Page 2: The Yamaha NS10 Story

NS10 Technology & Measurements

Fast forward to 2001 (ironically, the year in which Yamaha discontinued the NS10), when studio and monitor designer Philip Newell, Julias Newell, and Southampton University's Dr Keith Holland presented a research paper to the Institute of Acoustics that constituted probably the first objective investigation of the NS10 phenomenon.

The Newells/Holland paper was based on acoustic measurements of 38 different nearfield monitors, carried out in the UK's premier research anechoic chamber at Southampton University. The acoustic measurements taken included frequency response, harmonic distortion and time-domain response (how quickly

a monitor starts and stops in response to an input). At the end of the exercise it's no exaggeration to say that one monitor stood out like the proverbial sore cliché: the NS10. While its frequency response wasn't particularly flat, and its low-frequency bandwidth was restricted in comparison to many others, in terms of time-domain and distortion performance it was outstanding.

During my work with Acoustic Energy on its recently launched AE22 nearfield monitor, we repeated some of Newell's and Holland's time-domain measurements of the NS10 and found similar results, and I've reproduced some curves that illustrate it. The measured data was generated by Phil Knight using the MLSSA acoustic measurement and analysis package, together with a calibrated B&K measurement microphone and custom-made power and microphone amplifiers. The relatively small measuring environment allowed for acoustic accuracy only down to around 150Hz — so, in Figure 4, reproduced later in this article, data below that frequency was generated through analysis of the NS10's low-frequency electro-acoustic parameters and calculating its response (see the explanation below). Before I get deeper into the acoustic measurements of the NS10, however, I'll first touch on one fundamental reason, as Newells and Holland pointed out, why its time-domain response is significantly better at low frequencies than most nearfield monitors: it's a closed box speaker.

Thanks to its two reactive elements — the mass of the cone/coil and the combined stiffness of the driver suspension and the air in the cabinet — a closed-box speaker displays second-order (12dB per octave) high-pass filter characteristics. A reflex-loaded speaker, on the other hand, thanks to the extra mass element of the slug of air in the port and the slug's own reaction against the air in the box, behaves as a fourth-order filter (24dB per octave). All reactive filters display a delay in their response to an input that increases with their complexity. In-phase movement of the air in the port of a reflex-loaded speaker must occur a half cycle (180-degree phase shift) after movement of the driver cone. This kind of time delay is known technically as group delay; it's actually the phase change with frequency expressed as time.

Figure 1: Calculated NS10M low-frequency amplitude response (black) and group delay (blue).

To illustrate a comparison of a closed box and a reflex speaker I've generated two low-frequency response simulation curves showing frequency response and group delay. The simulation in Figure 1 is based on the cabinet volume and driver parameters of the NS10. The NS10's limited low-frequency bandwidth (-3dB at 70Hz), slightly humped response and slow roll-off are clearly apparent. The group delay reaches a maximum of around 3.5ms at 70Hz.

Figure 2 shows what might have happened if Akira Nakamura had decided to aim for maximum low-frequency bandwidth (retaining the characteristic slightly humped shape) when he designed the NS10. The simulation in Figure 2 is again based on the NS10's 12-litre box volume. I've had to tweak the driver parameters slightly to make the system viable, but they are broadly similar to the genuine article. So if Nakamura had decided to go all out for LF bandwidth (as many contemporary nearfield monitor designers do) he could easily have reached -3dB at 57Hz — but look what would happen to the group delay. It increases to just under 11ms at 60Hz, which is around three times that of the closed-box NS10.

Figure 2: Calculated reflex-loaded NS10M low-frequency amplitude response (black) and group delay (blue).

Group delay is not some imaginary construct that helps acousticians feel important, it's real — and it means, for the reflex-loaded NS10 option, that a bass-guitar fundamental at 60Hz will arrive at the listening position around 9ms after the second harmonic at 120Hz. Put another way, and expressed as a distance, the low fundamentals of the bass guitar (and parts of the drum kit) will sound as if they are nearly four metres behind the rest of the band (you can insert your own bass player gag here). Low-frequency group delay doesn't only influence mix decisions: it also varies widely between speakers and, unlike low frequency level, which can be adjusted via EQ, once its influence on tracking or mix decision has been 'printed' to the mix, it can't be undone.

A reflex-loaded NS10, however, would not just have had significantly delayed low-frequency output. As well as delaying the arrival of low-frequency output, reflex loading also results in its output continuing significantly after the input signal has stopped (something that takes time to get moving generally takes time to stop), and in a multitude of dynamic compression, pitch-accuracy, noise and distortion mechanisms that simply do not occur in closed-box speakers. These again are effects that come without an 'undo' function once a mix is printed — so one of the best decisions Nakamura made when developing the NS10 was to make it a closed box.

Closed-box loading explains why the NS10's time-domain response is good at low frequencies but, as Newells and Holland discovered, the excellent performance also continues into the vital mid-range. Figure 3 shows a 'waterfall plot' of an NS10M from 200Hz up to 20kHz. These plots illustrate how quickly the output from a speaker dies away after a full-range signal stops suddenly. Imagine instantaneously switching off a source of pink noise. That's not quite how the waterfall plot is generated — one of this type is actually generated by taking sequential windowed snapshots of the speaker's impulse response and applying a Fourier transform to each — but it's a useful mental image.Figure 3: Measured NS10M waterfall plot.

Time runs in the waterfall plot from back to front, and a perfect speaker would display just one line (equivalent to its steady-state frequency response) at zero milliseconds. At the left-hand side of the plot there's a combination of the tail of the NS10's low-frequency fundamental resonance (even closed-box speakers don't stop immediately), and the unavoidable artifacts of a relatively small measurement space. Moving to the right, there are a couple of obvious discrete features — one at just under 2kHz and one at just under 3kHz. These are resonances in the NS10's bass/mid driver cone (or possibly its surround or dust-cap), and while they may look a little ugly they actually die down very quickly, and in subjective performance terms are relatively innocuous. The second resonance is at around 3kHz, which is actually above the NS10's 2kHz nominal crossover frequency, and illustrates that driver performance is important, even outside its nominal operational band. And speaking of the nominal operating band, the NS10's unusually low crossover frequency of 2kHz (made possible by a larger than typical tweeter, able to operate at relatively low frequencies) provides another clue to its time-domain performance. Any paper-cone bass/mid driver such as that used in the NS10 will become pretty badly behaved, in terms of resonance, above about 2kHz. Above the 3kHz feature in the waterfall plot is an area of general hash: this is the bass/mid cone in what's known as break-up mode, where its output is really just the result of one resonance after another. If the NS10's crossover frequency was an octave higher, at 4kHz, this cone break-up region would reflect in the time-domain performance and the waterfall plot would look very much worse. Moving further to the right, the NS10's tweeter performs very well and shows very little delayed output. Generally, the NS10's waterfall performance reveals a speaker that achieves -40dB within 6ms. Most speakers will take twice that long and many, especially those designed to maximise bandwidth, longer still. With just two small resonant features in the waterfall plot up to 3kHz, Nakamura could justifiably consider his design for the NS10 bass-mid driver a success.

I suspect that the success came from the NS10's only really unusual feature: its iconic white bass/mid driver cone. The cone wasn't just unusual because it was white, of course, but thanks to the way it was manufactured. The vast majority of paper-based speaker cones are pressed from pulp using a mould — partly because moulding gives the designer the ability to specify a curved cone-profile, to enable a degree of tuning of the driver's frequency response and resonant behaviour. A cone with a curved profile will generally become less rigid towards its outer edge, so as frequency increases its effective radiating area and output level reduces. Designers often use this technique to delay the onset of directionality in bass/mid drivers, so allowing a higher crossover frequency than would otherwise be possible

The curl-and-join construction of the NS10's paper cone (look closely and you can see the glued seam) is not typical of most speaker designs — and is a factor in the NS10's success in the studio.

The NS10's bass/mid cone was not pressed but 'curled-up' from flat paper sheet and then glued (look closely at the picture and you can see the join). The cone is straight-sided as a result, and the curl-and-join technique had two consequences for the performance of the NS10 bass/mid driver. First, the straight-sided form generally results in a driver with a rising frequency response, and second, while straight sides maximise rigidity, which would normally result in a cone with a strong 'bell-mode' resonance, the glued join acts as a damper (imagine a bell with a glued sawcut down the side: it won't ring much).

The characteristic rising response of a straight-sided cone is clearly apparent in Figure 4, which illustrates the NS10's frequency response measured at one metre on an axis halfway between the bass/mid unit and tweeter. Figure 4 is correctly calibrated so the NS10's sensitivity for a 2.83V (nominally 1W into 8Ω) input can be read from the vertical axis — somewhere between 87dB and 92dB. The NS10's relatively restricted low-frequency bandwidth, and the low-frequency roll-off slope of 12dB per octave, can also be seen. The 15kHz 'suck-out' in the response is most likely caused by diffraction from the tweeter's wire grille and, as it makes only a fleeting appearance in the waterfall plot is probably of little significance (it fades away in off-axis measurements too, which suggests its root cause is one of geometric symmetry).

Figure 4: Measured and calculated NS10M amplitude frequency response. Measured at 1m on axis. Curve calibrated to 2.83V input.