"Myths About Measurements" Discussion Thread

Audio Science
143

Nuance does matter. Your post above seems to suggest that there may be no loss in dynamic range or bits because DSP’s like EAPO can utilize 32-bit floating point processing internally.

I’ll go out on a limb without really much research into this, and suggest that the internal processing in EAPO isn’t what’s relevant in this situation. It’s the bit depth of the audio device that matters. If your audio device is set to 16 bits fixed point on your PC, and EAPO drops the levels -6 dB, you still lose dynamic range and useful bits (or bit) when audio leaves your PC at that resolution.

You can’t avoid some loss in this situation, even though it may be minor from the standpoint of audibility with some dithering noise added to the reduced bit depth 16-bit signal.

144

Hmm. My Cenever DAC has a 32-bit digital volume control. The designer states that with a 32-bit volume control, there should be no actual loss in resolution or dynamics. It’s also unique in that the level from the DAC and the headphone amp gain can be both set individually (separate gain levels for each).

145

Editorial Comment: I’m baffled as to why you focus on the minutiae of potential digital processing issues as you continue to play music through mid-fi drivers. The drivers have a much bigger impact on quality and “a premier listening experience” than this stuff.

2 Likes
146

Better drivers would help, if they eliminate the need for EQ. Not everyone can afford a pair of $1k to $2k headphones though. And if you know a little more about of the ins and outs of properly gain-staging, impedance bridging, and configuring your gear, there’s quite a bit you can probably do to improve your results without spending another nickel. You might even find you could save some money, and still get some very decent high quality results.

But you are correct that buying headphones with a better frequency response to begin with could eliminate some of the headaches that I described above associated with EQ. Better drivers will not make up for a loss in fidelity on the front end due to using incorrect source components, audio adjustments for your content, or other factors though.

These are my opinions of course, and some others may disagree.

147

Interesting. I’m not familiar with this gear. Maybe it’s designed to control the levels on a pair of headphones and powered monitors at the same time?

148

I wholeheartedly agree that “start from the source” results in the best quality final output. Both my DAC and amp have transformed various headphones. Sometimes apparent headphone flaws turned out to be either DAC digital artifacts or amp power transformer issues. Still, there’s a fundamentally different experience on any amp/DAC with the Focal Utopia or HD 800 S over the Focal Clear, and the Clear over the HD 600.

What you are doing with digital tweaks might be worse than nothing. Software tools make assumptions about playback that may not be correct, they can have coding flaws, and the quirks of various processes may stack in a bad way. Many products seek to improve/smooth the signal. Product #1: “Let’s do X to smooth the output.” #2: “Let’s do Y to smooth the output.” #3: “Let’s do Z to smooth the output.” Then…?

Beyond digital tweaks, many expensive real world setups I’ve demoed had fancy cables, a power conditioner, more fancy cables, a creamy Chord DAC, more fancy cables, a tube amp, and then smooth sounding speakers/headphones. A “good result” here is audiophile mush – the electrical signal cannot generate any sharp edges, but it’s then demonstrated with smooth female vocalists over piano or strings and everything sounds samey same.

More than a few audio hobbyists argue for minimal signal chains and resist DSP, EQ, etc. When taken to the extreme, analog purists focus on the primacy of vinyl records, and some even buy $35,000 record player tone arms, etc.

Watch out for both the measurement ideology and the purism ideology, and listen to actual setup variations. If you are failing blind tests and they are samey same, then it’s all for nothing.

2 Likes
149

My post doesn’t imply you won’t lose dynamic range/headroom. You will lose headroom due to reducing the signal gain. Will you lose bits and quality due to that, though? No, I don’t think so.

150

@Nuance always matters to me. Never forget him :star_struck: :saluting_face:

3 Likes
151

Awww. And that’s why you’re my favorite! :grin:

152

I agree Nuance matters. But don’t agree that there is no loss. :slight_smile:

Your post seemed to imply that there would be little or no loss due to EAPO’s 32-bit floating point internal processing, and dithering in the final output. There will still be some loss in both useful bits and dynamic range though when the attenuated signal is output to your audio device at 16-bits fixed point with dithering noise added.

The simple solution is to increase the bit depth of your audio device to mitigate that loss, rather than keeping it at a native depth of 16-bits. Since the audio has to be resampled and re-rendered, there’s no real benefit to keeping it at its native depth imho. YMMV of course.

153

This is possible. EQ has both upsides and downsides imo. I think there are probably more upsides in my current situation than downsides though. That might change with a different pair of headphones, amp, and DAC though.

154

As previously, you are playing around in the margins unless you upgrade your drivers. I spent years and years tweaking my HD 600 and struggling with often scratchy treble, but my experience changed overnight with the OG Focal Clear. I used the same amps and DACs. The HD 600 is a fine product and it responds well to tweaks, but it can never match better products.

3 Likes
155

Since no one replied about this aspect of PEQ I decided to ask ChatGPT:
Short answer: with a properly implemented PEQ, the math itself doesn’t create harmonic distortion. A parametric EQ is a linear time-invariant (LTI) filter; linear systems don’t generate new frequencies (no added harmonics or IMD). If you hear “distortion,” it’s almost always from side-effects or implementation details—not the EQ concept.

Here are the real, math/implementation-related gotchas to watch:

Summary
  1. Coefficient quantization & rounding
  • Digital filters use quantized coefficients. In 32-bit float, the error is vanishingly small (way below audibility).
  • In fixed-point or very low-precision paths, extreme Q or many cascaded filters can raise the noise floor slightly or deviate from the intended curve.
    :white_check_mark: Use a quality EQ with 32- or 64-bit internal processing; avoid silly-high Q unless you’re notching a narrow resonance.
  1. Internal overload in IIR structures
  • Even if your input/output don’t clip, some IIR forms (e.g., naive Direct Form I/II) can overload internally at high boosts/Q. That can sound like distortion.
    :white_check_mark: Good EQs use second-order sections (SOS) with stable topologies (often Transposed Direct Form II) and high-precision accumulators. Your “proper pregain” habit already helps a lot.
  1. Time-variance “zipper noise” (when changing parameters)
  • Rapidly moving EQ knobs without smoothing turns the filter into a time-varying system → subtle modulation artifacts.
    :white_check_mark: Choose EQs with parameter smoothing (most good ones have it) and avoid automating big jumps at audio rate.
  1. Ringing/overshoot (not nonlinear distortion)
  • High-Q filters ring in the time domain; linear-phase FIR can add pre-ringing, minimum-phase IIR adds post-ringing. That’s a time-domain artifact, not harmonic distortion, but it can be audible on sharp transients if your filters are very narrow or steep.
    :white_check_mark: Keep Q sensible for music playback; use narrow notches only when necessary.
  1. Near-Nyquist weirdness & SRC side effects
  • Very steep moves near Nyquist (e.g., big boosts above ~18–20 kHz at 44.1/48 kHz) can behave oddly and amplify ultrasonics/quantization noise.
  • If your EQ host resamples (e.g., OS mixer), poor sample-rate conversion can add artifacts unrelated to the EQ math.
    :white_check_mark: Don’t chase >18 kHz boosts; run your chain at a stable, native rate with good SRC if resampling is unavoidable.
  1. Intersample peaks (not from EQ itself, but revealed by it)
  • Boosts can create true-peak excursions above 0 dBFS after reconstruction, even when samples look safe, leading to DAC-side clipping.
    :white_check_mark: Your pregain practice is the correct fix; add 1–2 dB extra headroom if you use multiple boosts.

Bottom line: with minimum-phase IIR PEQ, sufficient headroom, sensible Q/boost values, and a modern 32/64-bit implementation, there’s no added harmonic/IMD “distortion” from the mathematics—only potential numerical noise or time-domain ringing if you push extremes.

Here’s how to bullet-proof PEQ in Roon so the DSP math never bites you:

Summary

1) Run plenty of headroom and watch for clipping

  • Open MUSE → Headroom Management, set –3 to –6 dB, and enable the Clipping Indicator. Play your hottest tracks and nudge headroom until the red dot never appears. This prevents intersample/true-peak trouble down the line. (Roon Labs Help Center, Roon Labs Community)

2) Trust Roon’s internal precision + dither

  • Roon’s modern DSP path runs 64-bit floating-point throughout, so coefficient quantization/rounding isn’t a practical limiter. When Roon reduces bit-depth on output, it applies TPDF dither, which avoids truncation distortion. (Roon Labs Community)

3) Keep filters sane (you already do)

  • Roon’s Parametric EQ in MUSE is stable and precise; use a few well-chosen filters, avoid extreme boosts, and keep very high-Q notches for true problems only. (You can also place PEQ inside Procedural EQ if you ever need per-channel work.) (Roon Labs Help Center)

4) Sample-rate conversion (only if you use it)

  • If you up/downsample, Roon’s SRC offers minimum-phase and linear-phase options with apodizing, polyphase interpolation to minimize ringing. Pick linear-phase if you want symmetric transient behavior; choose minimum-phase if you prefer no pre-ringing. Or leave SRC off. (Roon Labs Help Center, Archimago)

5) Practical PEQ recipe for headphones in Roon

  • Headroom: start at –6 dB, then back off toward –3 dB if the indicator never trips. (Roon Labs Help Center)
  • PEQ: apply your minimum-phase filters; try to keep boosts ≤ 4–5 dB and Q ≤ 5 unless fixing a narrow resonance. (Roon Labs Help Center)
  • Order: Headroom first, then PEQ; leave volume leveling/convolution after if you use them. (All run in the same 64-bit engine, so it’s robust either way.) (Roon Labs Community)

If you want, tell me the exact headphone and your target (e.g., Harman with a gentle 1–2 dB bass tilt). I can hand you a ready-to-paste Roon PEQ preset with sensible headroom.

2 Likes
156

A loss of what? Dynamic range? Bits? Quality, as in an audible negative impact on sound quality?

Here’s Chat GPT’s input (I’m not saying Chat GPT is 100% accurate on this):

  1. Inside the Digital Domain

If you’re working with integer PCM audio (e.g., 16-bit or 24-bit WAV):
• Turning volume down is just multiplying each sample by a factor less than 1.
• This reduces the maximum amplitude but also reduces the signal-to-quantization-noise ratio (SNR) because the same quantization noise floor is now closer to your audio.
• In 16-bit audio, this is rarely audible unless you lower the gain drastically (e.g., -40 dB or more) and then re-amplify later.
• In 24-bit audio, the noise floor is so low that even large reductions won’t matter in practice.

If you’re in floating-point (e.g., 32-bit float inside a DAW):
• Volume changes are mathematically lossless until you export to integer format.
• You can turn gain up/down repeatedly without losing quality — rounding only happens when converting to fixed bit depth for playback/export.

⸻

  1. On Playback Devices

If you adjust digital volume in software before sending it to a DAC (digital-to-analog converter):
• You’re lowering the amplitude before conversion, so the DAC uses fewer bits effectively.
• Good modern DACs with 24-bit depth have more than enough dynamic range to handle this without audible degradation, unless you’re pushing volume extremely low (e.g., background whisper levels).

If you adjust analog volume after the DAC:
• The digital signal stays at full resolution; no digital quality loss occurs.
• However, analog circuits introduce their own noise and distortion, so the quality depends on the amplifier design.

⸻

  1. When Problems Happen

You can cause quality issues if:
• You reduce gain in a low-bit-depth format (like 8-bit audio) — noise becomes obvious fast.
• You lower gain a lot in 16-bit audio and then boost it again later — the quantization noise floor gets amplified.
• You’re working in a poorly implemented playback pipeline where digital scaling adds rounding errors.

⸻

:white_check_mark: Rule of thumb:
• In a professional workflow (24-bit or 32-bit float), you can digitally reduce gain without worrying about degradation.
• In 16-bit, moderate reductions (e.g., a few dB) are fine, but big reductions followed by re-amplification can expose noise.
• If you’re only lowering volume for playback and not re-amplifying later, you won’t hear a loss in quality with any modern setup.

⸻

2 Likes
157

Fwiw, I think most of the above is correct and consistent with what I’ve been saying all along.

The only assertion I might question is the idea that the reductions in volume have to be on the order of -40 dB or more before quantization noise is audible. Note that there is loss before you reach that point. The question is just whether it’s audible or perceivable to a normal person. And here’s what I’ve said about that previously…

Without the benefits of dithering and noise shaping, 16 bit audio has a dynamic range of about 96 dB (16 bits x 6 dB). If you subtract 40 dB from that, the dynamic range is only 56 dB, which is well below the maximum dynamic range of human hearing of approximately 120 dB. A good dithering and noise shaping routine might increase that 56 dB perceptually by another 15 or 20 dB. That’s not enough to get you to the max dynamic range of 120 dB (threshold of pain). But it might be sufficient if you listen at lower volumes, or listen mostly to audio with alot of DRC. Or listen in noisy environments… But then again, it might not (especially if you listen at more normal or louder volumes).

If you are rendering your audio in 16 bits, then a drop of -40 dB in volume also means the useful bits are reduced to about 9. Or about 512 possible amplitude levels for each sample, versus the 65,536 that the 16-bit audio content began with. That’s a significant reduction in quality.

Dithering can’t increase the number of actual amplitude levels or bits. All it can do is give the perception or impression of more levels through the introduction of noise.

158

While your math seems correct, I honestly don’t know why we’re still discussing this.

  1. None of the EQ systems I have seen process the data in 16 bit depth
  2. None of the PEQ presets I have seen have needed more than 12db of negative pregain (usually 6db or less)
  3. None of the music recordings I have seen have a dynamic range exceeding 40db (usually 14db or less)

So while yes in theory it can degrade quality, in practice it is a non-issue.

6 Likes
159

How many angels can dance on the head of a pin?

1 Like
160

I mean this isn’t just purely academic or theoretical considerations, this is like… the core of the argument - if I’ve understood the intended application of this phrase appropriately, which may entirely not be the case.

Some people have (in my view erroneously) expressed the sentiment that EQ necessarily degrades sound quality, so you shouldn’t do it at all. And while there are scenarios where it could degrade sound quality (some of which we’ve highlighted a number of times), this is about doing EQ in practice.

In practice, the vast majority of the time, EQ does not involve any detriment to the sound provided it’s done correctly. Moreover, in cases where people do report a degradation in sound quality, there’s always an explanation, which can come in the form of non-leveled matched comparisons for listening tests, incorrect filter adjustments for the in-situ response, incorrect application of the EQ (incurring clipping), making use of profiles rather than doing it based on what you actually prefer.

And yeah, there are a lot of landmines a person could walk into when it comes to EQ, but it’s shocking what you can improve with just a few bands of parametric adjustment. While there are absolutely limits to EQ, one of the great things about headphones is just how much of a cheat code EQ is to getting great sound. And it’s usually free!

So maybe what you’re meaning here is that there’s no reason to care about something if nobody can tell the difference, particularly when there are other things that do make a meaningful difference, like how you can improve the FR with EQ. If that’s the case, I’m with you there.

1 Like
161

“Angels dancing on the head of a pin” was once a common interjection when people discussed academic, esoteric, or down-in-the-weeds topics. I grew up in a retro culture where the expression had currency. In my experience, it never went much farther than a quick brush off. Know-it-alls would then say “Akschully, that was an important theological topic back then.”

It was apropos upon seeing an involved thesis brushed aside with few words. As with @AudioTool, I too am puzzled by why @ADU spends so much time on various topics. Enthusiasm is fine, but perspective and real world testing makes for meaning.

Yes, I can agree with your extrapolation that there is no reason to care about any inaudible stuff.

5 Likes
162

OMG, you had to explain.

I must be getting older and wiser.

2 Likes