Volume bug - can you really hear it?

[cbemoore]

Hi Dean,

Your change applies 8 bit accuracy to volumes louder than -35dB. What volume on the SB 0-40 volume scale corresponds to -35dB?

Chris

[sbjaerum]

Originally Posted by cbemoore

Hi Dean,

Your change applies 8 bit accuracy to volumes louder than -35dB. What volume on the SB 0-40 volume scale corresponds to -35dB?

Chris

I believe the volume changes in steps of 1.25dB.
That means that -35dB corresponds to a volume setting of 12 in the SB 0-40 volume range.

Steinar

[Andrew L. Weekes]

Can you tell me what FFT software you used to read in through the SPDIF of your sound card? Was it shareware or did you roll your own?

It's Spectralab.

Also, can you tell me what type of windowing you used on the input data stream when you measured THD?

Windowing for those measurements was Hanning which is generally good for distortion and noise measurements. Blackman being the other common choice for distortion only.

Andy.

P.S. I'll install the 6.5 nightly on the test PC later, but may not be able to make any measurements 'til Friday, as I'm busy tonight and tomorrow.

P.P.S. Sean / Dean, does this show on the D-Scope - it should?

[jhwilliams]

Originally Posted by Patrick Dixon

Whilst I'm happy to take the credit for being a pita over it, I reckon it was jhwilliams and his golden-eared girlfriend who picked it up first!

Ta! Hats off for your solution though.

Speaking of ears, I was planning another round of tests, but I have a bad headcold at the moment... so critical listening is off the agenda.

I did repeat Andy's test though for my own interest... I got very similar results. Has anyone else tried the same?

I'll see about also repeating with the patched version this weekend.

[usualsuspects]

FWIW - I have been following the "new volume" vs sound quality issue. I also thought that my SB2 lost some sound quality at some point. My setup is: SB2 analog -> Parasound HCA-750 -> Paradigm Studio 20 v3. I was using the 6.1 final release, and just installed the latest 6.5 nightly (slimserver-2005_11_26-1.noarch.rpm on Cent OS 4.1). My totally unscientific / not blind opinion is: the sound quality is back to what it was before the “new volume” issue raised its head. The best way that I can describe the “old” new volume sound was “flat”. I especially notice the difference in cymbals and other high frequency sounds – much better now! Thanks to everyone who is pushing this issue – it is real (to me anyway).

[Andrew L. Weekes]

Sorry for the delay in results, this has proved to be quite a hard nut to crack (and I've not cracked it yet)...

Firstly, I had some problems getting 6.5b1 to start, it doesn't like installing over 6.1.1, so I had to uninstall between server changes, which took a while to work out!

Secondly, the earlier distortion results I posted earlier are NOT representative, what I should have noted is that of course the old volume law and the new are different.

This means that for a given SB indicated volume the old and the new volume laws represent a different attenuation. An inherent problem of doing the volume in the digital domain is that you will always worsen the S/N and the THD figures as you lower the gain, since the noise floor is a constant, nominally. Since the gain changes with the old law are finer steps at the top end, it gives rise to lower distortion figures for a given indicated volume.

So to that end I re-ran results for Server version 6.1.1 / FW15, 6.2.1 / FW28 and 6.5b1 / FW28.

Since few if any of the data points correlate (i.e. for a given SB indicated volume, each Server versions gives different attenuation results, albeit slight for the latter versions) it's hard to compare results, one has to interpolate between values, which isn't ideal.

To that end measurement at present is inconclusive. If the issue is as supposed, which the math would seem to support, there should be dynamic range gains with the patched version, but at present I'm struggling to resolve that through measurement.

After some hours spent trying, I gave up and settled back for an hour or so in front of the HiFi and did some listening, connecting the SB2 to 6.2.1 and 6.5b1 in turn.

The benefit of the patched version is clearly audible to me, as it has been to others, the obvious effect being that the unpatched code has a harsher, sound to the treble end (cymbals being obviously less detailed with a harder sound) along with a loss of ambient, low level information throughout the mix.

You can hear this on the decay of cymbals, for example, where the decay of the sound is a longer event on the patched code as if the noise floor is lower. It's also noticeable in the sense of acoustic surrounding a good recording (be it genuinely live artifacts from room ambience, or artificially added echo, at the mixing desk) where the unpatched code sounds 'flat' as if the information below a certain level has been gated out.

I'm open to suggestions for a test technique to reveal what I can hear, but at present it's Ears:1, Test Equipment:0



Andy.

[Andrew L. Weekes]

That means that -35dB corresponds to a volume setting of 12 in the SB 0-40 volume range.

I believe you'll find the changeover is at 16 on the 0-40 range. That's what the measurements indicate - below this value the old and patched code measure identically.

Andy.

[sbjaerum]

Originally Posted by Andrew L. Weekes

I believe you'll find the changeover is at 16 on the 0-40 range. That's what the measurements indicate - below this value the old and patched code measure identically.

Andy.

The first checkin of the patch had the changeover at -35dB, this corresponds to a volume setting of 12. The final version of the patch changed the changeover to -30dB. As you correctly states, -30dB corresponds to 16 on the 0-40 range.

Steinar

[Patrick Dixon]

I think it could actually go to -32dB to completely maximise the benefit.

[Andrew B.]

Originally Posted by seanadams

Having dismantled a DAC-1 I can tell you it is a perfectly vanilla implementation of ASRC using a standard ADI chip, and it does perform as advertised WRT to jitter rejection - this feature is _inherent_ in asynchronous sample rate conversion.

ASRC effectively moves the source of jitter from the s/pdif input DAC's internal clock. That's fine but it's not necessarily an improvement - the latest standard s/pdif receiver chips from Crystal, ADI, TI all have very good jitter attenuation - in fact a good PLL design will inherently act as filter to clean up the recovered clock however bad it is.

However, ASRC while it obviously eliminates jitter in the source signal, does god knows what to the data coming through.

That said, the Benchmark's DAC and amplification stages appear to be carefully engineered, and measured performance here is great. Unfortunately real performance is harder to market, but ELIMINATES JITTER sells.

I bought the Benchmark DAC on the basis of the sound, rather than the jitter attenuation. My previous CD player cost over $10k so buying a "good" transport wasn't going to be an issue. It turned out that, for once, the hype is correct and it does sound fabulous. Not warm, not scratchy, not dull, not bright... just right!

I now use it with a very old Meridian CD player and a very new SB3

Steinar's link was interesting and suggests that ASRCs are inherently superior to dual PLL designs, provided that the implementation is done carefully.

Certainly the Benchmark is great value for money compared to the audiophile competition I have heard (standalone CD players and transport/DACs).

I haven't yet had enough experience of using it with my SB3 to be definitive about proclaiming the end of CD players as we know it - but I have spent the last two days reripping 200 of my CDs into FLAC, so perhaps that says enough...

Andrew