PS Audio UltraLink D/A processor Measurements

Sidebar 2: Measurements

The UltraLink was as impressive on the test bench as it was in the listening room; its measured performance was excellent. Unless noted, the following measurements were taken from the balanced outputs. Testing was performed on both balanced and single-ended outputs; any large deviation between the two is noted.

The UltraLink's output voltage when decoding a full-scale, 1kHz sinewave was 3.525V (left channel) and 3.568 (right) from the unbalanced outputs. This is nearly 5dB higher than the CD standard of 2V. At the balanced outputs, the voltage was 7.049V (left channel) and 7.135V (right), exactly the expected 6dB—greater level than the single-ended outputs.

Looking next at frequency response (fig.1), the UltraLink's curve was very flat, with just a hint of rolloff (a tenth of a dB) at 20kHz. Most digital processors exhibit about half a dB attenuation at 20kHz. De-emphasis error was virtually non-existent, owing to the fact that it is performed mathematically in the digital domain by the NPC digital filter.

Fig.1 PS Audio Ultralink, frequency response at –12dBFS into 100k ohms with data sampled at 44.1kHz (right dashed) (0.5dB/vertical div.).

Channel separation, shown in fig.2 plotted as a function of frequency, was exemplary. The UltraLink's crosstalk of –120dB at 1kHz was among the best measured (the Mark Levinson No.30 measured –133dB at 1kHz). The UltraLink's excellent channel separation is surprising in light of the fact that both channels are converted to analog within a single dual-channel DAC, the entire circuit is on a single pcb, and the two analog audio channels are right next to each other on the board. It is apparent that such excellent channel separation is intrinsic to the dual-channel UltraAnalog DAC, and achievable at the processor's output provided the signals are treated correctly in the anaog circuitry downstream. I am at a loss, however, to explain the spikes at a little over 200Hz. They do not appear as noise or artifacts on subsequent spectral analyses, as would be expected if they were power-supply–related junk.

Fig.2 PS Audio Ultralink, channel separation (10dB/vertical div.).

Fig.3 is a third-octave spectral analysis of the UltraLink's output when decoding a dithered –90dB, 1kHz sinewave. Again, the measured performance is excellent. Note the absence of power-supply noise at 60Hz and 120Hz and the freedom from distortion harmonics. Note also the generally low level of the entire frequency region below the test-tone. Further, the trace peaks at exactly –90dB, indicating good low-level linearity. The left- and right-channel traces precisely overlap, revealing that both channels within the DAC are performing identically.

Fig.3 PS Audio Ultralink, 1/3-octave spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS with 16-bit data (right channel dashed).

Performing the same kind of spectral analysis on the UltraLink's output when driven by "digital silence" (all data words are zero) produced the plot in fig.4. There appear to be no converter-produced artifacts–some Bitstream players produce idle tones in the audio band. Note also the drop in noise above the audio band, apart from a slight peak around the 44.1kHz sampling frequency.

Fig.4 PS Audio Ultralink, wide-band 1/3-octave spectrum with noise and spuriae of digital black with 16-bit data (right channel dashed).

Amplitude linearity as a function of signal level is shown in fig.5. This is excellent performance by any standards. The apparent "linearity error" below –100dB is likely noise, not converter misbehavior. Even if it is converter non-linearity, a fraction of a dB error at –100dB is insignificant.

Fig.5 PS Audio Ultralink, linearity error (right channel dashed) (2dB/vertical div.).

The UltraLink's reproduction of an undithered 1kHz, –90dB sinewave is shown in fig.6. The waveform's overall shape is excellent, revealing good converter resolution and uniform quantization-step size at the LSB.

Fig.6 PS Audio Ultralink, waveform of undithered 1kHz sinewave at –90.31dBFS, 16-bit data.

Fig.7 is an FFT transform of the UltraLink's output when decoding a mix of 19kHz and 20kHz at full scale. The plot was made with Audio Precision's 16-bit A/D converter rather than MLSSA's 12-bit converter which we used up to this issue's reviews. The ability to resolve very-low-amplitude intermodulation products is thus greatly improved. The UltraLink shows only a hint of the 24.1kHz aliasing product (44.1kHz sampling frequency minus the 20kHz test tone), low-level sidebands around the test frequencies, and virtually no 1kHz component (the difference between 20kHz and 19kHz).

Fig.7 PS Audio Ultralink, HF intermodulation spectrum, DC–30kHz, 19+20kHz at 0dBFS into 100k ohms (linear frequency scale).

Using a technique developed by Audio Precision's Richard Cabot and described in last February's Stereophile (Vol.15 No.2), I measured how the UltraLink's noise floor was modulated by varying input signal levels. In this test, the converter is driven by the code representing a –60dBFS, 41Hz sinewave. This exercises code transitions within the DAC. The analog output is then high-pass–filtered to remove the test signal, and the converter's noise level is plotted as a function of frequency. The test is repeated at five levels in 10dB increments, down to –100dB. Ideally, the converter should have a low noise level that does not vary with signal level. Additionally, the noise floor's spectral balance should remain constant with input level. The result is shown in fig.8. The lower the noise floor and the tighter the trace groupings, the better. The UltraLink's performance was excellent in this regard.

Fig.8 PS Audio Ultralink, noise modulation (5dB/vertical div.)

When I first tried to capture the UltraLink's reproduction of a 1kHz, full-scale squarewave, the unit turned itself off whenever driven by this signal. This was probably caused by the hum-sensitive front-panel switches reacting to the magnetic fields created inside the unit by this unusual signal. Repeating the test at a different location (which had perhaps less background hum), the UltraLink did not turn off when driven by a full-scale squarewave. In real-world use, I don't think users will encounter this potential problem. The squarewave is shown in fig.9, with the shape typical of the NPC digital filter.

Fig.9 PS Audio Ultralink, 1kHz squarewave at 0dBFS (one sample at 0dBFS, 44.1kHz sampling, 3ms time window).

Output impedance (unbalanced) was a low 39 ohms throughout the band in both channels, indicating the UltraLink will work well with passive level controls. Balanced output impedance was 78 ohms in both channels at any audio frequency. I measured no DC offset at either the balanced or unbalanced outputs, even with the voltmeter on the most sensitive scale (its threshold is 50µV). The UltraLink is the first processor I've measured with no DC offset, even more surprising in light of the unit's direct-coupled circuit and no DC servo.

Finally, the UltraLink does not invert absolute polarity from either the balanced or unbalanced outputs unless the front-panel inversion switch is pressed and the "invert" LED illuminates.—Robert Harley

Pothes's picture

Would you recommend this dac over msb analog and Hugo TT or it is old to to step up and take over these two ?

Will you review mola mola dac

Thanks Robert .

John Atkinson's picture
Pothes wrote:
Would you recommend this dac over msb analog and Hugo TT or it is old to to step up and take over these two?

The PS Audio truncates data with bit depths greater than 16 so is not competitive with modern DACs.

Pothes wrote:
Will you review mola mola dac?

No plans to do so at present.

John Atkinson
Editor, Stereophile