology. (i.e. what needs to be improved, what potential stumbling blocks you see, etc.)

In short, I believe that through the application of real-time analysis techniques, we can obtain data that not only quantifies the average sound pressure level, but allows for identification of specific objectionable sonic characteristics of a fan's noise as well as comparison of the sonic character of one fan with that of another, through a means that does not rely on subjective observation (which may vary from person to person; not everyone is equally qualified).

Buszka, Rory
7/21/2006

Method for Assessing Noise from Computer Fans

I. Background

The majority of personal computers in use today contain components which generate waste heat. If allowed to build up, this heat can damage components, decrease their longevity, or impair their proper function. To dissipate localized build-ups of heat in a computer system, cooling solutions are typically applied to the specific components in which heat build-up is most severe. Cooling solutions typically consist of a heat sink or heat spreader, which increases the effective surface area of convection. However, convection alone is sometimes not enough to sufficiently dissipate the waste heat. For these applications, a fan of some type is used in conjunction with the cooling solution to increase airflow across the surface of the radiating area, assisting convection. The use of fans makes it possible to achieve suitable results with smaller heat sinks or spreaders than would be required if convection alone were to be relied upon.

The introduction of these fans also introduces a source of noise to the work area, which many users find offensive. Scientific data confirms that increased levels of noise from personal computer systems decrease the user’s ability to concentrate, and increase the user’s stress level. Yet the increased performance of modern computer systems dictates the use of cooling solutions of ever-increasing capacity. The route followed by the majority of the computer industry has been to minimize the physical size of the heat sinks involved, while increasing airflow across the heat sinks by increasing the speed and size of the fans involved.

Currently, the computing industry finds itself responding to a backlash by consumers against noise in computers and computing products. This has been driven by two things. Perhaps most notable is the recent surge in popularity of “Home Theater PCâ€

Felger Carbon
Posts: 2049
Joined: Thu Dec 15, 2005 11:06 am
Location: Klamath Falls, OR

Post by Felger Carbon » Tue Jul 25, 2006 5:45 pm

An excellent suggestion, Rory! You have been anticipated - here is a review of a number of HSFs. Pages 35-42 include detailed FFTs of HSFs, about 5 per page - as for instance page 40.

Now, using this information, please tell me which are the good HSFs and which are the bad, noisewise of course, based on these FFTs? :D

Ayah
Posts: 15
Joined: Tue Jul 18, 2006 4:04 pm

Post by Ayah » Tue Jul 25, 2006 5:55 pm

Can someone summerize for me?

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Tue Jul 25, 2006 6:12 pm

Rory,

Your initiative is much appreciated. :) I presume you suggest that SPCR adopt your approach...?

Whether or not that's your intent, there are many aspects of your post which are valuable and give pause for reflection of SPCR's noise analysis methodogy. The search for a complete and objective metrology for assessing computer noise is certainly valuable.

a few details...

1) 10cm is too close, for reasons much discussed recently elsewhere in these forums (I'm sure someone will point out the threads). It's why we've changed to a mic with a lower noise floor, and moved to 1m and 30cm for recordings.

2) In terms of allowing various reviewers to contribute in the same way, it would really only work if the same high quality gear was used by all. Such gear is not cheap. We have >$1500 invested in our recording gear and another thousand on the SLM. Calibration should really be done not only of SPL level sensitivity but the freq response of the mics. Again the only realistic way to ensure that the sound is captured the same way by everyone is to use the same high quality gear (which has much lower sample-to-sample variance) -- along with similarly quiet rooms with similar acoustics.

3) We can actually take the current recordings and turn them into FFT graphs right now. This has been done in some reviews, as on page 6 of the Arctic Cooling Silentium T2. In fact, anyone with acoustics analysis software can download the WAV files and turn them into FFT graphs.

So why don't we do this routinely in our reviews? For the same reason we don't just report the SPL in dBA without comment: The information is not easy to interpret, especially when there are subtle differences, and often, even plainly audible differences (to us) do not show up obviously in 3d plots of frequency & amplitude.

They're most useful to those who have a very good understanding of noise measurements and units. Does the average SPCR reader have a good handle on what 26 dBA sounds like? How about 36 dBA or 18 dBA? Or what a particular FFT means, acoustically? I don't think so. We make and post the recordings so you can compare and contrast what you hear on them to how we say it sounds and measures.

If we were to plot a 10s selection from every recording (which 10s?) and show the FFT plot of it, perhaps regular readers would become educated over time of the correlation between the recordings, our descriptions, our SPL measurements and the FFT graphs. This would add another degree of complexity, and we'd have to spend time to ensure the graphs were indeed representative of what we perceived (and discuss them at least a little).

Would all this really be a valuable benefit? Or would really interested readers be better off investing in a better pair of headphones and/or sound card to hear more faithfully the recordings we're making right now?

Rory Buszka
Posts: 175
Joined: Tue Oct 04, 2005 4:58 pm

Post by Rory Buszka » Wed Jul 26, 2006 3:46 am

I agree that some aspects of the testing procedure would need to be tweaked, but I disagree that the FFT would be of no benefit. With an FFT histogram, one could go back with arrows and show which narrowband artifacts are produced by the extraneous motor and bearing noise, and which wide-band hump is produced by the fan's fundamental frequency.

Your argument against the FFT analysis method comes mainly from added complexity in educating readers to interpret the FFT histogram. There would indeed be much added complexity in making measurements of things like hard drives, but measurement of fans by this method should be relatively straightforward. I believe that many of our readers already have an intuitive knowledge of what the bars mean, being acquainted with the chintzy little spectrum analyzers on cheap Japanese mini-stereos, among other things. Windows Media Player has some visualizations based on spectrum analysis. I think that readers of these articles would already have some idea of what it means when bars at the far right end of the plot are taller than bars at the left end of the plot, or when one bar or a group of three histogram bars sticks up above the others. Small labels with arrows could be used to point out the "ticking", the "buzzing," and the "Whoosh". The greatest value that FFT analysis would have is in comparison between two fans, or two heatsinks with fans mounted to them. The intensity of things like motor noise would be relative to the overall turbulent noise floor. It would be of greatest value in articles where two or more competing products are evaluated, side-by-side. And this would also eliminate subjective differences in the perception of the reader listening to your recordings -- they could also see exactly what they were hearing, and compare it to others regardless of the playback equipment used by the reader of the article and how well it is calibrated to a specific playback level. Rather than demanding a further investment by your readers in order to gain the most value from your articles, FFT charts would give the reader a reference in their listening, no matter whether they're listening through headphones or a pair of $10 computer speakers. (Or in my case, $150 computer speakers -- I can't stand headphones.)

I originally suggested this in the thread about the fan roundup, but received no response from SPCR management, so I did not realize that it had been seen. I acknowledge that FFT analysis may not be very useful unless it is in the context of a "round-up" review -- where readers can see several FFT graphs side-by-side or one right after the other.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Wed Jul 26, 2006 8:19 am

Rory Buszka wrote:I agree that some aspects of the testing procedure would need to be tweaked, but I disagree that the FFT would be of no benefit. With an FFT histogram, one could go back with arrows and show which narrowband artifacts are produced by the extraneous motor and bearing noise, and which wide-band hump is produced by the fan's fundamental frequency.

Your argument against the FFT analysis method comes mainly from added complexity in educating readers to interpret the FFT histogram. There would indeed be much added complexity in making measurements of things like hard drives, but measurement of fans by this method should be relatively straightforward. I believe that many of our readers already have an intuitive knowledge of what the bars mean, being acquainted with the chintzy little spectrum analyzers on cheap Japanese mini-stereos, among other things. Windows Media Player has some visualizations based on spectrum analysis. I think that readers of these articles would already have some idea of what it means when bars at the far right end of the plot are taller than bars at the left end of the plot, or when one bar or a group of three histogram bars sticks up above the others. Small labels with arrows could be used to point out the "ticking", the "buzzing," and the "Whoosh". The greatest value that FFT analysis would have is in comparison between two fans, or two heatsinks with fans mounted to them. The intensity of things like motor noise would be relative to the overall turbulent noise floor. It would be of greatest value in articles where two or more competing products are evaluated, side-by-side. And this would also eliminate subjective differences in the perception of the reader listening to your recordings -- they could also see exactly what they were hearing, and compare it to others regardless of the playback equipment used by the reader of the article and how well it is calibrated to a specific playback level. Rather than demanding a further investment by your readers in order to gain the most value from your articles, FFT charts would give the reader a reference in their listening, no matter whether they're listening through headphones or a pair of $10 computer speakers. (Or in my case, $150 computer speakers -- I can't stand headphones.)
Fair enough, although imo, a further investment in headphones or other audio gear would have a big side benefits -- mainly better quality music reproduction & increrased enjoyment. ;)

Anyway.... we'll have to experiment with different ways to generate 3D FFT plots. The program we used to generate these....

Image

Image

...does not let us choose the vertical scale precisely. It'd be important that all the graphs be plotted in exactly the same (visually useful) way. I'm not keen on the 2D plots, because they require too much averaging or show only one point in time, when in reality the spikes and valleys over time are also indicative of consistency -- smoothness or lack there of.

breunor
Posts: 96
Joined: Tue Jul 12, 2005 10:18 am

Post by breunor » Wed Jul 26, 2006 10:36 am

Something all the charts/graphs/descriptions don't porvide, but the sound recordings do, is how these sounds will come out in a reader's personal setting. For example, I went from a carpeted office to laminate flooring, and it has changed the perceived noise in the room (much easier to track little dogs now - click click click !).

Having a sound recording to work with, and decent speakers, I can place a speaker in the general location where I'd expect to hear the product to determine the impact in my setting. For example products with a higher frequency are now more pronounced with the laminate than carpet. Others might have different enclosures or materials that affect the sounds in other ways.

Having the visual offers a "cool factor", and perhaps to a person well versed in interpreting it they can just look at the visual to interpret the audio without downloading. I've done a lot of work with the visually impaired who use speech output software like JAWS where these audio samples would be excellent for making a system that is not only quiet, but also making sure the noise it does emit doesn't interfere with what the user needs to hear.

Rory Buszka
Posts: 175
Joined: Tue Oct 04, 2005 4:58 pm

Post by Rory Buszka » Wed Jul 26, 2006 1:45 pm

Breunor, the main value of the FFT is not that they will help you anticipate how a fan will sound, but they will help you compare the intensity of noise artifacts (clicking, buzzing, ticking, etc.) between many different fans.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Wed Jul 26, 2006 2:00 pm

Rory Buszka wrote:Breunor, the main value of the FFT is not that they will help you anticipate how a fan will sound, but they will help you compare the intensity of noise artifacts (clicking, buzzing, ticking, etc.) between many different fans.
Whether they actually will is not clear at this point, not to me. Between two fans that are fairly different, no problem, but then you can hear that on our recordings pretty easily, too, so you don't need the FFT graphs for that. But between two close competitors? I don't know that the FFT will be very useful unless you really know what you're looking at/for -- the sound file, to me, would still be easier to tell the differences with, especially if you study them -- ie, listen many times closely, comparing back and forth.

Anyone else with comments on this? Do the sound files tell you lots or not? And for those of you who have some of the products we havce recorded, are our sound files indicative of what you hear / experience? (especially the ones made / posted since july 4, earlier this month.)

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Wed Jul 26, 2006 2:04 pm

breunor wrote:I've done a lot of work with the visually impaired who use speech output software like JAWS where these audio samples would be excellent for making a system that is not only quiet, but also making sure the noise it does emit doesn't interfere with what the user needs to hear.
This is pretty cool, btw, I hadn't really thought about that. My brother is blind and he uses JAWS. I built him a silent Shuttle Zen like the one in our sound lab... and he loves it, and some of his friends & techs (who occasionally visit for whatever...) can't believe how nice it is not to have the noise.

jaganath
Posts: 5085
Joined: Tue Sep 20, 2005 6:55 am
Location: UK

Post by jaganath » Wed Jul 26, 2006 2:42 pm

Those 3D FFT plots look very cool. :P Presumably the sharp peaks are harmonics of the blade-passing frequency?

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Wed Jul 26, 2006 3:05 pm

jaganath wrote:Those 3D FFT plots look very cool. :P Presumably the sharp peaks are harmonics of the blade-passing frequency?
Harmonics? I don't think so. They may be the effect of the blades -- the number of blades and RPM may dictate the rise/fall, but harmonics refer not to time-related variances, but rather, multiples of the primary frequency of sound.

If my theory is correct, then if you have a 7 blade fan spinning at 1000rpm, you'd see 7 peaks/valleys per revolution. Since there are 16.7 revolutions per second, this would mean 117 peaks/valleys per second. Which is obviously not the case -- the plots are for <5 seconds and there are nowhere near that many peaks/valleys.

So I don't really know the significance of the peaks/valleys in amplitude.

If we assume that with 7 blades and 1000rpm that the overall sound is the same as a single blade (like a platter) because the initial spike and following decay of each blade would all blend together too quickly for our minds to separate, then it's easy to calculate the fundamental frequency -- 1000 divided by 60 = 17 Hz.

Most of the regular sounds of the fan will be multiples (harmonics) of this frequency, and the particular mix of harmonics will depend at least partly to the firing frequency of the commutators in the fan, the resonant frequencies in the materials/shape of the fan parts, and the resonances of whatever the fan is attached to.

breunor
Posts: 96
Joined: Tue Jul 12, 2005 10:18 am

Post by breunor » Thu Jul 27, 2006 5:58 am

As long as the visual tool used has a set scale I can see where you could make comparisons. If the image scales to the greatest noise peak on each sample though, it would be a problem I guess. It would just seem a person would have to train themselves to mentally map noise to a charts bands so that you could look at an image and think "that will sound like this", rather than just listen to the sound itself. I expect I'm missing part of the equation; if the visuals improve the interpretation of the noise for someone, then it's a good tool for them. I'm not trying to knock it.
MikeC wrote:This is pretty cool, btw, I hadn't really thought about that. My brother is blind and he uses JAWS. I built him a silent Shuttle Zen like the one in our sound lab... and he loves it, and some of his friends & techs (who occasionally visit for whatever...) can't believe how nice it is not to have the noise.
I trained a guy who needed to use JAWS as well as Naturally Speaking, as he had severe motor impairments. He operated a laptop with his voice to query a spreadsheet to find information for customers. It was the ultimate test for sound quality-he had to speak to the system, hope it interpreted the commands :roll: and then JAWS would speak the information, which he would repeat to the customer. Ultimately the battery was the the killer, with all that stuff running we couldn't keep it running long enough for him. Even got that huge flat battery pack that a laptop can sit on (it's the size of a closed laptop itself), but couldn't get more than 3 hours back then. But that was maybe 7-8 years ago now, I expect the power use/battery life ratio would be much better now. I wonder if VIA's little mini-itx systems can handle JAWS? You'd get a lot of run time that way. :D

Post Reply