Philips CDR880 CD-R/RW CD recorder Measurements
Sidebar 4: Measurements
I performed a full set of measurements on the CDR880 as a CD player, followed by a selection of recorder tests, from both analog and digital inputs. The CDR880's maximum output was 1.9V RMS, almost 0.5dB below the standard 2V, sourced from a low output impedance of 200 ohms in the midrange and treble. Below 20Hz the source impedance rose to 622 ohms, which is still low. The output polarity was absolute-phase correct.
The CDR880 featured superb error-correction/concealment, playing through all the standard missing-data tracks on the Pierre Verany test CD without muting. It stumbled only on track 46, which features two successive 0.5mm data gaps. The only clue that the player was working hard on the tracks with a large gap in the data spiral was a slight increase in the measured distortion figure with more than 1mm of missing data.
The top trace in fig.1 is the replay-only frequency response at 0dBFS. The low frequencies start to roll off below 30Hz, reaching 0.4dB at 10Hz, and there is a very slight lift (0.2dB) at 20kHz. Neither should be audible. Below that response is the record/replay (analog in to analog out) response at 1dBFS; it has less top-octave rise but a greater LF rolloff, reaching 0.8dB at 10Hz. The bottom traces in fig.1 are the responses with de-emphasis, which show a maximum 0.65dB error at 4kHz. The replay-only crosstalk is shown in fig.2. It is very low, reaching 100dB at 1kHz and rising to just below 90dB at 20kHz.
Fig.3 shows the audioband spectrum of a dithered 1kHz tone played back at 90dBFS. There is a very slight negative-level error and a trace of second-harmonic distortion, but the plot is otherwise clean at high frequencies. Note, however, the peaks at low frequencies, due to the 60Hz AC line frequency and its harmonics. The CDR880 did seem quite fussy over how it was grounded to the Audio Precision System One test setup, and this and the other spectral analyses were taken with the grounding that gave the lowest level of power-supply noise. These artifacts are too low to be heard as hum, however.
Extending the measurement bandwidth to 200kHz and driving the player with data representing digital silence gave the spectrum shown in fig.4. The power-supply harmonics can again be seen, and there is an increase in noise apparent above the 20kHz edge of the previous graph. This is due to the noise-shaping algorithm used by the Bitstream D/A converter.
With its record-level control all the way up, the Philips CDR880's analog input clips (1% THD+noise) at 850mV at 1kHz. The meter's "0dB" lights were flickering at this level, though the "Over" lights remained dark. Backing off the level to 825mV dropped the distortion to 0.07%, meaning that the recorder's A/D converters overload quite abruptly. Dropping the input level by 1.2dB, to 740mV, turned off the "0dB" lights, leaving the "3dB" lights illuminated. (I would have preferred more resolution at the top of the meter's dynamic range window.) I then fed the CDR880 with an analog 1kHz tone at 82.5µV (80dB ref. the overload level). The resultant playback spectrum, again measured with a wide 200kHz bandwidth, is shown in fig.5. The power-supply components can be seen, as can the ultrasonic noise, but the trace is basically free from distortion products.
Fig.6 shows the CDR880's replay-level error plotted against absolute level. The DAC is obviously very linear down to well below 100dBFS. Spot-checking the level error for tones fed to the analog inputs revealed that the A/D converters also offered good linearity. Fig.7 shows the waveform of an undithered 1kHz tone at 90.31dBFS. The expected three distinct voltage levels can be easily made out, but note that there is some low-frequency noise present that makes the graph slope from right to left. This is the 60Hz-supply noise noted earlier. There is also a very small DC offset noticeable in this graph.
As far as steady-state distortion is concerned, the CDR880's analog stages were very linear. Fig.8 shows the spectrum of a maximum-level low-frequency tone, 61Hz, played back from the CBS Test CD. The distortion components are all very low in level, with the highest in levelthe second and third harmonicsboth more than 90dB down from peak level. Recording a 61Hz tone from the analog input at 1dBFS gave an identical spectrum, suggesting that the '880's ADCs do not add significant distortion.
Intermodulation was also very low. Fig.9 shows a spectral analysis of the '880s' analog output while it replayed data representing 19kHz and 20kHz tones, each at 6dBFS. There is no appreciable 1kHz difference component visible, while the second-order components are all at or below 85dB. This is excellent performance.
I used the Miller Audio Research jitter analyzer to analyze jitter on both the Test CD-R provided by MAR, and on a digital clone of that disc made at 44.1kHz on the Philips. One peculiarity of the jitter tests was that the results were more variable/less repeatable than is usually the case. I also did not get any statistically significant differences in jitter from the original disc, or from a clone made with the CDR880.
But a typical result is shown in fig.10. The absolute jitter level was 1130 picoseconds peakpeak, which is quite high. (The grayed-out trace in fig.10 is the Meridian 508.24's jitter, measured on the same basis, which is a very low 144.2ps.) Though a large number of higher-frequency sidebands can be seen, these are realtively low in level; most of the CDR880's jitter is present in the form of low-frequency sidebands, which can be seen clustered around the center tone and are marked with purple numerical indicators. Note also the strongly asymmetric sidebands, marked with a purple "30." These are all unknown in origin, but the low-frequency components may well contribute to the quality of the bass noted by WP in his auditioning, and which I heard in my own listening sessions.
Finally, the conjecture in the UK press that the earlier CDR870's sample-rate converter gave rise to audible problems led me to check out the '880's behavior when fed the same data but with different sample rates. I used the Audio Precision System One to generate a low-distortion 11kHz sinewave which I digitized with a dCS 902 professional A/D converter set to 16-bit resolution. The peak level was 6dBFS, the average level the same 10dBFS as the Miller test signal. I fed the dCS's digital output to the CDR880's coaxial digital input and recorded a couple of minutes of data with the dCS first set to a 48kHz sample rate, then to a 44.1kHz sample rate. For the former, the '880's sample-rate converter is in circuit, rewriting the data at the CD's standard 44.1kHz rate; for the latter, it is switched out, the recorded data being bit-for-bit identical with the original. Using the Miller analyzer to perform a narrow-band, high-resolution analysis of the player's noise floor around the tone's fundamental while the '880 played back both versions of the 11kHz tone, I generated the traces shown in fig.11. (Thirty two readings, each with a 32,768-point FFT size, were averaged for each spectrum to reduce the contribution of random noise.)
The foreground spectrum is with the 44.1kHz sampled data cloned directly. The same low-frequency sidebands can be seen as in fig.10; these have a fundamental frequency of 22.5Hzto what this is related, I have no idea. The central peak also has "skirts" added by the presence of low-frequency, noiselike jitter. The grayed-out spectrum in the background is of the same tone but sample-rate converted from 48kHz to 44.1kHz by the CDR880. The noise level is basically the same, but note how the skirts have spread either side of the central peak, covering the lowest-frequency sidebands. According to Paul Miller, such spreading tends to be perceived subjectively as a smearing of stereo images, and possibly as a "slowing" of the bass. Certainly, the switching out of the sample-rate converter for 44.1kHz-sampled clones appears to be a step in the right direction, measurement-wise.John Atkinson