Final System Measurements

All of the measurements shown in this section represent total end-to-end system response. This means that everything in the system is tested: the preamplifier, amplifiers, crossover networks, speakers, and (with one exception) the room itself. In addition to the listening test below, this section includes measurements on frequency response, square wave and transient response, cumulative spectral decay, and harmonic distortion. A good set of loudspeaker measurements can also be found on Lynn Olson's site.

The Ultimate Test - Listening With My Ears

I think the system sounds great. I hear things in CD's that I never heard before. The transients are terrific. The bass has a very satisfying punch. The speakers sound curiously neutral, and it did take a little while to adapt to this, but now I love it. Imaging is excellent. My earlier system uses Acoustat 1+1 electrostatic speakers, and on some classical music at moderate volume levels, I still might give a slight edge to the clarity of the electrostatics over my new system. For rock, jazz, and everything else, it's no contest.

Frequency Response

The frequency response was measured with the sound system installed in my music room. Commercial speaker manufacturers typically measure their speakers in an anechoic chamber. In most situations, room effects actually dominate the frequency response performance of a sound system, so the anechoic measurements can be misleading. The CLIO sound measurement system calibrated microphone was placed where my head usually is and carefully aligned (using the time-domain impulse response) to be exactly the same distance from each speaker; the measurement was made with both speakers on. These measurements are not time-gated. They show the full effect of the room. Virtually all of the major features in the frequency response are caused by the room, rather than the drivers or enclosures. This is the way I hear it, so this is the way I want to measure it. The "marketing department" and "engineering department" curves were measured under identical conditions. The marketing department curve looks a lot better because it:

Both curves use the standard format of a horizontal logarithmic frequency scale from 2 Hz to 20 kHz, and a vertical logarithmic system output response scale in dB. 0

Marketing department frequency response curve (24.5 kb)

Engineering department frequency response curve (28.2 kb)

I am quite happy with these results. The dominance of the effects of the room can be seen by comparing the engineering curve with the pseudo-anechoic measurement of the midrange and tweeter in the following section.

Back to the top

 Transient and Square Wave Response

My Holy Grail in this project was to achieve a good square wave response. (The driver cone motion required to produce a sound pressure square wave is described in another section). Square waves do not occur in real music, but if a system has a good square wave response, it will have a very good response to anything music can throw at it. A square wave is an acid test because the abrupt rise and fall (the jumps) severely test the transient and high-frequency response and the flat top (dead stop) severely tests for any tendency to ring. Ringing means the speaker continues to emit sound after the sound should have stopped. Ringing can arise in the electronics, particularly in crossover networks, or in the speaker structure. Both the system amplitude and phase response have to be good to produce a good square wave. Good phase response is particularly difficult to achieve, and most systems look God-awful when tested with a square wave, due to their non-linear phase response.

A few months after writing this I finally got around to actually testing if I could hear the difference between a good and bad square wave response. I can't. You can listen to sound demos yourself as described in the sound demo section. However it is still true that a good square wave response implies a good overall system response.

The room effects are removed for the square wave response data in this section. This is accomplished using the CLIO MLS measurement option. You can download Chapter 10 of the CLIO manual if you want further information, or see the section on maximum length sequences. The MLS technique provides the capability of slicing out one time segment of the measured response. Engineers call this time-gating. The time segment is selected to contain the sound that comes directly from the speaker, but to terminate before any reflections from the walls and ceiling arrive. Thus, a virtually reflection-free response is measured, similar to what one would measure in an anechoic chamber. (The actual measured time response is available two paragraphs below). One limitation of the MLS measurement is that the frequency response typically cuts off below 400 Hz or so. For this reason I only measured the midrange and tweeter, and shut off the woofer entirely. The midrange and tweeter follow a 1st order high-pass filter with a crossover frequency of 300 Hz. The measured response is compared to the square wave response of a "perfect" system, but one that similarly contains an ideal 1st order high-pass filter with the same crossover frequency of 300 Hz. The "perfect" system also contains an ideal low-pass filter with a crossover frequency of 20,000 Hz. The input is a 1000 Hz square wave. The horizontal axis is time, in milliseconds, and the vertical axis is the response measured by the calibrated microphone, in volts. The blue curve is the output of the "perfect" system with the ideal filters. If the filters weren't in the system, the top of the blue curve would be perfectly flat on the 1-volt line, and the bottom perfectly flat on the minus-1-volt line. The black curve is the measured response of my system. If you have never seen a measured square wave response of a loudspeaker, this might not look too good. But it looks terrific to me - I am real proud of it. It actually follows the ideal response quite closely, except for an overshoot where it goes above 2-volts, and below minus-2-volts.

Square wave response of my system and the perfect system (11.9 kb)

The CLIO system measures the impulse response, and converts it to amplitude and phase of the frequency response. The horizontal scale of the measured impulse response (17.9 kb) is time in milliseconds (ms), for the 1st 10 ms. The vertical scale is the response in volts, with an unspecified scale. It takes the sound almost 4 ms to travel a little over 4-feet from the speaker to the microphone. The microphone was moved this close to the speaker to extend the frequency range of the measurement. At 4 ms there is a nice sharp spike, followed by some ringing. At about 6 ms there is a reflection from a wall, and the yellow segment of the curve is selected to eliminate it and the following reflections. The red portion of the curve is not used. The derived amplitude response (25.3 kb) and phase response (25.4 kb) show generally very good behavior. The horizontal axis is frequency in kHz. The vertical axis for the amplitude is the dB scale on the left; the vertical axis for phase is the degree scale on the right. There is a glitch around 8 kHz that does not appear in the full system measured data, and my guess is that it is due to the microphone placement away from the "sweet spot." I am particularly proud of the smooth behavior through the crossover frequency of 3 kHz. I exported the frequency response into Matlab, and computed the square wave response from the frequency response. It is important to note that the phase measurement is a crucial part of the process, and the calculation would be meaningless without the phase data. Matlab was also used to compute the square wave response of the perfect system. Computing the square wave response from the frequency response, which is computed from the measured impulse response, which is not actually measured with an impulse input, may sound a little flaky. However, each step is in fact based on rock-solid mathematics. I recently have corresponded with Per Malare, who has directly measured the step response for his loudspeakers using CLIO. He kindly sent me the step signal he created, and so I can now directly measure the step response. First refer to the response of a perfect system [32.9 kb] with a first-order crossover at 300 Hz, and a low-pass 20 kHz filter. This ideal response is the blue curve in the figure. Without the 300 Hz filter the blue curve would be perfectly flat on top. In the same figure I have re-constructed the response using old measurements and the same theoretical process described above. This is shown as the black curve. Then compare this to the direct measurement using CLIO [62.4 kb]. The direct measurement is a little spikier, and indicates that I am getting a bit too much signal from the tweeter. However the time alignment is near-perfect.

Back to the top

Cumulative Spectral Decay (Waterfall Plot)

The CLIO system generates a cumulative spectral display plot (41.6 kb) of the system response. This measurement was made under the same conditions as the frequency response, using both loudspeakers. This plot includes 1/3 octave averaging, with the blessing of the engineering department, because it makes it much easier to see certain features of the response. The horizontal axis is frequency from 20 Hz to 20 kHz, and the vertical axis is response in dB, over a 40 dB range. This is essentially the same as the frequency response plot. The new axis that adds a third dimension is time in milliseconds, over a 312 millisecond range. When the sound system tries to play a note that suddenly stops, the sound does not suddenly stop. The main effect is that sound echoes around the room for 1/3 second or so. The loudspeaker enclosure walls, driver cone, etc. also continue to vibrate for a fraction of a second. The waterfall plot shows the relative frequency response of this lingering sound, as it decays. Room resonances persist the longest. The plot clearly shows room resonances at the lower frequencies. There is a particularly large ridge at 90 Hz, and another prominent one at 50 Hz. The 90 Hz peak is very evident in the engineering department frequency response curve as well. These resonances are discussed in the section on room acoustics. There are two resonances at 500 and 1000 Hz, which are somewhat obscured by the 1/3 octave averaging; they are pretty sharp in the un-averaged plot. My guess is that they are speaker cabinet resonances; I plan to investigate these in the future. Why do we care about any of this? For two reasons: (1) the change in the shape of the frequency response of the decaying sound adds coloration not present in the original music, and (2) resonances at the higher frequencies can also affect the transient response. An ideal waterfall plot would show a sequence of curves that are all smooth and flat across the frequency band, and that drop uniformly in amplitude to near zero in 1/3 second or so. A more rapid decay at higher frequencies may be preferable for good transient response. The waterfall plot is useful in identifying unwanted resonances, and sometimes something can be done to eliminate or reduce them. The response of my music room is actually pretty darn good, but I do plan on fooling around to see if I can reduce the 90 Hz resonance. I'll describe my success or failure in a future edition.

Back to the top

Harmonic Distortion

A measurement of the 2nd and 3rd harmonic distortion was made at a loud listening level. Only one speaker was used for this measurement. For two speakers, the sound level would increase 3-6 dB while maintaining the same level of distortion. (The 6 dB increase occurs at the lower frequencies where the sound pressure from one speaker adds to the other, and the 3 dB increase at higher frequencies where the combination is generally incoherent). The yellow curve is the fundamental tone. The red and pale blue dashed curves are the 2nd and 3rd harmonics, raised 30 dB compared to the fundamental. At a point where the harmonic curves cross the fundamental, the harmonic distortion is 3%.

Harmonic distortion curve (44.7 kb)

The microphone position for this measurement was again outside the "sweet spot," so the frequency response is a little more ragged than the curve in the frequency response section.

Back to the top

To the table of contents.