Let's develop an ASR inter-sample test procedure for DACs!

[AnalogSteph]

Neither do I still fully understand why Benchmark opted for using a potentiometer (prone to wear and eventually noise or even distortion) for the analog volume control in the DAC1. Even at that time, there were already DSP- and IC-based ones available - as commonly used by AVRs.

You can do the math:
For a playback system with no ifs and buts you want about 110 dB of dynamic range.
Back in the early 2000s, you'd realistically get about 117 dB(A) out of high-end DAC chips. Linearity at the extreme low end could also be a bit so-so at times, plus fun stuff like idle noise.
This leaves some wiggle room for amplifier gain / speaker sensitivity matching but really not a ton.
So in sum you definitely want some sort of analog gain setting for better level matching.

Pro-level converters of the late '90s would generally aim to accommodate a wide range of 0dBFS levels up to fairly high values (+26 dBu to even +29 dBu with Benchmark - you don't even need to show up to a Hollywood studio if you can't do +24 dBu, meanwhile +18 is common in project studios and +12 in home studios, and if you want to directly connect some monitors a bunch of those need +6 at best), often with some more or less fine-grained DIP switch settings. For example, the Prism Sound Dream DA-2 (some of the best you could buy in 1998, sporting a 114 dB(A) dynamic range) boasted

Calibrated maximum output level adjustment:

• +5 dBu to +28 dBu in 1 dB steps
• Fine trim in steps of 0.05dB

Big studio tech is something else, they don't mess around. Keep in mind this was a time when there still was lots of money in the music industry, as well as (public) broadcasting.

This kind of effort makes RME's mere handful of reference levels look downright built to a budget, even though +4 / +13 / +19 / +24 on their larger models should cover the vast majority of cases perfectly well.

Nowadays, with DACs boasting a 126-130+ dB(A) dynamic range being downright commonplace, going to extreme lengths in terms of analog gain selection obviously makes less and less sense. The very best DACs are still going to need some to cover the >142 dB(A) of total dynamic range found in existing models, but it could be something like 2 or 3 steps tops with the rest done completely in the digital domain and nobody would bat an eye.

 

[win]

I love benchmark as a company. Their products are top notch, and John does a great job interacting with the community. He makes great posts here, and if you've caught him on the audioholics YouTube channel, those are great too. The application notes section of their site also has great articles to review.

Some day I'll grab another AHB2, when I have a bigger house and my 85db sensitivity speakers don't cut it any longer with a single AHB2.

 

[tmtomh]

So the original .wav files they sent you were 16/44 or something else (16/48)? Just wondering if it was the effects of a SRC at the mastering for CD/disc manufacturing end.

Sorry, I should have been more explicit about that. Yes, the WAV files they sent me were 16/44.1 files. They are bit-identical to the files on the CD, including their sample rate and bit depth, except for those samples that are at digital 0.0. Those samples - and only those samples - are inverted on the CD. It's the first and only time I've ever seen that.

 

[danadam]

Sorry, I should have been more explicit about that. Yes, the WAV files they sent me were 16/44.1 files. They are bit-identical to the files on the CD, including their sample rate and bit depth, except for those samples that are at digital 0.0. Those samples - and only those samples - are inverted on the CD. It's the first and only time I've ever seen that.

I had a similar story. On the fist disc of "PentaTone The First 10 Years" box, the "Tchaikovsky - Serenade for Strings & Souvenir de Florence", there was a track with such artifacts:

I contacted them and I got the standalone disc, which had exactly the same artifacts, only the track had 15ms offset. Then I got the download which was OK. Comparison showed that for the first two and a half minute it was bit exact and for the next three minutes it had short bursts of differences four times per second (which is what I heard):

 

[VientoB]

I get that. Benchmark were of course on message, prosecuting their narrative in the comments/responses to GS's review linked.

They have a pattern. It probably works for them and likely for many potential customers of theirs.

Apparently it was just one guy at Benchmark who no longer works for them. If you believe that, it seems a little far fetched but who knows.

 

[restorer-john]

Sorry, I should have been more explicit about that. Yes, the WAV files they sent me were 16/44.1 files. They are bit-identical to the files on the CD, including their sample rate and bit depth, except for those samples that are at digital 0.0. Those samples - and only those samples - are inverted on the CD. It's the first and only time I've ever seen that.

That is truly amazing. Your experience and interaction with record company is likely unique.

 

[TNT]

And they mostly only happen at 11kHz - right?

//

 

[pma]

No. Fs/4 is just easy to make a test file.

 

[restorer-john]

And they mostly only happen at 11kHz - right?

Mostly. /s

 

[VientoB]

How much effect on sound would you guys say intersample overs could have? Could it make one DAC sound noticeably worse than another?

 

[little-endian]

Depending on who you ask, the "doctor's opinion" will severely differ. In case of Benchmark where they discovered that topic for themselves a decade or so ago (and didn't seem to care too much about it either with the DAC1), it's now a major issue of course.

At least, intersample clipping is probably less "voodoo-y" than the jitter fuss but certainly blown out of proportion as well.

Good news is that in principle, one can "fix it in post" with any DAC and to really answer your question, one should get a DAC known to be specifically bad at this, feed it with a "hot mastered" source (which there is hardly any lack nowadays) and compare whether lowering the input level to the DAC by using some digital gain control makes an audible difference. The challenge here as well is that one would have to compensate for the level difference again in order not to prever any version over the other by sheer SPL performance.

 

[AnalogSteph]

How much effect on sound would you guys say intersample overs could have?

That depends a lot on the material played, how exactly the DAC reacts to overs (ASRCs may spit out a slew of nasty anharmonics) and what the playback chain looks like.

When using RG with 4x oversampling using the SoX resampler DSP for peak scanning, the gnarliest loudness war era CDs in my collection (recorded prior to the use of oversampling brickwall limiters and true peak monitoring, generally in the late 2000s) are clustering around 1.35 - 1.38 - 1.39, there's even one a tad over 1.40. So in theory, about 3 dB of digital attenuation would entirely take care of this issue, and RG wants to attenuate those CDs by around 10 dB anyway. These days I'm rarely seeing more than about 1.15 on new releases. MP3s may peak higher, I've seen up to 1.59 in extreme cases.

It's those folks running CD players directly into a DAC who are the worst off, as they basically are at the mercy of DAC performance.

 

[Scytales]

How do you think a non-HDCD capable, ordinary, signal processor or DACs will behave when playing an HDCD encoded CD with peak extension as far as inter-sample overs are concerned ?

 

[tmtomh]

How do you think a non-HDCD capable, ordinary, signal processor or DACs will behave when playing an HDCD encoded CD with peak extension as far as inter-sample overs are concerned ?

It depends on the DAC and the mastering. An HDCD encoded with Peak Extend will play 4dB louder with a non-HDCD capable DAC or CD player than on an HDCD-capable one, because that's what Peak Extend does: it compresses/limits the loudest peaks, enabling the overall volume of the music to be raised, and then encodes the information required to reconstruct the original dynamics of those peaks into the least significant bit of the 16-bit CD signal.

In principle, a DAC with HDCD would have less of a change of spitting out intersample overs because the entire program would be turned down 4dB. However, the point of Peak Extent is to use that extra 4dB of headroom to reconstruct whatever loud peaks where originally in the signal. So it depends on how many of those reconstructed peaks there are and how close to the new digital 0.0 they are. And it also depends on how loud the peaks of the compressed, non-HDCD version of the musical information is - if that version doesn't have a lot of 0.0 peaks then it too might not produce a lot of intersample overs.

 

[John_Siau]

Depending on who you ask, the "doctor's opinion" will severely differ. In case of Benchmark where they discovered that topic for themselves a decade or so ago (and didn't seem to care too much about it either with the DAC1), it's now a major issue of course.

We discovered this issue while designing the DAC2. We made changes to the gain structure of the DAC2 before its release in 2012. It was too late to do anything about the DAC1. The DAC1 was released in 2002, long before we were aware of the issue. DAC1 production ended in 2012 when the DAC2 was released.

Benchmark has now been building high-headroom DACs for over 12 years (DAC2 and DAC3 family) and we have been talking openly about how the problem can be solved. Nevertheless, it is amazing that many manufacturers are still ignoring the DSP headroom issue.

There are several other manufacturers that added DSP headroom to their DACs, processors, or playback software, but they are still in the minority. On the other hand, there are some devices that produce more than 1% distortion when intersample overs are encountered. Some devices actually invert the polarity of the audio while the overload occurs.

This problem is real. This is not a marketing promotion. It is also not caused by "hot masters" or bad mastering techniques. Waveform peaks will always exceed the values of the samples unless the samples happen to occur on the waveform peak.

Until reviewers start testing for DSP overload problems, many products will continue to be shipped with inadequate DSP headroom and with digital volume controls that are in the wrong part of the signal chain.

 

[AnalogSteph]

BTW, an informal survey produced a record of almost +6 dB peaks, and I found two CDs with over +4 dB in my own collection.

True Peak may be a matter of definition...

I have been playing with the options for "True peak scan" in the Foobar2000 ReplayGain scanner lately, which allows you to use any resampler DSP installed for the oversampling. This made me notice a marked difference in results between "auto 4x oversample" - which I presume uses the filtering...

www.audiosciencereview.com

Using a mixed or minimum phase filter will aggravate the problem, much more so than just using a steeper slope. I was able to turn >+4 dBFS peaks into >+7 dBFS ones like that.

 

[little-endian]

It was too late to do anything about the DAC1. The DAC1 was released in 2002, long before we were aware of the issue. DAC1 production ended in 2012 when the DAC2 was released.

Yes, fair enough. At that time, virtually no one except for maybe the mathematical space had this in scope at that time, I guess.

True however is also that the DAC1 repeatedly got praised for its awesome sound quality and neutrality, obviously despite its less than ideal intersample peaks reconstruction capability which leads to the logical conclusion that either we all just didn't care and were simply used to their caused side effects or it wasn't that bad after all.

Nevertheless, it is amazing that many manufacturers are still ignoring the DSP headroom issue.

While it is a different issue and a different domain, it reminds me of my astonishment about how badly images and also digital videos often are still handled from a scientific, signal theory's point of view. For instance, most image viewers don't use decent (down)scaling algorithms and lack proper low-pass filtering in order to prevent aliasing or imaging respectively. I remember there where even DSLRs offered without an optical filter to create more (partly fake) sharpness. In audio, I don't know any manufacturer who would dare to leave away the pre-filtering before an A/D-converter (for end users at least, not transmission or testing scenarios where aliasing might be even desirable).

Also, there are tons of otherwise decent Blu-ray and even UHD Blu-ray releases which show ugly banding artefacts due to the lack of proper dithering. In the audio domain, all of that is a matter of course, but not so for its 2D counterpart where ignorance is still extremely widespread.

 

[VientoB]

Yes, fair enough. At that time, virtually no one except for maybe the mathematical space had this in scope at that time, I guess.

True however is also that the DAC1 repeatedly got praised for its awesome sound quality and neutrality, obviously despite its less than ideal intersample peaks reconstruction capability which leads to the logical conclusion that either we all just didn't care and were simply used to their caused side effects or it wasn't that bad after all.


While it is a different issue and a different domain, it reminds me of my astonishment about how badly images and also digital videos often are still handled from a scientific, signal theory's point of view. For instance, most image viewers don't use decent (down)scaling algorithms and lack proper low-pass filtering in order to prevent aliasing or imaging respectively. I remember there where even DSLRs offered without an optical filter to create more (partly fake) sharpness. In audio, I don't know any manufacturer who would dare to leave away the pre-filtering before an A/D-converter (for end users at least, not transmission or testing scenarios where aliasing might be even desirable).

Also, there are tons of otherwise decent Blu-ray and even UHD Blu-ray releases which show ugly banding artefacts due to the lack of proper dithering. In the audio domain, all of that is a matter of course, but not so for its 2D counterpart where ignorance is still extremely widespread.

Which image views use good algorithms?

 

[little-endian]

For Android and images, there is the AA Image Viewer, however not actively developed anymore.

For video, the renderer madVR features quite a lot filters such as lanczos, spline ... It also nicely visually proves that banding is not a necessary consequence at any bit depth if properly dithered, as shown in this comparison:

1 bit per channel

6 bit per channel

 

[VientoB]

Guys, Audioholics are doing a livestream on inter-sample overs on Wednesday 6th November evening (USA time):