Ambient levels and noise floor discussion

Discussion in 'General Audio Discussion' started by Serious, Nov 28, 2019.

  1. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    I acquired an ACO 7020 1" externally polarized measurement microphone which has the following special features:

    1. equivalent self noise specified at < 2 dBA - mine was measured closer to 0 dBA
    2. sensitivity of 100mV / Pa where 1 Pa = 94 dBSPL
    3. ±2 dB 3Hz to 10 KHz - mine measured ±0.25 dB to 5 KHz where it begins roll off to -2 dB by 10 KHz
    4. very well controlled omni directional polar pattern transition to uni-directional from low to high frequencies

    With 200V polarizing power supply, mic preamp body (impedance converter) and special mic body to power supply cables this is a $3K mic system capable of measuring close to 0 dBA. Should be exemplary for room ambient noise measurements.

    Calibration with Bruel & Kjaer 4231 mic calibrator:
    00 IMG_0626_small.jpg

    System calibration with Smaart
    RME ADI-2 Pro set for 0 dBFS = +24 dBu (lowest system noise for this measurement setup)
    P-Solo gain adjusted for 94 dB SPL = -16 dBFS
    01 20191217-1640 cal 4231 7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png

    Now testing the electrical equivalent input noise a 160 ohm resistor is substitute for the microphone after gain staging.
    -1.7 dBA electrical only, which is below the mic capsule self noise of close to 0 dBA.
    02 20191217-1605 P-Solo ADI-2 Pro ref +24dBu 160R -1_7 dBA  7020 100mV Pa gain 94dBSPL=-16dBFS.png

    Another set of residual noise measurements in both my lab and other rooms of the residence, which is in a reasonably quiet suburban residential neighborhood.

    Exceptionally quiet moment, sitting absolutely still and holding breath while approximately 1 meter from mic: 10 dBA!!!
    External noise contamination was limited to the rumble below 60 Hz that occurs in a suburban environment from trains, planes and automobiles at a considerable distance as evidenced in the rise to 20 dB below 100 Hz.
    03 20191217-1645 very quiet moment 10 dBA  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png

    A more typical level, still sitting absolutely still and breath holding: 19.6 dBA
    Greater external contamination present by vehicles passing on a street outside the development.
    04 20191217-1605 room residual 20 dBA  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png

    Same as the previous setup, but now the HVAC operating in heating mode: approximately 30 dBA
    My HVAC has a lower noise than most residences as you may have suspected. Wanting to stay warm / cool has noise penalties.
    05 20191217-1700 HVAC living room 30 dBA  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png

    Now everyone's necessary activity, normal breathing approximately 1 meter from the mic in the room: approximately 35 dBA
    06 20191217-1900 quiet breathing 100 cm 30 dBA  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png

    It doesn't take much to raise the room ambient noise to 30+ dBA. A simple drag of Norne Audio headphone cable across clothing while turning one's head will produce 30+ dBA as experienced at a distance of 1 meter from the noise source.

    Picture of 7020 located in lab.
    IMG_0627_small.jpg

    If I wanted to be very precise this would be located center of room and aimed vertically. It makes a difference. But in this room the difference is small due to acoustic treatments.
     
    • Like Like x 9
    • Epic Epic x 1
    • List
    Last edited: Dec 17, 2019
  2. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    When measurements lie and closed back headphones rule

    Today my neighbor (also an audio professional) is having his driveway cleaned with a gasoline engine powered pressure washer. One of the rooms in my house is adjacent to his driveway. Opportunity for a real world noise measurement. The measurements below used the same calibrated measurement system described in post #61 above.

    Smaart displays both spectrum and SPL calculated from spectrum. Usually the spectrum display is UNWEIGHTED while the SPL is either dBA or dBC. In this case 35.0 dBA is displayed and the spectrum is unweighted. The SPL weighting and response time is listed below the reading in small numbers, lower left. Max listed lower right is the loudest event since last time reset. Given I typically check a 0 dBFS = 110 dBSPL and forget to reset there will be 110 shown lower right below main SPL reading.
    21 20191218-0840 pressure washer outside 35 dBA  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png Now 35 dBA should be pretty quiet. In post #61 above normal breathing produced approximately 35 dBA. Yet the pressure washer is clearly louder and very annoying. How can this be?

    The great measurement lie: dBA.
    A-weighting filter with the curve shown on the Red line below is applied to the unweighted data for SPL to calculate dBA:
    SLM A-B-C weighting curves - small.png Sound level below 500 Hz is attenuated considerably, 20 dB at 100 Hz and by 50 dB at 20 Hz. Many indiscretions may be hidden with dBA. Electrical examples include amplifier mains noise, Johnson noise, etc. Acoustic examples include gas operated engine noise, automobiles, night club bass thumping, etc. Looking at the curve above we note C-weighting is very close to unweighted data.

    Gas operated pressure washers vary in spectral noise as the engine responds to changes in the spray output. Everyone should be familiar with the rising and falling pitch and sound level. In the following graphs data is captured as close to the same levels for comparison but some variation in the spectrum will be apparent with the different snapshots in time.

    Using C-weighting the same pressure washer is reported 53 dBC. This is approximately 20 dB higher than the dBA reading. Spectrum display is still unweighted,
    22 20191218-0855 pressure washer outside 53 dBC  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png
    53 dBC, nearly 20 dB higher. Remember dB is logarithmic and 20 dB represents 10x change. No wonder the gas engine pressure washer is so annoying while normal breathing isn't.

    In the following graph both SPL and Spectrum display have A-weighting filter applied. Note the drop in low frequency spectrum:
    23 20191218-0855 pressure washer outside 35 dBA  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png
    Considerable attenuation of low frequency information by A-weighting.

    Both SPL and Spectrum display are C-weighted in the following graph. Note rise in low frequency spectrum and SPL displays.
    24 20191218-0840 pressure washer outside 53 dBC  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png
    This measurement is much more representative of the actual experience.

    Direct comparison of A and C weighted spectrum displays demonstrating how much attenuation is applied by A-weighting as compared to C-weighting. Same noise source, different measurement results based on weighting filter used:
    25 20191218-0855 pressure washer 35 dBA 53 dBC  7020-11144 ADI-2 ref +24dBu 94dBSPL=-16dBFS.png

    A-weighted measurements have been tradition as long as audio and acoustic measurements have been in existence. To be comparable all vendors report data with A-weighting in dBA. It's similar to the loudness wars. Everyone wants to measurements to give most favorable presentation.

    Different scales, dBA, dBB and dBC on sound level meters are selectable for knowledgeable users knowing how SLM measurements relate to the ISO 226:2003 equal loudness contours. dBA was for use in 40 phon loudness and below, dBC for 100 phon loudness and above. The little used (but should be) dBB covers ranges around 60 to 80 phons.

    Likewise in electronic measurements, bandwidth of analysis filters are important when assessing audio measurement data. One amp with a particular measurement in dBA will have an advantage over another that is using 20 Hz to 20 KHz bandwidth. The numbers aren't comparable. Both need to be in dBA or unweighted. Many vendors can hide power supply and thermal noise sins in a dBA measurement.

    This is a perfect example of when Focal Stellia rules. Annoying pressure washer noise is diminished sufficiently without having to raise listening levels to continue musical enjoyment while writing this post.
     
    Last edited: Dec 18, 2019
  3. spwath

    spwath Hijinks master cum laudle

    Pyrate BWC
    Joined:
    Dec 13, 2015
    Likes Received:
    7,894
    Trophy Points:
    113
    Location:
    Madison, WI
    upload_2019-12-18_14-23-9.png
    I got similarly quiet measurements in our acoustics lab. These are the measurements from 4 different people.
    Don't have the NC curves overlaid, but this has to be about NC 20 or 25. These were taken in the reverberation room. Probably was even quieter in the anechoic chamber, but couldent find my measurements from that.
    Also was a few people in the room, and some breathing, and our measurement procedure was to hold the B&K 2250 in front of us moving it in a figure 8 pattern, so there could be some noise from clothes moving and such. But each person seemed to get very similar results.
     
  4. Biodegraded

    Biodegraded Friend

    Pyrate Contributor
    Joined:
    May 28, 2017
    Likes Received:
    7,983
    Trophy Points:
    113
    Location:
    Vancouver BC
    In the same vein: the second graph in my earlier post shows that even with dBA weighting, low-frequency noise (mostly electrical in this case) is considerable. Here's VSLM's Noise Criterion plot, which is unweighted, of the same data:

    [​IMG]

    The NC is 32, but it's down at the 250 Hz bin - and the 125 Hz bin is almost as high on the NC classification curve.

    This location is now known as 'The Harman Room' :D
     
  5. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    How much SINAD is sufficient for real world listening?

    From the ambient room measurement data previously presented it might be fair to set 30 dBA as a reasonable listening noise floor. Now suppose we wish to listen at a fairly loud 85 dBA. That is only a 55 dB range. Add 20 dB signal to noise below the room noise floor to be safe. Allow 10 dB headroom for crest factor above average listening level. The total is 85 dB of range. Want to be extra safe maybe make it 90 dB SINAD. Anything beyond is lost unless either we listen louder (not good hearing longevity) or we listen in a much quieter space (most won't be able to achieve this.) What meaningful audible difference exists between any two components, based on SINAD measurements, once past 85 to 90 dB?
     
  6. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Part 1: Reality check with a cheap SPL meter or why you should always double-check

    As mentioned in a profile post, I really regret what happened in this thread. I have since apologized to atomicbob and Marv in private. I don't expect them to forgive me, but (as mentioned in my profile post) I still want to add to this thread and I hope afterwards we can all be a little friendlier. This first post is likely not going to be very interesting for others, but it's necessary to get a baseline for my actual measurements in the following post. I expect this series to be 4 posts.

    By the way, I found out that my original question has been answered quite extensively already in a study I found linked in a relevant ASR thread.
    Turns out the average home is around 10-15dB above an exemplary hearing threshold, equating roughly to an average A-weighted noise level of 30dBA. In the upper midrange the quietest measurement was 25dB lower than that and the maximum measured was roughly 10dB higher than that.
    It also turns out that someone on the AVSForum already did exactly what I had done - used an LCT550 (calibrated vs an SPL meter) to roughly measure the noise floor of his home theater room, with very similar results.

    With the original question answered much more accurately than I could ever hope to, I wanted to instead focus on measuring dB SPL levels for various sounds and examining relationships driving sound pressure levels. I will also be sharing calibrated (to the best of my ability) sound files for listening. More on that later…

    I've mentioned this before - I bought a cheap SPL meter as a sanity check against my SPL calibration. It can be found as 'RGBs SPL meter':
    RGBS_SPL_meter.jpg
    The reasons I chose this one are as following:
    • The acoustic design seemed more well thought out to me than the other cheap SPL meters. Not as big of an edge behind the tube holding the microphone, tube is slightly longer, body more rounded, etc.
    • I wanted A and C weighting, fast and slow response and an analog output. Many didn't check all of the boxes, this one did.
    • It was advertised as having a 1/2" capsule, which I thought was going to give me lower noise. It was still specified as 30dBA minimum range, so I didn't expect much, but I was hoping it'd be at least somewhat better than others in this price range.
    • It can be had for very cheap.
    • Lastly, I didn't want it to have any gaudy colors.
    Well, as it turns out it has a 6mm capsule inside its 1/2" capsule body. In hindsight this should've been obvious as there aren't really any cheap 12.5mm electret capsules available. A capsule swap with one of the 10mm capsules might be a cool project in the future in order to get lower noise, but with its 28.2dBA self noise I seem to have gotten slightly lucky already. Others get 35dBA self noise from the same meter. Conversely its maximum SPL is also 2dB lower than specified now.

    So how does it compare against my UMIK calibration?
    Generally, using noise around the fundamental region and midrange region (where we would expect its frequency response to be close to linear) it reads values around 1.2-1.3dB LOWER than my UMIK calibration. Now normally I'd say the cheap SPL meter is to blame, but in this case I think there's an argument to be made that this may actually be more accurate due to the following reasons:
    • Measured sensitivities for the electret microphones I have measured would be closer to their specifications. Sample size: A couple dozen microphones
    • Similarly, measured noise levels for the same electret microphones would be closer to their specifications. More on both points later in another thread.
    • The sensitivity I measure for my LCT 550 and the noise is also closer to what is specified. My measured sensitivity would otherwise be 1.0dB higher at 1kHz than what is specified. I think it's likely the LCT 550 matches its specification well, since they are specifically sold as being a matched pair without actually matching them. Still, can't verify.
    I'm still not entirely sold that my UMIK calibration is 1.0dB too high, but I am going to use the lower value for now. However I will not be offsetting all my already posted measurements. When I can get a chance I will compare it against real measurement equipment in an anechoic chamber. Afterwards I may offset all my posted measurements.

    I also tried to measure the FR of my meter (which was quite a tedious process) because I saw that some of these meters have a horrible FR. It turns out it's not the worst, but not the best either. A bit of a 10kHz peak. In a proper free-field measurement you would normally see the reflections from the body as a comb filter (see here and in the attached B&K 2230 specification), but I haven't taken measurements to the required accuracy to see this effect (within the uncertainties of the measurement itself), yet. I'd expect this meter to fare worse than the B&K 2230 in that regard.
    RGBs SPL meter FR family.jpg

    The overall linearity is slightly better than the venerable Voltcraft 33-2050 (aka Radioshack SPL meter), but it still has a noticeable peak in the higher frequencies. What's kind of weird is that it seems to have a slightly different FR internally than on its output. The output has less HF rolloff than the display reads. Also the A-weighting isn't perfectly accurate, so it's more accurate when used in the C-weighting. A-weighting has somewhat more LF rolloff than it should have (see the dotted blue line). Probably just component accuracy in the high pass filter section.
    The foam windplug doesn't result in a large FR difference and reduces wind noise by more than 20dB. Holding the meter 90 degrees off-axis results in a FR difference like you'd expect from a 6mm capsule. Both are very good results.

    There's an excellent comparison of the Voltcraft SPL meters on audioXpress. Turns out the digital ones have a truly horrible treble peak, but even the regular ones have a noticeable elevation across the upper midrange and treble. I assume this is the reason why the digital ones are said to be so inferior to the analog ones - their frequency response is truly awful compared to the older analog models.


    The SPL meter also has a DC output, which is a much more useful feature than I imagined. It outputs 10mV per dB, so if the display reads 28.2dB it will output 282.0mV. The great thing is that it's not limited to 3-4 digits like the display, so it allows you to connect a multimeter and more accurately calculate SPL values that are below its noise floor. For example: When set in my room with all windows closed and no noticeable noises present it will read 28.22dBA. When turning on my AC on its lowest fan RPM it will read 30.00dBA measured 1.5m from its output. Now we can calculate (or use sengpiel's website which does this) that the A/C has to emit a noise of around 25.27dBA in order to result in 30.00dBA total noise. This is very close to what I measure with lower noise equipment. I have even approximated lower SPL values (in the 10-20dBA range) with surprising accuracy using this method, hence why it is useful to have 2 decimals places of precision.

    To be continued...
     

    Attached Files:

    Last edited: Jul 23, 2020
  7. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Part 2: Sound isolation of windows and doors

    In order to measure SPL values inside buildings we may differentiate between two sources of noise. Inside noise, which for a large part we can control and minimize and outside noise, which is nearly impossible to control. Since indoor SPL values may be at or near the noise limits of many cheap SPL meters, I think it makes some sense to measure outdoor noise levels and subtract a certain number to get an estimate for the resulting indoor noise level. When I mentioned that 25dB isolation for windows is a good assumption, Ravi rightfully called me an idiot, who doesn't know how acoustics actually work. I have since measured the isolation of multiple windows and it turns out that my original estimate of 25dB+ was far off the mark. For pink noise I measured an A-weighted reduction of 35-40dB for the windows I measured!

    It seems the sound reduction index is actually a very well documented parameter for windows. Many regular 2 glass pane tilt and turn windows (which are common in Europe) have an A-weighted sound reduction index Rw of 37dB. So we can estimate that the outdoor noise will be reduced by circa 37dB when using A-weighting.

    Here's my own measurement for the sound reduction index averaged from a few windows in this house. These are very ordinary windows installed decades ago and I think it's quite common to come across windows that isolate even more than that.
    Window sound reduction index measurement 60-12kHz.jpg
    The plots are limited between 60Hz and 12kHz, since the speaker I used lacks bass extension and for higher frequencies I started to get close to the noise floor.
    As you can see the sound reduction drops off very quickly for low frequencies.

    However considering equal loudness contours this is not such a problem:
    ISO 226 Equal-Loudness contours vs inverse weighting curves.jpg
    Note that according to the ISO 226 Equal-Loudness contours the A-weighting works well for sounds in the 60 Phon range. For lower SPLs we hear less bass than even the A-weighting. The B-weighting seems to work fairly well for higher SPLs and the C-weighting works well to check how much low frequency sound a source has. It's debatable which equal-loudness contours are the most accurate, however I think the ISO 226 isn't bad. Personally I think hearing 20Hz at 75dB is a lot more difficult than hearing 1kHz at 0dB, so the LF contour could possibly be even steeper for me (at least at the hearing threshold).

    I also tried to measure how much doors within the house isolate. Generally they are not nearly as good as windows. Here's a graph for that:
    Door sound reduction index measurement.jpg
    A front door should be a bit better than that. I also tried to estimate the sound transmission of a Microsoft anechoic chamber door based on their YT vid as I could clearly hear the bass coming through the closed door when listening to the video. I estimated roughly 20dB reduction in the 20-100Hz range. To get significantly better sound isolation for low frequencies one could use a vacuum enclosure, but that is only useful for small devices.

    As an example for a window's sound isolation my neighbors lawnmower is circa 58dBA inside my room when I have my windows fully open and it gets reduced to roughly 20dBA with all my windows closed. Of course ultimately the reduction depends on the spectrum of the emitted sound and it's largely coincidental that the lawnmower gets reduced similarly to pink noise. But many sounds are somewhat comparable to pink noise in their spectrum, so the pink noise figure is still useful.
    Special windows can achieve 50dB of sound reduction in the midrange. Those might be used for buildings near an airport. Or consider that a jet engine is often quoted as 120-140dB for distances in the 10m range, yet inside the cabin (at a similar distance) the noise may only be in the 70-80dBA range.
     
    Last edited: Jul 24, 2020
  8. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Part 3: SPL noise table

    It's been a while since I made the other two posts and truthfully I never wrote down exactly what I was going to write.

    I was initially going to make two posts equally lengthy as the two other ones, but I think I can make more succinct posts and still get across what I was trying to say…

    dB SPL table SBAF Serious.jpg

    As you can see I have tried to measure dB SPL values for many common sounds. The ones marked with an asterisk are figures I stole off the internet. I also tried to get recordings of some of those and have uploaded them on a shared OneDrive folder:

    https://1drv.ms/u/s!AlMZTB6OPpUMgZ9fo_iUOiFIzPvRlA?e=hbL98U

    I was going to record and upload more sounds, but I think it's really not all that interesting. Sometimes the values can be deceiving and one sound may seem quieter than another one despite a higher measured number. Especially around our hearing threshold I think the A-weighting isn't a very good fit. But none of this is anything new or surprising.
    Some of the quieter sounds which I hear frequently (say the haptics of a phone, CRT monitor hum, a quiet laptop fan) are very hard to record without much background noise at a distance of 1 meter, but we can record them at a closer distance and remove the nearfield effect with EQ to get a convincing result. I haven't done this yet for the files in the folder, since I don't think it's all that interesting. For sounds at larger distances we run into issues estimating the SPL since the air dampens higher frequencies more so than lower frequencies, which is why the fireworks may be even louder than 172dB at 1m.

    One sound that is suprisingly loud momentarily is a regular balloon being inflated to rupture. There's a paper which claims a value of 150 dB (link), but my own estimate from a recording at 4m distance is closer to 145 dB equivalent at 1m. (I would've clipped the soundcard input at 1m.) The difference between the quietest sound and the loudest sound in the folder is an impressive 130dB.

    I find it remarkable how much dynamic range microphones can have, not that it really matters when DACs and amps, let alone speakers and real-world listening conditions don't allow for it to be reproduced fully.
     
  9. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Part 4: Two examples of relationships driving SPL

    Another thing I did was to examine fan speed vs dB SPL and car interior noise at different speeds.

    NF-A15 Fan theoretical SPL vs RPM at 1m SBAF Serious.jpg
    The fan measurements match quite well with what silentpcreview got for the NH-D15 fans, just that I tried to estimate SPL values lower than 10 dB by subtracting background noise and by taking nearfield measurements. Take everything below 10 dBA with a grain of salt, though. Due to rubbing noise the trend may not continue as well into the very low RPM ranges.

    Still, it is quite clear that fans running at low RPMs are effectively silent. I cannot hear my fans running at 200 RPM. To provide some context, I measured the fan RPM for various CPU powers with my fan profile. See the attached image. My figures are roughly 2dB louder than the silentpcreview figures for the whole NH-D15 heatsink with two fans, but keep in mind that mine were done in-room (with reflections) and I also took the measurements at an angle slightly closer to on-axis, rather than the diagonal measurement of SPCR.

    speed noise table car SBAF Serious.jpg
    The car interior noise measurements were taken in a Mazda 6 GJ. Different road materials and tires are going to influence the results somewhat, especially in the 60-120 km/h range where tire noise seemed to dominate subjectively. On OPA it is subjectively much quieter, even if the measurements generally only get reduced by 2 dBA with no real apparent reduction in the dBC.
    One thing is clear: Driving at high speeds will generally be very loud from wind noise. More high end cars may fare better, but for driving long distances at high speeds on the Autobahn I'd recommend earplugs either way. Note that I don't have measurements for the 220-250km/h range, but you can expect the trend to continue. At 250km/h we're approaching the SPL of the interior of an aircraft at cruising altitude. Trains are much quieter at similar speeds (see the estimated ICE 3 figure above).
    Noise from transportation is mostly LF rumble, but I'm convinced that this is also quite damaging to our ears.


    That's pretty much it from me as far as the whole ambient noise and dB SPL discussion goes. I initially wanted to make these last two posts more in-depth, but could never find the motivation to spend even more time recording sounds, making prettier graphs and longer texts, so this is it. One thing I learned is that the dynamic range of sounds in real life is very high.
    (LOL, I wore double hearing protection when I popped the two balloons in the backyard. I don't even want to imagine how loud a gunshot is.)
     

    Attached Files:

    Last edited: Dec 6, 2020

Share This Page