Mark Levinson No.30.5 Reference digital processor Page 2
There are, however, several problems with FIFO buffers. For example, if the FIFO holds only a small number of audio samples, the buffer will quickly overflow if the source data has a slightly faster clock frequency than the output clock frequency. Data are read faster in than they are out. Conversely, if the source clock is slower than the output clock, the buffer empties, leaving no data available for output. This creates glitches in the audio signal.
To prevent these problems, the buffer must be very big. Even a slight difference in frequency between the input and output clocks accumulates over the 74-minute playing time of a CD, causing the buffer to overflow or run empty.
The problem with a big buffer is the time lag between the input signal and the buffer output. You'd have to wait many seconds after pressing "Play" on the transport to hear musicthe samples must work their way through the buffer. Similarly, music would continue for several seconds after pressing "Stop," as the music continues to be output from the buffer memory. Moreover, a digital processor with a big buffer is completely unusable for decoding video soundtracks because of the synchronization difference between the video and audiothe buffer's time lag would put the soundtrack out of sync with the picture. Even a slight difference between sound and image is disturbing.
Given these fundamental facts about data buffers, Madrigal has devised an ingenious solution that provides the jitter-rejection benefits of a buffer with none of its problems. They call it the "Intelligent FIFO."
Because the Intelligent FIFO buffer is very smallit stores only 2048 audio samplesit creates only a very short delay time. To prevent the buffer from overflowing or running empty, the buffer's fullness and the incoming clock frequency are monitored many times per second. This information is used to shift the output-clock frequency to match the long-term average of the source clock, allowing the buffer to absorb short-term deviations from that average (ie, jitter).
The microprocessor monitoring the buffer's fullness generates a control signal that determines the VCXO's frequency, and thus how fast data are read out of the buffer. If the source clock is a little fast, the microprocessor tells the VCXO to speed up; if the source clock is slow, the VCXO's clock is slowed down.
Typically, the VCXO will stabilize at some frequency just a few seconds after a transport is connected; the microprocessor "learns" the incoming clock's average "speed," and sets the VCXO frequency accordingly. In reality, the VCXO frequency isn't set in a linear way, but with a more complex algorithm that analyzes the rate of change to prevent any abrupt shifts in clock frequency.
If the transport's clock frequency is too far out of spec, the FIFO is taken out of the circuit, and the clock recovered by the Crystal CS8412 is used. Similarly, the Crystal CS8412 is invoked when decoding 32kHz input data. The VCXOs are so expensive, and 32kHz so seldom used, that Madrigal omitted a separate VCXO for 32kHz input signals. When developing the Intelligent FIFO, Madrigal found that it works with even the most inaccurate transport they've testedinaccurate in terms of departure from a true 44.1kHz word clockthe Radio Shack CD-3400 portable player.
The result? All the benefits of a buffer and none of the drawbacks. Note that the clock driving the DACs is generated by the VCXO, and is therefore completely independent of the incoming clock. This means that transport and interface jitter are unable to affect the VCXO's clock, and thus degrade the sound. Moreover, the word-clock jitter is the VCXO's intrinsic jitter, which Madrigal claims is less than 20 picoseconds (footnote 4). Despite this ingenious solution to the problem of jitter in digital processors, Madrigal claims that the No.30.5 still sounds different with different transportsalthough the differences are much smaller than those heard through the No.30.
Interestingly, Madrigal says that roughly half the No.30.5's sonic improvements over the No.30 were rendered by the FIFO, the rest by the collective changes in signal routing, digital filter, digital power supply, and other refinements.
Although I very much enjoy the Sonic Frontiers SFD-2 processor, it was a great pleasure to hear the No.30 again (it's been ensconced in JA's listening room lately). The No.30 has some very appealing characteristics.
In my room, it didn't take hours of analytical listening to discover that the No.30.5 wasn't just a marginal refinement of the No.30, but a dramatic and wholesale improvement. In many ways, the No.30.5's musical presentation was very similar to that of the No.30it was obvious that they were cut from the same cloth. But in other aspects of the music, the No.30.5 was different in ways that had a large musical significance.
Although the No.30.5's tonal balance was nearly identical to that of the No.30, the No.30.5 sounded much more warm, rich, sweet, and full. This difference in overall presentation was primarily a combination of two huge improvements: bass authority and treble liquidity. The No.30.5 had a much-improved bass presentation, giving the music more power, weight, warmth, drive, and authority. It didn't have more bass, but it did have a much more articulate, dynamic, precise, detailed, solid, and powerful low-frequency presentation. Bass guitar and double-bass took on a completely different feel through the No.30.5I heard much more bounce, detail, power, weight, and dynamics. The instrument sounded so much more like a string being plucked, with attack, decay, and a newfound wealth of inner detail. In an objective sense, the difference could be considered marginal, but the musical significance was profound. Music had more life, vitality, palpability, and drive.
Footnote 4: It was difficult to measure the No.30.5's clock jitter and compare it to that of the No.30. I had both units apart on the test bench, but found that Madrigal's unique method of sending the word clock to the towers precluded an accurate measurement. The word clock is sent differentially, but the Meitner LIM detector has no differential input. This uncertainty of results makes it prudent to avoid characterizing the No.30.5's jitter performance.