Digital Processor Reviews

Jim Austin auditioned the Huron firmware update in October 2017 (Vol.40 No.10):

PS Audio's DirectStream digital processor is unusual in that, via firmware, it can be upgraded not only functionally but sonically. This is possible because, instead of by hardwired, off-the-shelf chips, its digital processing is done by a field-programmable gate array (FPGA). Within the constraints of the hardware, the designer is thus free to explore new ideas by reprogramming the FPGA, and to implement the effective ideas by disseminating firmware updates. A few other companies offer firmware-upgradable DACs (footnote 1), but few permit such extensive modifications as does PS Audio's DirectStream.

These updates can significantly change how a processor works. For example, compared its predecessor Torreys, Huron, which is the most recent update to the DirectStream's firmware, doubles the upsampling frequency, from 10 to 20 times the single DSD rate; lowers noise in the FPGA's power supply; removes an upsampling filter; and adjusts low-frequency group delay. (I discuss a couple of other important changes below.)

What does this accomplish? PS Audio claims that the Huron update (I quote):

• "reduces jitter to the point where it is almost unmeasurable;

• "lowers in-band noise by 3dB;

• "lowers ultrasonic noise by a staggering 18dB, kept below the noise floor until 120kHz (footnote 2); and

• "improves the perceived frequency response of both extremely low bass and very high treble."

Those are impressive claims. In an e-mail exchange, I asked Ted Smith, chief designer of the DirectStream DAC and architect of its firmware, how they were achieved. I'll start with jitter, as what follows is a good illustration of just how radically an FPGA-based DAC can be reengineered with firmware. Previously, the input buffer—the stage at which the bitstream is effectively reclocked by throwing out the old timing information and assigning new timing information by detecting the digital waveform's rising edges—came after the upsampling filter. "In Huron, I moved the input buffer earlier so that the upsampling would be synchronous with the outputs," Smith told me. "This should lessen any possible effect of incoming signal timing (from clocks or data) on the output of the FPGA."

Meanwhile, the improvements in the noise (second bullet point) are made possible by doubling the DSD output frequency—the frequency at the final stage before it's filtered back to analog—from 2x to 4x the base DSD rate (footnote 3). PS Audio chose not to mention this change in their marketing copy because they didn't want customers to confuse the output rate with the input rate and conclude that the DirectStream can now accept quad-rate DSD inputs. It can't; it still maxes out at 128Fs, or 2x DSD.

I was fascinated by the claim in bullet point four: How can you extend "perceived" high- and low-frequency response while maintaining the correct actual frequency response, which presumably was already objectively flat? A candidate answer is: by ensuring correct signal phase. In discussions of frequency response, the focus usually is on how the magnitude of the signal varies with frequency, but "phase"— ie, when that frequency arrives relative to the other frequencies—is equally important: To reconstruct a waveform in the time domain, you need both magnitude and phase. There's scientific literature on the audibility of phase effects, and, while there's still some disagreement, there's plenty of evidence that phase effects are audible (footnote 4).

In my conversation with Smith, the p-word came up quickly. On the low end, he told me, there was "a tunable parameter that I could adjust for more accurate (very) low-frequency phase." The precise changes are proprietary, and not, he emphasized, a simple tuning of the frequency response. Rather, he altered the code to dedicate more FPGA power to handling low frequencies in a better way. "Does flattening the group delay below 1Hz answer things well?" he asked.

One hertz? No, it doesn't! On its lowest setting, my ceiling fan spins faster than that! Air vibrating at a frequency of 1Hz—one cycle per second—doesn't produce pressure waves that the human ear can experience as sound. Assuming that Smith had mistyped, I sought confirmation. He confirmed. He'd flattened the group delay below 1Hz.

That change does, however, affect higher frequencies too, including the magnitude (not just phase): Huron alters the frequency response, Smith told me, by 0.007dB at 20Hz—seven thousandths of a decibel at the extreme low end of human hearing. There's no way that's audible. In our conversation, it soon became clear that, in attributing the audible changes to his sub-1Hz adjustments, Smith is speculating: "I'm not convinced that that's what people are hearing [and] reacting to," he told me.

What, then? "I really suspect that people have tuned their system[s] empirically, and with Huron having less jitter and a blacker background, they need less euphonic twiddling. We all understand that more accurate bass isn't usually louder bass; instead the bass gets less emphasized and doesn't call itself out, but everything from bottom to top sounds more time accurate and hence there's a little better PRaT [pace, rhythm, and timing]." I like that explanation better.

Smith attributes the perceived improvement at the other frequency extreme to modifications to the DirectStream's upsampling engine. "[E]very time I lowered the order of or increased the cutoff on the final upsampling filter . . . people liked the sound better," he said. "Even though, on paper, the changes were only in the ultrasonic, I expected that, once again, doing less filtering of the very high frequencies would have an audible effect." He removed the final upsampling filter and things opened up, he told me—even though the frequencies he was no longer filtering out are above what people can hear, and even though those frequencies would later be removed by the analog filter (footnote 5).

Climbing Mt. Huron
Updating my firmware was far easier than understanding the impact of Ted Smith's changes. I downloaded the update file from PS Audio's downloads page and unZipped it. For good measure, I reformatted an SD card (footnote 6) that I'd previously been using in a camera and copied to it the firmware: four unzipped files. If you prefer, you can order an SD card with the files already loaded ($29) from PS Audio's website, or get one from your PS Audio dealer. (When was the last time you significantly upgraded your sound for $29?) All that's left to do is to turn off the DirectStream, stick the SD card upside down into the slot on the rear panel, and turn the power on. If all goes well—it did for me—in a couple of minutes your DAC will be upgraded.

My first impression of the Huron upgrade was of improved clarity—"blacker" backgrounds and all that. The sound was pristine. Highs were extended and noise-free, hence relaxed. I wouldn't say that the highs sounded more extended; rather, they seemed a bit smoother and more even, drawing less attention to themselves. I found it easiest to hear the change with natural-sounding recordings of acoustic instruments, which to me seemed a bit more present, more in the room.

Compared to earlier firmware versions, Huron may have a bit less of the obvious warmth that has led some listeners to characterize the DirectStream's sound as "analog-like." So it's possible that your Huron-updated DirectStream DAC will sound less like your record player. Are you okay with that? Some may regret it, but to me—and I hope I don't lose cred with vinyl lovers—it was a clear improvement, and in line with Smith's last observation: "the bass gets less emphasized and doesn't call itself out, but everything from bottom to top sounds more time accurate, and hence there's a little better PRaT."

What you think of these changes is likely, I think, to depend on your system, especially your loudspeakers and/or headphones. While I was auditioning Huron, I experienced a radical change in my system: my modestly sized DeVore Fidelity Nine speakers were replaced by a pair of Alta Audio Titanium Hestias—much more expensive, 135-lb beasts with oodles of bass—currently in for review. Through the DeVores, the Huron update registered as a loss of pleasant warmth in the low frequencies, although overall I still preferred its pristine, natural sound to the DirectStream before the Huron upgrade. With the larger speakers, when I switched between the two firmware versions, Huron cleaned things up. The improvement was unambiguous.

Huron also includes some functional changes, though as I write this the practical benefits of those changes have yet to be realized. Owners of DirectStreams know that PS Audio offers an optional Ethernet renderer, the Bridge II. The Bridge II can also be firmware-updated, independent of the DAC itself. PS Audio has promised an update to the Bridge II that will allow it deliver MQA content via Ethernet to the DirectStream DAC from Tidal and from Roon, the music-server software I use. Huron sets the stage, preparing the DirectStream DAC for the upcoming Bridge II updates.

PS Audio's Huron firmware upgrade for their DirectStream DAC is a clear improvement over Torreys, which was already very good. I'm hearing more of the music, with greater clarity and a greater sense of order and ease.—Jim Austin

---

Footnote 1: Among the digital audio companies offering firmware upgradability are Auralic, Aurender, Ayre Acoustics, dCS, EMM Labs, and Mytek. In some of these, the upgrades are only of features; others are more similar to the PS Audio in facilitating upgrades likely to affect the sound.

Footnote 2: In our e-chat, Smith was eager to qualify this point to ensure accuracy: Whether the digital noise rises above the analog noise at 120kHz or some other frequency depends on how you measure the noise floor.

Footnote 3: One problem with DSD is its intrinsic high-frequency noise. DSD is a one-bit system in which the bit has only two possible states or values, 1 or –1. In the frequency domain, this rapid switching takes the form of high-frequency noise. The faster it switches, the further above the audioband the noise shifts, and the more flexibility a designer has in filtering it out. Filtering out the noise is equivalent to smoothing the transitions in the time domain. When you do that, you get back music.

Footnote 4: Consider that the shortest possible sound—an impulse—has the same spectral content as white noise.

Footnote 5: Skeptical? Me too, a little. But it's in line with similarly odd-sounding things I've heard, or heard of—such as the apparent fact that a 32-bit data path can sound better than a 24-bit path, even though the very best digital processors approach but do not reach a resolution of 21 bits.

Footnote 6: If you reformat your SD card, use a FAT-16 or FAT-32 format, as neither NTSF nor Ex-FAT will work. If you have a DirectStream Junior, the firmware is loaded with a USB memory stick, not an SD card.