Linn Karik/Numerik CD player 1992 Measurements
The Numerik put out 2.135V when decoding a full-scale, 1kHz sinewave. This is only slightly higher (0.57dB) than the CD standard of 2V. Channel balance was the best I have measured, the right channel being just 0.004dB higher in output than the left. Digital converters often have a few tenths of a dB channel imbalance.
Frequency response (fig.1) was flat, rolling off an almost negligible 0.5dB at 20kHz. Note the virtual overlap of the left and right traces, confirming the almost perfect channel balance. Fig.2 is the Numerik's de-emphasis error. A slight positive error of 0.2dB at 4kHz can be seen. It corrected itself at 10kHz and had a very slight negative error at 16kHz. The HF rolloff seen in fig.1 is superimposed on this plot. The de-emphasis error is what remains after the frequency response is subtracted. Both channels exhibited the same error. These deviations are very slight and not likely to be audible. Channel separation, shown in fig.3, was excellent, measuring better than 120dB over most of the audio band. There was not a significant separation decrease at high frequencies, as is often seen.
Fig.1 Linn Numerik, frequency response (right channel dashed, 0.5dB/vertical div.).
Fig.2 Linn Numerik, de-emphasis error (right channel dashed, 0.5dB/vertical div.).
Fig.3 Linn Numerik, crosstalk (right-left dashed, 15dB/vertical div.).
Fig.4 is a third-octave spectrum analysis of the Numerik's output when decoding a -90.31dB, 1kHz sinewave. Note the very low noise floor, especially in the low-frequency range. A hint of the Numerik's superb linearity is provided by the 1kHz peak reaching exactly the -90dB horizontal division. Indeed, looking at the departure from linearity in fig.5, we see that it is virtually perfect down to -100dB. Below -100dB, the measurement becomes dominated by noise. The error of +0.47dB (left channel) and +0.37dB (right channel) is among the best measured. In fact, I suspect that this apparent "linearity error" is actually noise, not converter misbehavior. It should be noted that this superb low-level linearity performance is not dependent on correct MSB trimming. All Numerik processors will likely have this excellent performance, and will maintain the linearity without further adjustment over time.
Fig.4 Linn Numerik, spectrum of dithered 1kHz tone at -90.31dBFS, with noise and spuriae (1/3-octave analysis, right channel dashed).
Fig.5 Linn Numerik, departure from linearity (right-channel dashed, 2dB/vertical div.).
A 1kHz, full-scale squarewave is shown in fig.6. The slight ringing with minimal overshoot is typical of linear-phase digital filters optimized for low-level resolution. A very low level, undithered (-90.31dB) 1kHz sinewave is shown in fig.7, captured with 16-bit resolution by the Audio Precision Dual Domain. The three discrete converter transitions at this level (+1, 0, and -1) are partially visible, but the plot is overlaid with audioband noise.
Fig.6 Linn Numerik, 1kHz squarewave at 0dBFS.
Fig.7 Linn Numerik, waveform of undithered 1kHz sinewave at -90.31dBFS.
Looking at an FFT-derived spectral analysis of the Numerik when decoding a combination of 19kHz and 20kHz at full scale (fig.8), we can see an amazingly low level of intermodulation products, basically none sticking up above the noise floor of the MLSSA A/D converter. (The MLSSA hardware uses a 12-bit ADC.) It is common to see products at 1kHz (20kHz minus 19kHz) and at 18kHz and 21kHz (due to the interference between that 1kHz and the original tones), as well as a 24.1kHz signal (the sampling rate of 44.1kHz minus 20kHz). This is one of the best intermodulation plots I've seen.
Fig.8 Linn Numerik, HF intermodulation spectrum, 300Hz-30kHz, 19+20kHz at 0dBFS (10dB/vertical div.).
Output impedance was 64 ohms across the band, a low value. This suggests the Numerik will work well with passive level controls. The Numerik does not invert absolute polarity, and a negligible level of DC (1mV left channel, 2mV right channel) was measured at the analog outputs.
To test the reportedly improved error-correction capability of the new Sony chip used in the Karik, I played the dropout tracks on the Pierre Verany test disc. Disc #2 of this set has a series of tracks with intentional data dropouts of increasing length. The Karik could play track 33 with no problem, but stuttered on track 34. This is slightly better, but not significantly better, than most transports. This performance nevertheless exceeds the CD specification for dropout correction. Incidentally, none of the measurements changed with the sync cable connected.—Robert Harley