Mike, I really appreciate your writing. You include enough details to convince me that you know what you are talking about, but not so many that I get bored and skip to the next paragraph. It's very dedicated of you to devote an entire room of your house to the project, even if it is theoretically reversible. Thanks for taking the time to write all 11 pages. I read them all.

Avalanche --

Glad you enjoyed the read. I wonder if the air traffic & Van. geography info is over the top, but it all seemed apropos at the time.

I agree, to the point that I, personally, don't think this was even necessary (though very cool!). Your noise floor was already at least a little lower than mine, so any differences that you couldn't perceive, I wouldn't perceive. For me, that's good enough: finding the difference between 12 and 13dB is completely irrelevant. I greatly respect and admire the work that went into this, though. This definitely cements SPCR as the authority on PC noise reviews, since no one else has a facility like this that I'm aware of!

Wasn't better to buy this instead of the anemometer? It can measure the static pressure too. I asked just for curiosity the price of the smallest model and they replied $ 5500.

http://www.fantester.com/

If you're interested I can give you in pvt the 2 pdf they sent me. I don't want to seem to advertize this manufacturer...; )

Actually, it was and is necessary for SPCR reviews -- perhaps not as much for the improved ability to measure / differentiate at super low levels but for the improved insulation from external noise. The frequency & level of air and ground traffic is high enough to prevent/interrupt low level recordings and measurements throughout the day... which is why I did that whole analysis of air traffic / noise around the house and Vancouver. This is why, even now, I partly regret not having gone the whole 9 yards -- doing the room-in-a-room construction for better isolation from external noise.

I did catch that, but it seemed like you were also saying you had to raise it to another level of accuracy at low levels, as to advance the "professionalism" (for lack of a better word) of the site through its ability to be precise at low levels. My decision to actually post a comment was triggered by the guy who was concerned over .04dB, which just made me laugh.

Yea, .04 dB.... well, nothing more I can add to that.

But w/ regard to improving low level accuracy -- I do feel it's somewhat necessary because the best products were getting beyond SPCR's ability to differentiate with measurements. They're so quiet now that the old system wouldn't show any measurement differences, and even the recordings weren't really good enough to show them. We could still hear qualitative and sometimes simple amplitude differences between them, and we could describe these, but there was no way to measure or record them... except at much close than 1m distance, which is unrealistic -- most people don't sit that close to the PC. To most people, the differences might be too small to worry about, but for some, it's important. OK, I admit it, that it's obsessive to quantify the difference between a 13 dBA fan and a 15 dBA one when either would be swamped by even a laptop drive. But being able to do it gives more confidence about measurements a bit higher up in level, too.

Here's a practical example, btw: I'm working on a PSU review right now, and the fan at no load or idle (say to 65W) seems to vary ever so slightly. It starts usually at ~16 dBA/1m, then over the next 10-15 mins, drifts up to as high as 18 dBA. I recorded it at both levels and the difference is audible, the 18 dBA sound is higher pitched -- ie, the fan is spinning faster. But wait another 5-10 minutes and the sound has gone back down to 16 dBA, with a lower rumble.

I might have caught this difference by listening before, but chances are slim. I don't hang around listening to a noise for 10 minutes at a time... and I'm not sure that I could be certain of noticing the change even if I did. Certainly I would not have been able to catch it with any of the instrumentation before. Now, it's clearly discernible and recordable. Is it significant? I'm not sure yet but I'm glad to know I can discern, record and quantify it.

I am left with a challenge tho: Should this PSU's idle be stated as 16 or 18 dBA? Because the variation takes place so slowly, it is probably not discernible in any real use. Maybe I should just call it 17 dBA?

I've been in an anechoic chamber, it's so quiet that your own heartbeat sounds loud, you can even hear your blood pumping through your arteries, kind of spooky, you'd go mad in there after a while.

My point is that if your old system couldn't differentiate, I, at least, am unlikely to be noticing the difference myself.

Why not call the PSU idle 16-18? Do you really need a single number? If it idles in a range, it idles in a range, and that's what you should say.

Someone not far from Vancouver already did and they are selling. For the small sum of 575 000 $ it can be yours

http://www.oneofakindhouse.com/fortress.html

We do usually indicate oddities like this in our writing, but the purpose of the tables is to provide a quick source of data to the readers. The engineer in me says that we should often pick the "worst" data to show. If a product really does excel in what it does, the numbers will show that for itself.

The problem that Mike was trying to indicate was that the variation was over a period of 5-10 minutes. Before having the anechoic chamber, detecting subtle changes like that was all but impossible (especially for noise levels that we're talking about). I can tell you that the difference in perceived noise inside and outside the room is very different, and both are equally important for us to determine the quality of a product.

Your description of the testing room, along with all the thought you put into building it, made for an interesting read.

I was curious why you decided on a test room rather than a test chamber. It would seem a lot easier to build something just large enough to hold a PC with you and all your equipment on the outside.

I don't mean to answer for Mike, but I think the problem with enclosing PC type of equipment in a small volume is heat. Ambient temperature is pretty crucial to convective cooling, and controlling it would mean ventilating a test chamber, which would be difficult to do without affecting test results (though not impossible). That's my guess, anyway.

Yes, there is the issue of thermal/acoustic testing things like >500W PSUs at various loads up to full power for many hours on end. A small box (like the 1 meter cube Anadtech uses for their audio measurements) would be unusable.

More basic is the issue of boundary effects in acoustics.

The simplest way to explain this: As you get closer to a wall or intersection between two walls, the lower frequency sounds get boosted. This boost can start as high as ~300Hz, and its effect can be several decibels. You can easily check this yourself with your own voice, in most rooms, moving from the center slowly towards a wall or corner while talking. You can hear the bassiness or chestiness of your voice increasing. (In the corner, you sound like a typical late night FM radio DJ, right?) That lower frequency rise is not natural, it is imposed by the geometry of the room, the longer sound waves reflect & get reinforced by the boundaries. This is one reason why commercial anechoic chambers are usually quite large (much bigger than my original 10x12' room), and why almost always, the objects being tested/measured are in the center of the room -- minimal effect from boundaries. In a too small space -- like that 1m cube -- the boundary effect is unavoidable. (In that box, everywhere is within a boundary effect) To be safe, basically you want both the mic and the noise source to be at least a meter from all room boundaries -- including ceiling and floor.

Also related, the "bass cutoff" frequency for an anechoic chamber increases as room volume decreases. My chamber is echo-free only to ~150Hz, below which the reflections steadily increase so that by ~50Hz, there is no difference with an acoustically untreated room. (Have a look at the unweighted room curve on this page, see how the plot line rockets up below 150Hz -- http://www.silentpcreview.com/article876-page10.html) The 600 lbs of damping in the room is an acoustic filter which is really effective except at the lower frequencies. Very little gets through it (from within or outside the room), but the lower the frequency, the more easily sound goes through it. This is why the unweighted SPL or frequency sweep at 8am looks quite different than at midnight when there is hardly any car traffic within many blocks. Though the morning traffic noise on Main St 2 blocks away is not really audible, the <100Hz rumbling that is barely audible by ear is easily picked up by the ACO Pacific microphone. This is one reason we continue to use the A weighting, it eliminates most of the ambient level variations in the chamber during the course of a day, because it filters out the lowest frequencies (much like human hearing).

Hi Mike,

First: I don't think you're doing anything wrong. If you have the option of a whole room over a small chamber, obviously you go with the room. Also, SPCR is great (I'm a long time reader, despite my low number of posts) and I love the attention to detail in the writing here. I haven't purchased a case or power supply in half a decade without coming here first. You're definitely doing things the right way, in my opinion.

I'd like to play devil's advocate for a bit on this chamber vs room discussion. Since you're looking for an additional test environment, this might be relevant in SPCR's current situation.


Let's say you get a 3 dB boost at 300 Hz. How would that affect your measurements?

If you're using A-weighting, the weight at 300 Hz is -7 dB. So let's say we're measuring a very peculiar sound that is made up of tones that drive both the 315 Hz and 1 kHz 1/3-octave bands to 100 dB. Unweighted, our SLM would read 103 dB (100 + 100), but A-weighted it would read 100.8 dBA (100 + 93). Add in the +3 dB change at 300 Hz and our SLM reads 101.5 dBA (100 + 96).

So that's a pretty extreme situation (the most extreme, I think), and it gives us an error of only 0.7 dBA, within Type I accuracy. For a real measurement all bands would have energy, including all of those non-weighted and positively-weighted bands between 1 kHz and 10 kHz, further diminishing the effect the 315 Hz band has on the total value. The resulting change in dBA would almost certainly be less than the differences between individual measurements of identical conditions.

Then consider that the primary noise source being measured is a small fan, which will have broadband noise with most of its energy 500 Hz and above. If the noise you're measuring starts out -5 or -10 dB at 300 Hz (with respect to 1 kHz), and then you weight it a further -7 dB, a boost of 3 dB in that band isn't going to have any noticeable effect on your A-weighted numbers. Even a 6 dB bump wouldn't be detectable.

So that's something to consider.

Another thing to consider is that it would be much easier to keep environmental noise out of a chamber than out of a room, especially at low frequencies. A box within a box is much easier to build at scales of a few meters, and large wavelength sound waves would be less efficient at permeating a smaller object.

Food for thought.