This is a first – a fan with the leading edges of the impeller blades on the opposite side to normal ones. This is primarily done for a better view of “fans without stator struts” in cases with glass side panels. In addition, such an unconventional design also has quite clear and measurable advantages and disadvantages, also in terms of functional characteristics. Let’s take everything in turn.
33 dBA or 33 dBA
The noise level, given as a single dBA value, is good for quick reference, but it doesn’t give you an idea of exactly what the sound sounds like. That’s because it averages a mix of noise levels of all frequencies of sound. One fan may disturb you more than the other, even though they both reach exactly the same dBA, yet each is characterized by different dominant (louder) frequencies. To analyze thoroughly with an idea of the “color” of the sound, it is essential to record and assess noise levels across the entire spectrum of frequencies that we perceive.
We already do this in graphics card tests, and we’ll do it for fans too, where it makes even more sense. Using the UMIK-1 miniDSP microphone and TrueRTA’s mode-specific, fixed dBA application, we also measure which frequencies contribute more and which contribute less to the sound. The monitored frequency range is 20-20,000 Hz, which we’ll work with at a fine resolution of 1/24 octave. In it, noise levels from 20 Hz to 20 000 Hz are captured at up to 240 frequencies.
The information captured in the spectrograph is a bit more than we will need for clear fan comparisons. While you’ll always find a complete spectrograph in the tests, we’ll only work with the dominant frequencies (and their noise intensities) in the low, mid, and high bands in the comparison tables and charts. The low frequency band is represented by 20–200 Hz, the medium by 201–2000 Hz and the high by 2001–20 000 Hz. From each of these three bands, we select the dominant frequency, i.e. the loudest one, which contributes most to the composition of the sound.
To the dominant frequency we also give the intensity of its noise. However, in this case it is in a different decibel scale than those you are used to from noise meter measurements. Instead of dBA, we have dBu. This is a finer scale, which is additionally expressed negatively. Be careful of this when studying the results – a noise intensity of -70 dBu is higher than -75 dBu. We discussed this in more detail in the article Get familiar with measuring the frequency response of sound.
Strict acoustic safeguards are required to ensure that these measurements can be carried out with satisfactory repeatability at all. We use acoustic panels to measure the same values at all frequencies across repeated measurements. These ensure that the sound is always reflected equally to the microphone regardless of the distribution of other objects we have in the testlab. The baseline noise level before each measurement is also naturally the same. The room in which we measure is soundproofed.
Like the noise meter, the microphone has a parabolic collar to increase resolution. The latter is specially in this case not only to amplify but also to filter out the noises that occur whether we want them or not behind the microphone. We are talking about the physical activity of the user (tester). Without this addition, human breathing, for example, would also be picked up by the spectrograph. However, this is successfully reflected off the microphone sensor by the back (convex) side of the collar. As a result, the spectrogram only contains information about the sound emitted by the fan itself.
Any plans for tests of the non-reverse variant, and also spacers (like Noctua NA-IS1)? To me they would be the logical next steps for topics raised in this test.
We do not plan to test the standard variants of TR120 fans in the near future. In the long run, the basic plan remains the same – we have to compare all the fans that exist, haha. Only time will tell where we will really end up. 🙂
We’ve had the NA-IS1 frames in our editorial office since their introduction. Of course it would be really useful to get to them and one day it will come. I still can’t make space for them – there is always something “more important”.
Always looking forward to your tests, whatever they are🙂
One additional question though, could you further elaborate what’s going on with the hexagonal grille tests for this fan? First time I’ve seen a fan that somehow has parts of the data missing in the middle (33 and 36 missing, but not 31 and 39). You say unstable tonal peaks, so is the RPM unstable at those noise levels, or is it due to something else?
I think it’s happened before. But maybe it was for the quietest or loudest mode of normalised noise and then it can be attributed to another reason? Anyway, I will explain.
It is important to realize how we bring the fans to the “same noise level”. It’s not like we set the fan somehow and it’s stable at, say, 36 dBA. It’s that we set the fan so that the average of 30 samples ends up at 36 dBA. The interval of these samples can be in the range of 35,9–36,1 dBA during the measurements, for example, but possibly also in a much wider range, for example 32–38 dBA, and now I am not exaggerating. This latter case is similar to trying to get the TR120 into the 33 and 36 dBA modes, which failed. With no PWM setting (nor after very fine tuning of the pulse strength with voltage in single digits of mV), we could not set the fan on the grille so that the average of 30 samples corresponds to 33 dBA and 36 dBA, respectively. It was always more or less, i.e. not what was required. This is a topic that certainly makes sense to look at in more detail. Especially after Noctua opened this “Beat frequency theory” topic, which is a good basis for understanding the issue by a wider than very narrow spectrum of users.
–“Beat frequency theory”
…well, already during the 1st world war the company on the bridge had to stop marching in order not to shake the bridge.
Wave interferance
https://en.m.wikipedia.org/wiki/Wave_interference
In acoustic
https://en.m.wikipedia.org/wiki/Beat_(acoustics)
Thanks for your detailed explanation. The only other time I have heard the need of averaging noise samples is from ThermalLeft. What’s the sampling rate you’re using?
This phenomenon is definitely worth looking into some day, especially for instances like this where large deviations occur.
The sampling rate of the Reed R8080 is 1s. The settings for the individual modes normalized according to a fixed noise level are based on the arithmetic mean of a 30-second recording. This must always be exactly 31.0; 33.0; 36.0; 39.0; 42.0 or 45.0 dBA after rounding.