We are sometimes told that speaker development has stopped. We disagree. Yes: an awful lot was developed in the 1950s and 1960s, but development has not stopped. If we look at the last decade, we see plenty of interesting things. Think of new cone-materials, new magnet technology or interesting casting techniques for cabinets. Not to mention the active models… Manufacturers really do invest in R&D.
We also see the results of these developments in our measurements. Speakers are really improving. We see less and less distortion. Even in price classes where that wasn’t possible before. And that is a very positive development, of course. But let’s see how those measurements work.
Anechoic room
Acoustics is the biggest enemy while measuring loudspeakers, because we only want to measure the speaker; not the room in which it is placed. After all, the room itself also contains sound properties. Think of certain resonance frequencies. These properties influence the measurement, making it unreliable.
For this reason, major brands and research centres have started building so-called anechoic rooms. These are – often large – rooms that absorb all energy. So there is no reverberation at all. In doing so, they often isolate almost 100%. Including resonances from other sources. Think of traffic.
You will understand that a serious anechoic room is a substantial investment. It often involves sums of six zeros. After all, a lot of space is needed to do full range measurements and that entire room must also contain a lot of mass and be completely free of reflections/reverberation. A difficult and expensive approach.
It is possible to make them smaller, but then it is often not possible to measure full range. Incidentally, this is not always necessary. That is why you will also see smaller anechoic rooms.
Different approach
Those who cannot make a anechoic chamber and still want to make decent measurements have two options: a Klippel system that very ingeniously makes countless ‘close mic’ measurements and aggregates them, or measuring with a so-called ‘time window’ or ‘gate’.
Now, a complete Klippel measurement system also costs more than 100,000 euros if you select some interesting options. So for many users that one is also out of the question. That leaves a few systems. We opted for the Clio 12 with QC software. That comes to about 3500 Euros excluding microphones, cables and a computer. We use a DPA 4091 high spl microphone – it is dead straight from 20 Hz – 20 kHz – and have a central PC that we use for all measurements; very convenient.
Windowing
So how is it possible to take measurements in a non-anechoic space? That’s where smart software comes in. We can specify a so-called ‘window’ – also known as a ‘gate’ – in the software. This ‘window’ indicates the time period within which the software should measure. So we need to look at what distance the first reflections occur. In many cases, this is the floor or ceiling.
In our measurements, we usually measure at a distance of 1 metre from the speaker and keep this distance as the limit. If we convert that, we arrive at a window of 3ms. (Sound velocity is 343 m/s, which is about 3ms for 1 metre). Here you can find a handy tool for who’s a bit lazy :-).
Incidentally, in the Clio 12 we have recently bought (June 2023), we can do it a bit more handily by looking at an impulse and indicating there where the window should open and where it should close again. (In other words, where the measurement starts and stops). Many times we are then also around 3ms.
The advantage of this method of measurement is that we are no longer bothered by acoustics. After all: we exclude those. The disadvantage is that we cannot measure low frequency response, as the sine of a lower tone is too long and therefore too long for the time window of 3ms. So anything below 300 Hz is effectively truncated.
Now, though, we recently – June 2023 – have implemented a solution for that too: the Prism dScope turns out to contain a very good script for measuring phase and frequency response of loudspeakers. That script can calculate what the bass response is. And that turns out to be awfully accurate.
What do we measure in speakers?
Now we have switched to another system since June 2023 and are yet to develop routine within this system. However, by default we measure the following things with a loudspeaker:
- Frequency response (1 metre from the front, 1 metre off axis)
- Distortion (THD with harmonics, 1 metre from the front)
- Waterfall (resonances)
- Impedance (with the Sourcetronic LCR)
- Impulse response
- Group delay (since the CLIO 12)
- Phase behaviour
These measurements give a good overview of the speaker’s behaviour. Now, we usually find frequency response less interesting than distortion. This is because distortion says much more about quality than response; that is more the tuning of the speaker. And thus the manufacturer’s choice.
In the gallery above, you can see four screenshots of measurements. The first is a waterfall. This shows where the resonances are in the speaker. The shorter the better, as then the cabinet and unit do not resonate too much.
The second picture shows the frequency response at the top and an impulse response below. The frequency response has a window, so it drops off in the low end. You can see the window / gate below for the impulse response: orange is off, yellow is active. So the window runs from 2 to 6ms (and is 4ms in this case).
The reason a window causes low roll off (is actually not measured along) is that a frequency has a wavelength and therefore needs time. After all: time, distance and speed are linked. The time required by a frequency can be calculated by 1/T. In this case, 1/4ms. That comes down to 250 Hz. So within a window of 4ms, no wavelength of a 100 Hz tone will fit, to give just one example.
With the Prism, thanks to some clever algorithms, we can still measure what the bass response would be. That’s why we recently supplemented the measurements with a measurement via the Prism dScope. You can see below that we measure response down to 100 Hz with ease. This is a very small speaker, so -6 dB at 70 Hz is certainly realistic.
Finally, we have measurements of phase behaviour (acoustic phase) as well as impedance and electrical phase. The acoustic phase mainly indicates how well the units work together in terms of phase (time). Are the treble, mid and bass units in phase with each other? This is surprisingly difficult to get right, because both the filter and the placement of the driver in depth have an influence. In many speakers, you will see that the tweeter is slightly deeper than the mid-driver and woofer, for example. This is to correct the acoustic phase. Very luxurious high-end speakers allow it to be optimised to the room and listening position itself. Think of models from Wilson or Estelon.
The Prism can also measure acoustic phase. That is the red line in the measurement above. Ideally it should be straight, but that is practically impossible. Thereby, the phase behaviour will always change in a room due to the acoustics of the room (reflections, absorption, distance differences between speakers, etc). However, it is desirable to have a system that is relatively phase-correct to begin with.
Impedance and electrical phase
The impedance and electrical phase say a lot about the controllability of the speaker. How much current is needed to drive the speaker? Some brands are notorious for having very deep dips in impedance. Something that causes an amplifier to have to work very hard, since low impedance results in a lot of current having to be supplied. Not every amplifier likes that.
The chart above is of the Graham Audio LS8-1. This is actually a very friendly speaker with no mean dips. The blue line is the impedance. The orange line is the electrical phase.
The electrical phase shows whether voltage (volts) and current (amps) are in phase with each other, or whether they are ahead (positive) or behind (negative).
A positive phase shift of 90 degrees means that the current is a quarter wavelength behind the voltage. This is the result of an inductive circuit. With a high capacitance, it is the opposite. Then we are dealing with a negative phase and the current runs ahead of the voltage.
Now a quarter wavelength (360 / 4 = 90) doesn’t seem like much, but the result is that the current is at maximum and the voltage is at the zero point. See the graph above. The current – yellow – is maximum and the voltage – black – is at the 0 line. That’s basically a short circuit for the amplifier. And therefore very inconvenient, if not impossible. Fortunately, we don’t see this in practice.
A phase shift of 45 degrees is already pretty intense for an amplifier, since a lot of power is delivered and very little – a quarter approximately – is processed by the speaker. In practice, a lot is lost in heat. A shame. And it means that an amplifier may have to do its best to remain stable.
You will understand that a combination of a low impedance AND a substantial shift in phase can be a very challenging load for an amplifier. Above you can see an example of a Bowers & Wilkins 702 that can pose a considerable challenge to an amplifier and, next to it, a TAD E2 that is actually very friendly.
Concluding
So measuring speakers can be done in a variety of ways. A anechoic chamber, or a system like the Klippl is optimal, but very expensive. A system like the Audiomatica Clio 12 makes the threshold a lot lower, but you need to know what you are doing as well as what you are looking at.
The most important variable is the time window. This has the consequence that the bass response and information is therefore not measurable, because it is eliminated by the time-window.