When Jason Victor Serinus visited the Bluebird Audio room on the Venetian's 35th floor, he mentioned that Chord was demonstrating its Blu Mk.2 CD transport ($11,788) along with the Dave DAC that I reviewed and was impressed by last June. I chatted with Chord's digital guru Robert Watts (above in photo) about the new transport and he mentioned that it incorporated his latest WTA (Watts Transient Aligned) digital filter with a million taps! (The more taps there are, the closer a DAC can reproduce the timing information in the reconstructed analog signal—see my DAVE review for why Robert feels why this should be so.) I was puzzled, as a digital reconstruction filter belongs in a DAC, not a transport.
"Think about how a digital filter works," said Robert. Okay, I will. Simply put, a digital filter comprises a series array of multiplication units or taps, each separated in time by a single sample delay, and with a summing unit fed the outputs of every multiplier to create the filtered output data fed to the DAC. As the first word is fed to the first multiplier, it is multiplied by the coefficient stored for that multiplier, and is then fed to the second multiplier while the next data word is fed to the first one. And so on until the data exits the final multiplier. (One of the first articles I wrote for Stereophile, in September 1986, discussed this process; if you plot this series of coefficients against time, you get the familiar filter impulse response that I publish in our reviews of digital processors.)
As the summer adds the outputs of every multiplier for every clock cycle, the more coefficients there are, the harder it has to work. The mathematical operations are most commonly performed within the DAC chip but some companies, like Chord, use an FPGA (Field-Programmable Gate Array) as this allows the DSP engineer almost unlimited flexibility in creating filters. Unless you want to keep increasing the number of filter coefficients, which Watts finds always gives a sonic improvement. The DAVE, for example, uses a Xilinx FPGA that allows a filter with 164,000 coefficients to be programmed. But even 164,000 taps was not enough for Watts, and the availability of a new Xilinx FPGA, the X7A200T, which has no fewer than 740 DSP cores, 215,360 logic cells, and 16MB of memory, allowed him to increase the number of taps to one million—actually 1,015,808!
So why is the filter in the Blu Mk.2 rather than the DAVE, I asked. "The FPGA draws up to 10A of current," Watts replied, "and DAVE's power supply just can't supply that much current without compromising the noise floor." The Blu can, he explained, so it made sense to put the filter in the Mk.2 version of the transport and feed the filtered data stream to DAVE.
Now I understand!
