The worst distortion gets hidden exactly where standard testing never thinks to look.
DAC buyers are trained to look for a few familiar numbers, especially THD+N and signal-to-noise ratio. John Siau, Director of Engineering at Benchmark Media Systems, argues that those figures can leave out the part of a ladder DAC’s behavior that matters most.
His concern is not simply that R2R designs produce distortion, though. He says standard tests can miss where the largest errors occur, allowing a converter to look cleaner on paper than it performs across real signal levels.
In fact, a documented Philips design shows that the mechanism itself is real. So, the bigger question is how far that evidence can be applied to modern R2R DACs.
To answer that, we looked at the measurements, the architecture, and the limits of Siau’s claim.
How a DC Offset Moves the Major Transition
In a conventional binary ladder DAC, the most demanding code change often occurs as the waveform crosses zero. Many bits switch state at once, allowing small resistor mismatches to combine into a larger linearity error at that transition.
However, this behavior is not inherent to every R2R design. Sign-magnitude, segmented, factory-trimmed, calibrated, and actively compensated architectures can reduce or avoid the same all-bits-at-once transition. Siau’s criticism applies most directly to conventional ladders that retain it.
That’s why a standard THD+N measurement may not tell the whole story because it usually tests one sine tone under a fixed set of conditions rather than mapping the DAC’s accuracy across its full range of output levels.
According to Siau, some ladder-DAC designers add a small digital offset that moves the major transition away from zero, where the resulting error can be masked by a stronger part of the signal.
A historical example and the limits of the evidence
The idea is not merely theoretical. A Stereophile technical account describes how Philips’ SAA7220P/B digital filter added a small value to each digital word before sending it to the TDA1541A ladder DAC.
Used in products including the Marantz CD12/CDA12 and Philips LHH1000, the system shifted the major transition from zero to roughly -63 dB, where the surrounding signal could better mask the error.
But while this provides a documented commercial example of the mechanism Siau describes, it does not prove that current R2R manufacturers use the method to improve their published specifications. Philips presented it as a treatment for zero-crossing distortion, and no named modern product has been independently shown to use an offset specifically to produce a better THD+N figure.
This is why a single headline measurement is insufficient. Low-level linearity sweeps show whether conversion accuracy changes with signal level, while IMD and multitone tests reveal nonlinear interactions under more complex signals.
The 16-Bit Ceiling
For Siau, the major-transition error is only part of the problem. A conventional, uncompensated R2R ladder also depends on its resistor ratios remaining accurate across thousands of output steps. So reaching 16-bit linearity requires precision of roughly one part in 65,536, which he argues is already near the practical limit of passive component matching.
Still, his argument becomes less absolute once the resistor ladder is treated as one part of a complete converter. Factory trimming, segmentation, calibration, and active correction can compensate for errors that passive matching alone cannot prevent. A finished DAC may therefore achieve better system-level linearity than its raw resistor tolerances would suggest.
What makes this harder to spot is that R2R DACs are genuinely quiet when they’re not playing music. With no signal passing through the resistor ladder, there’s nothing to mismatch, so noise floor measurements and signal-to-noise ratios can look excellent.
That’s why temperature still matters because the ratios can shift as components warm, though the result depends on how closely the resistors track one another and how the design manages heat and compensation. Some converters may show measurable changes after reaching operating temperature, while others are built to keep those changes small.
The same distinction helps explain why an R2R DAC can post an excellent noise floor without proving equally strong linearity. At a fixed output code, the ladder is not repeatedly crossing the transitions that expose signal-dependent errors. Once the music begins moving through low-level codes, those inaccuracies may become easier to measure.
Therefore, calling the result “16-bit performance” can be misleading. Noise floor, usable resolution, linearity, monotonicity, and distortion describe different parts of a converter’s behavior, and no single one of them defines its overall accuracy.
The Sigma-Delta Problem Siau Says Is Often Misdiagnosed
Siau applies the same scrutiny to sigma-delta DACs, particularly the harshness often described as “digital glare.” He argues that much of it comes from inadequate reconstruction headroom rather than the conversion architecture itself, estimating that 80% to 90% of DACs are vulnerable to intersample clipping.
The thing is, every stored sample can remain at or below 0 dBFS while the continuous waveform reconstructed between those samples rises higher. These intersample peaks can exceed full scale by as much as 3.01 dB.
So, when a DAC provides no room above 0 dBFS before interpolation, the reconstructed peak is clipped before it reaches the analog output.
Benchmark also found the issue repeatedly in commercially mastered music. Its analysis of Steely Dan’s “Gaslighting Abbie” counted 1,129 intersample overs across a little more than five minutes, or roughly 3.7 per second.
By Siau’s account, the fix requires only 3 dB of headroom before the interpolation stage. Drop the signal level before it reaches the DAC chip, and the between-sample peaks have room to reconstruct cleanly.
If that’s all it takes, the migration toward R2R DACs starts to look like a mass reaction to a problem that needed three decibels, not a different architecture.
Where Siau’s Case Breaks Down
The strongest parts of the argument apply to conventional, uncompensated binary ladders. But once trimming, calibration, alternative architectures, and independent measurements enter the picture, the claim of a universal R2R limitation becomes much harder to defend.
The 16-bit ceiling is too absolute
Describing performance beyond 16 bits as “physically impossible” makes sense only when discussing raw resistor matching in isolation. A complete converter can correct those physical errors through trimming, calibration, segmentation, and other compensation methods.
One commercial example is the Analog Devices AD5791, a 20-bit R2R DAC specified for ±1 LSB integral nonlinearity, guaranteed monotonicity, and temperature drift of 0.05 ppm/°C.
Those figures do not guarantee perfect 20-bit audio performance in every product built around the chip, but they show that system-level accuracy can exceed what uncompensated resistor tolerances would allow.
Independent audio measurements make a single 16-bit ceiling even more difficult to defend.
During Stereophile’s testing of the Denafrips Terminator, switching from 16-bit to 24-bit source data lowered the noise floor by about 35 dB, implying resolution close to 22 bits. Its linearity sweep also produced an unusual pattern, with the error alternating by approximately +1 dB and -1 dB at lower levels.
As a result, reviewer John Atkinson could not determine what that behavior meant for sound quality.
Substantially stronger linearity results from the Holo Audio May add another complication, although the manufacturer has not publicly explained whether active correction contributes to its performance. Together, these examples show why noise floor, usable resolution, linearity, monotonicity, and distortion cannot be reduced to one universal “bit performance” figure.
Nor does a measurable error automatically establish an audible defect.
Its significance depends on the distortion’s level and spectrum, the music, playback volume, listening environment, and the rest of the signal chain. None of the measurements cited here proves that listeners can reliably identify these specific R2R errors under controlled conditions.
The major-transition problem is not universal
Whether the zero-crossing criticism applies also depends on how the ladder is organized. Sign-magnitude designs handle positive and negative signal values separately, avoiding the same all-bits-at-once transition around bipolar zero found in a conventional binary ladder.
For instance, Texas Instruments used this architecture in the PCM1704 and added factory laser trimming to reduce differential-linearity and gain errors.
Designers such as Soekris follow a similar principle in discrete ladders, while segmented and actively compensated designs manage transition errors through other methods.
None of these approaches guarantees perfect conversion. They do show that the major-transition problem cannot be treated as an inherent weakness in every R2R DAC. But the evidence supports a narrower conclusion: conventional, uncompensated binary ladders can suffer meaningful matching and transition errors, but the same weakness has not been established across most current R2R implementations.
Mosquito Farts and Jet Engines
Once the universal claims are stripped away, the dispute becomes less about whether measurements matter and more about what listeners expect from a DAC.
For instance, Benchmark treats transparency as the goal, while some buyers choose R2R converters because they prefer the way those designs present music, even when measurements show a departure from the source.
Preference does not make the engineering irrelevant, though. It simply means that a measurable error and an undesirable result are not automatically the same thing. Plus, the opposite problem also applies: an excellent THD+N or signal-to-noise figure cannot prove that a converter remains equally accurate across quieter signals, complex tones, or reconstructed peaks above 0 dBFS.
That’s why the architecture label tells buyers less than the completed design does. Low-level linearity shows whether accuracy holds as the signal falls, while IMD and multitone testing can reveal nonlinear behavior that a single sine wave misses.
So the useful question is not whether R2R or sigma-delta is inherently better. It is whether the finished converter performs consistently across the tests that expose its particular weaknesses, and whether any measured errors are significant enough to matter in use.
Mike Moffat, the Schiit Audio co-founder who helped design one of the earliest standalone DACs, offered a blunter description of that final judgment.