Quickly back to the crafting work. At the microphone input (but not only there) I now use the Soundblaster AE-9 including external panel, which is simply more practical and the bias voltage is also correct. The important thing is that there is as little noise or distortion as possible, even if the levels are of course far above the background noise. The AE-9 even delivers better results here than the Steinberg interface, and the headphones are even better off on the front of the AE-9 than on the Asus Xonar Essence STU. Keyword impedance and maximum voltage gain.
The Harman Curve
The so-called Harman curve is an (optimal) sound signature that most people prefer in their headphones. It is thus an accurate representation of how, for example, high-quality loudspeakers sound in an ideal room, and it shows the target frequency response of perfectly sounding headphones. Thus, it also explains which levels should be boosted and which should be attenuated based on this curve. This also explains the term “bathtub tuning”, which is often quoted, but in which the Harman curve is completely overused and exaggerated.
For this reason, the Harman curve (also called the “Harman target”) is one of the best frequency response standards for enjoying music with headphones, because compared to the flat frequency response (neutral curve), the bass and treble are slightly boosted in the Harman curve. This “curve” was created and published in 2012 by a team of scientists led by sound engineer Sean Olive. At the time, the research also included extensive blind tests with different people testing different headphones. Based on what they then liked (or disliked), the researchers found and defined the most universally popular sound signature.
Headphone tuning can be really problematic because of the human anatomy. Everyone has a slightly different pinna and ear canal, which affects how individuals perceive certain frequencies. In extreme cases, there is a few dB difference from person to person, which then explains the small differences in some measurements with artificial ears. Furthermore, if the sound is not absorbed, it is additionally reflected by other surfaces. Theoretically, a torso could also be included in the test setup, but that would be far too time-consuming.
A comparison with Oratory
I followed the advice and my measurements and based on the 1/8 rule for output impedance with the Sound Blaster AE-9, I use an amplifier with under a 1 ohm. And why all this? The impedance of the headphones should be at least 8x the impedance of the amplifier to avoid possible sound changes, which can be especially negative in the bass (level drop, changed curve). If you also want to measure headphones with 16 Ohm impedance, then we are already at 2 Ohm as maximum for the amplifier! This is then the so-called impedance curve, which, however, does not play any role with my electrostats, for example, so that one can already compare and judge quite well there.
For the first comparison and as a basis for discussion, I had measured the popular and well-known Beyerdynamic MMX300 with 600 ohms and grabbed the appropriate measurement from the database at Oratory. Unfortunately, it does not say which generation with which impedance was measured there, but the curves already look very similar. I use the second generation for my measurement, but here is the reference from Oratory for now:
In the meantime, the measurement was done on the Beyerdynamic A20 behind a very good USB DAC (192 kHz), the curves were normalized to zero (in contrast to Oratory) at 1 kHz as reference point. The curve up to the upper midrange is quite similar to Oratory’s measurement result. The other peaks are a bit apart, even though the character still follows the Harman curve quite well:
Moment of truth: Exemplary test of a gaming headset
Hi-Fi is one thing and since I compared other curves (and they coincided amazingly well), I am now quite sure that everything is as it says and you can also measure something cheaper. So at least I know that the curves are really so bad in an emergency and that it is not due to the measurement setup. Thus, we come to the first measurement in the cheaper segment, because you can see errors much more clearly here. You can see the quite well-developed bathtub very nicely, but it weakens a bit in the bass. I also adjusted the chart graphic, because I don’t want to copy Oratory.
However, the problem with this headset is not so much the drop in the low bass below 40 Hz, which is actually still easily tolerable. What is really annoying here is the very pronounced upper bass, which leads to a somewhat unpleasant “cardboard sound” from 200 Hz. Insiders know this from the party basement, where cheap Chinese scratch boxes boom and drone. I would have preferred to see the hump of the curve a bit more to the left, but ok, it’s a gaming headset. The dark curve is the Harman target, see above.
The lower midrange collapses a bit because the bass pushes everything into the background a bit. The dip at around 550 to 600 Hz isn’t really noticeable, though, even if it spreads some coolness because the fundamental frequencies weaken a bit towards the top. At 2.5 to 3 KHz, we see the hearing-related level increase, which is slightly stronger than the ideal line of the Harman curve. The drivers also jitter a bit in the super high frequencies at about 10 KHz, but you will hardly notice that subjectively in gaming. But we will see this area again in a moment. It is also interesting that the drivers still play cleanly up to 20 KHz, which is also not a matter of course. So you can also listen to music with them, even if they are not hi-fi headphones. But it’s not a toot either, just a bit top heavy on the upper bass. You can like that, or not (like me).
Cumulative spectra (CSD, SFT, Burst)
The cumulative spectrum refers to various types of graphs showing time-frequency characteristics of the signal. They are generated by sequentially applying the Fourier transform and appropriate windows to overlapping signal blocks. These analyses are based on the frequency response diagram already shown above, but additionally contain the element of time and now show very clearly as a 3D graphic (“waterfall”) how the frequency response develops over time after the input signal has been stopped.
Colloquially, such a thing is also called “fading out” or “swinging out”. Normally, the driver should also stop as soon as possible after the input signal is removed. However, some frequencies (or even whole frequency ranges) will always decay slowly(er) and then continue to appear in this diagram as longer lasting frequencies on the time axis. From this, you can easily see where the driver has glaring weaknesses, perhaps even particularly “clangs” or where resonances occur in the worst case and could disturb the overall picture.
Cumulative Spectral Decay (CSD)
Cumulative spectral decay (CSD) uses the FFT and a modified rectangular window to analyze the spectral decay of the impulse response. It is mainly used to analyze the driver response. The CSD typically uses only a small FFT block shift (2-10 samples) to better visualize resonances throughout the frequency range, making it a useful tool for detecting resonances of the transducer. The picture shows very nicely the transient response and some present bass resonances in the upper bass. It has to come from somewhere. The diaphragm resonates slightly below 250 Hz. Lousy, highly compressed MP3 files or lousy YouTube streams are crystallized a bit in the treble by the peaks, but with very good recordings this is already a bit too much for me. You can love it, but you don’t have to hate it. Actually, it fits quite well.
Short-time Fourier Transform (STF)
The Short-time Fourier Transform (STF) uses the FFT and Hanning window to analyze the time-varying spectrum of the recorded signals. Here, one generally uses a larger block shift (1/4 to 1/2 of the FFT length) to analyze a larger portion of the time-varying signal spectrum, especially approaching application areas such as speech and music. In the STF spectrum we can now also see very nicely the work of the drivers, which afford various weaknesses in some frequency ranges. This “dragging” at the lower frequencies below 500 Hz is then repeated and at about 2.5 to 3 kHz and then there is the jittery whip in the super high frequencies at about 10 kHz.
Burst Decay
In the CSD, the plot is generated in the time domain (ms), while the burst decay plot used here is represented in periods (cycles). And while both methods have their advantages and disadvantages (or limitations), it’s fair to say that plotting in periods may well be more useful for determining the decay of a driver with a wide bandwidth. We see a strong resonance oscillation in the upper bass with the maximum at about 200 Hz, a few small lags around 2 to 3 kHz and later a peak between about 8 and 12 kHz. But at least the treble is nothing that is subjectively perceived as a real negative. However, the bass is not only audible, but also visibly a bit too muddy and pop-like here.
These are the metrics that I will continue to use in the upcoming tests, of course, based on the new test setup and Arta as the software.
65 Antworten
Kommentar
Lade neue Kommentare
Mitglied
1
Urgestein
1
Neuling
Urgestein
1
Veteran
1
Mitglied
Veteran
1
Mitglied
Veteran
Mitglied
1
Urgestein
Mitglied
1
Alle Kommentare lesen unter igor´sLAB Community →