Sidebar 3: Measurements
I used my Audio Precision SYS2722 system to examine the HiFi Rose RD160's behavior on the test bench. I used AES3 and optical S/PDIF data for the testing as well as USB data sourced from my MacBook Pro. The RD160 offers a choice of six reconstruction filters, three oversampling modes and three without oversampling; a choice of several fixed output levels or a variable output; and both balanced and single-ended outputs. I performed a complete set of tests using the balanced outputs with the variable output level option, the oversampling bypassed, and the default "Short Sharp" reconstruction filter. I then repeated some of the tests with the other filters, with the fixed output levels, with the single-ended outputs, and with the oversampling options.
The AES3, coaxial S/PDIF, and optical TosLink inputs all accepted data sampled at rates up to 192kHz. Apple's AudioMIDI utility revealed that the USB port accepted 16-bit and 32-bit integer data sampled at all rates from 44.1kHz to 768kHz. The USB Prober app identified the RD160 as "RD160-DAC" from "HiFi ROSE" and indicated that the USB port operates in the optimal isochronous asynchronous mode. The processor's front-panel display indicated that the firmware versions were 1.26 (CPU) and 1.0.1 (MCU).
The HiFi Rose DAC's volume control operated in accurate 0.5dB steps, though there was no change in level between the top two steps. With the control set to the maximum, the output level with a 1kHz signal at 0dBFS was 9.45V from the balanced outputs, 4.67V from the unbalanced outputs. These output voltages were the same with the fixed level set to 9V; the output voltages were 5.32V and 2.62V, respectively, with the fixed level set to 5V; with it set to 2V, the output voltages were 2.1V and 1.02V. Oversampling to DSD64 reduced all the output levels by 6dB. The RD160 preserved absolute polarity, ie, was noninverting, with all level settings and from both output types.
The balanced output impedance in both variable and fixed output modes was a low 97 ohms at 1kHz and 94 ohms at 20kHz but rose to 2760 ohms at 20Hz, suggesting that the output is capacitor-coupled. As expected, the unbalanced output impedances were half these values. As long as the RD160 is used with a preamplifier or power amplifier with a balanced input impedance of at least 20k ohms or an unbalanced impedance of at least 10k ohms, the processor's low frequencies won't sound lightweight.
Fig.1 HiFi Rose RD160, "Short Sharp" filter, impulse response with one sample at 0dBFS, 44.1kHz data, 4ms time window).
Fig.2 HiFi Rose RD160, "Short Slow" filter, impulse response with one sample at 0dBFS, 44.1kHz data, 4ms time window).
Fig.3 HiFi Rose RD160, "Low Short" filter, impulse response with one sample at 0dBFS, 44.1kHz data, 4ms time window).
The RD160's impulse response with the default "Short Sharp" reconstruction filter is shown in fig.1; it is typical of a long minimum-phase filter: All the ringing follows the single sample at 0dBFS. The "Sharp" reconstruction filter's impulse response (not shown) revealed this to be a long linear-phase filter, with equal amounts of ringing on either side of the sample at 0dBFS. The "Slow" filter was also a linear-phase type but with only a small amount of symmetrical ringing. Fig.2 shows the impulse response with the "Short Slow" filter, fig.3 the impulse response of the "Low Short" filter. Both are hybrid types, with different amounts of ringing on either side of the high-level sample.
Fig.4 HiFi Rose RD160, "Super Slow" filter, impulse response with one sample at 0dBFS, 44.1kHz data, 4ms time window).
Fig.5 HiFi Rose RD160, "Super Slow" filter, impulse response with one sample at 0dBFS, 44.1kHz data oversampled to DSD64, 4ms time window).
The odd man out was the "Super Slow" filter (fig.4), which, other than the small amount of symmetrical ringing from the Audio Precision's A/D converter sampling at 200kHz, is close to a time-perfect delta function. Peculiarly, while the "Short Sharp" and "Super Slow" filters' impulse responses were preserved with oversampling set to 44.1kHz, it changed to a long linear-phase type with both DSD64 and 705.6kHz PCM oversampling (fig.5).
Fig.6 HiFi Rose RD160, "Short Sharp" and "Sharp" filters, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan) into 100k ohms with data sampled at 44.1kHz (20dB/vertical div.).
Fig.7 HiFi Rose RD160, "Slow" and "Short Slow" filters, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan) into 100k ohms with data sampled at 44.1kHz (20dB/vertical div.).
The magenta and red traces in fig.6 show the HiFi Rose's wideband spectrum with 44.1kHz white noise data at –4dBFS and the "Short Sharp" filter. (The spectrum was identical with the "Sharp" filter.) The response rolls off sharply above the audioband, reaching full stopband attenuation at just above half the sample rate, indicated in the figure by the green vertical line. The image at 25kHz of a full-scale 19.1kHz tone (cyan, blue traces) is completely suppressed, and the harmonics of the tone lie at or below –70dB (0.03%). Fig.7 shows the result of the same test with either the "Slow" or "Short low" filters. The white noise spectrum (magenta, red traces) starts to roll off above 10kHz, with a peculiar notch at exactly half the sample rate. The image at 25kHz of the 19.1kHz tone (cyan, blue traces) lies just 12dB below the level of the tone.
Fig.8 HiFi Rose RD160, "Super Slow" filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan) into 100k ohms with data sampled at 44.1kHz (20dB/vertical div.).
Fig.9 HiFi Rose RD160, "Low Short" filter, wideband spectrum of white noise at –4dBFS (left channel red, right magenta) and 19.1kHz tone at 0dBFS (left blue, right cyan) into 100k ohms with data sampled at 44.1kHz (20dB/vertical div.).
As expected from the "Super Slow" filter's impulse response, the ultrasonic rolloff with white noise data is very slow (fig.8), with nulls at 44.1kHz and 88.2kHz. With a 19.1kHz tone at 0dBFS, a large number of aliased products are present, both in the audioband and above. The image at 25kHz of the tone is attenuated by just 3dB. The "Low Short" filter (fig.9) offers a compromise between the extremes shown in fig.6 and fig.8.
Fig.10 HiFi Rose RD160, "Short Sharp," "Sharp," and "Low Short" filters, frequency response at –12dBFS into 100k ohms with data sampled at: 44.1kHz (left channel green, right gray), 96kHz (left cyan, right magenta), and 192kHz (left blue, right red) (1dB/vertical div.).
Fig.11 HiFi Rose RD160, "Short Slow" and "Slow" filters, frequency response at –12dBFS into 100k ohms with data sampled at: 44.1kHz (left channel green, right gray), 96kHz (left cyan, right magenta), and 192kHz (left blue, right red) (1dB/vertical div.).
The RD160's frequency response with the "Sharp," "Short Sharp," and "Low Short" filters with 44.1kHz, 96kHz, and 192kHz data (fig.10) was flat in the audioband and followed the same basic shape at all three sample rates, with a very sharp rolloff just below half of each rate. This graph was taken from the balanced outputs with the variable level set to the maximum; the responses were the same at lower volume control settings, with the fixed output levels, and from the single-ended outputs. The response with the "Slow" and "Short Slow" filters and 44.1kHz data (fig.11, green, gray traces) was down by 9dB at the top of the audioband. I haven't shown the frequency responses with the "Super Slow" filter because the rolloff at each sample rate was corrupted by the presence of aliased energy.
Fig.12 HiFi Rose RD160, "Short Sharp" filter, spectrum of 1kHz sinewave, DC–1kHz, at –12dBFS with volume control set to the maximum (left channel green, right gray) and with 1kHz sinewave at 0dBFS and volume control set to –12dB (left blue, right red, linear frequency scale).
Channel separation was superb, at >120dB in both directions below 3kHz and still 112dB at the top of the audioband. The low-frequency noisefloor with the volume control set to the maximum (fig.12, green, gray traces) was very low in level and free from power supply–related spuriae. Reducing the level by 12dB or setting oversampling to DSD64 lowered the level of the noisefloor by 6dB (blue, red traces).
Fig.13 HiFi Rose RD160, "Short Sharp" filter, spectrum with noise and spuriae of dithered 1kHz tone at –90dBFS with 16-bit data (left channel green, right gray) and 24-bit data (left blue, right red) (20dB/vertical div.).
Fig.14 HiFi Rose RD160, "Short Sharp" filter, waveform of undithered 16-bit, 1kHz sinewave at –90.31dBFS (left channel blue, right red).
Fig.15 HiFi Rose RD160, "Short Sharp" filter, waveform of undithered 24-bit, 1kHz sinewave at –90.31dBFS (left channel blue, right red).
Fig.13 shows the RD160's output spectra with 16- and 24-bit dithered data representing a 1kHz tone at –90dBFS with the "Short Sharp" filter and oversampling bypassed. (The spectra were identical with the other filters.) The increase in bit depth lowered the noisefloor by 30dB, which suggests a superb measured resolution of 21 bits. Fig.14 shows the waveform of undithered data representing a tone at exactly –90.31dBFS; the waveform symmetry shows that changing all 16 bits in the digital word gives exactly the same change in the analog output level as changing just the LSB. With undithered 24-bit data, the HiFi Rose output a clean sinewave despite the very low signal level (fig.15).
Fig.16 HiFi Rose RD160, "Short Sharp" filter, spectrum of 50Hz sinewave, 24-bit data, at 0dBFS, DC–1kHz, into 100k ohms (left channel blue, right red; linear frequency scale).
The HiFi Rose RD160's distortion signature primarily comprised the third and fifth harmonics (fig.16), but these were very low in level, lying at –99dB (0.001%) and –109dB (0.0003%) respectively. While other harmonics are present, these all lie at least 10dB lower in level. Fig.16 was taken into 100k ohms; when I repeated the spectral analysis into 600 ohms, the levels of all the harmonics increased by around 8dB but were still very low in absolute terms.
Fig.17 HiFi Rose RD160, "Short Sharp" and "Sharp" filters, HF intermodulation spectrum (DC–30kHz), 19+20kHz, 24-bit data, at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Fig.18 HiFi Rose RD160, "Short Slow" and "Slow" filters, HF intermodulation spectrum (DC–30kHz), 19+20kHz, 24-bit data, at 0dBFS into 100k ohms (left channel blue, right red; linear frequency scale).
Intermodulation distortion with 24-bit data representing an equal mix of 19 and 20kHz tones, each at –6dBFS, with the "Short Sharp" and "Sharp" filters is shown in fig.17. The difference product at 1kHz lay at –100dB (0.001%), and its level didn't increase when I repeated the test with the demanding 600 ohm load. However, the higher-order products were higher in level, those at 18kHz and 21kHz lying at –76dB (0.015%). Other than the presence of aliased images at 24kHz and 25kHz, the intermodulation spectrum with the "Low Short" filter was identical to that shown in fig.17. As expected, a lot of aliased images were present with the "Slow" and "Short Slow" filters (fig.18), but actual intermodulation products were lower in level than they had been with the sharp rolloff filters. Even more spurious tones and aliased products were present with the "Super Slow" filter (not shown).
Fig.19 HiFi Rose RD160, "Short Sharp" filter, 16-bit optical data, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Fig.20 HiFi Rose RD160, "Short Sharp" filter, 24-bit optical data, high-resolution jitter spectrum of analog output signal, 11.025kHz at –6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz (left channel blue, right red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
When I examined the RD160 optical odd-order harmonics of the undithered low-frequency, LSB-level, 16-bit squarewave all lay close to the correct levels, shown by the green line in fig.19. A pair of low-level sidebands of unknown origin was present at ±620Hz; these were slightly lower in level with AES3 and USB data, and the levels were not affected when I connected a wire from the grounding terminal on the processor's rear panel to the analyzer's chassis ground. They were also present with 24-bit J-Test data (fig.20).
In almost all respects, the HiFi Rose RD160's measured performance was superb, with state-of-the-art resolution and very low harmonic distortion and noise. However, that performance is dependent on which of the six reconstruction filters is used. As always, the better-behaved a filter is in the time domain, the worse it behaves in the frequency domain, and vice versa.—John Atkinson
