Why Headphone Target Curves and THD Don’t Tell the Whole Story
Introduction
After weeks of measuring, presenting, and debating headphones at events including the AES International Conference on Headphone Technology and xMEMS Live Asia, one thing is clear: headphone specifications rarely tell the whole story. Despite endless measurements, a lack of standards means there’s little consistency in the way frequency response, sensitivity, and THD are measured between manufacturers, making results nearly impossible to compare.
Why Frequency Response Specs Mislead
Frequency response measurements are hard to compare as most include the unique response of the ear simulator, yet it is rarely specified what simulator was used. Similarly, test level, corresponding sound pressure level, and the number of times the headphones were repositioned on the ear simulator and averaged is rarely stated. A clearer picture would come from showing frequency responses from several measurements to demonstrate how strongly the responses shift due to ear seal leaks and high frequency directional characteristics when the headphone is repositioned.
There would also be value in sharing individual harmonic curves so it’s clear how much of the THD is even versus odd order harmonics. This information reveals a lot about the type of distortion, cause and audibility. Specifications should also state whether measurements were made with ANC on or off, and wired or via Bluetooth connection.
The Problem with Target Curves
There was a lot of talk about target response curves at the AES Headphone Conference – unsurprising since Dr. Sean Olive was a keynote speaker. The legendary Harman Target Curve gained widespread recognition as the first widely-shared target curve based on listener preference using many subjects. It is understandable why this is a preferential design target over a diffuse field correction curve, which it is often compared to. That said, the diffuse field correction curve was designed as a reference curve, not a design target, so it’s not really comparing like-for-like.
Target curves also have other limitations. Since a target curve is simply an ideal targeted frequency response, as with headphone frequency responses, information is rarely provided on the specific ear simulator used to develop the curve, making it challenging to reliably compare your own device to the target. The original Harman target curve was measured on a modified GRAS 43AG ear simulator with a custom pinna. This makes any comparison imperfect, since any other coupler will introduce its own signature.
Target curves differ for in-ear and over-the-ear headphones, and although they are based on large amounts of empirical data, they are by definition, an average, and not specifically attuned to any one particular person. Furthermore, a specific target curve is not universally aspirational. For example, the gaming industry places more emphasis on factors such as spatial cues, bass, or clarity for voice than the frequency response that is ideal for listening to music.
Target curves are also one-dimensional, ignoring factors such as phase matching, distortion, time variance due to signal processing (such as that found in ANC), and other factors that contribute to listener perception.
The onboard DSP now available for headphones means that the natural progression of a target curve based on many people is a customized target curve. There were several papers at the conference, such as those by Nielsen and Hamasaki that addressed the evolution of customized target curves based on auditory models. This allows frequency response to be customized based on the listener’s ears, but how then is that presented in a headphone specification? How does an independent reviewer even describe the sound quality, when each person will configure it differently?
ANC Headphones: Quieter but not always better
Active noise cancellation is another area where specifications do not fully represent listener experience. Manufacturers boast about noise reduction levels but rarely disclose how it is measured and how ANC affects the sound quality. Conventional measurements such as THD usually demonstrate that ANC reduces distortion, yet most audiophiles claim the opposite. In my paper “Active Noise Cancellation Distortion in Headphones” at the AES conference, I demonstrated that although THD is indeed reduced when ANC is activated, this measurement alone is an incomplete picture. Time domain analysis reveals time and phase variance showing where perhaps the experience of poorer sound quality comes from. Surprisingly, the best correlation to listener perception was found with the POLQA algorithm – a communications quality algorithm that is certainly not designed for measuring headphone performance! Clearly more work is needed to arrive at a similar measurement method that truly captures the effects of active noise cancellation in a way that correlates well with human hearing.
Ironically, although R&D engineers struggle with producing headphone specs that correlate to listener experience, we’ve made significant headway with perceptual production line test methods that align well with human hearing to identify manufacturing defects such as perceptual rub & buzz and perceptual transient distortion detection.
Why Conventional THD fails for headphone measurements
A couple of weeks later, I presented at xMEMS Asia live events in Taipei and Shenzhen, where I talked about small drivers being used in headphones and wearables such as smart glasses and smart watches. This time I was addressing a specific issue with the way many people measure loudspeaker THD.
Conventionally, harmonic distortion is compared to the fundamental at the stimulus frequency. This is not totally accurate as linear distortion is superimposed on the non-linear distortion. This often goes un-noticed in conventional speakers where the highest distortion is usually at low frequencies near resonance, and the fundamental response is typically fairly flat since it is not measured with an ear simulator. The target response for headphones, however, is not flat and their fundamental response influences their harmonic responses.
Conventional THD measurements are also a poor choice for piezo-MEMS speakers. These tend to be very linear at low frequencies, exhibiting distortion peaks at higher frequencies above 5kHz. For such devices, distortion that lies outside the audible range can be misrepresented as audible when referenced to the fundamental at the stimulus frequency, creating a misleading picture that doesn’t match perception.
Since 1993 I have been promoting the use of frequency normalized distortion measurements. With frequency normalized distortion, the harmonic distortion is plotted at its measured frequency and compared to the fundamental response at the same frequency before plotting difference at the stimulus frequency which separates non-linear distortion from linear distortion. When piezo-MEMS speakers are measured using this method, it becomes clear that the higher distortion is just a consequence of their linear response or rising sensitivity versus frequency which is equalized to be flatter as part of the product design. It also makes resonances far clearer, allowing their effects to be mitigated, for example, by damping.
With the explosion in the use of MEMS speakers for headphones, hearing aids and wearables, I suspect that normalized distortion will at last become more widely accepted as the best way to measure THD. After all, accurate measurements are the key to good product design.
Closing the Gap Between Specs and Perception
So where does this leave us with headphone and headphone component measurements?
If we want to be able to tell what a headphone sounds like from its specifications, the industry needs well-documented standardized tests such as IEC 60268-7 and measurement methods that go far beyond frequency response and THD. As these devices become more feature-rich with onboard DSP, ANC, custom response profiles and more, the industry must develop and standardize new measurement methods that correlate with human perception.
Until then, we’ll just have to rely on the subjective opinion of the many reviewers out there, together with the objective measurements that they have made. At least when review sites make objective measurements, they use consistent equipment, making results comparable, even if they don’t align with manufacturer specifications.
