Question for Ted regarding Directstream and headroom

I was reading a review of the Benchmark Audio DAC3 today on and in the review, Doug Schroeder talked about the utilization of the headroom option in Roon and how enabling the function on the DSP settings and changing the value markedly changed the sound. Apparently this is because of design of most DACs that are overloaded during upsampling. This is the explanation provided by Benchmark’s John Siau:

“It is a gain reduction function that reduces the signal level before any DSP operations occur. This also reduces the signal level going to the DAC. This reduction provides enough headroom to prevent DSP overloads. Most importantly, it prevents the gross DSP overloads that can occur when an upsampling function reconstructs the audio waveform between samples. All sigma-delta D/A converters use oversampling or upsampling. The headroom function prevents the digital clipping of intersample overs (signal peaks that exceed 0 dBFS). Benchmark converters have an extra 3.01 dB of headroom in the entire digital path and the Roon headroom function should be turned off (or be set to o dB). Virtually all other D/A converters will benefit from this function. In my opinion, this should never be set less than about 1.5 dB and it should be set at the full 3 dB if MP3 files will be played. I have written six papers on the topic of intersample overs. The people at Roon and JRiver are aware of the intersample over problem, but sadly, most converter manufacturers have ignored this issue. Intersample overloads generate false high frequency percussion sounds. This defect is entirely preventable using the headroom function or a DAC with built-in headroom.”

So it got me to wondering if the Directstream was immune to this problem. So I went into the DSP of Roon and picked a particularly hot section of Santana’s Black Magic Woman ripped from the Ultradisc Abraxas. I record in Vinyl Studio with the option on 0 dB headroom since I’ve never noticed distortion during playback. According to Roon, if there is distortion, the signal path light will go red. Hopefully you would also hear distortion while this is going on, obviously. No red light and no distortion to my ears. I then gave it -3 headroom and it was not only a bit quieter, but also slightly less dynamic, but this might be my imagination. I then dug out some of my harder hitting music recorded in MP3, files of which which I haven’t listened to in forever. I tried Soundgarden Superunknown MP3 multiple songs and got no distortion, visually or audibly. Likewise Nirvana’s Nevermind in MP3 was free of distortion. My guess is that the Directstream is immune to this problem John Siau speaks of. When he says “Virtually all other DACS will benefit from this function”, I guess the rare exception would be the Directstream. And, I guess the Benchmark DAC3.

Here’s the link of the section of the review that Mr. Schroeder talks about this:

I will note that he talks of being able to change the headroom from -3 to +3. My version of Roon only allows -3 to 0. I’m not sure if this is an error on his part or an idiosyncrasy of DSP settings with his DAC.

So what do you say, @tedsmith? Am I correct in thinking that the Directstream is immune to these problems? Thanks!

Intersample overs can indeed be a problem with many DACs. It is why, when mastering a CD, I leave some extra headroom and never approach 0 dBFS. But an intersample over rarely results in immediately obvious digital distortion even though there technically is clipping. Immediately obvious clicking/crackle occurs only if there are a good deal of successive overs. Usually if you hear anything it is a comparative lack of transparency.

Intersample clipping/distortion was of concern to audio engineers and addressed in pro DACs long before audiophiles even recognized the issue.

DSD exhibits 3dB head room above 0dBFS. SACDs often peak up to +3dB. Bit this can only exist briefly before becoming noticeable.

I assume intersample overs are not an issue for the DS as PCM is transcoded to DSD before conversion and this nice bit of extra headroom avoids the problem. This is where I need Ted’s help . . .

Indeed SACDs can reach about +4dBFS. When some (technically illegal PCM) is upsampled the new samples can reach above 0dBFS. I say illegal because if the input were truly bandlimited and in range then it still will be when upsampled. But we all know that people have a tendency to “normalize” their tracks which (typically) means they scale the track up until some sample reaches the max (or min).

Also, FWIW because of the Gibbs phenomenon a bandlimited square wave (or step) “overshoots” the top by about 9%. This is independent of the sample rate and is caused just by sampling. This means test tones that have square waves (or other tones with “instantaneous” hops) are illegal because they aren’t bandlimited. When they are bandlimited they can’t get to 0dBFS.

The DS does two things - it keeps (at least) an extra bit on the high end of all processing which allows both up to 6dBFS peaks, perhaps because of upsampling adding points higher than 0dBFS. The analog hardware also has about 3dB headroom above that 6dBFS. Actually not quite that much if people use the volume at 106. So short peaks above 0dBFS shouldn’t be a problem (they can happen a lot in some SACDs.)

(There’s a bug in Redcloud which can crackle at high peaks, but that bug isn’t related to the above.)

Another resource about inter-sample peaks:

1 Like

Thanks Ted!

Robert Hutchins