- Written by Dr. David A. Rich
- Published on 10 May 2012
The Phase Technology PC-3.5 Speakers On the Bench
I identified crossover frequencies at 950Hz between the woofer and midrange, and 4,500Hz between the midrange and tweeter. The crossover occurs when the acoustic output declines to 0.5 (-6dB). After the transition band, the slopes are approximately 24dB/octave. These measurements imply a fourth-order filter response (electrical plus acoustical). In a properly designed system with even-order slopes, the drives are in-phase at the crossover. Since both drivers are in-phase, the total response is the summation of the magnitude responses. Each driver supplies half the total acoustic output (0.5 + 0.5 = 1.0). The proprietary solid piston flat woofer cones enhance the in-phase property at the crossover to the midrange. Now you can appreciate why the company took the name "Phase Technology."
In-phase drivers at the crossover are critical. Figure 3 shows odd order crossovers are not in phase. Typically these would be first- and third-order Butterworth filters. The drivers are separated by 90 degrees at the crossover. The total response is not only the summation of the magnitude response; the phase difference must also be considered. The summation of the response is, in fact, 45 degrees out of phase with each driver.
Note the 0.7 amplitude of each signal from the drivers in figure 3, not one-half as with the even-order network. We need the extra amplitude, given the drivers' phase offset, to approach unity gain with both drivers active.
For most speakers, as with this one, the drivers are not coincident in the vertical plane. The amplitude response changes in the vertical direction as the arrival time of the wave fronts of the two drivers change and a phase offset emerges. With an even-order network, the result can only be a reduction in amplitude since each driver is producing half the acoustic output on axis to the drivers. With an odd-order network, it is possible to have an increased amplitude response with a change in the vertical offset (0.7 + 0.7 = 1.4). An advantage of the odd-order networks is they produce flat power response which is generally considered less important for speakers designed for home use.
The combination of the fast acoustic rolloff (24db/octave) of the crossover, which minimizes interaction of the drivers outside the crossover region, and the in-phase response at the crossover yields minimal changes in the vertical radiation pattern as the offset angle is increased. More details on how the crossover design affects the vertical radiation patterns can be found in an article I wrote for Sensible Sound (pages 12 – 19 of the Dec 2005 issue). The text is online in The Free Library (www.thefreelibrary.com) but the figures are only in the physical magazine.
Note that the speaker PC-3.5 is not linear phase. For high-order crossover networks, the phase of the far-field response varies with frequency. This effect is rarely audible in double-blind tests reported in AES conference papers above 200Hz. A speaker with poor radiation patterns, however, can be easily discerned by the ear. I was able to bring the speaker into linear phase using the Trinnov room EQ, which does phase and amplitude correction. I heard no audible difference with the PC-3.5s, which yield a near-perfect impulse response with phase equalization enabled (not shown).
Cascading a digital correction filter in front of a passive network is not the optimal way to use DSP. An active speaker with a DSP crossover provides much more flexibility in the optimization of the speakers performance in the time and frequency domain. This is why the next line up in the Phase Tech lineup is an active DSP based system.
Figure 4 shows the NRC (National Research Center of Canada) listening window (direct summed with +/- 15 degrees of the tweeter axis vertical and horizontal). My measurements with the Acoustisoft R plus D software are at 1 meter. Moving farther than 1 meter raises the frequency to which these quasi-anechoic measurements are accurate since room reflection corrupts the measurements in my test room. At 1 meter, the measurements are good to 300Hz, which approximates when the room characteristics start to dominate over the speaker characteristics at the listener's seat (Toole Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms, Focal Press 2008).
A discussion of the pioneering work of Dr. Floyd Toole at the NRC (long before his association at Harman) would be appropriate here, but the space is too limited, and it is best to read his book or for the deep dive, the original AES papers. The Cliff Note version: the NRC team developed a method to statistically quantify the sound of a speaker using a trained listening team and double-blind testing procedures. The measured response characteristics of the speakers used in the listening tests were next examined with the goal of identifying design criteria to yield superior sound quality in the controlled listening tests. Upon achieving the goal, the design criteria were published.
The sequence of events is not arbitrary. Intentionally, the NRC knew the proper order was to first have listeners qualify the attributes of a good sounding speaker, and then attempt to characterize why they sounded good. The date of appearance of the classic AES papers makes this clear:
Toole, F. E. "Subjective Measurement of Loudspeaker Sound Quality and Listener Preference" Journal of the AES Vol. 33, Feb 1985
Toole F. E. "Loudspeaker Measurements and Their Relationship to Listener Preference" Journal of the AES Vol. 34, Part 1, April 1986 April and Part 2, May 1986
Looking at the PC-3.5 it is unclear if Phase Technology engineers have been influenced by the NRC papers in their development work, or established their guidelines for quality-sounding speakers independently. A Phase Technology 3 way floor standing speaker introduced back in 1989 answers the question. The design and independent measurements from Stereo Review show a fully realized modern speaker. 1989 is too early for a production product to have been designed from the NRC AES papers.
Ken Hecht writes: "This was done on our own. We were using FFT analysis back then to analyze diffractive effects and radiation patterns".
I perform many of the tests advocated by NRC as important in identifying a good sounding speaker within a controlled environment. One significant omission from my test set is power response. I cannot measure a speaker's power response, which is also used to calculate directivity indexes with frequency.
Since I am limited to 300Hz, I set the gate time to 45msec for figure 4. The curve is high resolution at 0.1 octave smoothing to highlight significant (and audible) high Q resonances. (See Dr. Toole's text as a reference for this statement). Unfortunately, since this is my first speaker review for Secrets, I have no comparable to present on this website; nonetheless, this result is excellent as the total span of the curve covers no more than 21dB in the Y axis. The response aberrations at 900Hz are associated with the crossover and not a resonance. The midrange also is admirable, with only a small peak at 2.5kHz and perhaps something around 4kHz, although this deviation is close to the crossover and its exact origin is hard to pinpoint. Above 5kHz, in the range of tweeter the curve is as smooth as glass. These results are not magic, but the result of nearly 50 years of soft dome development at Phase Technology.
Figure 5 shows the NRC listening window as typically presented in some magazines and manufacturer literature One-third octave smoothing does not show the resonances clearly. This curve also has a span in the Y axis of 42dB, which is more typical of the range in which these curves are presented in some magazines and manufacturer literature. You can see with the loss of information content as smoothing and Y axis span are altered.
Figure 6 shows the horizontal radiation pattern. This is again a 1 meter near field measurement. Each curve represents a step of 15 degrees from the center of the speaker. These are at one-third octave smoothing to make the overlaid curve easier to read. I did look at these curves at higher resolution smoothing, but found nothing additional to report on driver performance. The curve shows measurements to a very wide 75 degrees. Examining wide angles is critical to account for early wall reflections in the room. Work at the NRC informs us a good sounding speaker (trained listeners in a double-blind test) needs to have monotonic off-axis response with an increasing downward slope as the angle increases. In practice, some non-monotonic behavior occurs around the crossovers as one moves from a narrowing dispersion pattern of the larger driver to the wider dispersion pattern of the smaller driver.
We can see only a small non-monotonic trajectory at 800Hz-1000Hz and a slightly more significant deviation at 1.7kHz– 3kHz at the widest angles (green, blue and pink curves). Again, since this is my first speaker review at Secrets, there are no comparable curves on this website; but this is a truly excellent result.
The NRC research tells us reflections from the ceiling and floor also contribute to the sound at the listening seat. These reflections are dominated by the vertical radiation patterns that can look worse than the horizontals. The crossover dominates and this dynamic exemplifies the company's name. All the drivers are in-phase at the crossover. Repeating from above, the response deviation around the crossover will be minimized and any response deviation should be negative.
Figure 7 shows the vertical radiation pattern. Each curve steps from -10 degrees below the tweeter at 1 meter to 15 degrees above the speaker. Moving back to 9 feet, the listener would be 2.5 feet above the tweeter axis to be the equivalent of 15 degrees above the tweeter at 1 meter. That is an average person standing with the speaker on a 24 inch stand (tweeter at 36 inches). -10 degrees is 1.5 feet down or 1.5 feet above the floor. That is the on-axis response of a child sitting on the floor, but floor bounce is the primary concern at this angle.
Figure 7 shows us virtually nothing at these angles. At negative angles the sight reduction in amplitude in the curves below 500Hz is a measurement artifact. For negative angles the measurement microphone moves closer to the floor and picks up a floor reflection. Ideally the speaker should be rotated instead of moving the microphone but this is not workable for me.
Dual 6.5 inch woofers are equivalent to 13 inches in the vertical direction, so they are somewhat directional above 500Hz and the curve returns to baseline as the 1.5 inch midrange takes over. The crossover leaves few visible artifacts in the curve. No peaking is seen in the curves as expected from the in-phase crossover. The same story holds at 4.5kHz. Extraordinary!
I have had many bad experiences with defective speakers or poorly designed speakers that caused me to waste time listening before identifying the defects in the measurements. The key measurement for finding a speaker's weakness is the in-room response. Measuring this first also permits one to identify the optimal placement in the room and, in the case of a bookshelf, the optimal stand height. I make no apologies now for measuring first since many PC-based room EQs enable the listener to make measurements to improve speaker placement.
Figure 8 shows an averaged measurement at 9ft back at the listening seat. This is a full range curve with a 500msec gate time. The measurement incorporates all the early reflections in the room. The span of the Y axis is again 21dB. This is a spatial average of nine points around 1.5 foot square. Again, it is a 0.1 octave smoothed curve.
Figure 9 shows the nine single shoot curves used to produce figure 8 One sees the single-shot curves are too corrupted by local effects to be usable at 0.1 or finer resolution.
Note how close the in-room curve (figure 8) is to the 1 meter listening window (figure 4). The most prominent deviation is the notch at the 950Hz crossover point. It is not clear why it is more prominent in this curve than the listening window (figure 4). I studied the audibility of this type of notch and found it relatively inaudible, in double blind tests, if the width is narrow.
Cochenour B., and Rich, D. A., "A Virtual Loudspeaker Model to Enable Real-Time Listening Tests to Examine the Audibility of High-Order Crossover Networks" AES 115th Convention, Oct 2003, Preprint 5908
Compared to the listening window (figure 4) the 2kHz peak is not apparent in figure 8, although the 4kHz is retained. Other than this effect, the figure depicts a resonance free result for both the midrange and tweeter. Please note this speaker is not voiced. The response is absolutely flat: the high-end roll off matches the near field, an immediate consequence of the wide dispersion of the drivers' range. The bump in the bottom end in figure 3 is the room gain, not a bump in the speaker's anechoic response. Different rooms, speaker placements, and seat placements will produce different response below 300Hz. The objective is a speaker with a flat anechoic low-end response so speaker placement can achieve a good low-end in-room response. Unfortunately, speakers with an intentional mid-base bumps move more quickly off the showroom floor.
Without the typical dip in the 1kHz – 3kHz range and a more extended top end roll off, the speaker sounds differently than a speaker with intentional voicing. The difference between the best possible passive cone-based speakers selling for under $10,000 and a speaker in six figures is skillful intentional voicing that fools people into thinking the speaker is worth the price of the average house. It appears the higher the price, the more complex the voicing with more peaks and dips intentionally introduced. When compared with the PC-3.5 at matched levels, the voicing is obvious.
Some room EQs allow the listener to put these voicings back in the speaker when desired. It does not work the other way. If the speaker's in-room response has been manipulated, it cannot be brought back to flat because a properly functioning EQ system measures only the in-room response (figure 8). If the direct anechoic response (figure 4) does not match the in-room response, the anechoic response gets messed up as the EQ adjusts the in room response. A room EQ also has no knowledge of the quality of the radiation patterns (figures 6 and 7) and can do nothing to correct sub-optimal ones.
Those who make $200,000 speakers might suggest otherwise, but a speaker's sound is the result of listening window response, radiation patterns, directivity indices, driver resonance, distortion, and in room response. I do not have the equipment to measure distortion to which others have access. A common source of distortion around the tweeter crossover arises as a result of the tweeter getting too much energy below the crossover. First-order crossovers are bad news here.
The speaker does have a low end limitation. After all, it is two 6.5 inch woofers in a bookshelf enclosure. I measured the speaker in a number of locations to examine the low frequency limit. It looks to be -6dB down at 35Hz in-room consistent with the near-field response of the woofer and port (not shown). Floor standing three-way speakers in this price range (not to be confused and, in my opinion, never to be purchased 2.5 way systems) will go slightly lower and produce less distortion in the bass at higher SPLs. This is the advantage of more cone and box area. Higher SPLs can be achieved with the PC-3.5 by adding a subwoofer. The ideal crossover is about 70Hz. Forget subwoofers if a good room EQ is not present in the chain (at this point that is the Anthem ARC outside of 5 figures); otherwise, there sound quality between 50Hz and 90Hz will be compromised.