Celestion SL600si loudspeaker & DLP600 digital equalizer 1992 part 2
Two families of digital filters are realizable with such a DSP chip: FIR and Infinite Impulse Response (IIR) types. An FIR filter, which in the real world of limited hardware resources is best suited to high-frequency processing, basically multiplies each input data sample by each of a set of coefficients (numbers) in turn, the output being the sum of the results of each operation. By contrast, though an IIR filter uses similar multiplying and adding operations on the audio data, it feeds some of the output back to its intermediate stages. The IIR type is theoretically more versatile, therefore, and can be used to process lower frequencies than an FIR type, but is both more complex to implement and design for an arbitrary response shape, and runs the risk of instability under some conditions. The FIR type, however, can produce complicated response shapes as easily as it can classical low-, band-, and high-pass filters.
As Celestion's engineers decided only to apply correction above 1kHz, leaving the speaker's low-frequency rolloff alone, they were able to implement the filter for the SL600Si as an intrinsically stable FIR type. They also decided to accept the fact that there would be some time delay in the process which could not be compensated for; the target response for the speaker would therefore be "linear-phase"; ie, the phase increasingly departs from zero with frequency in a linear manner.
The 81 coefficients for the DLP600's FIR filter are held in an EPROM and are loaded into the DSP engine on power-up. The EPROM also holds a set of coefficients representing no equalization. These are also loaded into the DSP; pressing the EQ In/Out button actually chooses between the two sets of filter coefficients, the DSP operating on the signal in both conditions.
As there are 81 coefficients operating at a sampling rate of 44.1kHz, this means that the impulse response of the DLP600's FIR filter can be 81 x 22µs long: 1.83ms, which should be enough to compensate for early reflections of the sound. To determine what these coefficients should be, the SL600Si's impulse response is measured at a distance of 2m. The axis chosen (although I didn't know this until I had performed all my measurements) was 10 degrees laterally off the listening axis. This impulse response is transformed to the frequency domain with the Fast Fourier Transform, and the inverse response above the minimum frequency of interest calculated. Performing an inverse FFT and time-windowing the result gives the impulse response of the desired filter, from which the coefficients can be derived. These are loaded into the DSP and the resultant speaker response auditioned, at which time the designer decides whether the result is better or not. If not, it's time to reconsider the choice of target response.
Because the Fourier Transform implicitly assumes a perfectly linear system, the loudspeaker must not add distortion in the frequency range of interest. In addition, it must not have deep notches in its response, nor must it significantly roll off below 22kHz. These last factors would demand large boosts in the correction EQ which will compromise the overall performance.
After processing by the ADSP-2105, the data are re-encoded with quite a large amount of high-speed TTL logic chips and connected to the DLP's output socket via a pulse transformer. There is therefore no electrical ground connection between the DLP600 and the user's D/A processor, even if the coaxial feed is used.
Celestion's original SL600 minimonitor of 1983 was a ground-breaking design in that its cabinet used 0.5"-thick Aerolam, an aluminum-honeycomb material extensively used in the aerospace industry. This is effectively inert in the midrange—see Stereophile, Vol.15 No.6, June 1992, pp.206-207—and results in a superbly transparent presentation from its 32mm, electroformed copper-dome tweeter and 6.5" woofer. A revised version, the SL600Si, was released in 1988, differing from the older model in having two sets of 4mm input sockets—two for the HF leg of the crossover and two for the bass/midrange—to allow bi-wiring or bi-amping, and a revised layout for second-order, 12dB/octave crossover with star-grounding.
I had a small problem with the first sample of the DLP600. While the VTL and Audio Alchemy processors had no trouble locking on to the DLP's output at both 44.1kHz and 48kHz rates, neither the CAL System 1 that TJN reviews elsewhere in this issue nor our Audio Precision System One Dual Domain would recognize its datastream. The CAL processor, for example, would just continually cycle through its three sampling-rate options without locking. I had no trouble with a second sample, so I assume that the first DLP600 was a trifle idiosyncratic.
I spent a week using the SL600Si without the DLP600 to refamiliarize myself with its sound. When I reviewed the SL600Si three years ago, I was struck by the clarity of the rather depressed high frequencies and the way in which the speaker managed to produce the illusion of good low-frequency extension. "It is astonishing to hear the clarity with which the Celestions can present kick drum, coupled with a suitable degree of weight," I noted in that review, adding that a good second-order, sealed-box loading seems inherently to have a cleaner mid-upper bass presentation than a typical reflex design. The '600Si also, in the vernacular, boogied, though the lower midrange sounded rather opaque and a little hard at very high playback levels. The transition from the upper bass to the lower midrange was also a little uneven, and male spoken voice sounded too chesty.