Unimpeachable reference quality digital source

[wakibaki]

Up until very recently I thought of near-perfect reproduction as being unreachable and unaffordable and (in a way) a bit unnecessary.

Recently I came across this

http://www.linkwitzlab.com/reference_earphones.htm

Which led to this

http://www.johncon.com/john/amp/

I have ordered a set of the Shure in-ear phones on the strength of this. The Sony phones are available at very reasonable prices so I’ll probably get a set of these too. Inevitably I will build the amp, as the cost is trivial.

If you read the rest of what these guys have exposed on the web there's good reason to think that these items (the phones and amp) in combination will provide a quality of reproduction against which others can be judged.

Of course in-ear phones are not ideal for all listening, but it would be nice to be able to tell exactly where the flaws in something played back are coming from. For example, is it the recording? You can play the same source through different chains to get this kind of insight, but it is timeconsuming.

Now all I need is an unimpeachable digital source...

I have read a lot of posts about digital sources. While there is considerable excitement about higher definition recording, the majority of recorded material currently available is 44k1 16-bit and indeed many authorities insist that the standard is fully adequate if exploited to the full. Increasing numbers of people seem to accept that computers make excellent sources for audio. Where there are concerns they seem to centre around jitter. I find this difficult to understand, since this seems to me easy to eliminate, but without investigating the circuits of the commercially available USB and other DACs it is difficult to know to what extent the concerns are justified.

In order to sidestep these issues, and in the interest of providing a well understood and documented process whereby the digital information is converted to analog, I am thinking of constructing a player.

This would be a simple device stripping uncompressed linear PCM from a flash memory device and feeding it to a DAC. A high-quality low-phase-noise TCXO would be the clock source and should eliminate any concerns about jitter. No operating system as such would be required, a simple vending-machine style finite state machine could be implemented in a small FPGA or CPLD.

It is a shame that sufficiently fast devices are not available since a full-resolution PWM DAC could otherwise also be accommodated and would thoroughly satisfy my taste for knowing exactly what is going on in a particular implementation.

Be that as it may, given the construction of the device, data could for example be read from a CD on a PC, cross checked against a second CD if this was felt to be necessary, and uploaded to a flash memory which could then be played back in as near to jitter-free a fashion as imaginable without resorting to extraordinary measures.

I was impressed by the low noise performance of Tektronix hand-held scopes when run on their internal batteries, so I would probably aim at running the whole shebang from batteries.

Anyway I figure I can lash this up fairly readily.

If a suitable DAC is selected then the system might have its usefulness extended to accommodate future formats including 24-bit and 192kHz although a second TCXO may be required. Given that the logic is reprogrammable this need not be addressed immediately and prior provision could be made on the PCB.

There is also a case for a NOS 16-bit device dependent on linearity and other factors given that the intrinsic SNR of most program material is 96dB anyway. Any comments about reconstruction filtering will also be appreciated.

Flash memory of 16G is available, so if a FAT reader and some user interface were provided the device would be useful for more than just test purposes, although any additional functionality adds to the problems of keeping the system-generated noise low.

None of these components are exorbitantly expensive.

I am prepared to hear arguments that, for example, an iPod of a particular vintage with lossless compression is adequate to my requirements, however I would like to hear any recommendations as to which (chip) DAC might be employed.

Or perhaps someone else knows if there's any way of getting what I want, or close to it, without spending an unconscionable amount of money?

w

[mlihl]

Have you checked out the Slim Devices/Logitech Squeezebox? It's a LAN music player for less than $300 and already does what you want. You could mod it with a better PSU and probably tap I2S directly to feed an external DAC.

[wakibaki]

OK, this is a suggestion at least.

I am aware of the existence of the Squeezbox and its big brother the Transporter (prices starting at $2k US).

What I'm aiming at is more a cheap tool than an entertainment system.

There's too much room to argue about packets lost over the network with this kind of streaming device, and the operating system is too complex, even if full details were available, for the whole process to be readily transparent even to someone with considerable expertise. I have little doubt that the Slim Devices products are well enough engineered not to have any problems with buffer underruns or jitter, but my confidence is not equivalent to being able to say that the device design is well documented, or even highly regarded by purchasers.

I'm talking about building something so simple its operation could be understood almost as readily as a turntable by observation.

Buyers are confronted with a difficult decision when considering purchases. Devices generally come with a technical specification, and are tested for compliance before release. (Release test for some items is becoming increasingly perfunctory.) Other than this there are reviews and a great number of anecdotes published on the web.

Given that numerous producers are competing for business and that marketing departments are occasionally unfussy in the strategies they employ, it is unfortunately necessary to beware that not everything that one reads is written in good faith.

Is the Transporter really audibly better than the Squeezebox? Who can tell without buying both? ...not really an option as far as I'm concerned.

I am fairly sure that a number of the available DACs have performance errors below the level of perceptibility as determined by tests on a human population. This is, however, not my area of expertise, although I am prepared to make it so.

If a well documented system were available, then listening and other tests of any system against the reference system could eliminate concerns about its adequacy.

There are very many DACs on the TI site alone. Perhaps somebody knows which ones have a poor reputation?

w

[Wingfeather]

I know this is only picking on a tiny subsection of your enquiry, but:

Originally posted by wakibaki
It is a shame that sufficiently fast devices are not available since a full-resolution PWM DAC could otherwise also be accommodated

What do you mean, there? Pretty much any modern FPGA is fast enough to make a 24-bit resolution PWM DAC. I imagine one of the cheap Digilent dev boards would be suitable for this, and could just be included in your main enclosure - or, if you have the know-how (I don't), you could roll your own custom FPGA board. Either way, excellent results could definitely be attained.

[Javin5]

The idea to read the audio samples from a RAM, which is clocked by the same low jitter clock as the DAC, is indeed a way to essentially eliminate the jitter problem. And large flash memories are now very affordable,

Teac is using a 128 MB RAM-Buffer in its D-70 DAC unit, see www.teac.com/esoteric/D-70.html. I also read about at least one other DAC which reads the entire CD into RAM before playing it, but have unfortunately lost that link.

As far as DACs are concerned, the new ESS Sabre chips are state of the art; there is a thread right here in the digital section. Chips from Wolfson are also highly regarded.

Kurt

[ezkcdude]

Originally posted by wakibaki


In order to sidestep these issues, and in the interest of providing a well understood and documented process whereby the digital information is converted to analog, I am thinking of constructing a player.

This would be a simple device stripping uncompressed linear PCM from a flash memory device and feeding it to a DAC. A high-quality low-phase-noise TCXO would be the clock source and should eliminate any concerns about jitter. No operating system as such would be required, a simple vending-machine style finite state machine could be implemented in a small FPGA or CPLD.

Just make sure you don't reinvent the wheel:

Die, SPDIF ! Design of the Ethernet DAC

[tinears02]

ultimate source


..... with a TCXO/VCXO ?

[mako1138]

Originally posted by Wingfeather
What do you mean, there? Pretty much any modern FPGA is fast enough to make a 24-bit resolution PWM DAC. I imagine one of the cheap Digilent dev boards would be suitable for this, and could just be included in your main enclosure - or, if you have the know-how (I don't), you could roll your own custom FPGA board. Either way, excellent results could definitely be attained.

What are you thinking, a sigma-delta design? My DSP skills are pretty weak, but if you point me in the right direction I can try something out on a dev board.

[Wingfeather]

Originally posted by mako1138
What are you thinking, a sigma-delta design?

No no, Sigma-Delta is a type of Pulse Density Modulation - altogether different. Sigma-Delta is really quite ugly, mathematically problematic, and almost always unstable. I'm not a fan.

At its most basic, PWM can be implemented as Uniformly-sampled PWM (UPWM) with a counter:

- Pre-load the counter with the current sample's value,
- Set the PWM output HIGH at the beginning of the current sample,
- Wait for a bit,
- Set the output LOW once the counter reaches zero,
- Wait until the end of the current sample, and repeat.

This doesn't work for full-resolution audio because the counter clock rate has to operate at 2^N times the basic sample-rate, for N-bit audio. 2^16 * 44100 is 2.89Ghz - not feasible. Additionally, UPWM generates truly silly levels of harmonic and intermodulation distortion, the amount of which gets worse as the input frequency approaches the switching rate. Nobody uses PWM like this.

The simplest and most obvious approach is to oversample and noise-shape the input data. Oversampling raises the switching rate, and so keeps the input signal much lower than it - it thus helps a bit with distortion; noise-shaping allows a large dynamic range within the small bandwidth of interest by pushing the quantisation noise up into the new regions created by oversampling. This sort of scheme makes PWM at least a workable paradigm. A 100MHz internal clock rate is entirely suitable and around 16-bit performance can be achieved at 44.1kHz with this.

For higher levels of performance, you need to directly attack PWM's distortion via some sort of precompensation system, and lots of methods have been developed in the AES (that are far too complicated to describe here) - with excellent results. At the top-end, one scheme achieves a theoretical 26 bits of distortion-free modulation performance in the passband, with either a 4-bit or a 6-bit (can't quite remember!) UPWM modulator. Impressive stuff!

If you have access to the journals I can give you some interesting references to check out. They're all (or, nearly all) protected by patents - but I don't imagine that's a problem for personal use.

[mako1138]

Yes, I have access to journals (though it requires actually going to the library ). I'll have to review my discrete-time signals and systems material, urgh. Thanks for the pointers, I really appreciate it.