Fixed Integer Playback

[ted_b]

but I also have experienced a "big difference" with the unreleased integer support in PM, and I attribute it simply to the fewer low level routines that are needed (no 32 bit floating point conversion) to get the signal to the DAC. Why this is so, when one would assume it is a not a big cpu deal in the first place...well, maybe it's not the computation expense as much as the conversion anomalies...dunno.

 

My take on integer playback is that it brings a more resolving, blacker, more dynamic playback than standard hog mode. As is the case with all that is computer audio, some may not like the differences in his/her system.....and that's great cuz it's just a checkbox in the audio setup panel.

 

[Mercman]

I will agree with ted_b that integer reduces computer processing that seems to lead to improved sound quality. As Ted pointed out, it's easy to try if your dac supports integer.

 

[damien78]

MarkS, I'll be extremely interested by your issues reports, as I'm in the last phase of the beta phase of Audirvana, hunting the very last bugs before making 1.0 release (and moving to new shiny features). And the recent feedbacks are more on the rock solidity side... So I'd like to know what you have uncovered.

 

That said, I still think PM is an excellent player. Additionally, Audirvana was extremely buggy for me, crashed frequently, and often created an audible right channel crackle and hiss. PM has been rock solid for me for several builds now.

 

Damien

 

PS: Regarding the advantages on Integer mode, a strong one is that the CoreAudio mixer is bypassed (in addition to removing the float to int conversion). More in an upcoming paper I'm writing on OSX for audiophiles.

 

[goldsdad]

Hi Damien,

 

"Regarding the advantages on Integer mode, a strong one is that the CoreAudio mixer is bypassed (in addition to removing the float to int conversion)."

 

Unfortunately, I can't use integer mode, but the subject is very interesting and something to consider for a future DAC upgrade.

 

Do you hear a consistent and unmistakable improvement in sound quality from enabling integer mode even when connecting via async USB to your DAC(s)?

 

 

[Jud]

@goldsdad -

 

It'll only be one subjective data point, but I have the V-Link which does work with Audirvana in integer mode, and should be able to find time to do some listening tests regarding this tomorrow.

 

[ericuco]

Dropped a note to Metric Halo on supporting integer playback and here is the response:

 

"In response to recent requests, we have looked at, and will be implementing fixed integer playback with the next software update.

 

There is no timeline for the release of the next update yet, however, so please stay tuned."

 

I downloaded Audirvana and was able to play music through my LIO-8 with the "Integer mode" checked although I doubt, based on earlier comments, that the LIO-8 is using this without doing some conversion. Still, the Audirvana sounds REALLY nice.

 

For me, one of primary reasons for using Pure Music is the digital crossover piped to the LIO-8 on 4 channels via Firewire so I will mostly likely stick with Pure Music and look forward to the upgrades in both Pure Music and LIO-8.

 

[jdubs]

Eric, I just received the same response from MH. Awesome news!!

 

-Jim

 

[antipodes_audio]

I have played around with this stuff extensively and expensively. I don't think it is possible to be definitive, but came to some conclusions for my own use.

 

The main conclusion was that you don't want the computer or your DAC to have to do more computational steps than are necessary during playback. When you remove the computational steps required to be performed during playback, the sound improves. The fact reported earlier that the MH was bit perfect after manipulation misses the point - which is that unnecessary manipulation during playback appears to my ears to be a bad thing.

 

Some posts above have referred to removing computational steps in the computer, and this is critical for good sound, as is shutting down unnecessary services that may run during playback. It is not so much how much computation resources they suck, but about the fact they are there at all, interfering with the process of streaming the digits.

 

Additionally, there is the issue of upsampling DAC chips. A lot of DAC chips will upsample on the fly in order to do the DAC conversion step at the resolution (and with the filter) that it is set to operate at. While the DAC conversion step is probably optimised, it is negatively affected by having to do the upsampling computations at the same time. And then there is the issue that phase errors in the treble region can be reduced by upsampling.

 

So one strategy is to ensure your 16/44.1 files stay at that resolution throughout the process and have your DAC convert at that resolution.

 

But in my experience there is an alternative strategy that sounds better. What I prefer is to convert my 16/44.1 files to 24/88.2 files offline. That is, I convert them using software like izotope or Sample Manager and store them at that resolution. I then play them through a system that keeps the data at that resolution throughout - as much as that is possible with any given DAC.

 

Whether this alternative strategy is better for anyone else is impossible to say. I have to use larger hard drives, the conversion algorithms will have an effect, the computer and DAC have to do more work during playback because the bitstream rate is so much higher, and gentler digital filters can be employed.

 

So if your DAC, for example, upsamples everything to 24/192 before conversion, then you may be best to convert your 16/44.1 files to that resolution and store them. And then use integer playback from end to end with your computer/dac system. For what it is worth, with 3 different systems (MH LIO-8, Apogee/DCS, and Weiss DAC202) what seemed to sound best was offline conversion to 24/88.2 WAV files first, but a number of factors are at play here. Also, this only applies to playing 16/44.1 files. Higher resolution files seem best to be left alone, or the advantages of conversion to be minor - with the exception of conversion to WAV (as opposed to AIFF, FLAC or ALAC).

 

The point I am making is that if the conversions are done offline with good algorithms, then upsampling can be beneficial, and that integer playback can be applied at that resolution to achieve the benefits of integer playback. I am certain that integer playback has real advantages, but it does not have to be applied at 16/44.1. Provided the upsampling is offline, you can get benefits from both upsampling and from integer playback.

 

[ted_b]

them. You need to make a USB cable.

 

I wonder what the "sweetspot" is for my Antelope Gold DAC (i.e what a good effective converted-from-redbook sample rate would be to relieve the Gold of some of its work) and wonder if upconverters as esoteric as the Weiss Saracon (I have access to one) would make a difference over, say, good ol DBPoweramp. I might do some experimenting.

 

Oh, and I'm glad I;m not too alone in the "wav" theory.

Ted

 

 

[antipodes_audio]

Hi Ted,

 

I wonder whether WAV files sound better because they don't have tags and other info in them?

 

Can't help on the Antelope but suspect that converting to the max resolution of the DAC chip is generally safe to ensure no on the fly upsampling, but that might result in some huge files with some chips, and the faster chips work the more errors they make.

 

The DAC area is confusing - note that some people suggest integer playback is possible with a benchmark DAC1 - which seems odd to me when it uses over-sampling to map jitter to broadband noise (ie. changes the bits).

 

Sample Manager does a good job for $79, but would love to compare it with a file made using Saracon. But I fear I would then have to buy it if it sounded better.

 

Thanks for relaying the feedback on the cables - it is always good to hear. No plans for USB cables. I know this sounds utterly crazy but it typically takes me 2 years to come up with a new fundamental cable design, so at this stage ICs, SCs and Digital (SPDIF/AES) only. The reason it takes me so long is that I address every piece of the related science with extensive experimentation, and absolutely nothing is taken as read. I haven't yet heard a USB interface that comes close to the good firewire interfaces, and I don't have enough faith in the longevity of USB or Firewire for audio yet, but that could change.

 

[elcorso]

How long ago is Integer working on Pure Music, Mercman?

 

I'm really admired!

 

I was following your posts here on CA and the Audio Asylum regarding Pure Music and Integer playback, it's amazing...

 

Then, since Damien Plisson, Audirvana creator, release his 0.9.1 version with Integer implementation for a large variety of DAC's and USB to SPDIF interfaces recently, but already working since a real long time for good interfaces like the Halide Bridge, suddenly you announce "Channel D had integer working a long time ago. I really can't believe it.

 

Channel D will profite for Damien long work?

 

Channel D will profite for the long DACs & USB interfaces stated that works under Integer on Adirvana thread, that you tried to move to Pure Music 1.8 not even delivered to us expensive payed app?

 

Channel D will profite since you state Audirvana is 'beta' and open source (and then could be copied)? Or Audirvana is the 'beta tester' for Pure Music integer playback?

 

You try to look very kind and diplomatic by saying: "I agree with wgscott that Damien has done a great deal of work". But you look to me like the advertising guy for Channel D.

 

I like to finish my reply with what Steve Jobs uses to say: "Redmond, start your photocopiers!"

 

Roch

 

 

[barrows]

I am really not sure what you are talking about regarding "upsampling":

 

"For what it is worth, with 3 different systems (MH LIO-8, Apogee/DCS, and Weiss DAC202"

 

None of the listed DACs actually do D/A conversion at the native rate (44.1, or your converted 88.2 files). They all employ oversampling digital filters (in the case of MH and Apogee), async upsampling+oversampling (in the case of the Weiss DAC202) and conversion to very high sample rate (mHz range) and 5 bits (in the case of dCS). All of these DACs are doing conversions on the data rates, on the fly, before the actual D/A conversion.

Perhaps you might be a little more clear on the point you are trying to make?

 

[Panelhead]

My favorite dac is FireWire and the Lyngdorf Room Perfect operates at 96k/24. So it upsamples everything.

My guess is none of my dacs; KRK Ergo, Apogee Duet, TC Impact Twin, and HRT Music Streamer will allow INT.

If PM 1.8 makes it easy to check operation, it will not take long to see a more complete listing of Fixed Integer Processing capable dacs.

Next dac purchased will definitely be INT capable.

 

George

 

[antipodes_audio]

Yes, sorry for the confusion.

 

I have and do use integer playback on my Debussy and Cosecant 3.

 

The point about the MH, Apogee and Weiss is a separate but related point that there were clear audible benefits to the offline conversion from 16/44.1 to 24/88.2 even when those DACs do further upsampling/oversampling - and I didn't make that clear.

 

My key point is that the benefits of integer playback are not confined to playing 16/44.1 at 16/44.1, such as if you use offline conversion. In my opinion the benefits of integer playback are not so much about the lack of conversion but about the lack of conversion during playback.

 

[Miska]

So if your DAC, for example, upsamples everything to 24/192 before conversion, then you may be best to convert your 16/44.1 files to that resolution and store them. And then use integer playback from end to end with your computer/dac system.

 

Most modern DAC chips interpolate 44.1 kHz by 8x (you can call it upsampling too) to 352.8 kHz first, this is still from 24 to 32 bits resolution. Then they oversample it by 64x to 22.5792 MHz in the SDM stage, where the number of levels is reduced to equivalent of 2 to 5 bits which is then converted to analog. This lowers cost and improves low level linearity.

 

Modern Burr-Brown DAC chips are hybrids such way that they split the data after the first 8x stage, six bits of 24 go to sort of multibit DAC and rest go to the 64x ~3-bit modulator stage. Output of these two stages is then combined at integration stage.

 

The first 8x stage is the most quality critical, and it's sonic impact can be reduced by feeding higher rate data. Whether this rate conversion happens on the fly or offline doesn't matter from technical perspective. Any possible adverse sonic effects of on the fly conversion are elsewhere and usually due to poor noise isolation between audio electronics and computer. Some "upsampling" DACs perform this operation on the fly within the DAC device on a separate chip or DSP, but it can be as well performed within the computer too.

 

 

[Forehaven]

Well from what I'm reading, I'm glad to hear MH is working on integer playback.

 

But Barrows, you said "None of the listed DACs actually do D/A conversion at the native rate (44.1, or your converted 88.2 files). They all employ oversampling digital filters (in the case of MH and Apogee)"

 

I'm not following, using PM, when I play a 16/44 file, it plays at 16/44, and so forth. At the native rate. Can you help me understand how the LIO 8 uses oversampling filters?

 

Thanks!

 

[Mercman]

Roch,

 

I'm not advertising for anyone. I have been a beta tester for Amarra, Pure Music, Fidelia, and Decibel.

 

I have complimented Damien a number of times for his work on Audirvana. Certainly expectations for perfection are much higher for a paid program. If you doubt this, look at the comments made for Amarra, Pure Music, Fidelia, and Decibel.

 

I can assure you that the code for Pure Music is very different from that of Audirvana. And they do sound different.

 

Here is my post From Feb. 2011 at AA about Pure Music Integer:

 

http://db.audioasylum.com/mhtml/m.html?forum=pcaudio&n=85807&highlight=pure+music+integer&search_url=%2Fcgi%2Fsearch.mpl%3Fauthor%3DRandHat%26user_id%3D24364%26forum%3DALL%26sortRank%3DNone%26sort%3Ddate%26sortOrder%3DDESC

 

Not that long ago I was accused of being in bed with Sonic Studio. And so it goes.

 

 

 

[barrows]

"I'm not following, using PM, when I play a 16/44 file, it plays at 16/44, and so forth. At the native rate. Can you help me understand how the LIO 8 uses oversampling filters?"

 

Read Miska's post above. He has a much better technical understanding of Delta Sigma DAC chips than I do. The MH ADC/DACs use AKM Delta Sigma DAC chips, which convert from digital to analog at very high data rates.

You send your digital audio out from PM at 16/44.1, but then it gets converted to a much higher rate inside the DAC chip before the actual conversion to analog.

 

 

 

[ted_b]

oversampling that is done before the signal becomes analog. Yes, your DAC and mine both handle native sample rates natively, i.e they "end up" at 16/44 for a 16/44 file...but a lot happens in the meantime as they oversample to hyper-levels in order to use gentle digital filters to remove supersonic noise.

 

All of these reminders make me thing that I am not going to try desperately and upconvert all my redbook. The small theoretical advantages (if any) of offline conversion is outweighed by the hassle factor, the larger file sizes and the question of whether my upconversion is "cleaner" than what the DAC needs to do regardless. In the past I've not noticed differences, and I'll probably still do a test, but these reminders portray a "pissing in the ocean" level of improvement.

 

[Jud]

@goldsdad -

 

OK, I've now had a chance to do some listening. In my system the change isn't earth-shaking (the music still sounds very good with integer mode off), but I do feel I hear small but consistent differences that make integer mode more involving. (All the following is with revision 157 of Audirvana built from source.)

 

"Good Night Old World" from Emmylou Harris' latest, "Hard Bargain," relies on a contrast between the verse, which seems to be recorded nearly in mono, and a chorus where her voice and the backing instrumentals are bigger and more reverberant. Harris' dry country voice on the verse gives that part of the song a lonely, Dust Bowl quality. Then when the chorus comes in, she's perfected an almost Aaron Neville-like echoing fadeaway at the ends of her lines, which with the bigger instrumental backing turns the denouement of the song from the feeling of isolated loneliness into more of a classic heartbreak tune.

 

With integer mode off, the dry isolation at the beginning of the song was exaggerated to the point that Emmylou's voice sounded almost as if it were coming from a table radio. But there was less drama to the chorus, less of a change in the sound. I'd say low-level dynamics, especially the last lower-level reverberations at the ends of echoes, were being reproduced less well; and also that the difference in soundstage from the near-mono beginning to the bigger, more spread-out chorus was less apparent.

 

With integer mode on, the dry, lonely feeling of the verse was still there, along with the near-mono soundstage, though not quite to the extreme degree that it was with integer mode off. But the change when the chorus came was more dramatic, with lots of echo to the vocals and a higher, wider soundstage.

 

"Oh My Heart" from REM's latest, "Collapse into Now," is a love song to New Orleans, very French in its waltz rhythm and accordion backing. With integer mode off, it's a very nice tune. With integer mode on, the soundstage is bigger and the attack on the strings of Peter Buck's mandolin is more apparent, both of which help the romantic feeling. And there's a critical difference in the vocals. I heard a radio interview with Michael Stipe last week where he talked about how painfully shy he was when he began his career, and how shy he still is by nature. He has a little quaver in his singing voice, which you can hear with integer mode off. But with integer mode on, the micro-dynamics are better and I swear with that quaver you can almost tell it's a shy man singing.

 

[wgscott]

I thought that new REM song was about impending myocardial infarction.

 

[goldsdad]

Jud, thanks very much for the detailed report. Sounds similar to improvements one hears when comparing hi-rez and Redbook versions of some high quality recordings. Hopefully, I'll have an opportunity to check for myself soon.

 

 

[Forehaven]

Thank You very much Barrows, and Ted_b for describing how the LIO dac works. I had no idea of the use of oversampling an digital filters were being used to produce the native rates. Interesting how diff. dac chips do the conversions differently.

 

And Ted, I too have upconverted several redbook cd's to 24/192 files using Wave Editor's iZotrope converter. I get more detail/microdynamic music, but the 'life' of the music is lost some over the redbook. I prefer the 16/44 cd's sound. Maybe I did an overkill going that large. Perhaps had I only doubled it, to 24/88, I could have both the incr. detail/microdynamics and still retained the 'life' that's on the redbook. But I agree, it's a major pain and not something that I want to do, for questionable improvement. It's all so subjective.

 

[DaveLew]

Last time I looks, double precision floating point was an 11 bit exponent and a 52 bit mantissa. I'll bet that if you converted floating point conversions to integer and compared those purely fixed point operations, you'd be hard pressed to find any numbers that differed, and if they did, the difference wouldn't extend beyond the least significant bit. No ones hearing is good down to 2 to the 24th!

 

 

 

 

If you don't understand how things work, everything matters. If you understand how things work, you focus on things that matter.

 

[Mercman]

We all agree that no information is lost with the 32 bit float conversions. The computer processing is reduced with integer. Most of us can hear a significant difference in sound quality. Try it and let us know what you hear.