Thiel CS3.6 loudspeaker Measurements
The following measurements were performed by JA. I saw the results and discussed them with him only after I'd finished writing my listening impressions.
The CS3.6's impedance magnitude and phase plot (fig.1) reveals a very low impedance value. The loudspeaker is under 3 ohms through most of the range, dropping to a minimum of 2.3 ohms at 3.6kHz (the cursor position). The low impedance value explains the CS3.6's need for the iron-fisted Mark Levinson No.23.5 to provide control in the bass; the CS3.6 would appear to be current-hungry. There is, however, a consistency to the impedance curve that makes the CS3.6 look much like a resistor to the amplifier, rather than an impedance that varies greatly with frequency. The phase angle—represented by the dashed line—is quite benign. The passive radiator tuning can be seen as the valley centered at 29Hz. Both the bass peaks are low in amplitude.
Fig.1 Thiel CS3.6, electrical impedance (solid) and phase (dashed) (2 ohms/vertical div.).
Fig.2 is the CS3.6's impulse response on the midrange axis with the grille in place. A B&K 4006 microphone, calibrated to be flat on-axis up to 30kHz, fed DRA Labs' MLSSA system via an EAR preamp. The symmetrical impulse shape is typical of first-order crossovers. There is a minimum of overshoot and ringing, seen in the impulse's sudden decay and low amplitude about the time axis after the impulse. Speaker systems with a fourth-order crossover have a characteristic hump after the impulse, rather than the clean decay seen in fig.2.
Fig.2 Thiel CS3.6, impulse response on tweeter axis at 45" (5ms time window, 30kHz bandwidth).
The CS3.6's step response—which ideally should look similar to a right triangle—is shown in fig.3. This was measured on the tweeter's axis, the most realistic listening height, but at a microphone distance of 45", which Thiel does point out is too close for this design (footnote 1). The step response is not quite time-coherent, on this axis, as can be seen by the bump beginning at the 4ms vertical division. This is the woofer lagging behind the midrange and tweeter. A much better step response was obtained when measured at the passive radiator (fig.4), the loudspeaker then becoming time-coherent due to the woofer output moving forward in time; but 24" is an unrealistically low listening axis. I tried listening close to this axis by sitting on the floor, but the tonal balance was sufficiently skewed to render meaningless any improvement in the time response. Nevertheless, a slightly lower than normal listening axis appears to be optimum.
Fig.3 Thiel CS3.6, step response on tweeter axis at 45" (5ms time window, 30kHz bandwidth).
Fig.4 Thiel CS3.6, step response on passive radiator axis at 45" (5ms time window, 30kHz bandwidth).
Footnote 1: The problem with measuring speakers in real, as opposed to anechoic, rooms with a technique like MLSSA or TDS is how to push the reflections of the sound from the room boundaries far enough back in time that the anechoic section of the speaker's impulse response is long enough to give good resolution down into the midrange.
I place the speaker under test on a stand so as to place its tweeter about halfway between the listening room's floor and ceiling and midway between the sidewalls. I then construct an acoustic "black hole" on the floor between the speaker and microphone out of graded layers of acoustic foam and fiberglass damping material. This kills the floor reflections from tweeter, midrange, and woofer, meaning that the primary reflection in the MLSSA's calculated impulse response is that of the tweeter from the ceiling, which arrives about 4ms after the direct sound. This 4ms anechoic time window results in a measurement valid down to 300Hz or so. However, this is with my standardized 45" mike distance. Were I to place the microphone farther away, which is what is needed with speakers with first-order crossover slopes and widely spaced drivers, like the Thiel, the reflection from the ceiling would arrive much sooner after the direct sound and the measurement resolution would suffer as a result.
It's a beautiful Zen situation: the best room in which to use quasi-anechoic measurement techniques like MLSSA is, in fact, an anechoic chamber; in which case, you can perform a real anechoic measurement.—John Atkinson