Bun in the oven. In December 2013 a white paper landed in my inbox – it detailed a new DAC to be launched the following year by Boulder, Colorado’s PS Audio. It was to be called the Theorem.
“Attached please find a confidential document on Theorem, PS Audio’s revolutionary new D to A Converter scheduled for release in April. We will announce this to the world in March.”
“We will have a first production run of only 10 Theorem DAC’s next month, January 2014, and we have reserved one for you. It is housed in the same chassis as the PerfectWave DAC is now.”
“This news is VERY CONFIDENTIAL. We will not release any news to the public until March 2014.”
Having spent considerable time with their original and MKII PerfectWave units, I already had a handle on what to expect – or so I thought.
When the formal announcement rode into town in April there came with it a whiff of difference. At US$5995/AU$6699 the new model would sell for more than the PWD.
It wasn’t called Theorem either. It was called DirectStream, presumably to more effectively communicate the core message: all inputs are transformed to DSD prior to analogue conversion.
With the advent of SACD Sony and Philips re-branded pulse-density modulation (PDM) encoding methods as “Direct-Stream Digital” – DSD.
Big-upping the ante, PS Audio’s press release promised the DirectStream would rescue PCM from the doldrums.
This new model isn’t the brainchild of PS Audio’s CEO Paul McGowan either. It’s a designed licensed from ex-Microsoft software engineer Ted Smith. The story goes that mutual friend Gus Skinas – whose mastering and audio engineering skills spans decades – was so impressed by Smith’s breadboard prototype that McGowan’s presence was demanded pronto.
The DirectStream doesn’t use an off-the-shelf DAC chip from the likes of ESS, AKM, Burr Brown or – as per PS Audio’s previous PerfectWave models – Wolfson. Instead, a field programmable gate array (FPGA) converts all PCM inputs to PDM/DSD allowing a passive output stage (with low-pass filter) to handle digital to analogue conversion.
An FPGA in and of itself isn’t the key to success. It is only one part of the equation. The real magic comes from the code that is contained therein.
Designer Ted Smith tells it this way: “I now know that there are many paths to great enjoyment, but still the simplicity of a passive output filter to convert digital to analog was and is very appealing. I started out designing a DSD DAC and since I’m a software guy using an FPGA was a natural way to avoid repeated board turns. After I had an FPGA on board I threw on PCM inputs almost as an afterthought and assumed I could then code the conversion of PCM to DSD.”
Credit must go to Ted Smith for taking the time to expand on the technical ins and outs to non-engineer types like myself. Smith has also been prolific in responding to numerous reader questions both over at Computer Audiophile and on PS Audio’s own website forum.
Smith earns a tip of the cap here for providing explanations to my questions that exuded patience (with my questions) as well as pride and humility (in his own engineering work). Extensive replies to each of my emails came within twelve hours.
Let’s delve into the nitty gritty of how the DirectStream processes data.
Irrespective of sample rate all incoming PCM data is converted to a single high sample-rate PCM stream – ten times that of DSD (28.22MHz).
At this point much of my technical understanding tops out so Ted Smith’s words will take us deeper: “This up-sampling is the same math that most PCM DACs do these days. There are multiple ways of doing things, but in general the process is to stuff zeros between the input samples and filter. For example to go from 44.1 to 88.2 you put in a zero between each set of samples then you filter out the aliasing between 22.05k and 44.1k. That filtering keeps any frequencies higher than 22.05k from interfering with the real audio, but in particular it filters out the bumps from each zero to an input sample then back to the next zero then to the next input sample. That “bouncing” clearly has a frequency of 44.1k and is filtered out, ideally leaving an exact copy of the original signal at the sample points and replacing the zeros with the correctly interpolated values.”
After this process sample rates are brought back down to twice that of DSD (5.64MHhz). Volume control is then applied before a sigma-delta modulator converts the stream to 1-bit 2xDSD and buffers it.
Smith again: “We need to convert that high rate PCM into a DSD stream (a stream of 1 bit samples). That math is weird and not easy to visualize. But the thing that helped me the most when visualizing the process was to know that DSD is a signal that when low pass filtered gives you the PCM signal you want. By its very construction all it takes to convert DSD to analog is a low pass filter.”
“There are a lot of ways of finding a bit stream that when low pass filtered gives the desired output so first I’ll first describe an almost silly way. ‘Look thru all possible bit streams for the bit stream that when low pass filtered gives you the analog you want.’ I wrote software to do essentially that: I ignored most bit streams that clearly differed too much from the input, but I ran thousands of bit streams in parallel and the threw away the worst ones so far and took the best ones so far and added one more bit and checked them again. Repeating that process gave me a DSD bit stream that when low pass filtered was the input signal. It works, and clearly works correctly, but it’s very slow.”
“There’s a process called sigma-delta modulation that does something similar in a much more efficient way. One way of thinking about it is that it filters the input signal so far and also filters the output bit stream so far and then compares the results. If the filtered output so far signal is currently too high it outputs a zero and if the filtered output signal so far is too low it outputs a one. Then it does the whole process again to produce the next bit to output.”
“Let’s take a different view of sigma delta modulators: Let’s say you are getting paid for lots of little odd jobs during the day, each job returns small returns so you wait until the end of the day to get paid and get paid in whole dollars. The bookkeeper subtracts off the whole dollars and keeps track of the difference between what you’ve been paid and what you are owed. To keep you up to date on the average rather than having say 90 cents sitting around over night they might even round up and pay a little early knowing that you’ll do more work tomorrow.”
“Further to make this simpler and more like our audio example let’s assume that the amount of work you do per day on the whole averages to less than a dollar per day.”
“This is a sigma delta process. The only output is whether you get a dollar on a given day or not (a one or a zero), the inputs can be almost anything less than about a dollar, and on average you have the correct pay for your work in your hands. To get a more accurate idea of what you have at any given moment you filter your output zeros and ones of dollars. And to guess when to pay you ahead of time to even out your pay they filter the raw input. We just subtract the two filtered values to see if you should get another dollar or not.”
“In audio we want an output that’s fairly accurate so we use a filter that’s more complicated than just using an average. And since audio is between -1 and 1 instead of 0 and 1, instead of only getting zero or one dollar we get +1 or -1s as the output (tho we encode them as ones and zeros), but the idea is the same.”
The DirectStream eschews all input clock data in favour of clocking out the data just prior to the output stage; by minimising distance between clock and output stage the window for jitter to creep back into the picture is kept vanishingly small. A VXCO handles the exit timing of ones and zeroes from the buffer before being filtered and (finally) seeing daylight via output transformers.
Of course, PS Audio isn’t the first manufacturer to substitute mass-produced DAC silicon for an FPGA + code recipe. The UK’s Chord Electronics have been doing this for some time most recently trickled down into – and ‘popularised’ by – the more affordable Hugo.
Neither is PS Audio the first to play advocate for DSD. Andreas Koch’s white paper detailed here makes for a compelling argument as to its theoretical superiority over PCM.
With a paucity of legally-available (or otherwise!) DSD downloads, turning to DSD technology for improving existing PCM content isn’t without intuitive appeal. Conversely, one cannot hope to restore to PCM encodes that which was stripped out during recording studio (post-) processing.
McGowan’s DirectStream promotional logic is this: if PCM can be converted to DSD prior to decoding the ensuing analogue results will exhibit some of the audible traits heard when decoding a DSD-encoded source file directly.
What are those audible traits you wonder?
Despite there being slim pickings in the DSD download-sphere for the gentleman with more contemporary music tastes (read: not classical, not jazz, rock, electronic), I did manage to pit Acoustic Sounds’ DSD encode of Steely Dan’s Gaucho up against its HDTracks-sourced hi-resolution PCM cousin some months back. The DAC used was a Resonessence Labs Herus. You can read the specifics on how that played out here.
I also listen to my vinyl collection as either DSD or hi-res PCM: a VPI Scout 1.1 with Dynavector 10×2 feeds another PS Audio device, their NuWave Phono Converter. The latter’s ability to digitize to PCM or DSD on-the-fly and the output the digital audio datastream via S/PDIF to a nearby DAC has me effectively listening to (what could be) needle-drops in real time. The DAC of choice of has been Resonessence Labs INVICTA Mirus if only because it’s capable of decoding DSD over it coaxial input. The AURALiC Vega only handles DSD over USB and the Aqua La Scala MKII with its old school R-2R PCM1704UK conversion chips can’t do DSD at all.
The PS Audio NPC allows for near-instantaneous switching between PCM and DSD encoding which has allowed me to note differences between the two over a considerably longer period than with Gaucho vs. Gaucho on the Herus.
Paul McGowan recently featured the following as part of his daily blogging activity that goes out under the Paul’s Posts banner:
“There are more than several people I respect that have presented the following ‘fact’ to me: ‘PCM sounds better than DSD. Even original DSD files, converted to PCM, sound more lifelike than the original DSD files.’ “
“And here’s the thing: I have personally heard and verified this ‘fact’. It’s true. Only, it isn’t a fact. The only fact here is the observation. It is a ‘fact’ we can make these experiments, repeat them time and again getting the same results.”
Cynics might point to McGowan’s square dancing around evidence, in the process attempting to shift the objective to the subjective, as yet another cog in the PS Audio marketing machine. However, I felt he had a point. Do we want our audio gear to sound ‘lifelike’ (as true as possible to the original) or do we simply pursue the illusion that’s most pleasing to our ears and brains?
My off-the-cuff and of-the-moment reply:
“PCM sounds livelier, more energised than DSD. The leading edges of transients are three espressos in by the time they arrive at your ear. DSD on the other hand sounds like it’s been taking advantage of recent changes to Colorado state law (if you catch my drift). It leans towards cotton ball softness, whereas PCM can sound harder.
“Facts? Possibly. Observations? Definitely. Can you tell which one I prefer? Probably not. I’ve purposely sidestepped words like ‘best’ and ‘better’.”
The point here is not just about how the sound of DSD compares to that of PCM. It’s about the value placed on those observations. Some might prefer PCM’s zingier transient delivery whilst others might feel such caffeination crowds out DSD’s greater opulence with spatial information and image specificity. To these ears, DSD definitely connotes a mellower vibe.
I’ll give the final introductory words back to Ted Smith: “I’m pretty sure nothing I say will end the discussion about PCM vs DSD. It’s often more of a religious debate or a confusion of terms than a practical discussion.”
The review unit arrived early July inside the same innovative packaging as its predecessor: a cardboard exoskeleton supports the sandwiching of the DAC between two layers of plastic film thus suspending it in the centre of the inner box which is then wrapped in a outer box for additional protection – common practice among the serious players.
Because the DirectStream has been designed to fit inside the existing PerfectWave chassis the aluminium and steel structure with its too-shiny-to-touch, piano gloss (MDF?) top plate have been retained. That means existing users don’t have to throw out the baby with the bath water and can upgrade the internals of their existing units. The DirectStream kit replaces ALL of the PWD’s innards. PS Audio have filmed YouTube videos that detail the installation process – ideal for users who can’t/won’t call on their local dealer for advice or assistance.
On the back side of this moon sits balanced and single-ended outputs, an SD card slot for firmware upgrades, a (covered) slot for the optional Network Bridge, a pair of HDMI-accommodating I2S input sockets as well as the more familiar coaxial, toslink, AES/EBU and USB digital inputs.
The remote control is pretty much the same as the PerfectWave allowing one to avoid sullying the touchscreen with grubby fingerprints and control volume, source selection, filters from the listening chair. To switch between high and low output voltages one must touch the cog wheel in the upper left corner of the touchscreen; a nice touch if DAC-direct into power amplifier is your favoured approach. I still prefer the meatier sound of a traditional pre-amplifier in the chain but one cannot deny the simplicity, not to mention cost-savings, of using the DirectStream’s onboard volume control.
And thus with several high/er end D/A converters to hand as well as some long-term PCM vs. DSD experience under my belt, the scene is set for listening impressions. However, Part 2 doesn’t just feature my own take – mirroring PS Audio’s recent Sprout Kickstarter campaign, there’ll first be crowd-sourced data points from around the globe. Click here to move onward.
Further information: PS Audio