Zanden 5000 Mk.IV/Signature D/A converter & 2000 Premium CD transport Measurements
First, it must be noted that Zanden doesn't have enough dealers to qualify for a full review in Stereophile. At the time the review was organized I was under the impression that the Zanden gear was available from five US dealers, the minimum we require. But as we went to press, we were informed that Zanden has just three US dealers. It wasn't possible that late in the production to abort the review. My apologies to the magazine's readers for this inadvertent exception to our rules.
For almost all of my tests, I drove the 5000 Signature D/A processor from the 2000P CD transport using the supplied I2S datalink, though I did perform some tests using audio data sourced from my PC with an AES/EBU link from an RME soundcard with a digital output. (Peculiarly, though the 2000P has a word-clock output on a BNC jack, the 5000S doesn't have a separate word-clock input and there is no mention of using the word-clock connection in the manual for either unit.) But whether driven by the Zanden 2000P transport or by an external digital source, the Zanden 5000S put out a maximum of 1.6V at 1kHz, almost 2dB below the CD standard's 2V RMS. This was the left channel; the right channel was 0.5dB higher in level. The output impedance was a high 2.5k ohms at 1kHz, this dropping a little to 2k ohms at 20kHz, and rising slightly to 2.6k ohms at 20Hz. As the owner's manual recommends, loads below 10k ohms should be avoided.
The Zanden DAC inverted signal polarity with the front-panel switch set so that the green LED was illuminated, and preserved absolute polarity with it illuminated red—the opposite of what I would have expected. Assessed with the Pierre Verany Test CD, the 2000P's error correction was superb, the transport not flagging uncorrectable errors until the gaps in the data spiral were 2mm long.
The combination's frequency response was disappointing, with a noticeable rolloff of low frequencies that reached –3dB at 65Hz and –9dB at 24Hz. The top octave also rolled off prematurely (fig.1, top traces). With a pre-emphasized CD, the treble rolled off even lower in frequency (fig.1, bottom traces). This is simply poor engineering. Crosstalk was buried under the noise floor below 1kHz (not shown), but channel separation decreased to 72dB at 10kHz, due to the usual capacitive coupling. The noise floor was higher than I usually find with 16-bit DACs, as can be seen from the spectral analysis of the 5000S's analog output while it decoded data representing a dithered 1kHz tone at –90dBFS (fig.2). You can also see relatively high AC-supply components at 120Hz and 240Hz in both channels and at 60Hz in the left channel. No matter how I arranged the grounding between the 2000P, the 5000S, and my Audio Precision System One, I couldn't eliminate these spuriae.
Fig.1 Zanden 2000P-5000S, frequency response at –12dBFS into 100k ohms (top), with de-emphasis (bottom). (Right channel dashed, 1dB/vertical div.)
Fig.2 Zanden 2000P-5000S, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS (right channel dashed).
Note that the peak representing the 1kHz, –90dBFS tone rises above the –90dBFS line in fig.2, suggesting the presence of some positive linearity error. This was confirmed by fig.3, which plots the level of a dithered 16-bit/500Hz tone as it drops in level from –60dBFS to –120dBFS. Linearity error is low down to –85dBFS, but then increases to a maximum of +7dB at –100dBFS; ie, a tone at this level actually reproduces at –93dB. The error is then increasingly dominated by noise. I haven't seen a DAC behave this poorly for many years; in fact, I remember Philips' TDA1541 as being rather better in this respect.
Fig.3 Zanden 2000P-5000S, right-channel departure from linearity, 16-bit CD data (2dB/vertical div.).
With its poor linearity and high noise, it's not surprising that the 5000S didn't fare well at the task of reproducing an undithered 1kHz tone at exactly –90.31dBFS (fig.4). Instead of the three clearly defined voltage levels you can see in the graph showing how the Musical Fidelity X-DAC v3 did with this test (fig.4), the Zanden produces a dirty, indistinct waveform, the AC noise introducing an overall slope to the oscilloscope trace. Feeding the D/A processor a 24-bit version of this signal via its AES/EBU input produced an identical waveform, showing that the 5000S truncates incoming hi-rez data to 16 bits. (To be fair, Zanden doesn't claim otherwise.)
Fig.4 Zanden 2000P-5000S, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit data.
Measurement of the Zanden combination's distortion was made problematic by the high level of ultrasonic energy present in its output, this due to the images of the audioband signal present at either side of the multiples of the sampling frequency. I got consistent results with two different measuring systems, however, so I believe the results shown are representative of the products' intrinsic behavior, and not to any interaction between this energy and the analyzer. Fig.5 shows an FFT-derived spectrum of the 5000S's output while it drove a full-scale 1kHz tone into 8k ohms, this a little lower than the load recommended by Zanden. Even so, the THD (actual sum of the harmonics) is respectably low, at 0.03% left channel and 0.066% right, and the subjectively benign second harmonic is the highest in level in both channels. Dropping the signal level to –90dBFS (the same test tone used to generate the 1/3-octave smoothed graph in fig.2) gave the spectral analysis shown in fig.6: the third harmonic is now the highest in level, and the distortion is higher in level than I see with the best DACs. (The TDA1541 chip clearly shows its age on this test.)
Fig.5 Zanden 2000P-5000S, spectrum of 1kHz sinewave at 0dBFS into 8k ohms (linear frequency scale).
Fig.6 Zanden 2000P-5000S, spectrum of 1kHz sinewave at –90dBFS into 8k ohms (linear frequency scale).
I got a surprise when I repeated these tests using a full-scale low-frequency tone: Even though the test load was now a benign 100k ohms, the FFT spectrum was littered with distortion harmonics (fig.7) and the THD+noise was ridiculously high at 25.4%. This was for the left channel; the right channel was somewhat better, at 21%. Even so, the Zanden 5000S seems incapable of reproducing high-level low frequencies without introducing high levels of distortion, though the fact that the response is down almost 5dB at 50Hz might reduce the distortion's audibility. I did wonder if the 5000S was one of those rare products that is actually more linear into low impedances than the 100k ohms of my analyzer. I therefore repeated the test into impedances ranging from 50k ohms down to 1k ohm and got the same results, other than the lower levels into the lower impedances.
Fig.7 Zanden 2000P-5000S, spectrum of 50Hz sinewave at 0dBFS into 8k ohms (linear frequency scale).
So what level of low frequencies will the Zanden reproduce in a clean manner? Fig.8 repeats the spectral analysis shown in fig.7, but with the signal level dropped by 30dB. The THD level was 2.5% left and 1.1% right, and the second harmonic is the highest in level, at –33dB/–39dB, respectively. It's possible that this THD level and this harmonic content will slip below the ear's threshold in this region.
Fig.8 Zanden 2000P-5000S, spectrum of 50Hz sinewave at –30dBFS into 8k ohms (linear frequency scale).
Fig.9 plots the THD+N percentage against digital input level at two frequencies, 50Hz and 1kHz, for the left channel; and at 50Hz for the right channel. The decreasing percentage of THD+N in the 1kHz trace below –20dBFS suggests that the distortion harmonics are buried in the noise below this level. (A constant amount of noise becomes a larger proportion of the total as the signal decreases in level.) There is a slight rise in THD between –10dBFS and full scale, but at 1kHz, at least, the Zanden is a quite linear device. At 50Hz, however, the picture is quite different: the noise dominates the analyzer reading only at very low signal levels; the 5000S's circuit becomes increasingly nonlinear above –35dB. Looking at the waveform on the oscilloscope screen revealed that the negative-going halves of the waveform become increasingly rounded off as the signal level increases. Perhaps the DAC's interstage transformer is inadequately specified, but whatever the reason, this is pathological behavior.
Fig.9 Zanden 5000S, THD+N (%) vs digital signal level (dBFS) into 100k ohms for (from top to bottom): 50Hz, left channel; 50Hz, right channel; 1kHz, left channel.
The Zanden 5000P's behavior on the high-frequency intermodulation test was also disappointing. Not only was the second-order difference product relatively high, at 0.2% left channel and 0.09% right, but the graph was dominated by a slew of audioband intermodulation and aliasing products (fig.10), and the image of the 20kHz tone at 24.1kHz was almost as high in level as the fundamental!
Fig.10 Zanden 2000P-5000S, HF intermodulation spectrum, 19+20kHz at 0dBFS peak into 8k ohms (linear frequency scale).
As usual, I checked the Zanden combo's rejection of word-clock jitter by playing a CD on which had been recorded a high-level tone at exactly one quarter the sample rate, over which had been laid the LSB toggling at approximately 230Hz, which is the sample rate divided by 192. Because both signal frequencies are exact integer fractions of the sample rate, the signal is free from quantizing noise, and any artifacts other than random noise that appear in the analog output of the device being tested will be due to its misbehavior. The two Zandens were joined by the I2S cable, and I examined the 5000S's analog output with the Miller Jitter Analyzer, a software suite that runs on a PC fitted with a National Instruments DSP card.
The result is shown in the narrowband spectral analysis in fig.11, plotted on a linear frequency scale symmetrically about the central spike, which represents the 11.025kHz tone. Correlating with the results of earlier measurements, the noise floor in this graph is 8–9dB higher than the best 16-bit D/As I have measured. The jitter level was high, at 1018 picoseconds peak–peak, which is more than four times the jitter of the best-performing components on this test. Most of the jitter comes from sideband pairs at ±230Hz and its odd-numbered harmonics (red numeric markers). However, there are also sideband pairs at the AC supply–related frequencies of ±60Hz (brown "2") and ±120Hz (blue "3"). Considering that the I2S connection is supposedly jitter-free, as the word-clock signal is not multiplexed with the audio data, this is poor performance.
Fig.11 Zanden 2000P-5000S, high-resolution jitter spectrum of analog output signal (11.025kHz at –6dBFS sampled at 44.1kHz with LSB toggled at 229Hz), CD data. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Turning to the AES/EBU connection, driving the 5000S with data sourced from the PC gave a significantly worse result: almost 4.4 nanoseconds! Again, this was due to major data-related sideband pairs and some AC supply components at ±60Hz and ±120Hz (not shown). Concerned that there was some unexpected interaction between the Zanden and my test gear, I repeated the test using a more recent version of the Miller Jitter Analyzer, running on a different PC with a different National Instruments card.
The result, taken with 48kHz-sampled data, is shown in fig.12. The jitter level was even higher than before, at 4.8ns peak–peak, and while I normally plot the effects of jitter on a graph with an expanded vertical scale, I used a full 120dB scale in this graph to show how severe the Zanden 5000S's problem is when it's fed data via its AES/EBU link. Given that this input also truncates data with a bit depth greater than 16, and won't lock on to sample rates greater than 48kHz, this might well disqualify the 5000S for use as a standalone DAC without the partnering 2000P transport.—John Atkinson
Fig.12 Zanden 5000S, high-resolution jitter spectrum of analog output signal (12kHz at –3dBFS sampled at 48kHz with LSB toggled at 250Hz), external 16-bit data. Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.