You are here

Designing & Measuring Reference Headphones

Heads Up

Headphones are central to music-making and music listening. But can we trust what they’re telling us? We ask the experts...

There has been a revolution in consumer music listening. Loudspeakers have been dethroned, and headphones rule in their place. From tiny earbuds to luxury wood-clad hi-fi models, global revenues for manufacturers are now estimated to top $10 billion.

As we spend more money on headphones, we expect to get a good quality product in return. This is especially true of music producers, who need our headphones to act as a reference, inspiring confidence that our music will translate to other pairs.

But what, exactly, does ‘good quality’ mean when it comes to headphones? What factors make one pair better than another, and what are the limitations on headphone sound quality? These are all questions that seem simple on the surface, but turn out to have hidden depths. To help SOS readers navigate them, I spoke to some leading lights in headphone design, testing and measurement.

Flat Earth

There is broad agreement that a reference monitoring system, be it speaker-based or headphone-based, should be tonally ‘flat’ or neutral. Yet beneath this bland statement lurks a family-sized can of worms, which is opened when we ask what exactly we mean by ‘neutral’.

It seems reasonable to define a neutral loudspeaker as one having a uniform frequency response across the audible spectrum. However, in any normal listening environment, reflection and absorption ensure that the output of such a speaker won’t be at all ‘flat’ by the time it reaches the listener. By contrast, headphone drivers are placed right up against the listener’s ears. There is no room environment to absorb, reflect and otherwise distort the sound; but at the same time, other acoustic phenomena become much more important. So, even if we could hypothetically use the same device as a speaker and a headphone earpiece, it couldn’t possibly be ‘neutral’ in both applications.

Most of the experts I consulted agreed that the ideal voicing for a pair of reference headphones would recreate the frequency balance — at the ear — presented by ideal loudspeakers in a perfect listening environment. This is the principle behind Sonarworks’ popular headphone calibration system, as the company’s Rūdolfs Putniņš explains. “From a music-production perspective, speakers are still the predominant sound reference system and headphones are secondary. From this perspective, we have to define reference sound on studio monitors. In layman’s terms, sound coming from ‘flat’ headphones should be perceived the same as from a ‘flat’ pair of speakers in a good room. In theory that is easy; however, in reality, much more complicated.”

Dr Drag Colich is Chief Technology Officer of Audeze.Dr Drag Colich is Chief Technology Officer of Audeze.Photo: Chris ParkinsonDr Drag Colich, the Chief Technology Officer of premium headphone manufacturers Audeze, concurs: “Headphones have to be corrected using a known preferred house curve, verified with a known measuring rig, with a sound level set up to be the same as in speaker sound evaluation. Audeze’s philosophy for a preferred house curve is to create headphone response similar to well-tuned high-end speakers in a good, acoustically treated room. The average correction curve was then verified by multiple qualified listeners and fine-tuned. A similar approach has been adopted by a few other headphone manufacturers, but each of them has their own preferred ideal curve.”

Like the other manufacturers we spoke to, Audio-Technica describe their goal in designing high-end models like the ATH-R70x as creating a ‘flat’ response. Unlike others, though, they choose not to define ‘flat’ with reference to loudspeaker listening.Like the other manufacturers we spoke to, Audio-Technica describe their goal in designing high-end models like the ATH-R70x as creating a ‘flat’ response. Unlike others, though, they choose not to define ‘flat’ with reference to loudspeaker listening.Not all manufacturers define a ‘flat’ response in terms of loudspeaker listening, however, and Hiroyasu Suzuki, of Audio-Technica’s Product Development Division, offers a different understanding of the term. “The meaning of a ‘flat’ response for Audio-Technica is to pull out the audio source without degradation. That sound has a characteristic to [allow the listener to] be able to enjoy music.”

In Situ

As Rūdolfs Putniņš says, measuring headphone frequency response is much more difficult than is the case with loudspeakers. This is primarily because meaningful measurements can only be made when the headphones are actually placed on a human head, or suitable substitute — and everyone’s head is slightly different. “Measuring headphones in a way that would reflect how we actually hear them is a multiple-step process that is not ‘plug and play’. Difficulty arises if one attempts to measure what ‘flat’ means on headphones, because to this day there is no real standard on how to measure headphone tonality and thus their voicing is much more nebulous, compared to speakers. It took us a few years to find out that there isn’t a way to reliably measure headphones with commercially available devices, which is why we developed our own methodology and tools.”

Drag Colich agrees, reporting that he too found it necessary to develop a proprietary measurement system in order to evaluate headphone response. “Microphones for speaker measurements are working in free space and can easily measure loudspeaker response correctly. Once the microphone is placed inside the headphone measuring rig, however, a flat response is not possible to achieve due to internal geometry and resonances. We can safely say that all headphone measuring rigs provide relative measurements, and most of them are drastically different!

“There is no universal standard for headphone measurements which will directly show the correct ‘flat’ curve. It is a rather complicated process of relative measurements using specific rigs and manual correction, interpreting the measurements. One can spend some time listening to the flat equalised speakers and try to correct headphone response to sound the same.”

Daniel Knighten is Vice President of Business Development at Audio Precision.Daniel Knighten is Vice President of Business Development at Audio Precision.Most testing procedures incorporate a fixture or stand that’s designed to mimic the acoustic properties of an ‘average’ skull, pinnae, ear canal and so on. Audio Precision, leading manufacturers of audio test equipment, recently introduced a new range of headphone measurement tools including the AECM206 Headphone Test Fixture. Yet the company’s Vice President of Business Development, Daniel Knighten, cautions that it might not even be possible to define a universal ‘flat’ headphone response, since there is no standard shape to the human head and ears.Audio Precision’s AECM206 Headphone Test Fixture is designed to mimic the acoustic properties of the average human ear.Audio Precision’s AECM206 Headphone Test Fixture is designed to mimic the acoustic properties of the average human ear.

“This is currently a subject of great debate within the industry,” he says. “Since the interaction of a given ear speaker with a given individual’s pinna, ear canal, ear drum and concha is unique, many people believe there is no ‘flat’ or uniformly desirable headphone response curve. On the other hand, Sean Olive with Harman, following in the footsteps of Floyd Toole, has done extensive listening tests and has arrived at a set of response curves for over-the-ear, on-ear, and in-ear headphones that he proposes correlate well with listener preference. Sean strongly believes that these curves should be the target response curves for headphones, in the same way that a flat line is the target on-axis response for a speaker in a free field.

“Whenever you have a target response curve for headphones, you can use software such as AP’s APx500 audio measurement software to refer the measured response to the target response by inverting the target curve and applying it as an EQ curve to the measured response. Using this method, the corrected (referenced) response of a headphone that perfectly matches the target response would be a flat line at 0dB.”

Heads You Win...

If our own human heads don’t exactly match the test fixtures used by headphone manufacturers, what sort of variations in sound can we expect? “It depends on the headphone, and how far one’s head/ear geometry falls from the bell curve,” responds Sonarworks’ Rūdolfs Putniņš. “Generally, loss of acoustic seal affects low frequencies, which is something to keep in mind if one wears headphones on top of glasses. Overall, large-earcup, over-ear, open-back headphones change their tonality very little due to placement changes, while other headphones change their tonal response much more.”

“The form factor of human ears and the HRTF (Head Related Transfer Function) do have an effect on the sound perception,” says Audio-Technica’s Hiroyasu Suzuki. “Horizontal earcup pressure is strongly related to the distance from the ears to the driver unit. The slight squashing of the ear pads through the pressure affects the quality of the seal and the HRTF, which is related to the ear-pad capacity. In terms of sound it will especially affect the mid to low end.”

AP’s Daniel Knighten likewise reports that differences in positioning, ear and skull geometry and earcup pressure can affect frequency response: “Each of these factors can have a significant impact. The most noteworthy of these is that in-ear monitors and closed-back headphones are generally dependent on a good seal to reproduce low-frequency sounds. This can be easily upset when a user’s pinna or head shape does not fit the model the device was made around. Furthermore, above 10kHz the length and shape of the ear canal makes a very large difference in response, so much so that all standards for ear simulators are currently undefined above 10kHz.”

Drag Colich, too, emphasises the importance of the ear pads in over-ear headphones. “All of this is with the assumption that the ear pads form a very good seal around the ears. If the pressure of the headband is low and headphones are positioned incorrectly, it is quite possible to have an opening around the ear pad, which can cause severe low-frequency extension loss and response anomalies in upper registers. Headphones would sound tinny and anaemic, with pretty poor imaging.

“In addition to the seal and positioning, the pinnae and ear canal affect the sound reaching the ear drum by way of their geometry. This is expected and is not very different from why measurements obtained from different measurement rigs differ. However, a big difference between a measurement rig and the human ear is the ability of the brain to compensate for one’s own ear geometry and enable us to perceive a flat response. Despite differences in ear geometry, with a proper seal and correct positioning, most people with normal hearing perceive sound similarly.”

Sweet Spot

Drag goes on to explain that correct headphone positioning is particularly important with the planar magnetic headphones that are Audeze’s trademark. “We use quite large diaphragms in our headphones. This has a lot of positive effects, like high sensitivity, very high SPL, very low distortion, excellent imaging, but also with some challenges. Plane sound waves generated by a large diaphragm have to be channelled into the ear canal with minimum sound degradation. If headphones are positioned in a relatively fixed position on the head, frequency response is quite consistent and repeatable. If headphones are moved more than a quarter of an inch in any direction, response fluctuates quite a lot above 1kHz. To get a good average response, we typically make five measurements within half an inch’s diameter of movement from the ideal centre position. If the relative movement of the headphones is larger than this recommended adjustment diameter, the response can be quite ragged, with large peaks and troughs above 1kHz which can colour the sound and skew the imaging.

“Normally, people tend to position the headphones in very predictable positions, and this can be quite repeatable. What is not predictable, though, is the absolute position of the headphones on customers’ heads, which may lead to non-optimal positioning and problems described above. This easily explains very different impressions of people using the same headphones.

“With smaller drivers, position sensitivity is less of a problem, but it always exists.”

An Open & Shut Case

Speaker designers have employed varied design philosophies in their attempts to create the ideal reference monitor. For example, some advocate the use of sealed cabinets, but others prefer to use ported cabinets or transmission lines; some employ ribbon tweeters while others use more traditional moving-coil designs, or electrostatic drivers; and some feel the advantages of dual-concentric drivers outweigh their disadvantages, while others disagree. There’s an equally wide range of different design approaches in the headphone world, and, just as with loudspeakers, they all have their pros and cons.

Most studio headphones are either over-ear (left) or on-ear (right) designs.Most studio headphones are either over-ear (left) or on-ear (right) designs.

Most high-end phones designed for critical listening are over-ear rather than on-ear or in-ear models. Within this category, some designers have sought to perfect the mainstream moving-coil design, while others have employed alternative technologies such as electrostatic and planar magnetic drivers. Perhaps the most fundamental distinction, however, is between closed- and open-back designs. In closed-back headphones, the driver radiates into a sealed chamber formed by the earcup on one side, the ear pads around the edge, and the listener’s head on the other side. Open-back headphones effectively remove the rear side of this chamber by making the earcup from an acoustically transparent mesh.

Popular wisdom has it that open-back designs are inherently better from a sound-quality point of view, with the advantages of closed-back models being mainly to do with isolation. Drag Colich says that this is certainly true of planar magnetic designs. “Open-back planar headphones allow sound waves generated by the rear side of the diaphragm to radiate to the outside without any real interactions and reflections from internal components of the earcups. In closed-back types, rear waves are trapped in the volume of the earcup and severely influence the direct sound radiation. The rear housing can create resonances and reflections, resulting in a very different frequency response compared to the open-back version — typically there is a large hole in the response in the low mid-range, with a worse phase response. Also, there are some reflections from the earcups coming back through the thin diaphragm, creating peaks in the upper mid-range and high frequencies, making sound sharper and more aggressive. Sound gets coloured and imaging starts collapsing. If the earcups are made of non-resonant materials and proper acoustical treatment is implemented, these problems can be drastically minimised, but they will never be eliminated.”

“Open-back headphones do not distort so easily compared to closed-back headphones,” agrees Audio-Technica’s Hiroyasu Suzuki. “As you might know, open-back models do not have the air mass on the rear side, which is a problem factor for the diaphragm.” However, he also points out that it is not a one-way street. “Regarding the dynamic aspect of sound quality, the closed-back model is much better.”

Rūdolfs Putniņš agrees that “if we look at the very best the industry can deliver then most of them would be open-back,” but he too cautions against assuming that open-back designs are more accurate in every respect. “It depends how one defines accuracy. Open-back headphones usually have better mid-range and treble response, as they’re less prone to resonances and reflections. However, due to their dipole radiation mechanics, they rapidly lose sensitivity at low frequencies. Closed-back headphones tend to have a better low-end response, at the expense of sensitivity at higher frequencies.”

These diagrams from Sonarworks’ laboratory measure the frequency response of the same two pairs of headphones in seven slightly different positions. As you can see, the response of the open-back model is relatively independent of position, but that of the closed-back phones varies significantly.These diagrams from Sonarworks’ laboratory measure the frequency response of the same two pairs of headphones in seven slightly different positions. As you can see, the response of the open-back model is relatively independent of position, but that of the closed-back phones varies significantly.Closed back over-ear responses.

Perfect Pairs

In an ideal world, the performance of the left and right drivers in a pair of headphones would be identical. In the real world, there are potentially at least two reasons why this might not be achievable. One is the matter of manufacturing tolerances. As Audio Precision’s Daniel Knighten explains, high-end headphone manufacturers go to some lengths to ensure that left and right drivers are matched, but this can be a costly goal to achieve. “A primary challenge resides in the incremental time investment to first measure, and then match, individual drivers (eg. after incoming inspection in the case of passive designs), with that time multiplied by their manufacturing volume. For high-end headphone producers, this investment in selection, or matching, is very common, but for mass-market consumer headphones the manufacturers usually tolerate as much as 3dB of imbalance.”

As you’d expect from a manufacturer of phones costing hundreds and even thousands, Audeze pay close attention to left-right matching. Drag Colich: “The final response variation between transducers is quite tight, but we still go further and match the overall response of transducers within ±0.5dB, creating matched pairs to be installed in headphones. Unfortunately, this still doesn’t guarantee that the final response will be as tightly matched like this. Ear pads crush when headphones are worn, and headphone position on the head or headband height adjustment can cause much larger response variations. Using the same headphones, for one person the sound can be almost perfect, and for someone else not as great.”

Hiroyasu Suzuki of Audio-Technica says that the company likewise apply a stringent standard for matching sensitivity across all their models: “There is a standard within Audio-Technica, which is a sensitivity difference between left and right of 2dB (measured at 500Hz and 1mW). This is, in engineering terms, not too difficult to achieve.”

However, he adds that matching sensitivity at a specific reference frequency and power input does not guarantee that headphones will be matched across the entire frequency response — especially as this can be affected by other factors in their design: “Concerning the frequency response, it is much more difficult, as it is not just dependent on the driver unit itself. One of the challenging parameters is the tolerance or difference of the whole earcup (housing) for left and right.”

House Points

This brings us to the other factor that can cause discrepancies between the left and right sides of a pair of headphones: intentional design differences between the two housings that have unintended consequences for the sound. The most common example is where there is a single cable entry point, located within the housing of one earcup. “We find that in general left and right matching is not a problem — but beware of headphones with asymmetrical enclosure designs,” cautions Rūdolfs Putniņš. “Some closed-back headphones with single-side cable entry have been plagued by driver imbalance: due to different acoustic loading, the drivers in these headphones can perform differently at the low end.”

The presence of a cable entry port on one earcup can create significant asymmetry in headphone frequency response, as in the unnamed phones measured here by Sonarworks.The presence of a cable entry port on one earcup can create significant asymmetry in headphone frequency response, as in the unnamed phones measured here by Sonarworks.

Overall, Rūdolfs is insistent that frequency-response anomalies are by far the most significant failings in headphone design: “We would say that 90 percent of the overall perception of sound is linear distortion, or, simply put, problems with the frequency response. Luckily, linear distortions can be corrected with signal processing. However, when that is done it still matters how good the overall headphone design is, because non-linear distortions are still going to be present. When frequency response is corrected, one will start noticing non-linear distortions like harmonic distortion, for instance. Unfortunately these cannot easily be corrected with signal processing.

“The source of both types of distortion can be attributed to the driver itself and enclosure design. A good example would be a poorly designed closed-back headphone, where the enclosure creates resonances as well as reflections that would interfere with the driver. Unfortunately, there is no easy way of telling a good design from bad by simple visual inspection, as most of the design nuances are inside the headphone.”

The Human Touch

Headphone design is an area of continuous ongoing development. Measurement systems are becoming ever more sophisticated, and advanced technology is being used to overcome practical design limitations. Digital equalisation is now widely used in studio monitor speakers to achieve the desired ‘flat’ response, and as headphone frequency response becomes better understood, the same techniques are being applied here too.

Now that music is mostly consumed using headphones, reference phones are becoming a vital tool for producers, engineers and mastering engineers — like Stu Hawkes of Metropolis Mastering.Now that music is mostly consumed using headphones, reference phones are becoming a vital tool for producers, engineers and mastering engineers — like Stu Hawkes of Metropolis Mastering.Photo: Chris Parkinson“The bottom line is: frequency response is the most important sound aspect of a headphone, and can be corrected with signal processing,” says Sonarworks’ Rūdolfs Putniņš. “There is no perfect headphone that would not benefit from frequency response correction, and only when the frequency response is reasonably corrected can there be any point of discussing other parameters.”

Audeze’s Drag Colich agrees: “While we design all our headphones to sound as close to our preferred response as possible, we believe all headphones can be further enhanced through DSP, breaking free from physical constraints. This is not very different from why loudspeakers can benefit from some EQ even after good room treatment. While headphones are more challenging to equalise, once a good frequency response is achieved, it can be enjoyed almost universally. In the case of loudspeakers, the corrective equalisation is only good for a specific room. We have made EQ profiles available for all our headphones through our Reveal plug-in.”

Despite progress in headphone measurement and correction, however, the designers and product specialists I spoke to emphasised that nothing can replace the judgements of trained human listeners in evaluating headphone sound quality. Audeze’s Drag Colich states: “Human intervention is required to get a proper transfer function and correct the headphone response to sound like ‘flat’, equalised speakers. In this case we need to trust ‘golden ears’ and accept their suggestions for necessary corrections.”

Or, as Rūdolfs Putniņš insists, “Only a listening test will tell you if headphones are tonally neutral.”

Which is why SOS has no plans to do away with human reviewers just yet...  

Impedance, Sensitivity & Loudness

Running off batteries, and with their output capped by EU legislation, the headphone amps in mobile phones can struggle to drive some headphones as loud as we’d like.Running off batteries, and with their output capped by EU legislation, the headphone amps in mobile phones can struggle to drive some headphones as loud as we’d like.Many of the devices on which we listen to music have relatively low-powered headphone amplifiers, and the output volume of portable music players (including mobile phones) bought within the European Union is also ‘capped’ in the interests of protecting listeners’ hearing health. In essence, the voltage swing that can be generated at the headphone output is limited to a level that will produce a sound pressure of 85dBA in an arbitrarily chosen ‘typical’ pair of phones; users are permitted to raise this threshold after acknowledging a warning, but only to a nominal 100dBA.

Now, few of us would want to listen to music at a continuous 100dBA for any length of time, but this situation gives rise to two problems. One is that different headphones can generate very different sound pressure levels for a given voltage input, and the other is that the need for headroom means the signals we listen to in music production are often much quieter than the mastered music that legislators assume we’re listening to. So, in a music-production context, the combination of a low-powered or capped headphone amplifier with the ‘wrong’ set of headphones can cause real difficulties. The situation is not helped by confusing terminology and a lack of consistency in the ways different manufacturers publish specifications for their headphones.

Two specifications are key here. One is the sensitivity or efficiency of the headphones, which records the sound pressure level generated in response to a fixed amount of electrical power dissipation (usually, though not always, 1mW produced by a 1kHz sine-wave signal). The other is their impedance, which (in conjunction with the source impedance of the headphone amplifier) dictates the voltage swing that is necessary to produce a given amount of power dissipation. Broadly speaking, the lower the impedance and the higher the sensitivity of the headphones, the louder they’ll be, at least with a typical low-power headphone amp such as those found in laptops and mobile phones.

Hiroyasu Suzuki of Audio-Technica also suggests that high impedance in a pair of moving-coil headphones is intrinsically problematic: “High impedance means basically that there are more winding turns on the voice coil, and the higher the number of winding turns on the voice coil, the greater the acoustic sound output for the same input level [ie. the greater the efficiency]. However, it also means we have a higher weight on the diaphragm itself, and to get this heavy diaphragm moving we need to have a high output level [from the amplifier]. Due to the inertial force required to move a heavy voice coil, the movement of the diaphragm will be complicated and the overall response will be degraded. To overcome this issue we can develop a diaphragm with higher stiffness, but this will be less reactive to long waves, and therefore the low-frequency response will deteriorate.

“There are also good and bad matches between headphones and playback systems. This has nothing to do with subjectivity and personal preference.”

A Stitch In Time

In loudspeaker design, achieving a linear frequency response is only half of the story. It’s also vital, and just as difficult, to ensure that a reference loudspeaker has a uniform and accurate time-domain response that is free from unwanted reflections.

You might expect this to be less of a challenge for headphone designers, since headphones use a single driver that is much smaller and lighter than loudspeaker drivers, with much less inertia and no phase-smearing crossover filters to worry about. Rūdolfs Putniņš of Sonarworks says that his company’s research supports this view. “Almost all [over-ear and on-ear] headphones use a single full-range driver, hence the time-domain performance is very good. We’ve tried to correct time-domain performance in headphones, but the results were hardly audible.”

Audeze’s Drag Colich, however, takes a different tack, arguing that: “Time-domain performance is very important for headphones, too — but for a different reason compared to loudspeakers.

“Typically, headphones are full-range and in close proximity to the ear canal, so time-domain problems introduced due to factors such as room acoustics are almost non-existent. Some reflections and resonances can be experienced within the ear pads and the ear, but these are much smaller problems than room response.

“Unlike loudspeakers, there is minimal crosstalk in headphones between the left and the right channels, and headphones do not capture the complete HRTFs. This results in a much smaller stereo image compared to loudspeakers. Then a fast-settling transient response (ie. time-domain response) becomes very important. A transient response that approaches a Dirac impulse (ie. a perfect impulse response) would result in increased transparency, and better depth and separation between instruments, helping deliver a holographic image. Our ultra-thin planar drivers deliver a fast transient response due to their very low inertia, and the transient response is also very well-controlled and fast-settling due to our magnetic circuit design that is optimised to exert a uniform force over the diaphragm.”