My Custom Designed Music Server for under $500 that matches $3,000 Servers

[iago]

As for the power supply chip, I think my diy design is better; it uses discrete analogue components instead of IC's..here is the link again for reference:

 

The σ11 regulated power supply

 

 

I have seen the σ11 design before, actually it measures quite well (10 uV ripple) and it can definitely deliver higher currents, even without external heat sinks. What puts me off is the number of bespoke parts and the fact that you have to source them yourself.

 

The TI LDO actually is slightly better than the σ11 at 4.5 uV residual ripple. But limited to 1A of output current, so σ11 might be the better option ...

 

There is also the John Svenson PSU, it was designed for 5V, but could be adapted. I have never seen any measurements of this layout, so I don't know how it compares.

 

I applaud your choice of the Juli@ - that has long been on my recommended list of components for music servers. What are you outputting to with the Juli@?

 

Actually the acquisition of the Juli@ goes back to it's use in the early versions of C.A.P.S on this site, so the credit belongs to Chris Connacker :-)

 

I use two output paths from mpd, one via USB into WaveIO into Mutec MC-3+ into WLM Gamma Reference DAC into Krell S300i and Martin Logan Ethos. The Juli@ outputs via analogue out into F2A11 tube monoblocks and Fostex (FE103 Sigma) back loaded horns. Since the second chain does not have volume adjustment, I employ the hardware volume control on the Juli@ for that.

[DarekLogic]

Darek here from Logic Supply. Looks like a fun project Ron. Thanks for sharing your build and for the shout out about our MC600 Mini-ITX case. We'll be keeping an eye out for any future upgrades you decide to do.

[randytsuch]

Very cool project.

 

Few things I can share with you on the power supply.

 

have a highly tweaked PC which uses a combination of batteries and Linear supplies. The batteries actually act as capacitors, smoothing things out.

 

My linear is pretty simple, a pre regulator and then another 3 pin regulator, and without the battery it was too slow on the 5V rail. PC would not boot up. During boot up, I think there are large variations in current draw on +5, and my supply was too slow, so voltage droop too much, and cause a reset. No idea how fast the AMB is to large variations in current demand.

 

Other thing is the power on sequencing. Strictly speaking, ATX has requirements for timed sequencing of the voltages and signals when you turn a PC on. But some motherboards are more forgiving than others about this. I found a MB that didn't care about the power on sequence, so it works fine with my homemade linear PS. I use relays to power it up, but they are for convenience, not for power on timing.

 

And I'd be interested in trying your linux if you ever make it available. I'm running Windows Server 2012 R2 right now.

 

Good luck

Randy

[rrstesiak]

Reply to Pneumonic:

 

I actually did to FFT comparison analysis on a much more sonically variable project over the summer and sadly concluded my MacBook Air and it's built-in microphone were insufficient to render significant enough differences in the waveforms to be useful. I later consulted with a veteran audiophile and they indicated I would need to purchase a special microphone that is calibrated and designed strictly for measuring audio performance. At this point I just don't have the extra $400 or so to justify the expense and have resorted to detailed note taking simply using my own ears as a guide; performing A/B testing.

 

Best Regards,

 

Ron

[rrstesiak]

reply to iago:

 

Thanks for the validation on the sigma11...I need higher amperage (a little under 4 to be more exact)...and good use of the Juli@!

 

Best,

 

Ron

[rrstesiak]

Reply randtsuch:

 

Yes..batteries would be the next step past linear!

 

As for sequencing, I too have read of this but am pretty certain I am lucky in that current board has a DC in, so it's good to go.

 

And the linux..I am just using the Debian "Wheezy" network install (all extras, including x-windows GUI stripped out)...and installing the 64-bit realtime kernel. I use it commandline only, and it is bitperfect and very low latency as very very few extra system daemons/process are running. I am also aware many commercial audiophile manufacturers use a pre-built flavor of linux much in the same way.

 

Respectfully,

 

Ron

[rrstesiak]

Darek:

 

Thanks for the recognition. The Logic Supply case is perfect with two extra PCIe slots and very well designed for these ultra low power embedded systems. I may be interested in getting a front panel with the LCD but could not find pricing on the website. perhaps you could provide a link?

 

Best Regards,

 

Ron

[DarekLogic]

Darek:

 

Thanks for the recognition. The Logic Supply case is perfect with two extra PCIe slots and very well designed for these ultra low power embedded systems. I may be interested in getting a front panel with the LCD but could not find pricing on the website. perhaps you could provide a link?

 

Best Regards,

 

Ron

Hi Ron, the optional LCD front plate for the MC600 isn't an option we offer on the front end of our website, and it's only provided as full kit (with the case included). That said, if you were interested in special ordering just the front plate, our sales team may be able to provide you with a quote. Here is a link to the full kit product: Logic Supply Compact Case with LCD Display | Logic Supply.

 

If you did want to move forward with an order, you would need to give our sales folks a call (802-861-2300 option 1) and let them know that you already have the MC600 case and are interested in purchasing just the front plate and LCD for the MC603.

[rrstesiak]

Reply to Darek:

 

Thank you for the link -

 

I am now deciding on whether or not to stick with my original, minimalist two-line LCD display already purchased; or to go to Logic Supply's 4 line display with controls.

 

Best Regards,

 

Ron

[rrstesiak]

All:

 

Being new to this forum, I figured it a good idea to literally show what my listening setup is currently:

 

 

The components are listed in my signature; except for the Bowers & Wilkins P5 v.2 headphones on the bottom middle of my music rack. As currently configured, the speakers are the furthest out from the wall I listen, with zero toe-in. The tone controls on the Creek integrated are bypassed, and the subwoofer is on. I find this setup to be the best at rendering sound stage width and depth.

 

This is essentially a near-field setup.

 

Best Regards,

 

Ron

[jabbr]

Jud:

 

I've actually just today implemented playback from memory. I currently have 1 8GB module, with room for another, so that should be plenty.

 

As for HQPlayer, I have heard good reviews; especially in the DSD domain.

 

Currently, I do not own DSD files but I would like to explore that format.

 

However, I am a little confused by the logic of claiming software performs better than a hardware DAC. Please elaborate...

 

As with everything, the details tend to be important. There is no such thing as a "software only" DAC. That said, most so-called "hardware DACs" are themselves programmed with software. So this distinction is not real, just depends on where the processing takes place. Specifically, most of these DAC chips both do their own upsampling and then delta-sigma modulation e.g. see this application note from Analog Devices .

 

So you can do this math using a relatively cheap, resource constrained DAC chip, or use your latest generation multicore Intel Xeon to do the math.

 

I am sure if you have done anything like computational neuroscience that you will recognize this issue of special purpose vs. general purpose CPU going back decades. However, then realize that these DAC chips are not all that sophisticated in terms of processing capabilities ... no one has ever suggested building a supercomputer on these :)

 

So consider that HQPlayer can use better filters and better upsampling than some cheap DAC chip. HQPlayer can also take your FLAC PCM files and do the upsampling/delta-sigma modulation/filtering in software and then send the, say DSD256 stream direct to the DAC which then implements the DAC function itself. The DSC1 is an open hardware design which shows you exactly how simple this function is.

 

My design principles are that the LESS processing of the music file on the music server the better... in fact, my goal of my server is to not perform any processing, including volume, filtering, conversion, etc. on the FLAC files. A specific point to support this view is even changing the volume via software reduces the resolution! I have read many different places confirming this approach as "Best Practices" when implementing an audio server WITHOUT dac. The linux OS then feeds this signal to the USB/SPDIF converter hardware, which in turn feeds it to the DAC, where HARDWARE specifically designed for the processing of digital files performs any filtering, re-sampling, etc. In my case it is a Bryston BDA-1, at one point a $3,000 DAC.

 

The open-source hardware DAC looks interesting.. I need to first come up to speed on it before I can comment further.

 

The breakdown in your reasoning is your assumption that a $3000 DAC has better facilities for digital filtering,re-sampling etc. than a $3000 computer. Not.

 

Your $3000 DAC ought spend its $$ on the analog output.

 

Now there is one reasonable counter argument, that being ladder, or "R2R" DACs which operate on PCM streams directly without converting to "DSD" i.e. there is no delta-sigma modulation. If you mean to say that you prefer "native" PCM DAC e.g. ladder/R2R then OK, and also you may prefer to manipulation of the PCM stream e.g. NOS. That is a defensible argument ... BUT ... don't assume that a DAC IC is somehow "better" at digital math than a general purpose CPU or GPU for that matter because they simply aren't.

[Pneumonic]

Reply to Pneumonic:

 

I actually did to FFT comparison analysis on a much more sonically variable project over the summer and sadly concluded my MacBook Air and it's built-in microphone were insufficient to render significant enough differences in the waveforms to be useful. I later consulted with a veteran audiophile and they indicated I would need to purchase a special microphone that is calibrated and designed strictly for measuring audio performance. At this point I just don't have the extra $400 or so to justify the expense and have resorted to detailed note taking simply using my own ears as a guide; performing A/B testing.

 

Best Regards,

 

Ron

That's unfortunate. This seems like a great opportunity to graph the spectra in order to see what the changed distortions look like and, if they are at audible levels or not.

[Jud]

Reply to Pneumonic:

 

I actually did to FFT comparison analysis on a much more sonically variable project over the summer and sadly concluded my MacBook Air and it's built-in microphone were insufficient to render significant enough differences in the waveforms to be useful. I later consulted with a veteran audiophile and they indicated I would need to purchase a special microphone that is calibrated and designed strictly for measuring audio performance. At this point I just don't have the extra $400 or so to justify the expense and have resorted to detailed note taking simply using my own ears as a guide; performing A/B testing.

 

Best Regards,

 

Ron

 

That's unfortunate. This seems like a great opportunity to graph the spectra in order to see what the changed distortions look like and, if they are at audible levels or not.

 

Don't know if there was something else besides the calibrated mic included in that $400, but see whether this might work:

 

Cross·Spectrum - Calibrated MiniDSP UMIK-1 Microphones for Sale

[rrstesiak]

jabber:

 

I am very familiar with software vs hardware.

 

in fact, you've left out a very cool hybrid approach used in a few high end DACS: FPGA.

 

I feel FPGA is actually the best of all worlds; just can't afford an FPGA based DAC and am very happy with my Bryston.

 

in addition, something very critical you are leaving out that is of huge benefit of hardware/FPGA based systems: realtime processing.

 

I don't have to boot up a powerful workstation to do any file conversions/processing/etc... I just have a huge library on my NAS, with a GUI front end on my iPad showing album art, etc...controlling my ultra low power and completely silent Linux music server.

 

So I simply pull up any artist/album/song I want and hit play - all instantly available.

 

I also prefer to have my DAC leave the native sample rate on the incoming FLAC file..again, the theory the less processing the better. My goal is to re-create the source music as close as possible; not corrupting it with artificial filters, effects, etc. Many audiophiles agree with this approach.

 

I think the software algorithms and processing you're referring to come into play with DSD and similar media...which is relatively new and bleeding edge...and it makes sense there exist software solutions. I feel as it matures; however, outstanding hardware platforms that can do realtime playing and processing will become more widely available / cheaper.

 

As as for me, I haven't checked out these formats but am curious and intend to after I finish my music server project.

 

in short, we are both "right". For the mature technology of FLAC/WAV/ALAC etc playback, hardware solutions are plentiful and already optimized. But since DSD/DXD etc are relatively new, software solutions make total sense.

 

I also appreciate the article you linked and found it educational. In addition to computational neuroscience and finance, I have an undergraduate degree in Electrical Engineering and so can follow such articles and discussion.

 

Respectfully,

 

Ron

 

Btw... As far as "some cheap DAC chip", that's simply incorrect. As an example, the widely used and comman Texas Instruments 1704 goes for $35. An Intel i3 goes for $100. I couldn't find pricing on the newest and more popular Ess Sabre chips, but I wouldn't be surprised if they approached the Intel in cost.

[rrstesiak]

Jud:

 

Regrettably, the $400 was just the cost of the microphone... I would also need to purchase an ADC!

 

Having said that, I followed the link you provided and that microphone looks highly compelling and does not require an ADC.

 

I may run it by my friend who is an audio engineer...I'm just concerned they "cut corners" somewhere but it may be totally sufficient for my purposes and I thank you for the tip.

 

Best Regards,

 

Ron

[Sal1950]

Ron

Just wanted to compliment you on the outstanding work you've done on your server design, I'm highly impressed! The tech level is miles beyond my station but I've enjoyed reading through you posts in any case.

Best Regards, Sal

[Jud]

Now there is one reasonable counter argument, that being ladder, or "R2R" DACs which operate on PCM streams directly without converting to "DSD" i.e. there is no delta-sigma modulation. If you mean to say that you prefer "native" PCM DAC e.g. ladder/R2R then OK, and also you may prefer to manipulation of the PCM stream e.g. NOS. That is a defensible argument ... BUT ... don't assume that a DAC IC is somehow "better" at digital math than a general purpose CPU or GPU for that matter because they simply aren't.

 

XXHighEnd does oversampling in PCM exclusively and is designed to work with the Phasure NOS R2R DAC, so the same arguments apply even if there is only sample rate conversion and no format conversion.

[Jud]

jabber:

 

I am very familiar with software vs hardware.

 

in fact, you've left out a very cool hybrid approach used in a few high end DACS: FPGA.

 

I feel FPGA is actually the best of all worlds; just can't afford an FPGA based DAC and am very happy with my Bryston.

 

For some reasonably informed discussion of how FPGAs fit into all this, here are a few links:

 

http://www.computeraudiophile.com/f8-general-forum/regardless-who-has-superior-algorithm-heavy-lifting-tasks-can-be-performed-much-more-efficiently-and-much-higher-precision-chips-and-fpgas-well-known-fact-25177/index2.html#post446578

 

http://www.computeraudiophile.com/f8-general-forum/upsampling-computer-or-digital-analogue-converter-25953/#post469427

 

http://www.computeraudiophile.com/f8-general-forum/regardless-who-has-superior-algorithm-heavy-lifting-tasks-can-be-performed-much-more-efficiently-and-much-higher-precision-chips-and-fpgas-well-known-fact-25177/index2.html#post446737

 

http://www.computeraudiophile.com/f8-general-forum/regardless-who-has-superior-algorithm-heavy-lifting-tasks-can-be-performed-much-more-efficiently-and-much-higher-precision-chips-and-fpgas-well-known-fact-25177/index2.html#post446530

 

http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/munich-2015-new-dacs-revealed-24426/index3.html#post429172

 

in addition, something very critical you are leaving out that is of huge benefit of hardware/FPGA based systems: realtime processing.

 

Timing for software upsampling is adjustable, and in fact HQPlayer on Linux uses a kernel with realtime patches.

 

BTW, I'm not at all trying to get you to change anything you like. Your project is very interesting and well done and your enthusiasm is great (and you're a fellow Pitt alum ). Just throwing info out there and if you're curious about any of it, go for it.

[jabbr]

 

I don't have to boot up a powerful workstation to do any file conversions/processing/etc... I just have a huge library on my NAS, with a GUI front end on my iPad showing album art, etc...controlling my ultra low power and completely silent Linux music server.

 

Likewise for me, because my workstation stays on 24/7 ... and my "music server" is a completely silent Linux Q1900M based machine.

 

So I simply pull up any artist/album/song I want and hit play - all instantly available.

 

I also prefer to have my DAC leave the native sample rate on the incoming FLAC file..again, the theory the less processing the better. My goal is to re-create the source music as close as possible; not corrupting it with artificial filters, effects, etc. Many audiophiles agree with this approach.

 

Well ... most audiophiles who use 'modern' NOS DACs, also use software upsampling. Unfortunately the PCM1704 is no longer made, but an excellent example of a DAC using this chip (actually 8 of these), is the Phasure NOS1a ... which also uses a highpowered workstation class machine to do the upsampling and filtering.

 

The vast vast majority of DACs both do upsampling as well as sigma-delta modulation.

 

I think the software algorithms and processing you're referring to come into play with DSD and similar media...which is relatively new and bleeding edge...and it makes sense there exist software solutions. I feel as it matures; however, outstanding hardware platforms that can do realtime playing and processing will become more widely available / cheaper.

 

DSD is not what I would call 'bleeding edge', its fairly common, and considering that *most* DACs do sigma-delta modulation, this means that the vast majority of DACs actually convert the PCM into 'DSD' form as part of the digital to analogue conversion process.

 

By 'cheap' I mean that, using the PCM1704 as an excellent example, the $35-75 cost of the chip is mostly devoted to the outstanding R2R ladder and not to any DSP function. Indeed this chip is mono and is usually paired with a DSP, but the point is that say there is $5 or even $35 of processing on the chip, that can hardly compete with a $300-1000 i7 or Xeon (needed to do high quality upsampling and filtering).

 

But you say that you don't want to do any upsampling or filtering ... uh you *aren't* listening to redbook CD by any chance are you? Tell me what type of low-pass output filters are you using on your DAC, and are you concerned about brick-walls at 44khz?

 

Regarding FPGAs ... these days the lines between these and general purpose CPUs have become increasingly blurred and sure this is a reasonable option, but these *aren't* DACs

[ElviaCaprice]

Doesn't Chord use FPGA in there DAC's? In fact the Hugo has a 26,000 custom coded tap length filter. You get the full FPGA when streaming redbook, but if you resample in the PC redbook to say hi Rez, you end up with a smaller tap length filter. Thus you don't want to resample from the PC to the Chord Dac, at least for redbook.

[Jud]

Doesn't Chord use FPGA in there DAC's? In fact the Hugo has a 26,000 custom coded tap length filter. You get the full FPGA when streaming redbook, but if you resample in the PC redbook to say hi Rez, you end up with a smaller tap length filter. Thus you don't want to resample from the PC to the Chord Dac, at least for redbook.

 

Number of taps is the new megapixels. I see a couple of manufacturers marketing numbers of taps - 15,000; 26,000....

 

While number of taps is some indication of the sophistication of a filter and the resources available to run it, PC/Mac resampling software has fewer limitations than DAC chips or FPGAs regarding number of taps. For example, iZotope sample rate conversion software bundled with Audirvana Plus allows something around 2.5 million taps if I remember correctly.

 

Like nearly everything, a larger number of taps has both advantages (the aforementioned use of a more sophisticated filter making use of greater computational resources) and disadvantages. The primary potential disadvantage is that, because taps can be thought of as the number of times a filter acts on a signal, then with all else being equal more taps means more filtering - signal cut - is applied. Greater signal cut (the steeper the filter) results in more ringing of the filtered signal, so the transient response of the filtered signal can suffer somewhat in comparison to a signal that hasn't been cut so steeply.

 

This isn't to say you may not like the filtering in the Chord better than some sample rate conversion software. But it is a situation where you may want to try alternatives rather than assuming the outcome will fall a particular way.

[ElviaCaprice]

This isn't to say you may not like the filtering in the Chord better than some sample rate conversion software. But it is a situation where you may want to try alternatives rather than assuming the outcome will fall a particular way.

 

I did try HQP and the Chord filtering sounded better without the HQP. Now with a different DAC?? I agree, the results may be different.

 

So in the case of using a Chord DAC I would recommend the low power PC build. Unless doing multi media, which I do, then you need something a little more powerful for quality graphics.

[Jud]

I also prefer to have my DAC leave the native sample rate on the incoming FLAC file..again, the theory the less processing the better. My goal is to re-create the source music as close as possible; not corrupting it with artificial filters, effects, etc. Many audiophiles agree with this approach.

 

I think the software algorithms and processing you're referring to come into play with DSD and similar media...which is relatively new and bleeding edge...and it makes sense there exist software solutions. I feel as it matures; however, outstanding hardware platforms that can do realtime playing and processing will become more widely available / cheaper.

 

 

I'll write about this once more to try to clear up any misconceptions, and then I'll leave off the particular subject of upsampling, because I like this thread and the last thing I want is to be an irritant.

 

- We already established in some detail with reference to the datasheets for the chips used in your DAC that doing upsampling in software would not make the signal undergo any additional processing. Software upsampling from 44.1 -> 176.4KHz would substitute one-for-one for the DAC chip's internal upsampling.

 

- For most current DACs, software upsampling actually results in less processing. This is because most DAC chips run the signal through their upsampling filters three times - once from 44.1 -> 88.2, a second time from 88.2 -> 176.4, and a third time from 176.4 -> 352.8. This 352.8KHz bitstream is then sent to the DAC chip's internal sigma-delta modulator, and from there to the final (usually analog) filter to be converted to an analog musical signal. Sample rate conversion software, on the other hand, will convert 44.1 to the desired PCM rate (usually 352.8) directly in one pass rather than having to do three iterations, so for most DACs two processing passes are cut out of the signal path if software upsampling is used.

 

- There is nothing "artificial" about software filters running on computers, and nothing "natural" about filters running on DAC chips or FPGAs. They are all algorithms running on hardware, performing sample rate conversion (interpolation) and anti-aliasing filtering. The only salient difference is that there is more freedom in programming filters running on a computer due to the greater computational resources available.

 

- There are two stages of rate/format conversion and filtering, whether we are talking about software running on a computer or FPGAs or DAC chips. One is interpolation and anti-aliasing filtering; the other is sigma-delta modulation. Interpolation (oversampling) and the necessary filtering have been going on in DAC chips since the 1980s. DAC chips performing sigma-delta modulation have been around since the 1990s, and for the past couple of decades have dominated the market. It is *very* rare to find a PCM-only DAC these days; you might be able to count the number of manufacturers who build such DACs on two hands, and perhaps have fingers left over. So there is nothing at all bleeding edge about sigma delta modulation, even though it is what is used to create DSD files (which have themselves been around for at least a decade). If you ask a chip manufacturer to show you DAC chips, for the past couple of decades that is what you would be shown - a chip that internally does interpolation and filtering to achieve 352.8/384KHz rates, then sends that bitstream to an internal sigma delta modulator to obtain a DSD-like signal to be sent to the final filter and converted to analog. Again, the only differences between your everyday normal DAC chip and sample rate conversion software are (1) that the algorithms are running on a computer rather than inside the chip, and (2) that there may be more sophisticated algorithms running on the computer due to the greater computational resources available.

 

And finally, the usual sign-off for this topic: None of this means you must or will like sample rate conversion and filtering done on a PC better than that done inside your DAC. You may not notice a difference, or you may like what your DAC does internally better. But you do want to be aware of what goes on inside your DAC, and that using software running on a PC is not doing anything to the signal that won't happen in your DAC anyway.

[RubenV]

Very cool project Ron! Since the relative low price I would to try this out when it works with a PPA USB card. I wonder how this minimalistic approach in hard- and software compares against my dual streamer setup.

 

@Jud: thanks for your explanations. Very interesting to read. We are lucky to have paople with all this knowledge in forums like these.

[rrstesiak]

Jud:

 

You seem to bring a wealth of good knowledge and it is highly appreciated!

 

Though we do not agree on a few concepts, I think much can be discussed and learned.

 

As far as sample conversion, I also do not do that. I allow the FLAC files to pass through unaltered.

If they are 44.1, then 44.1, if 192, then 192. etc.

 

However, I do find I get the best performance the higher the native sampling rate of the original file. As an example, my recording of Miles Davis: Kind of Blue on 192/24 is sonic bliss.

 

I think where we differ is primarily in source material - yours seemingly consisting of a large portion of DSD and similar, and mine of FLAC.

 

The techniques and products you mention are top notch and I have no disagreements there.

 

I welcome further dialogue and discussion.

 

 

Best Regards,

 

Ron