New Algorithm, New DAP Database

As of February 2022, HypetheSonics’ DAP database has been updated to reflect recent changes and improvements to soundexpert.org’s difference-level error algorithm. The new version 3.3 algorithm produces more detailed output, with each technical signal graphic now showing error as a function of frequency (ranging from 20 Hz to 20 kHz on the vertical axis). This is the reason for the appearance of horizontal banding in basic signals such as 1 kHz and 12 kHz sine waves, triangular and square-wave signals). Histogram medians from the old and new algorithms aren’t completely identical, which necessitated updating all slides in the database. (Note that the earlier version 2.4 database is still available in the legacy database section: https://www.hypethesonics.com/daptiv2p4-database.)


The need for this post created an opportunity to explain what tests are performed on these DAPs and what the improvements are in the new slides.

Our testing procedure remains the same. This involves having the device under test (DUT) play back a series of signals, including a signal with/without load to test output impedance, a series of standard technical signals, followed by a series of actual music tracks. These test tracks are described in detail on the soundexpert website: http://soundexpert.org/test-signals

One crucial observation on measurements from a large number of devices is that playback performance of various test signals usually doesn’t correlate well with one another, or with that of complex waveforms, i.e., actual music tracks. Most standard audio tests evaluate the ability of a DUT to accurately reproduce sinusoidal waveforms (single sine waves for THD/SINAD, or dual sine waves for IMD). While that may be intrinsically interesting, the fact that DUT performance on these simple test vectors correlates poorly with performance on more complex waveforms means that total harmonic distortion and noise at 1 kHz isn’t particularly relevant, because nobody spends their time listening to 1 kHz sine tones. The most important metric is the statistical, long-time aggregate error over a large sample of real music. As a result, the most important number to read on the df slides is the histogram median, i.e., the median error over the complete set of music test vectors. (Of the technical signals, the most relevant is the BS EN 50332-1 program simulation noise, which is designed to mimic the spectral content of actual music and voice, and this usually correlates better with the main histogram median.)

Besides the histogram (df) median value, another important number on the slides is the device’s output impedance. Many modern in-ear monitors have low and sharply-varying impedance curves (i.e., their electrical resistance changes with driving frequency). When coupled with a high output-impedance device, this skews the frequency-response of the headphone, relative to the FR created from a near-zero output-impedance source. The effects of source output impedance can be displayed from our main headphone-repository tool (https://www.hypethesonics.com/iemdbc/). In the tool menu, the output impedance can be entered in the ‘Z-out’ entry box. By default, this value is zero, since all headphones are measured using a near-zero output-impedance source. By adjusting the Z-out value to match the output impedance of your DAP, you can see how the frequency response of your headphones will shift. (Note that changing Z-out is a global effect that applies to all headphones and will change the results of any rank or profile search.) Ideally, one would want a DAP to have an output impedance as low as possible, or at least below 1 Ohm. Unfortunately, it’s common for many modern devices to still have a Z-out of 2 Ohm or higher (particularly on balanced-outputs), which is capable of significantly changing the frequency response of certain headphones.