Keywords: "Burr-Brown, Advanced Segment, sound quality"
After the Zen DAC V2 is powered up for 24 hours [1] by means of a 5V 2A AC-to-DC adapter plugged into the Zen's External Power input, it sounds unbelievably good, with the advantages of analog and digital, and without the disadvantages of either. The source material seems to be the limiting factor, and while playing the United We Swing CD by the Wynton Marsalis Septet, I heard high-end detail that would probably impress any audiophile.
The Zen also has incredible low-level detail, due to the unique design of the TI/Burr-Brown Advanced Segment DAC-chip (ASD) used in the Zen (details below). As I wrote this, I was listening to the Monk Quartet with John Coltrane at Carnegie CD [2], and for the first time heard Monk quietly humming as he played.
After hearing the Zen, I can't stand my previous DAC, a Topping D10s, because I realized that it has no low-level or high-end detail, and produces lifeless "toy" audio with a noisy/harsh, artificially bright high end. This might be due to the Sabre DAC's relatively high jitter sensitivity (DAC output noise vs. clock jitter), combined with a relatively high-jitter clock. Regular delta-sigma DACs apparently require clocks with very low jitter to sound good, and such clocks aren't cheap.
After hearing CDs through the Zen, I realized that they have the POTENTIAL to sound better than LPs. However, some copyright-holders want to retain control over the best versions of their recordings, so they reserve them for LPs since CDs can be copied exactly and easily. But most CDs are good enough for me, and some are fantastic.
High-res sounds no better than CDs, according to High-Resolution Audio: Does it Sound Better? https://goldmund.com/does-high-resolution-audio-sound-better/ by Goldmund Acoustic Laboratory. The higher sampling rates might have helped some ancient DACs with lousy digital interpolation filters, by reducing the need for digital interpolation, but only about 16 of the 24 bits are used for consumer releases. High-res formats were originally intended just for recording and processing.
TI/Burr-Brown Advanced Segment DAC-chips
What is probably the biggest breakthrough in digital audio playback history occurred in January 2001 with the introduction of TI/Burr-Brown model PCM1738 Advanced Segment DAC-chip (ASD). The Zen DAC uses the DSD1793 ASD, which is one of the most recent ASD-models. The ASD architecture (described below) is a more precise implementation of the Current Segment DAC (CSD) architecture, which is used in such models as the PCM1690, an 8-channel DAC-chip which according to Archimago's Musings' test of the Onkyo TX-NR1009 as a DAC, is an impressive performer.
But TI has never given this breakthrough the publicity which it deserves (such as small ads on audio-review websites with an endorsement from an audio expert and a link to a website with details), so that even now few people are aware of the existence of ASDs or CSDs, and even fewer recognize their significance. Their sound quality is typically attributed to their direct conversion of DSD (as if this would help PCM), by using names such as True Native or Bit Perfect. I didn't know about them until I got a Zen in 2023, wanted to know what makes it sounds so good, and investigated.
The attached block diagram and waveform-illustration show the basic idea behind the ASD architecture, which is more properly known as the "Advanced segment DAC interpolated by sigma-delta," for reasons explained below. ("Sigma delta" is interchangeable with "delta sigma.")
The current-segment DAC (CS-DAC) in the block diagram (not to be confused with the current-segment architecture mentioned previously) is the actual DAC which converts digital inputs into analog outputs. It consists of 66 nominally equal current sources which are differentially switched between two outputs (positive and negative). The ICOB-decoder converts Bits 1-6 into a 62-bit "thermometer-code" [3], and the delta-sigma modulator provides 4 such bits, but at 64x the sampling rate of the other 62 bits. When all are bits are off, the output (taken across the differential outputs) is -33 (-2.5 mA), when 33 are on, output is 0, and when all 66 are on, the output is +33 (+2.5 mA). So, it has a total of 67 levels, but since the 4 inputs from the sigma-delta modulator are switched at 64 times the rate of the others, they can represent any value between -2 and +2, with 24 binary-bit resolution, when filtered. This allows the total output to represent any value between +/-2.5 mA with 24-bit resolution.
Not all ASD-models use precisely the same configuration shown in the block diagram. For example, the PCM510xA family uses just a 32-bit sigma-delta modulator to drive the CS-DAC, probably because gear-manufacturers want to win the spec-war. However, the waveform-illustration probably still applies since it represents the fundamental concept behind every variation of the ASD architecture.
For purposes of the waveform-illustration, the ICOB-decoder (see the block diagram) is presumed to have an output of only 7 levels (0 and +/-3), when in reality it has 63 (0 and +/-31), i.e. 6 binary bits of resolution or (2exp6 - 1) levels.
The DWA block is a complex scrambler to randomize the connections between the CS-DAC's data sources (the ICOB-decoder and the sigma-delta modulator) and the DAC's inputs, in order to average-out mismatches between the CS-DAC's segments to obtain 24-bit linearity.
The characteristics of the on-chip digital interpolation filters (DIFs) [4] are also important, but the DIFs in the latest ASD-models sound great, although I'm not certain how they were in the first model, the PCM1738, introduced in 2001. The 1738 was used in some high-end gear, but it might have been combined with external digital filters.
Technical papers on ASDs
The attached waveform-illustration of the basic concept behind the advanced segment DAC is from a 2001 paper entitled "A 126dB D-Range Current-Mode Advanced Segment DAC" by Norio Terada and Shige Nakao of TI-Japan. A 2000 paper, entitled "A 117dB D-Range Current-mode Multi-bit Audio DAC for PCM and DSD Audio Playback" (apparently by the entire ASD development team) also contains useful information.
A potential explanation for TI's cryptic ASD/CSD sales pitch
Every ASD/CSD-model's data sheet includes the statement "[model #] uses ... TI’s advanced segment DAC architecture to achieve excellent dynamic performance and improved tolerance to clock jitter." I'm not entirely certain what they mean by "dynamic performance," but dynamic range specs are listed under that heading in ASD data sheets. It might also have something to do with the ASD's 24-bit linearity and its clean transitions from one sample to the next, with minimal switching- and jitter-related noise in the output, as a result of the CS-DAC's number of levels, its differential nature, and the design of its differential switches.
The 2001 paper includes a jitter-sensitivity analysis which indicates that the jitter sensitivity of a multi-level DAC is approximately proportional to the full-scale range and inversely proportional to the number of levels, which in the case of the ASD's current-segment DAC is 67 levels (0 and 33 levels above and below). The paper doesn't mention the sigma-delta modulator's jitter-sensitivity, but it has five levels, and its amplitude-range at the CS-DAC's output is only about 6% of the total range (4 levels of 66 total), so it probably doesn't make a significant contribution to clock-jitter-induced noise at the ASD's output. I suppose that its small amplitude-range also gives it tighter control over low-level signals compared to a full-range DS-DAC. This, combined with its low clock-jitter-induced noise, would explain the incredibly clean low-level detail I've heard from the Zen.
The 2001 paper also includes a photo of an impressively clean -120dB, 5 mVp-p 1kHz sine-wave, produced by an ASD and displayed on an oscilloscope. The clock-jitter isn't specified, although the paper mentions that 200 pS was typical and 100 pS was about the minimum in audio gear at the time. Commonly-used modern crystal oscillators apparently have about 10 pS. RME's Steady Clock is rated at 0.1 pS.
ASD-endorsements by high-end audio-gear designers
From Bel Canto's chief engineer John Stronczer in the Stereo Times review of the Bel Canto DAC 3.7 :
"At Bel Canto Design, WE STICK WITH THE CLASSIC PCM1792 DAC BECAUSE IT REPRESENTS THE BEST DAC PERFORMANCE ACHIEVABLE TODAY. This holds true for every aspect of the DAC - noise, distortion, dynamic performance and even the way the analog output electronics interfaces with the PCM1792 DAC is superior to other options."
Thorsten Loesch, the head of R&D at AMR/ifi when the Zen DACs were being developed, endorsed ASDs here (see the Reader-view page):
"R2R DACs also have problems with low signal levels [due to linearity errors, which severely distort low-level signals], which is what Delta Sigma DACs avoid. Conversely Delta Sigma (including DSD) technology performs relatively poorly at high signal levels in comparison to multibit solutions.
That’s why we prefer hybrids; multibit architecture for high signal levels, (the so-called upper bits or the MSBs), and high speed (DSD256 or higher) Delta Sigma topology for lower signal levels. The Burr-Brown DSD1793 we’ve been using so much is one such hybrid and one of the best options for us."
According to "MLGrado" on the regular (non-Reader View) version of the aforementioned web-page:
Essentially the BB DSD1793 behaves like an R2R DAC.
[...]
The Burr Brown 'segment DAC' ... was thought by its makers to be A BREAKTHROUGH ADVANCEMENT OVER PURE R2R DACS OR PURE DS DACS. IT AIMS TO GIVE YOU THE BEST OF BOTH WORLDS, WORKING AROUND INHERENT WEAKNESSES. [emphasis added]
So one may indeed consider going to R2R a step backward. In any case, you would not have iFi's TRUE DSD conversion. R2R must convert DSD to PCM.
[end of excerpts]
ASDs are used in everything from Sirius XM receivers, CD players, DVD players, TVs, etc., to extreme high-end DACs. My cheap Vizio TV's "analog" sound quality puzzled me for a long time, but now I realize that it probably uses ASDs. Audio Research's latest DAC, the approximately $8K DAC9, uses the PCM1792A ASD. Bel Canto also uses them in their latest gear. In a Steve Hoffman forum entitled "Enamored by the Burr-Brown PCM1796 DAC chip," someone claimed that a Pioneer DVD player's audio section with ASDs blew the doors off of an expensive discrete R2R DAC. A Zen DAC Signature V2 review on Amazon claims that the Signature version sounds better than a $1500 DAC which uses R2R chips. A lot of companies with a reputation for audiophile sound on a budget, such as Music Hall, Musical Fidelity, Denon, and Onkyo use ASDs in some of their gear. Yamaha's WXAD-10 streaming-DAC uses PCM5121 ASDs and reportedly sounds "superb" and "engaging." Pro-ject uses ASDs in their recently-introduced CD players. The $130 Soundavo HP-DAC1 DAC-amp also uses ASDs and got a rave review on Amazon by a 70-ish vinyl junkie who wasn't impressed by digital previously. So, it's not very difficult to find DACs, streamers, receivers, etc. which incorporate ASDs and sound great. But almost nowhere, except in TI data-sheets for individual ASD-models, are they identified as ASDs, which is their most significant aspect as far as sound quality is concerned.
Regular sigma-delta ADC/DAC chips are used in studio-grade ADCs and DACs, so the standard sigma-delta technology can obviously provide reference-grade sound quality. (There apparently are no ADCs which use ASDs, which might mean that the ASD architecture is incompatible with the requirements of ADCs.) However, studio-grade ADCs and DACs typically cost thousands of dollars, partly because they have clocks with extremely low jitter. The $1300 RME ADI-2, which is popular among recording engineers, has less than 0.1 pS of jitter and sounds great with either AKM or Sabre DAC-chips.
Zen's Output Stage
The Zen's output stage, as far as I can tell from various vague descriptions, consists of an OV2637A amplifier module, and associated circuitry to configure the module as an I-V converter and anti-aliasing filter in the Zen DAC. The module includes four single-ended amplifiers, each consisting of a JFET-input op-amp followed by a buffer such as a TI BUF634A (which can drive headphones), and each of which is used as one side of a differential pair, forming two differential pairs. It takes the form of a tiny circuit board with pins protruding from one end, and chips attached directly to it. Some of these chips are surface-mount devices, and others are bonded in place and connected to circuit-board traces via wire-bonds. The wire bonds and associated chips have to be protected, such as by epoxy encapsulation or with a cap bonded to the circuit board.
To provide sufficient supply voltages for the op-amps and buffers, the Zen has a voltage-doubler (a flip-flop, diodes, and caps), and the resulting dirty 10V supply is regulated down to about 9V, which is capacitively bypassed to provide high instantaneous current capacity. With a 9V supply, the maximum single-ended output would be about 8.5Vp-p, and the maximum differential output would be about 17Vp-p, which is a hefty signal from USB power or a 5V wall-wart. The Zen has single-ended and differential outputs for headphones and amps, so it's basically pro-grade.
Zen shows off good CDs and exposes flaws in bad ones
Esperanza Spalding's Radio Music Society CD didn't particularly thrill me when I listened to it through my delta-sigma-based DACs. But when I played it through the Zen DAC, I was amazed by the CD's sound quality, and mesmerized by the intricate music and Spalding's enchanting voice.
Another great CD is Alive in America by Steely Dan, whose main recording engineer Roger Nichols used it as an opportunity to prove how good he could make a CD sound, since it's a live CD and he didn't have to reserve the best version for LPs. So, he recorded and mixed it in analog, digitized the mixer's output with an Apogee AD-1000 (20 bits, 44.1 kHz), and used an Apogee UV-22 to convert the 20-bit data to 16-bit data that sounds like 20-bit data. Sign In Stranger sounds particularly good.
However, the Zen's excellent low-level detail is a two-edged sword, because it makes any flaw stand out. For example, the high end of my Kamakiriad CD (Reprise 9 45230-2) vaguely bothered me through the D10s, but through the Zen is obviously deliberately degraded. The LP probably has a great high end, and perhaps a more recent CD-release has a better high end.
Miscellaneous audio information
There is no need for precision CDs or expensive transports, due to the error-correction system built into the CD record/playback system. Precision CDs might contain better recordings than the ones used for regular CDs, because the superior sound quality can be attributed to the physical precision. But if the copyright holder reserves the best versions for LPs, no digital release will sound as good as the LP, although MP3's might be made from the best version since they use lossy compression.
Separate CD-transports cannot possibly reduce the CORRECTED error rate compared to that obtained from a cheap transport in a CD player, which is a few ppB for CDs in decent shape. If you doubt that CD read-errors can actually be corrected, explain how software can be stored on CDs (simply re-reading the erroneous data won't work). Using separate transports just adds compromises or cost to the task of synchronizing the transport's clock and the DAC-chip's clock. Ideally, the clock should be next to the DAC-chip's convert-command pin, to minimize jitter in the DAC's convert-command, and the clock's signal would be sent to the transport, which can tolerate higher levels of jitter. Any buffer-memories between the transport and the DAC-chip could also tolerate a higher level of jitter, as long as the DAC's data-inputs can tolerate the jitter in the buffer's output.
CD treatments don't reduce the corrected error-rate - they reduce laser-servo hunting, and thus reduce laser-servo-induced noise on the transport's power supply. The only CD players to benefit from CD-treatments were cheap early players with poor isolation between the transport power supply and the analog power supply, which allowed some of the laser-servo noise to get into the output.
Software is stored on CDs & DVDs, and it can't have any errors, so it's obviously possible to truly correct CD read-errors. Software has an extra layer of correction compared to audio, but it's rarely needed.
Some CDs, although very few, have inverted audio polarity, probably deliberately, in an attempt to prevent us from being able to rip the best version. So, if a CD just doesn't sound "right," try flipping the polarity. Flipping polarity doesn't make much difference if a recording is incoherent or has inconsistent polarities within it, or if the playback gear is incoherent. Examples of CDs with inverted polarity are Night by John Abercrombie, the CD layer of Michael Brecker's Pilgrimage SACD, and Bartok String Quartets 1-6 by the Alban Berg Quartet. Audio polarity can be inverted by flipping the polarity of each speaker connection, assuming that the speakers are passive, or by ripping the CD and using an audio-editing program to invert each file.
Some high-end gear, such as Audio Research preamps and Benchmark DACs, have polarity/invert switches, as does all sorts of pro-grade gear. Some DAC-chips, including some ASDs, have invert-functions. The fact that so little consumer-grade digital playback gear has polarity-control capability, although it would be cheap and easy to add it, supports my suspicion that inverted polarity is being used as a means of degrading the sound quality of some CDs.
To get the best sound from your system, periodically disconnect and reconnect all connections, including internal connections, AC connections and breakers (turn off everything that uses a significant amount of power before cycling breakers). Also use canned air to blow dust out of ventilated components, because dust acts as a parasitic circuit, especially in high humidity. In severe cases, it might help to scrub with alcohol and something like a toothbrush around components to eliminate all parasitic paths.
The optimal turntable configuration is belt-drive with a pivoting arm with a fixed head-shell. Direct drives have a feedback loop and their speed oscillates around the ideal speed, and tangential tonearms are a scam. Use a record cleaner such as a Spin-Clean, even if you have to scrimp on the TT or exceed your budget, because a really clean record sounds much better, and you can't undo record wear.
Notes
[1] Experts recommend leaving digital playback devices powered up all the time (unless they use a lot of power or have tubes which can't be shut down independently of the rest of the circuitry), and letting them warm up for at least 24 hours before judging them. For example, see Audioquest's technical paper entitled Evaluation of Digital Devices.
Any functional 5V 2A wall-wart with a 5.5mm OD/2.1mm ID barrel connector output can be used as a power source for the Zen. (See Prodigit's page entitled "How to test Noise from the Output of Power Supply" ) Some claim that higher-quality supplies improve the Zen's sound quality further, which I suppose might be possible.
When power is applied to the Zen's External Power input, the LED next to the input glows, and the USB connection isn't used for power. The LEDs on front indicate whether the USB input is active.
[2] The recording on the Monk Quartet at Carnegie CD was made on 11/29/57, using tube gear. ("Tube sound" strikes me as very coherent and alive. According to the VTL site, tubes sound better than other devices because they have a more linear gain, and need less feedback.) The tape was poorly labeled, put in storage, and forgotten, until it was discovered 48 years later in 2005, and digitized and digitally processed (see Recently Discovered Jazz Jewel Restored by Hip-Hop Legend). The resulting sound quality is amazing considering the age of the tape when it was digitized, and the performance was excellent.
[3] In a thermometer-code, each single-level increase is achieved by simply activating another bit, so that as the signal-level increases, more bits turn on, and none off, and as the signal level decreases, more bits turn off, and none on. Likewise, in a liquid thermometer, the liquid rises and falls with temperature, with no gaps.
[4] DIFs basically calculate samples to be inserted between the samples provided by a CD, for example, thus reducing the size of the steps in the "staircase" waveform at the output of the DAC, so that a simple, clean, and cheap high-frequency analog filter which doesn't affect the audio range can completely filter/smooth out the staircase.