Musical Fidelity kW DM25 CD transport & D/A converter Measurements

Sidebar 3: Measurements

The Musical Fidelity kW DM25 DAC had a maximum output level from its solid-state output jacks of 2.236V at 1kHz, 1dB above the CD standard of 2V RMS. The output from the tubed output jacks was 0.175dB lower, at 2.2V. Both outputs preserved absolute polarity (ie, were noninverting). Despite the different circuit topologies, both tube and solid-state outputs had the same usefully low output impedance of 47 ohms in the midrange and treble, rising slightly to a still low 56 ohms at 20Hz. Error correction of the kW DM25 TR was good rather than great by current standards, the transport muting its output when trying to cope with 1.5mm gaps in the data spiral (assessed with the Pierre Verany test CD).

At a 44.1kHz sample rate, the kW DM25 DAC's frequency response was identical from both sets of outputs for CD data fed by the transport and for external data fed to its Aux input, and was down 0.3dB at 20kHz (fig.1, middle pair of traces). However, when I played a pre-emphasized CD in the transport, a broad, 1dB-deep depression appeared in the treble (fig.1, bottom traces). By contrast, no de-emphasis was applied when the kW DM25 DAC's Aux input was fed pre-emphasized data, the DAC not recognizing the appropriate flag in the S/PDIF datastream. This results in a treble boost maxing out at just above 8dB when the Aux input is fed data from pre-emphasized discs (not shown). Fortunately, this situation should almost never arise.

Fig.1 Musical Fidelity kW DM25, tubed output, CD frequency response at –12dBFS into 100k ohms, with de-emphasis (bottom below 1kHz) and without (middle); Aux input, 96kHz frequency response at –12dBFS (top). (Right channel dashed, 0.5dB/vertical div.)

The DAC's Aux input had no problem locking to data with 44.1kHz and 48kHz sample rates. However, it was very fussy when fed double-speed datastreams, and would not recognize 88.2kHz or 96kHz data from any of my computers, which made measuring its performance with these sample rates problematic. It would lock to high-sample-rate data when connected to a professional A/D converter, such as models from dCS and Benchmark that I had to hand, presumably because these sources have more stable clocks (footnote 1). The top traces in fig.1 show the DM25 DAC's frequency response when fed data from a Benchmark ADC-1 converter. The small rolloff at the top of the audioband with CD data continues with 96kHz data up to the sharp cutoff above 43kHz.

Because the DM25 DAC has dual data inputs for connecting to the DM25 TR transport, the left- and right-channel data sampled at 96kHz and carried on individual links, I had originally thought I could test the DAC with these inputs, driving it from a dCS 972 digital-to-digital converter to do the necessary upsampling and demultiplexing of CD data. However, Musical Fidelity uses the links between the DM25 transport and DAC with the XLR genders reversed; ie, it uses input jacks as outputs, and vice versa. I didn't have time to fabricate gender-changing links before this report went to press, unfortunately.

Fig.1 reveals excellent channel matching from both the solid-state and tube inputs. Channel separation was different between the two pairs of outputs, however. Crosstalk was buried in the noise floor below 1kHz from the solid-state jacks, though it rose at 6dB/octave above that frequency, due to the usual capacitive coupling between channels. Channel separation was a still-excellent 97dB at 20kHz from the solid-state outputs (fig.2, bottom traces). From the tube outputs, however, the channel separation was around 25dB lower across the band (fig.2, top traces). It is still good in absolute terms.

Fig.2 Musical Fidelity kW DM25, channel separation of solid-state outputs (bottom) and tubed outputs (top). (10dB/vertical div.)

DC offset was very low from both the DM25's analog outputs, at around 10–20µV. The kW DM25 was also very tolerant of system grounding. Usually, to get the lowest level of hum and noise, I have to experiment with the grounding in my lab when I test audio gear, particularly products that have only unbalanced connections. But whether or not I floated the grounds of my analyzer input, or the ground connection of the DM25 DAC's or DM25 TR's AC cords, the Musical Fidelity's noise floor remained consistently low, suggesting that these products have their internal ground paths well sorted.

The solid-state outputs had a low enough noise floor to be able to take good advantage of high-resolution data. The top pair of traces in fig.3 show a 1/3-octave spectral analysis of these outputs with the kW DM25 DAC fed CD data representing a dithered 1kHz tone at –90dBFS. The traces are free from harmonic distortion, and other than a slight power-supply–related bump in the right channel's spectrum at 120Hz, the spectrum is dominated by the recorded 16-bit dither noise. Increasing the word length to 24 bits drops the noise by 10dB, implying about 18-bit resolution, which is not quite as good as that of Musical Fidelity's inexpensive X DACV3. However, the DM25 DAC will still resolve a dithered 24-bit tone at –120dBFS (fig.3, bottom traces). This wasn't true for the tubed outputs, where the –120dBFS tone was buried beneath a slightly higher level of background noise (not shown) and the increase in word length dropped the noise floor by just 5dB. For CD playback, of course, this will be academic.

Fig.3 Musical Fidelity kW DM25, solid-state output, 1/3-octave spectrum, with noise and spuriae, of: dithered 1kHz tone at –90dBFS, 16-bit CD data (top), dithered 1kHz tone at –90dBFS, 24-bit Aux data (middle); dithered 1kHz tone at –120dBFS, 24-bit Aux data (bottom). (Right channel dashed.)

Linearity error with CD data (fig.4) was basically due to the effect of the recorded dither. With 24-bit data (not shown), the error was minimal down to –115dBFS. Fed CD data representing an undithered 1kHz tone at exactly –90.31dBFS, the DM25 DAC's solid-state outputs clearly distinguished between the three DC voltage levels of this signal (fig.5), though there is more noise apparent than with the X DAC, which offers textbook performance on this test (again, see the measurements). With 24-bit data, the waveform was a recognizable if noisy sinewave (fig.6). The tubed output, however, had enough noise present to obscure the high-resolution waveform (fig.7).

Fig.4 Musical Fidelity kW DM25, solid-state output, right-channel departure from linearity, 16-bit CD data (2dB/vertical div.).

Fig.5 Musical Fidelity kW DM25, solid-state output, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit CD data.

Fig.6 Musical Fidelity kW DM25, solid-state output, waveform of undithered 1kHz sinewave at –90.31dBFS, 24-bit Aux data.

Fig.7 Musical Fidelity kW DM25, tubed output, waveform of undithered 1kHz sinewave at –90.31dBFS, 24-bit Aux data.

Musical Fidelity products have traditionally had very low levels of distortion, and the DM25 was no exception, from both its tubed and solid-state outputs. Fig.8 shows an FFT-derived analysis of the tubed outputs while the DAC decoded 24-bit data representing a maximum-level 1kHz tone. The measured THD (true RMS sum of the harmonics) was just 0.0009% (left) and 0.001% (right). The slightly higher level in the right channel was due to the second harmonic reaching –101dB rather than –106dB! THD was even lower from the solid-state outputs, with the third harmonic the highest in level at –104dB (left) and –109dB (right). With 24-bit data representing a dithered tone at –90dBFS (fig.9), the distortion harmonics present in the tubed outputs are almost buried within the noise. The slight drop in noise from the solid-state outputs (not shown) reveals that the second harmonic is the highest in level with this low-level signal, at –35dB, which is excellent performance. Even from the tubed outputs, intermodulation was vanishingly low in level (fig.10).

Fig.8 Musical Fidelity kW DM25, spectrum of 1kHz sinewave at 0dBFS into 4k ohms, DC–10kHz (24-bit Aux data, linear frequency scale).

Fig.9 Musical Fidelity kW DM25, spectrum of 1kHz sinewave at –90dBFS into 4k ohms, DC–10kHz (24-bit Aux data, linear frequency scale).

Fig.10 Musical Fidelity kW DM25, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 4k ohms, DC–24kHz (24-bit Aux data, linear frequency scale).

Finally, I examined the DM25 DAC's rejection of word-clock jitter using the Miller Jitter Analyzer. This performs a high-resolution spectral analysis of the device under test's analog output while it decodes 16-bit undithered data representing a high-level tone at exactly one quarter the sample rate, over which has been laid the LSB toggling at a low frequency that is also an exact integer fraction of the sample rate. The results for the DM25, with the transport as the source, are shown as the solid black trace in fig.11. The measured jitter level is 394 picoseconds peak–peak—low, but not as low as the best products I have measured. Data-related sidebands (red numeric markers) are at the residual level in the test signal, which is good. However, as well as some high-frequency sidebands of unknown origin (blue "17," purple "15" and "25"), a number of low-frequency sidebands can be seen (purple numeric markers). In addition, the central spike in this graph, which represents the 11.025kHz tone, is broadened at its base, implying the presence of some random LF jitter components. This kind of behavior generally correlates with a "big"-sounding soundstage, I understand.

The grayed-out trace in fig.11 is a repeat of the spectral analysis with a PS Audio Lambda transport feeding the DM25 DAC's Aux input. The measured jitter level has dropped by a third, to 297ps p–p, and while some LF sidebands can still be seen, these are generally lower in level and, perhaps more important, the central spike is much more narrow at its base. It is important not to read too much into this behavior, but I do wonder if the drama that AD noted regarding the DM25 combination's presentation is the result of the slight spectral spreading, which I imagine is a byproduct of the double pass through the upsampling process (once in the transport and once in the DAC).

Fig.11 Musical Fidelity kW DM25, high-resolution jitter spectrum of analog output signal when driven by kW DM25 TR transport (11.025kHz at –6dBFS sampled at 44.1kHz with LSB toggled at 229Hz, 16-bit CD data). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. Grayed-out trace is with the Aux input fed by a PS Audio Lambda CD transport.

Overall, the kW DM25 offers superb measured performance and more-than-competent audio engineering. I share AD's very positive impression of the DM25's sound quality, having used the DAC for two nights of musical presentations at Colorado dealer Listen-Up in March. I am starting to think of Musical Fidelity as the "Quad" of the 21st century, in that regardless of how they sound, the company's products are universally well-engineered and solidly made.—John Atkinson

Footnote 1: The DM25 DAC could be made more tolerant of data sources with sloppy timing by widening its input PLL (Phase-Locked Loop) acceptance window. However, this would compromise its rejection of word-clock jitter.—John Atkinson
Musical Fidelity
US distributor: Signal Path International
1824 30th Avenue NE, Suite 1
Bellevue, WA 98005
(704) 391-9337