Digital Processor Reviews

John Atkinson returned to the PS Audio DirectStream MK2 in July 2024 (Vol.47 No.7)

When I reviewed the new version of PS Audio's classic D/A processor (footnote 1) in the June 2023 issue, I enjoyed its sound quality, but some aspects of its measured performance concerned me. The impulse and frequency responses, the behavior of the reconstruction filter, the channel separation, the low intermodulation distortion, and the balanced and single-ended output impedances were all beyond reproach, and though harmonic distortion was higher in level than I was expecting, it was still sufficiently low. However, the level of ultrasonic noise was higher than I had found with PS Audio's original DirectStream processor running the "Windom" FPGA firmware: The MK2 processor running the v2.3.6 firmware had 210mV of ultrasonic noise in its balanced outputs, with a center frequency of 370kHz, compared with the earlier version's 11mV of noise.

This noise will be due to the processor synchronously upsampling the input data to a 30-bit word length running at 56.4448MHz, followed by a digital-domain volume control. The resultant 50-bit, high–sample-rate data are resampled to quad-rate DSD and converted to analog with a low-pass filter. The problem was that enough of this noise leaked into the audioband to compromise the measured linearity and jitter rejection and limit resolution to 16 bits, even with 24-bit data.

Then, in the spring of 2024, PS Audio introduced the Mount Massive FPGA firmware for the DirectStream DAC MK2. The claimed improvements with this firmware were the lowering of ultrasonic noise by 30dB and audioband noise by 10dB. The Digital Lens FPGA buffer and user interface were also updated.

A follow-up review would be justified, so as our 2023 review sample had long since been returned, PS Audio sent me a sample with the new firmware. The processor's display indicated that this had the serial number 031053 and was fitted with the v3.1.0 firmware (the main OS firmware) and the 233-Massive firmware (specifically for the FPGA).

After some preliminary listening using Roon and the DirectStream's USB input—unlike the original DirectStream processor, the MK2 does not have an Ethernet port for audio data—I took the new MK2 into the test lab, where I performed a full set of measurements using first my Audio Precision SYS2722 system then the magazine's higher-performance APx555 analyzer.

The MK2's maximum balanced output level at 1kHz was similar to the 2023 sample's, at 3.71V, but now a lower level of ultrasonic noise was present in its output: 167mV. This can be seen in fig.1, which shows the wideband spectrum of white noise at –4dBFS with the v2.3.6 firmware (left channel blue, right cyan) and with the Mount Massive firmware (left red, right magenta). The level of the earlier firmware's ultrasonic noisefloor starts to rise above 40kHz, with the levels of the FFT bins reaching –90dBFS at 90kHz. By contrast, the Mount Massive's ultrasonic noisefloor doesn't start to rise until 70kHz and is up to 20dB lower below that frequency than it had been with the 2023 firmware.

What about the audioband noisefloor? Fig.2 shows the spectrum of the new DirectStream's low-frequency noisefloor while it drove a full-scale, 24-bit, 1kHz tone from the balanced outputs, with the volume control set to its maximum of "100." I used the APx555's brickwall input filter set to 20kHz to minimize the influence of the ultrasonic noise on the measurement. The individual elements of the PS Audio's random noisefloor lay around –122dBFS. To put this measurement into perspective, fig.3 shows the low-frequency noisefloor of the Weiss Helios processor that I reviewed in the March 2024 issue. Measured under identical conditions, the Helios's noisefloor is 30dB lower in level than the DirectStream's. (Note the expanded vertical scale in this graph.)

How does this level of random noise affect the PS Audio's measured resolution? Fig.4 shows the spectrum of the 2023 DirectStream's output while it decoded dithered 24-bit data representing a 1kHz tone at –90dBFS. The level of the noisefloor is similar to that with the full-scale 1kHz tone in fig.2. However, when I repeated this test with the Mount Massive firmware, the level of the noisefloor dropped by up to 6dB (fig.5).

Despite this improvement, the DirectStream MK2's resolution is still limited to around 16 bits. I checked the spectrum shown in fig.5 against the same measurement that had been performed by PS Audio and is shown in fig.6. (Note the different aspect ratio of this graph.) My noisefloor looks cleaner than PS Audio's due to the fact that I average 32 FFT measurements to reduce the peak voltage of the noise whereas the manufacturer's appears to be a single capture. Nevertheless, the average level of the noisefloor appears identical in both graphs (footnote 2).

Given that all recordings of music have a dynamic range that varies between 13–14 bits at low frequencies and 20 bits in the top audio octave, due to the presence of room and microphone noise, maximizing the measured resolution might be thought to be academic. I therefore spent a month using the Mount Massive DirectStream MK2, fed USB data from Roon via an AudioQuest JitterBug FMJ, as my daily driver. The rest of the system was the pair of Parasound JCA100 Tribute monoblock amplifiers that I reviewed in the June 2024 issue driving either KEF LS50 or GoldenEar BRX loudspeakers. I used Roon's parametric DSP to flatten and extend both loudspeakers' midbass region and controlled volume with the DirectStream's control. After this followup review had been submitted for publication I spent 10 days using the PS Audio with the GoldenEar T66 loudspeakers that I will be reviewing in the August 2024 issue.

My impressions of the PS Audio's sonic character were very similar to those in my original review. The overall balance was easy on the ear, the stereo imaging was precise, with good soundstage depth, and while the DirectStream MK2's retrieval of recorded detail and its sense of low-frequency drive weren't in the same class as those of the MBL N31 that I have been using as a reference D/A processor, my involvement in the music didn't seem compromised.

So, yes, the Mount Massive FPGA upgrade does lower the levels of both ultrasonic and audioband noise, but not to the point where the measured resolution competes with that of the highest-performing processors. Nevertheless, I concluded my 2023 review by writing "for want of a better word, the PS Audio DAC preserved each recording's 'musicality.'" Mount Massive offers the same benefit.—John Atkinson

---

Footnote 1: The DirectStream MK2 costs $7999 and is sold direct in the US. PS Audio, 4865 Sterling Drive, Boulder, CO 80301. Tel: (800) PSAUDIO, (720) 406-8946. Web:

Footnote 2: Note, also, that the presence of odd-harmonics in the PS Audio graph suggests that the source was not 24-bit dithered data but 16-bit undithered data. This doesn't affect my comments regarding the DirectStream MK2's resolution.