LDAC is certified Hi-Res at 990kbps ... Which lossy codecs could meet or exceed the same standard?

[SivaKM]

I tested with REW using AAC, MP3 and Vorbis codecs (qaac, neroaac, mp3lame, libvorbis) at high quality settings with 48kHz, 64kHz and 96kHz sample rates (powershell script attached).

1) Sweeps
a. Generate REW mono measurement sweep from 20Hz to 36000Hz and save to 32bit 352kHz pcm wav including timing references.
b. Resample with sox to target sample rates 48kHz, 64kHz, 96kHz using high quality settings.
c. Encode with mp3, qaac, neroaac, vorbis at high quality settings i.e. mp3lame cbr 320, qaac tvbr 127, neroaac q1, vorbis q10.
d. Resample the results with sox to 32bit 352kHz pcm wav using high quality settings.
e. Load into REW as measurement sweeps.

2) White Noise
a. Generate REW full range white noise and save to 32bit 352kHz pcm wav.
b. Load into REW as RTA measurement taking the peak reading.

The Sweep distortion results on the original file seem to show a limitation of 24bit processing with readings at about 148dB and better i.e. 144dB SNR from 24bits and ~6dB from noise shaping.

Sweeps indicate that MP3 has very low distortion at 0.0035 (89dB) but the white noise check shows that MP3 makes use of adjacent frequencies unlike other codecs which vary the original frequencies slightly up or down.

For me the best option is Nero AAC 96kHz with a distortion peak of 0.02 (74dB) at ~1222Hz (the peak seems like a design choice as the remainder is considerably lower). Nero AAC 64Khz isn't that much different though.

LDAC comparison is problematic. One method is playing the measurement sweep file to a BT receiver which sends an optical output to a PC and saving to file. The sender would use 24bits 96Khz 990kbps LDAC and maybe 32bits 48kHz or other combos. Another way would be using the github libraries ldacenc and libldacdec.

REW_MDAT

MediaFire is a simple to use free service that lets you put all your photos, documents, music, and video in a single place so you can access them anywhere and share them everywhere.

www.mediafire.com

 

[voodooless]

LDAC comparison is problematic

They are all problematic. Lossy audio codecs are not meant to encode frequency sweeps, let alone white noise. Neither of these will give you a good indication of the actual ability to transparently encode music.

 

[Vincent Kars]

LDAC comparison

Try this one: https://archimago.blogspot.com/2023/08/part-i-comparison-of-bluetooth-fidelity.html

 

[SivaKM]

They are all problematic. Lossy audio codecs are not meant to encode frequency sweeps, let alone white noise. Neither of these will give you a good indication of the actual ability to transparently encode music.

Through LDAC it has been established that sub 1000kbps codecs can perform well enough to receive HiRes certification.

Transparency would involve a SNR of say 96dB (CD) or better. The bar seems a lot lower with lossy codecs e.g. 74dB.

 

[SivaKM]

Try this one: https://archimago.blogspot.com/2023/08/part-i-comparison-of-bluetooth-fidelity.html

Thanks, I've seen LDAC testing with 1kHz and multitones but hadn't considered reproducing them with the various audio codecs as I don't have the equipment.

But looking at the testing of Topping BC3 it appears that LDAC has distortion better than 100dB. Perhaps this methodology could improve results for the various lossy codecs.

Topping BC3 Review (Bluetooth Receiver) & BT CODECs

This is a review and detailed measurements of the Topping BC3 Bluetooth receiver and headphone amplifier. It was sent to me by the online seller, Yaoyaotiger HIFI Audio Store and costs US $69. This is a follow up to our previous tests of Bluetooth receivers and investigating performance of...

www.audiosciencereview.com

 

[EddNog]

Isn't LHDC the only true competitor to LDAC?

-Ed

 

[voodooless]

Transparency would involve a SNR of say 96dB (CD) or better. The bar seems a lot lower with lossy codecs e.g. 74dB.

No, transparency would involve not being able to distinguish the encoded version from the original.

 

[danadam]

can perform well enough to receive HiRes certification.

That can mean anything. What are the exact criteria to get it?

 

[voodooless]

That can mean anything. What are the exact criteria to get it?

I just checked, it basically involves being able to reproduce audio up to at least 40 kHz… whatever that means. I would assume encoding bits could better be spent encoding content people can actually hear. High-res certification is marketing nonsense.

 

[Vincent Kars]

That can mean anything. What are the exact criteria to get it?

Japan Audio Society | Hi-Res Audio Logo

オーディオの普及に向けた活動や情報提供を行い、「オーディオ・ホームシアター展」等のイベントの開催など、オーディオに関わるイベントを主宰する日本オーディオ協会の公式WEBサイト

www.jas-audio.or.jp

 

[SivaKM]

Japan Audio Society | Hi-Res Audio Logo

オーディオの普及に向けた活動や情報提供を行い、「オーディオ・ホームシアター展」等のイベントの開催など、オーディオに関わるイベントを主宰する日本オーディオ協会の公式WEBサイト

www.jas-audio.or.jp

Summarizing ...
a) For analog 40kHz+ capable, For digital 24bit/96kHz
b) For all "Listening Tests"
c) The following wireless codecs have been cerified Hi-Res: LDAC, LHDC, SCL6, LC3plus, SHDC, aptX Adaptive

MP3 (lame cbr 320) has a 20kHz cutoff limit
Neroaac (q1 96kHz) has a 30kHz cutoff limit with distortion at 74dB (0.02%) for Mids approaching 87dB otherwise
Vorbis (q10 96kHz/192kHz) has configurable cutoff or none with distortion at 60dB (0.1%) for Bass, Low Mids approaching 96dB otherwise

So none of them qualify but I'd say Neroaac comes the closest, with 0.02% distortion being acceptable.

 

[restorer-john]

Japan Audio Society | Hi-Res Audio Logo

オーディオの普及に向けた活動や情報提供を行い、「オーディオ・ホームシアター展」等のイベントの開催など、オーディオに関わるイベントを主宰する日本オーディオ協会の公式WEBサイト

www.jas-audio.or.jp

With no limits to actual measured/minimum performance? LOL.

 

[SivaKM]

I just checked, it basically involves being able to reproduce audio up to at least 40 kHz… whatever that means. I would assume encoding bits could better be spent encoding content people can actually hear. High-res certification is marketing nonsense.

We have the Nyquist factor of sampling at twice a signals bandwidth (for audio 44.1kHz or 48kHz is considered more than enough) and we have bit depth which defines the accuracy of the sampling. With infinite bit depth we would have a perfect signal when sampling at twice the bandwidth of the material. But we don't have perfectly bandwidth limited material (as analog has infinite bandwidth) so there is going to be extra information forming distortion (aliasing) i.e. we need infinite sampling to have a perfect signal.

We don't have infinite bit depth or infinite sampling so we compromise (starting with 16 bits 44.1kHz CD-quality) and rely on digital methods to retain audio quality (dither, noise shaping, filtering). We can measure distortion and phase-shifting to see how well we're doing and adjust our efforts accordingly. Another factor is that ultrasonic frequencies interact with the "system/environment" to produce audible effects some of which may be desirable. It also seems intuitive that accurate ultrasonic material could be beneficial to transducers by providing a less discrete input facilitating more "natural" movement.

Theoretically a high quality signal processed to be perfectly bandwidth limited to 24kHz and downsampled to 128bit 48kHz would be accurate but practically it would take up the same space as 32 bit, 192kHz sampled material which is easier to achieve and more rewarding to work with e.g. because on the DAC side of things we have similar considerations, where processing using greater bit depths and/or sample rates during output increases audio quality so the input is resampled during processing. Content delivered already at the higher bit depth and/or sample rate is going to bypass the DAC resampling stage and provide the best possible quality for that DAC, and also the higher sample rates include potentially useful ultrasonic material.

I'm amenable to high sample rates however I'd say there's little reason to exceed the CD bitrate for audio quality considering the distortion measurements of lossy encoding methods e.g. 0.02% distortion from ~900kbps could be considered acceptable.

 

[voodooless]

But we don't have perfectly bandwidth limited material (as analog has infinite bandwidth) so there is going to be extra information forming distortion (aliasing) i.e. we need infinite sampling to have a perfect signal.

That’s not how the sampling theorem works. Obligatory watch:

Another factor is that ultrasonic frequencies interact with the "system/environment" to produce audible effects some of which may be desirable. It also seems intuitive that accurate ultrasonic material could be beneficial to transducers by providing a less discrete input facilitating more "natural" movement.

Except that that one study could never be replicated and was highly suspect. Ultrasonic content if even present is like 60 to 80 dB down from the main signal, and then you’ll need some tweeters to reproduce it, and most can’t properly do that anyway. At best the content will be below the rooms noise floor. Good luck getting any audible effects from this.

Theoretically a high quality signal processed to be perfectly bandwidth limited to 24kHz and downsampled to 128bit 48kHz would be accurate but practically it would take up the same space as 32 bit, 192kHz sampled material which is easier to achieve and more rewarding to work with e.g.

128bit? WTF? Do you even realize how silly that is? You cannot just exchange bit depth for sample rate.

Content delivered already at the higher bit depth and/or sample rate is going to bypass the DAC resampling stage and provide the best possible quality for that DAC

No it won’t. It can’t. You must oversample to apply brickwall filtering.

I'm amenable to high sample rates however I'd say there's little reason to exceed the CD bitrate for audio quality considering the distortion measurements of lossy encoding methods e.g. 0.02% distortion from ~900kbps could be considered acceptable.

Like I said before: these distortion tests of lossy codecs are pointless: you have no idea how they behave with real music. That is what they are created to encode.

 

[bmc0]

and we have bit depth which defines the accuracy of the sampling.

It defines the minimum amplitude step size. The resulting error from the rounding is called quantization error or quantization noise. Signal to noise ratio for a full-scale sine wave can be calculated with a simple formula: 20×log(sqrt(3/2)*2^N) → 6.02N+1.76 dB, where N is the number of bits. Undithered 16 bit therefore has a theoretical SNR of ~98dB, while 24 bit is ~146dB. Dithering replaces the (often) correlated quantization error with uncorrelated error, resulting in a uniform noise floor.

With infinite bit depth we would have a perfect signal when sampling at twice the bandwidth of the material.

If you assume a theoretical noiseless signal, yes. However, all real signals have noise so infinite bit depth is not required.

But we don't have perfectly bandwidth limited material

Perfect bandlimiting can be done with periodic signals, though that is irrelevant to music signals.

so there is going to be extra information forming distortion (aliasing) i.e. we need infinite sampling to have a perfect signal.

Aliasing is not the same as harmonic distortion. With a good enough anti-aliasing filter, aliases are below the audible threshold so there is no need for infinite sampling rate in practice.

We can measure distortion and phase-shifting to see how well we're doing and adjust our efforts accordingly.

A properly dithered digital signal has zero harmonic distortion regardless of the bit depth. Anti-aliasing/-imaging filters are commonly linear phase. I don't see how measuring "phase-shifting" is relevant.

Another factor is that ultrasonic frequencies interact with the "system/environment" to produce audible effects some of which may be desirable.

I've never seen this satisfactorily demonstrated to be audible, let alone desirable.

It also seems intuitive that accurate ultrasonic material could be beneficial to transducers by providing a less discrete input facilitating more "natural" movement.

The final output signal is not "less discrete" with higher bit depth and/or higher sampling rate; it's simply lower noise (assuming the noise of the original input was sufficiently low) and/or wider bandwidth (again, assuming sufficient bandwidth of the original input). There are no "stair steps" in the analog output signal if the digital signal was dithered.

Theoretically a high quality signal processed to be perfectly bandwidth limited to 24kHz and downsampled to 128bit 48kHz would be accurate

128 bit integer has a theoretical SNR of 770dB. The difference between the quietest perceptible sound and a shock wave (SPL of 1 atmosphere, where the pressure minima are hard-clipped at zero) is less than 200dB.

Content delivered already at the higher bit depth and/or sample rate is going to bypass the DAC resampling stage

Not the case for basically any DAC. With few exceptions, a DAC will oversample any input signal to a large multiple of the original rate.

I'm amenable to high sample rates however I'd say there's little reason to exceed the CD bitrate for audio quality considering the distortion measurements of lossy encoding methods e.g. 0.02% distortion from ~900kbps could be considered acceptable.

Harmonic distortion has nothing to do with bit depth or sampling rate, as explained previously. Further, the performance of lossy audio codecs cannot be usefully evaluated in this way.

 

[danadam]

a) For analog 40kHz+ capable

So -10 dB @ 25 kHz and -50 dB @ 40 kHz is still good, as long as there's still "something" there?

a) For digital 24bit/96kHz

"I/O (Interface): Input/output interface with the performance of 96kHz/24bit or above" - this "performance of 96kHz/24bit" sounds rather vague to me.

"Decoding: File playability of 96kHz/24bit or above (FLAC and WAV both required)" - so if I only support ALAC or Wavpack but not FLAC then no hi-res sticker for me?

b) For all "Listening Tests"

Interesting there is no mention of what those "listening tests" are supposed to achieve.

c) The following wireless codecs have been cerified Hi-Res: LDAC, LHDC, SCL6, LC3plus, SHDC, aptX Adaptive

MP3 (lame cbr 320) has a 20kHz cutoff limit
Neroaac (q1 96kHz) has a 30kHz cutoff limit with distortion at 74dB (0.02%) for Mids approaching 87dB otherwise
Vorbis (q10 96kHz/192kHz) has configurable cutoff or none with distortion at 60dB (0.1%) for Bass, Low Mids approaching 96dB otherwise

So none of them qualify but I'd say Neroaac comes the closest, with 0.02% distortion being acceptable.

They only name which codecs are certified, they never tell what are the criteria to be certified.

 

[danadam]

(as analog has infinite bandwidth)

This sprang into my mind: Assume a spherical cow in a vacuum ...

Another factor is that ultrasonic frequencies interact with the "system/environment" to produce audible effects some of which may be desirable.

If they in fact produce audible effects, i.e. in audible range, then those effects will be captured, in the audible range, during the recording.

On the other hand, it seems undesirable to have have those effects reproduced twice, once from the recording content and once from the playback's ultrasonic frequencies interacting again with the system/environment.

It also seems intuitive that ...

Beware when something "seems intuitive" in digital audio

 

[voodooless]

If they in fact produce audible effects, i.e. in audible range, then those effects will be captured, in the audible range, during the recording

This is not a given. Microphones are not ears. They can perceive demodulated ultrasound in different ways. For instance, look at those obfuscation devices for meeting rooms that can overpower microphones with ultrasound, making records sound like noise, while humans will not hear anything.