PDA

View Full Version : Yet Another DIY DAC (work in progress)



Tam Lin
05-11-2008, 17:47
The objective is to compare different DAC chips and topologies.

A local oscillator and synchronous counter provide all timing and export clocks to slave a CDP or PC sound card. Analog output via passive I/V, transformer, and deemphasis and reconstruction filters. Digital to analog conversion is non-oversampled, 2x or 4x oversampled with null insertion and/or analog interpolation, or 8x oversampled with digital interpolation. There is also a hybrid mode that combines 2x and 8x oversampling.

Additional features and options include:
· 256Fs or 384Fs system clock.
· Slow or fast digital filter roll-off.
· Input to the clock divider from MCLK instead of the oscillator for occasional use with unsynced digital sources.
· S/PDIF input via transformer-coupled coax or via twisted pair using the same differential transceivers that transmit the CDP clock; i.e., one Cat5 cable between the DAC and CDP.
· The sample position within the sub-frame is optimized for best sound. Right-shifting the sample reduces the maximum and RMS sample value, which reduces the maximum and RMS output current of the DAC. Right shifting also reduces the maximum and RMS step size, which reduces di/dt and DAC settling time.
· With different DAC and fan-out modules almost any kind or number of DAC chips can be evaluated.

I’ve designed two DAC modules so far: One uses four PCM1704K that can be configured as two parallel or differential pairs and the other uses two PCM1794A that can be configured as two singles or one pair. Each DAC module has a corresponding fan-out module: 1:16 and 1:4. Summing the output of many parallel DACs improves S/N, dynamics, and reveals more low-level detail. The observed effect is greater than the theoretical +3dB S/N per doubling because human hearing and perception are not simple linear processes. This technique is common practice in astrophotography.

The project started with the idea of comparing NOS with 8x digital interpolation using four PCM1704 and a DF1704. It then grew in successive stages:

· Added shift register to time-align the left and right sub-frames. (This is an issue only when using mono DAC chips.)
· Added shift register to optimize sample position in the sub-frame.
· Added 2x OS with null insertion. (This is the simplest form of OS and only required the addition of a mux.)
· Added fan-out for 16 PCM1704 modules.
· Added 2x OS with analog interpolation. (Requires more shift registers and twice as many DAC chips as null insertion.)
· Added 4x OS with null insertion or analog interpolation. (Requires more shift registers and twice as many DAC chips as 2x OS.)
· Added PCM1794 module. (Required a different interface to include signals for system clock, deemphasis, and reset.)
· Added fan-out for 4 PCM1794 modules.
· Added hybrid 2x & 8x OS. (This was lurking in the design and only needed a unique 2x word clock to make it usable.)

I achieved 4x oversampling by double pumping the DIR and demultiplexing the stereo data stream.

Double pumping the DIR means running the BCK and WCK inputs at twice the incoming sample rate. The DIR responds by outputting each sample frame twice. This requires a DIR configured in slave mode and a source that is also slaved to the DAC clock otherwise samples may be missed or repeated.

The sample frames coming from the CDP looks like this:

| L1 R1 | L2 R2 | L3 R3 | L4 R4 |

Where Ln and Rn are the left and right samples for frame n. The vertical bars represent the frame boundaries (WCK).

The outgoing frames from the DIR looks like this:

|L1 R1|L1 R1|L2 R2|L2 R2|L3 R3|L3 R3|L4 R4|L4 R4|

That 2x stereo stream is demultiplexed into two mono streams.

The demultiplexed frames look like this:

|L1|L1|L1|L1|L2|L2|L2|L2|L3|L3|L3|L3|L4|L4|L4|L4|
|R1|R1|R1|R1|R2|R2|R2|R2|R3|R3|R3|R3|R4|R4|R4|R4|

With that stream and four DACs per channel, all permutations of NOS, 2x, and 4x OS are possible. (For clarity, the following stream diagrams omit the L/R channel prefix.)

NOS:
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|

2x null insertion:
|1|1| | |2|2| | |3|3| | |4|4| | |
|1|1| | |2|2| | |3|3| | |4|4| | |
|1|1| | |2|2| | |3|3| | |4|4| | |
|1|1| | |2|2| | |3|3| | |4|4| | |

2x analog interpolation:
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|
|0|0|1|1|1|1|2|2|2|2|3|3|3|3|4|4|
|0|0|1|1|1|1|2|2|2|2|3|3|3|3|4|4|

4x null insertion:
|1| | | |2| | | |3| | | |4| | | |
|1| | | |2| | | |3| | | |4| | | |
|1| | | |2| | | |3| | | |4| | | |
|1| | | |2| | | |3| | | |4| | | |

4x analog interpolation:
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|
|0|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|
|0|0|1|1|1|1|2|2|2|2|3|3|3|3|4|4|
|0|0|0|1|1|1|1|2|2|2|2|3|3|3|3|4|

Combining null insertion with analog interpolation yields many possible combinations limited only by the multiplexer configuration. Two examples:

|1| |1| |2| |2| |3| |3| |4| |4| |
|1| |1| |2| |2| |3| |3| |4| |4| |
|0| |1| |1| |2| |2| |3| |3| |4| |
|0| |1| |1| |2| |2| |3| |3| |4| |

|1| |1| |2| |2| |3| |3| |4| |4| |
|1|1| | |2|2| | |3|3| | |4|4| | |
|1|1| | |2|2| | |3|3| | |4|4| | |
|1|1|1|1|2|2|2|2|3|3|3|3|4|4|4|4|

The hybrid 2x & 8x mode sends a 2x data stream to each of two PCM1794 chips, which do the 8x digital interpolation.

No interpolation method if perfect. Null insertion attenuates the signal and increases sampling noise. Analog interpolation attenuates the high frequencies. Digital interpolation adds enharmonic ringing to transients and music is full of transients. Perhaps combining 4x null insertion with 4x analog interpolation will be preferable to either one alone because null insertion better approximates the ideal sinc function and provides some high frequency boost.

Filterlab
06-11-2008, 11:26
You say you achieved the 4x oversampling rate by double pumping the DIR and demultiplexing the stereo signal (I assume into two mono streams), does the demultiplexing aspect have a negative effect on the quality of the signal when compared to the non-demultiplexed 2x oversampled signal? It is essentially adding an additional decoding process to the stream after all, and that (in my mind anyway) would make it immediately inferior, be it only marginally and perhaps inaudibly.

I thoroughly appreciate this sentence by the way, it proves you're really in touch with how the result matters in the real world and it also means that you haven't made that oft made error of 'numbers is everything':


...The observed effect is greater than the theoretical +3dB S/N per doubling because human hearing and perception are not simple linear processes.

More please Tam, this is first class stuff and heavily technical. :)

Filterlab
06-11-2008, 11:33
Also, I guess you're listening mainly and measuring secondly, may I ask what equipment you're listening through?

Tam Lin
06-11-2008, 18:50
You say you achieved the 4x oversampling rate by double pumping the DIR and demultiplexing the stereo signal (I assume into two mono streams), does the demultiplexing aspect have a negative effect on the quality of the signal when compared to the non-demultiplexed 2x oversampled signal? It is essentially adding an additional decoding process to the stream after all, and that (in my mind anyway) would make it immediately inferior, be it only marginally and perhaps inaudibly.

I’m not quite sure what you are getting at. Of course the mono data streams are fed to mono DACs. And, like mono amps, mono DACs don’t suffer from cross-talk as stereo DACs do. The only potential problem that comes from demultiplexing the stereo samples is not time-aligning the left and right sub-frames. You also might look at demultiplexing as undoing the damage that was inflicted on the signal by multiplexing it in the first place.

In the digital recording process the signals from the stereo microphones are sampled simultaneously at precise intervals, converted to binary numbers, and recorded. At that moment, the samples are divorced from time and become featureless binary numbers. In the digital playback process, the samples are put in their original order and, at precisely the same intervals at which they were recorded, the left and right samples are simultaneously converted to analog signals that are more or less identical to the original signals from the microphones.

Until the samples are reunited with time during the A/D process, what happens to the bits has no effect on the sound they represent. Almost all digital recording and transmission methods are serial and the left and right channel data is necessarily multiplexed. Not only that, but the bits themselves may be reordered and translated into different encoding schemes for different media.

Contrary to popular belief, the data on the CD is not the ones and zeros of the original samples that are sent down the S/PDIF pipe as they are read from the disc. To create the CD data image the samples are dissected, interleaved, combined with error correction syndromes, track ids, and other bits necessary for the CD reading hardware to stay on track. In fact, only half the “bits” on the CD actually contain sample data. That conglomeration of bits are then converted into an 8:14 encoding scheme and recorded as a sequence of pits on the surface of the CD disk.

When the CD is read, the 8:14 data is decoded into octets, which are buffered in local RAM inside the CD reader hardware. The bits from each sample are scattered across many octets. Misreading a single octet only effects 1 bit in each of a number of samples and one-bit errors are always corrected. If an octet contained 8 bits from the same sample, error correction would not be possible and the entire sample would be invalid. The samples are then reconstructed by gathering bits from different octets in the RAM buffer. Finally the samples are assembled in to a frame with parity and other information, translated to bi-phase encoding, and transmitted via S/PDIF. In a one box CDP, only the last step is omitted. Note: Recording on a hard disk involves a similar process.

If, in your estimation, my demultiplexing the stereo signal might degrade its quality, how could any digital audio signal survive the massive multiplexing/demultiplexing, encoding/decoding, rearranging, and other manipulations it goes through before I get it? The reason is very simple: Until the samples are locked in time and converted to analog, there is no audio and no sound. They’re just bits. As long as the bits going into the DAC are identical to the bits that came out of the ADC and the sampling interval is identical, nothing has changed other than the anomalies produced the A/D and D/A processes.

In the DAC box, the bi-phase S/PDIF data is decoded and the sample bits are rearranged to match the format needed for the DAC chips, e.g., left-justified, right-justified, 16-, 18-, 20-, or 24-bits, or I2S. In a stereo DAC chip, the left channel sample is received first and is latched. After the right sample is received and latched, both samples are converted to analog simultaneously. True mono DAC chips have no concept of left and right; they convert the sample they receive when they receive it. (As with all things, there are exceptions but lets keep this simple, for now.) If we send the stereo data stream to two mono DACs and instruct one to convert the first (left) sample and the other to convert the second (right) sample, there will be an 11 us delay between the changing of the left and right samples. All Audio Note DACs that use the AD1865 exhibit this defect but nobody seems to care. (The AD1865 is essential two, independent mono DACs in the same chip.) They say the delay is exactly the same as moving one of your speakers forwards or back a couple of millimeters and nobody can hear that. Oh, yeah? If no one can hear an 11 us timing anomaly, how can anyone hear the effects of jitter. After all, jitter is exactly the same thing as rapidly moving your speakers forward and back a few nanometers, a distance shorter than the wavelength of visible light, 44100 times a second. Anyone who claims to hear jitter but doesn't hear an 11 us inter-channel delay has no credibility with me.


Also, I guess you're listening mainly and measuring secondly, may I ask what equipment you're listening through?

I think, study, and analyze first; build/modify and measure second; listen and evaluate third. Then, if I am not satisfied, I repeat the entire process. I don’t subscribe to the notion that you can start with a defective, ill conceived, or sub-optimal design and turn it into something exceptional by substituting different capacitors, tubes, and opamps, or by sprinkling it with fairy dust and snake oil. Nor do I subscribe to the notion that a theoretically perfect design will always sound perfect: Witness digital audio – perfect sound forever. (I’ll save that rant for another time.)

I listen through a Hovland HP-100, Art Audio Jotas, and Avantgarde Trios. I also have an SME-30 with an SME IV/Vi arm and Cardas Heart MC cartridge, and 8 or 9 other DACs to serve as reference as well as numerous real, live, acoustic, musical instruments. All interconnects and cables are my own design.

Filterlab
06-11-2008, 19:24
Ahh, I see. Your first and sixth paragraph have cleared things up in my mind. I'm not anywhere near as knowledgeable of digital as you, but I am very interested in learning as much as I can about the entire process of obtaining the best results from converting a digital signal into a very high quality audio signal. It's a subject that becomes more relevant by the week as more and more advance is made in digital based technology in all areas of the arena, and now with another digital source entering the audiophile world (i.e. computers) it's ramped up the requirements for very high quality DACs.

I may well throw more seemingly odd questions at you, bear with me. :)


...Witness digital audio – perfect sound forever. (I’ll save that rant for another time.)...

Oh, you may have a large discussion on your hands there mate, but I know exactly which angle you're viewing it from. :)

Tam Lin
06-11-2008, 20:11
Oh, you may have a large discussion on your hands there mate, but I know exactly which angle you're viewing it from. :)

Hmmm. Which angle might that be? :scratch:

Filterlab
06-11-2008, 22:02
The angle of someone who truly understands digital and how if it's done correctly it can be a perfect way to store music.

Tam Lin
08-11-2008, 02:15
That’s an interesting perspective. Many people who know far more than I have been trying to do digital audio “correctly” for many years and, in my opinion, have come up short. One of the reasons is that the performance bar was set so low. Whereas analog recording techniques, theoretically, can be improved without an upper limit, digital audio is constrained by the limits of the mutually agreed upon sample rate and sample width. The 44.1K samples per second and 16-bit samples chosen for Red Book CD presents a severe limitation. It’s amazing RBCD can sound as good as it does.

In the 1970’s I was involved with digital music synthesizers. The DAC chips of the day were only 10- or 12-bits wide and very slow. Obviously, the sound quality was not very good. The big question was: How fast and how wide did the sampling need to be to make digitally recorded music indistinguishable from live music. Stanford University’s newly endowed Center for Computer Research in Music and Acoustics took up the challenge. Their study found that the minimum requirement was 32-bit samples with a 1 MHz sample rate. At the time, 32-bits at 1 MHz was like asking the Wright brothers to build an airplane that could exceed the speed of sound.

Today some would argue that the minimum requirements set by the CCRMA study are too severe and that today’s DVD-A and DSD are sufficient. I’m not so sure. There is always room for improvement and until we can do 32-bits at 1MHz, we will never know.

Togil
08-11-2008, 11:30
The big question was: How fast and how wide did the sampling need to be to make digitally recorded music indistinguishable from live music. Stanford University’s newly endowed Center for Computer Research in Music and Acoustics took up the challenge. Their study found that the minimum requirement was 32-bit samples with a 1 MHz sample rate. At the time, 32-bits at 1 MHz was like asking the Wright brothers to build an airplane that could exceed the speed of sound.

.

How did they prove this, without the technology ?

Tam Lin
08-11-2008, 16:19
The CCRMA study’s conclusion was not a formal proof but a reasoned conjecture. Matters of human perception are rarely 100% provable. Remember, the question the study attempted to answer was: What is the minimum sample width and rate that would make digitally reproduced music indistinguishable from the real thing. When developing the CD a dozen years later, Sony and Philips approached the question from the other side and asked: What is the minimum sample width and rate necessary to reproduce 20-20,000 Hz with low distortion and 90dB S/N. There is little doubt, although the first generation of digital recordings met the Sony/Philips requirements, the sound was easily distinguishable from the real thing.

I should also point out that the CCRMA study looked at PCM only and did not consider the effects of digital filters, which have become so prevalent these days. In my opinion, the sonic signature of digital filters makes the whole question moot.

Togil
08-11-2008, 16:33
This is an extremely interesting thread.

Bob Stuart of Meridian once claimed that 44.1 was a bit mean ( 96 would have been better ) but the 16 bits can be improved on with the correct recording techniques and a player that doesn't truncate - can you comment on that ?

Tam Lin
08-11-2008, 17:54
Bob Stuart of Meridian once claimed that 44.1 was a bit mean ( 96 would have been better ) but the 16 bits can be improved on with the correct recording techniques and a player that doesn't truncate - can you comment on that ?

Not really. Whether 44.1K/16 or 96K/16 is “good enough” is a matter of opinion. Many people are satisfied listening to MP3 through ear buds.

If the goal is to reproduce music that is indistinguishable from the real thing, in my opinion, DVD-A comes the closest. But, because so few titles of interest are released in that format, it does not warrant further consideration. Of the remaining formats, RBCD (without digital filters) has the most appeal because it has the most titles and does the least damage to the aspects of music that are important to me: Harmonic balance and dynamics.

Tam Lin
18-11-2008, 05:17
I gave your question another look and took it to my oracle, a DAC simulator I wrote a while back. For each sample point, the simulator compares the analog output level of a DAC with the level of the original analog signal prior to ADC/DAC conversion. Using a theoretically perfect DAC, sample width trumps sample rate every time.

Bits Rate Freq dB
16 44 20 92
16 44 20k 92
16 96 20 92
16 96 20k 91
24 44 20 140
24 44 20k 140
24 96 20 140
24 96 20k 140

The simulator doesn’t consider the effects of the reconstruction filter, the quality of which depends on sample rate. Higher sample rates ease the requirements for the reconstruction filter and that is the sole reason for oversampling. The simulator does consider the effects of dither, jitter, and settling time. Repeating the above sequence adding TPD high-pass dither and 200 ps sample jitter yields:

Bits Rate Freq dB
16 44 20 89
16 44 20k 87
16 96 20 89
16 96 20k 87
24 44 20 137
24 44 20k 92
24 96 20 137
24 96 20k 92

Repeating the above sequence adding 200 ns settling time yields:

Bits Rate Freq dB
16 44 20 89
16 44 20k 43
16 96 20 89
16 96 20k 43
24 44 20 137
24 44 20k 45
24 96 20 137
24 96 20k 45

The first time I saw the effects of settling time was an eye-opening experience. For the last 30 years, while everyone was worried about jitter, the real monster in the room was settling time. The effects of both settling time and jitter are related to the step size with high frequencies having the largest steps. The typical settling times for parallel DACs is 1,000 times larger than the worst jitter. That’s when I started investigating how I could reduce settling time.

Tam Lin
18-11-2008, 05:26
Bob Stuart of Meridian once claimed that 44.1 was a bit mean ( 96 would have been better ) but the 16 bits can be improved on with the correct recording techniques and a player that doesn't truncate - can you comment on that ?

I gave your question another look and took it to my oracle, a DAC simulator I wrote a while back. For each sample point, the simulator compares the analog output level of a DAC with the level of the original analog signal prior to ADC/DAC conversion. Using a theoretically perfect DAC, sample width trumps sample rate every time.

Bits Rate Freq dB
16 44 20 92
16 44 20k 92
16 96 20 92
16 96 20k 91
24 44 20 140
24 44 20k 140
24 96 20 140
24 96 20k 140

The simulator doesn’t consider the effects of the reconstruction filter, the quality of which depends on sample rate. Higher sample rates ease the requirements for the reconstruction filter and that is the sole reason for oversampling. The simulator does consider the effects of dither, jitter, and settling time. Repeating the above sequence adding TPD high-pass dither and 200 ps sample jitter yields:

Bits Rate Freq dB
16 44 20 89
16 44 20k 87
16 96 20 89
16 96 20k 87
24 44 20 137
24 44 20k 92
24 96 20 137
24 96 20k 92

Repeating the above sequence adding 200 ns settling time yields:

Bits Rate Freq dB
16 44 20 89
16 44 20k 43
16 96 20 89
16 96 20k 43
24 44 20 137
24 44 20k 45
24 96 20 137
24 96 20k 45

The first time I saw the effects of settling time was an eye-opening experience. For the last 30 years, while everyone was worried about jitter, the real monster in the room was settling time. The effects of both settling time and jitter are related to the step size with high frequencies having the largest steps. The typical settling times for parallel DACs is 1,000 times larger than the worst jitter. That’s when I started investigating how I could reduce settling time.

Marco
18-11-2008, 13:14
Hi Tam Lin,

I meant to ask, are you from the US/Canada region rather than the Carterhaugh in Scotland? Where exactly is the Carterhaugh where you live? I think we got it wrong! :doh:

:)

Marco.

Tam Lin
18-11-2008, 13:44
My virtual self is in Scotland. My corporal self is in the US.

Marco
18-11-2008, 13:47
Ah, I see...

You also write somewhat 'American', too; certainly not like a Scotsman! ;)

So which country are you native to? If I spoke to you in person would it be a case of "Och aye, Jimmy" or "Hi, how y'all doin'?" :)

Marco.

Tam Lin
18-11-2008, 14:28
Born and raised in the USA. I was caught up in the US folk music revival in the 1950’s. The roots of US folk music are ballads from the British Isles. I like the Tam Lin story, which was later immortalized by the incomparable Sandy Denny and Fairport Convention.

Marco
18-11-2008, 14:49
Thanks for that :)

So what's your first name, then? I thought it was Tam! :lol:

Apologies for my ignorance of US folk music...

Marco.

Filterlab
18-11-2008, 15:09
Thanks for that :)

So what's your first name, then? I thought it was Tam! :lol:

Jeez! I thought he was Japanese, understandably backed up by his superior knowledge of electronics. How wrong one can be!

Humourously (or not in some eyes) I wondered if Tam's surname was Boreen. :lol:

I'll get my coat.

Tam Lin
18-11-2008, 15:50
I've been called many different names, some not so complimentary. I think Tam is appropriate for this venue.

Filterlab
18-11-2008, 15:59
Tam it is then. :)

shane
18-11-2008, 20:20
Marco, have you never heard Sandy Denny? You really should sometime. Without doubt the greatest female voice to come out of the UK last century. (You'll have heard her on Led Zep IV, but that doesn't count...).

Marco
18-11-2008, 20:38
I'll check her out then Shane - got any good links? :)

Marco.

snapper
18-11-2008, 21:07
Hi Tam Lin,

I meant to ask, are you from the US/Canada region rather than the Carterhaugh in Scotland? Where exactly is the Carterhaugh where you live? I think we got it wrong! :doh:

:)

Marco.

Tam Lin

I forbid you maidens all that wear gold in your hair

To travel to Carterhaugh, for young Tam Lin is there



Marco, have you never heard Sandy Denny? You really should sometime. Without doubt the greatest female voice to come out of the UK last century. (You'll have heard her on Led Zep IV, but that doesn't count...).


I'll check her out then Shane - got any good links? :)

Marco.


I'll bring some down.

:smoking:

shane
19-11-2008, 09:26
I'll check her out then Shane - got any good links? :)

Marco.

There are a couple of websites dedicated to her, but they don't get across the sheer emotional power of her voice. My favourite tracks are "Who Knows Where The Time Goes?" from the Fairport Convention album Unhalfbricking, "One More Time" from Rising For The Moon, and "Banks of the Nile" from the eponymous album Fotheringay. Sounds like Snapper is in a better position to complete your education, though!

Not forgetting The Battle of Evermore on Led Zep IV, the only time a guest vocalist ever appeared on a Zep album.

Marco
19-11-2008, 09:45
Great stuff, guys, - cheers!

I like Fairport Convention so I'm sure SD will hit the spot :)

Yep, I'm sure Snaps will duly educate me on all matters Sandy Denny over the next few days :smoking:

I do like folk music.

Marco.

Cotlake
19-11-2008, 21:50
I do like folk music.Marco.

Hmmm, bring some along to the next Owston to be sure not to ruffle any feathers, PP or SE regardless. You know it's always safe to play safe ;)

Telstar
26-12-2008, 11:48
Today some would argue that the minimum requirements set by the CCRMA study are too severe and that today’s DVD-A and DSD are sufficient. I’m not so sure. There is always room for improvement and until we can do 32-bits at 1MHz, we will never know.

DSD is so flawed in its implementation that i wouldnt consider it at all. Audiophiles likes it because it rolls off HFs, and sounds smooth, thats it.
DVD-As, which are 24/96 pcm come much closer to the ideal. Unfortunately the lack of good transports (the only one that i could define such is the esoteric ux-01) and quality DACs compromised its diffusion and now its a dead standard.

Computers come to the rescue. Digital, master-quality, uncompressed tracks are (finally) a reality now, and from direct and undirect experience i can say that already 24/88.2 comes out close to the ideal.

In one study made by Chord they conclude that around 700khz sampling rate is required for lifelike reproduction. Above that the human ear perceive only tiny feelings, but those go up to 1,5Mhz. I tend to believe them. The point is how to reach that sampling rate without digital filters that do more bad than good.

Another important point, often ignored, is that the upsampling, if we want to do it, has to be done by an integer, i.e. the redbook standard should be upsampled at 88.2, 176.4, 352.8, 705.6. and i would stop there. 16x direct interpolation is too costly to obtain. But if already 176.4 khz sound so damn good (listen to reference recording hrx files), 8x would be my ultimate goal.

I dunno about bitrate. I think that real 24 bit can be enough, but thats not reality yet (dont confuse with wordlenght, real 24bit means 144dB sn/r).
I'm not sure what 32 bit gives us more with the dacs currently available. There is no R2R 32 bit dac that we could use, altough in military applications those exist.

OK, i think i have said enough for my first post, and i have yet to finish reading this thread.

Tam Lin
27-12-2008, 19:02
The point is how to reach that sampling rate without digital filters that do more bad than good.

Easy. Just record at the same sample rate you want to use for playback. What do digital filters have to do with it?

StanleyB
27-12-2008, 23:16
Another important point, often ignored, is that the upsampling, if we want to do it, has to be done by an integer, i.e. the redbook standard should be upsampled at 88.2, 176.4, 352.8, 705.6. and i would stop there. 16x direct interpolation is too costly to obtain. But if already 176.4 khz sound so damn good (listen to reference recording hrx files), 8x would be my ultimate goal.
I have my doubts about upsampling. The playback reminds me of a reel to reel tape that has been played that has been played for so long that the magnetically stored information loses its sharpness.
I consider oversampling to be a far better sounding solution, and 8X is possible from today's chips.

Telstar
28-12-2008, 21:33
Easy. Just record at the same sample rate you want to use for playback. What do digital filters have to do with it?

It is not available.
Max sample rate used in recording is 192khz. Correct me if i'm wrong. The next step in digital media will likely be 210khz. Still too few.

If i have to choose between poor oversampling and NOS i choose the first, but If i could have 4x or 8x good oversampling, i'd go for it (and that's my plan).

Telstar
28-12-2008, 21:41
Have you heard 8x or 16x direct (linear) interpolation? Linear interpolation attenuates high frequencies and more stages means more attenuation.


True, but it rolls off right where the noise remains are. I think 16x would attenuate too much, while 4x would be ideal, 8x maybe.
I'm looking at 4x 1704 per channel, sans using the ubiquitous digital filter.
Feeding the DAC with high-res material (native or software upsampled in the computer), 4x would be enough reaching 705,6 or 768khz which are damn close to the limits of the human ear.
As as far as I know, this has never been done in a commercial, 24 bit dac. Sigma delta, filters and asrc have been used and abused instead.

This player is what comes closer to my idea in a commercial product:
http://www.sixmoons.com/audioreviews/zeroone2/mercury.html

4x D.I. instead has been done here,
http://www.diyaudio.com/forums/showthread.php?s=&postid=1697593#post1697593

whith the old good 1541, achieving excellent results (better than 16x and 8x with an older design so not exactly A/B)
(read the last 4-5 pages, as they explain the current status)

At which status is your project?

Mike
28-12-2008, 22:52
reaching 705,6 or 768khz which are damn close to the limits of the human ear.

Bloody hell!.... those are pretty damn good ears! :eyebrows:

<places tongue firmly in cheek>

Tam Lin
28-12-2008, 23:10
This player is what comes closer to my idea in a commercial product:
http://www.sixmoons.com/audioreviews/zeroone2/mercury.html


To each his own. The last thing I want in my audio system is a PC with multiple muffin fans and a big switching power supply. It is possible to build a full-featured PC without fans.


At which status is your project?

As of last week the design was finished and I started the final PCB layout when I decided I really didn't like the divide-by-three circuit I had: It was a few nanoseconds shy of perfect symmetry. So I did a redesign there and, while I was at it, I added a few more features. That added three ICs to the BoM. Now I am looking for ways to reduce the parts count without degrading the performance.

Telstar
29-12-2008, 11:56
To each his own. The last thing I want in my audio system is a PC with multiple muffin fans and a big switching power supply. It is possible to build a full-featured PC without fans.


The pc should be powered from a different outlet or its plug be filtered.
Besides, the referral of the zero was related only to the DAC part of the design.
I believe that powering the lynx card from a different power source eliminates most of the RFI and other noise. My solution is more drastic, mac mini (basically noiseless and cheap) + firewire pro card such as rme fireface 400 -> dac with AES input.

I know that you are pursuing traditional transport to dac route, but i want to ask which digital inputs are you planning? I dunno if there is an AES to i2s converter, but i know that i2s can be feeded directly to the pcm1704 (like has been done in the zero unit above).

Thanks for the update, i'll be following this thread closely.

Telstar
29-12-2008, 11:58
Bloody hell!.... those are pretty damn good ears! :eyebrows:

<places tongue firmly in cheek>

:)
About half of the max according to the Chord "study".

Another important point would be to SEPARATE the XOs, one to work with multiples of 44,1k and the other for multiples of 96k (or 48k) [for Tam, think of dvd players and quality universal transports such esoteric xu-01; for the others think about 24/96 highres files, this sampling rate is the most common nowadays]

StanleyB
29-12-2008, 12:47
If i have to choose between poor oversampling and NOS i choose the first, but If i could have 4x or 8x good oversampling, i'd go for it (and that's my plan).
Shall I put you down for my 7510+ then;).

Tam Lin
29-12-2008, 13:55
I know that you are pursuing traditional transport to dac route, but i want to ask which digital inputs are you planning? I dunno if there is an AES to i2s converter, but i know that i2s can be feeded directly to the pcm1704 (like has been done in the zero unit above).

My digital input is S/PDIF, either coax or differential. It is the most universal digital audio interface and can be found on most CD players, transports, and PCs. Unlike I2S, it provides data integrity checking from the source to the receiver and it includes deemphasis status and other indicators. I use a local oscillator close to the DAC chips and slave the source, CD, PC, or other, to that oscillator. I am not concerned with the quality of the recovered S/PDIF clock. All I ask of S/PDIF is to deliver the sample bits and it does that very well.

I2S cannot be fed directly to a PCM1704 unless you like the sound of noise. I2S is a left justified, stereo format with two, 16-24-bit samples per frame. Input to the PCM1704 is right justified, mono with one, 20 or 24-bit sample per frame.

Telstar
29-12-2008, 17:54
Shall I put you down for my 7510+ then;).

ops typo. i meant the latter.

Telstar
29-12-2008, 18:05
My digital input is S/PDIF, either coax or differential. It is the most universal digital audio interface and can be found on most CD players, transports, and PCs. Unlike I2S, it provides data integrity checking from the source to the receiver and it includes deemphasis status and other indicators. I use a local oscillator close to the DAC chips and slave the source, CD, PC, or other, to that oscillator. I am not concerned with the quality of the recovered S/PDIF clock. All I ask of S/PDIF is to deliver the sample bits and it does that very well.


How do you handle the RF noise?
Did you consider AES/EBU at least?



I2S cannot be fed directly to a PCM1704 unless you like the sound of noise. I2S is a left justified, stereo format with two, 16-24-bit samples per frame. Input to the PCM1704 is right justified, mono with one, 20 or 24-bit sample per frame.

"This is a multi-bit design using the famous Burr Brown PCM1704 chip. There are no digital filters or upsampling chips on the DAC board. All the required digital filtering and upsampling is done in the transport section with our custom software. The digital audio is fed directly to the PCM1704 chips by internal I²S running at sampling rates of up to 24 bit/192kHz."

How they do it? My guess would be oversampling to 24 bit before feeding the 1704 with i2s.
Also, using spidif wouldnt limit the pcm1704s to 96khz?
The datasheet does not seem updated as iirc i have seen 16 bit feeded to the 1704 w/o problems.

Tam Lin
29-12-2008, 22:41
How do you handle the RF noise?

By careful circuit design with adequate bypassing, termination, and RF suppression techniques. By carefully partitioning the design and isolating major functional groups: power supply, all digital, all analog, and mixed signals. By careful PCB layout with adequate power distribution, ground planes, controlled trace impedance, and special attention to signals that enter and leave each module. By careful positioning of each PCB and component to minimize mutual interference. It’s not rocket science. I spent 30 years working with high tech startups in Silicon Valley and learned a little about high-speed digital circuits in the process.


Did you consider AES/EBU at least?

Why? This is a DIY project I’m pursuing for my own amusement, education, and eventual use. I am not designing a product to satisfy the wants and desires of anybody else. The only digital sources I have are redbook S/PDIF. If and when I get higher bit-rate sources I’ll modify what I have or make something new. As it is, all I have to do to support 24/96K or 24/192k is change the oscillator and change a tap on the clock divider. The design is very modular with each functional unit on it’s own PCB with its own power supplies. If at sometime I want to try an AD1955, for example, I’ll do what I did to add the PCM1794: Make a couple of small PCBs just large enough to hold two DAC chips, voltage regulators, and connectors; then add it in place of the similarly small PCM1704 and PCM1794 modules.


"This is a multi-bit design using the famous Burr Brown PCM1704 chip. There are no digital filters or upsampling chips on the DAC board. All the required digital filtering and upsampling is done in the transport section with our custom software. The digital audio is fed directly to the PCM1704 chips by internal I²S running at sampling rates of up to 24 bit/192kHz."

How they do it? My guess would be oversampling to 24 bit before feeding the 1704 with i2s.
Also, using spidif wouldnt limit the pcm1704s to 96khz?
The datasheet does not seem updated as iirc i have seen 16 bit feeded to the 1704 w/o problems.

That’s all marketing BS. First, a digital filter is a digital filter whether it’s implemented in hardware or software. Second, I already told you the PCM1704 does not support the I2S interface. If you don’t believe me, read the data sheet. If you don’t believe the data sheet, call TI and talk to the chip’s designer. If you don’t believe him and insist on feeding your PCM1704s with 16-bit I2S, I have nothing more to say to you. I learned long ago that you can’t win an argument with a fool.

Marco
29-12-2008, 23:51
If you don’t believe him and insist on feeding your PCM1704s with 16-bit I2S, I have nothing more to say to you. I learned long ago that you can’t win an argument with a fool.


Ahem! Can we have less of that sort of talk, please? Thank you! Debating is fine but keep it from being personal.

Marco.

Tam Lin
30-12-2008, 01:42
Ahem! Can we have less of that sort of talk, please? Thank you! Debating is fine but keep it from being personal.

Debate?? There is nothing to debate. Whether or not the PCM1704 was designed to accept an I2S data stream and/or 16-bit samples is a matter of fact, not opinion. I don't suffer fools who try to tell me how DAC chips work when they haven't even read a data sheet.


The datasheet does not seem updated as iirc i have seen 16 bit feeded to the 1704 w/o problems.

Telstar obviously knows more about the workings of the PCM1704 then I do and more then the engineers at TI/Burr-Brown who designed the chip and wrote the data sheet. I stand corrected and will refrain from further technical discussions in this forum.

StanleyB
30-12-2008, 07:33
Telstar obviously knows more about the workings of the PCM1704 then I do and more then the engineers at TI/Burr-Brown who designed the chip and wrote the data sheet. I stand corrected and will refrain from further technical discussions in this forum.
I know more about the workings of the PCM1716 than the engineers at TI/Burr-Brown. The PCM1716 datasheet is only accurate for chips made before RoHS compliance and in DIP package. Some of my product upgrades are based on my discoveries of undocumented features in the 1716.

So it wouldn't surprise me if the 1704 and its datasheet suffers from the same problem.

leo
30-12-2008, 15:11
The manufacturers of various dac chips do tend to leave a lot out of the datasheets, TDA1541 is also an example, theres huge amounts of details missing from the datasheet for this chip, modders over the years have found some secret stuff not easy to find which really boosts the performance
Theres a lot going off internally with some of the chips we just don't know about

Marco
30-12-2008, 17:50
I don't suffer fools who try to tell me how DAC chips work when they haven't even read a data sheet.


Tam, we value your contributions, and you're obviously a knowledgeable chap, but we don't suffer "fools" either who won't respect our ethos. You are not here to judge anyone. So either shape up or ship out - this is not your personal playground. Everyone is entitled to an opinion, technical or non-technical, whether you agree with it or not. You must learn to respect other people's views whether you deem them as 'correct' or otherwise. Please don't make me repeat myself!

Marco.

leo
30-12-2008, 21:28
Yes, the amount of times I've had to bite my tongue after reading some posts over the years:lol:

Marco
30-12-2008, 23:19
Leo, how would you fancy being in charge of the D.I.Y room? We'll give you a custom title of your choice :)

You can organise this section of the forum and lay it out the way you want, create an FAQ, categorise it into a reference area for projects/technical data - do whatever you like :smoking:

Let me know what you think.

Marco.

Filterlab
31-12-2008, 09:37
He's the man for the job!

StanleyB
31-12-2008, 09:57
He's the man for the job!
It's like putting an alcoholic in charge of a pub;).

Filterlab
31-12-2008, 10:15
Free beer. :)

Telstar
01-01-2009, 18:26
This is a DIY project I’m pursuing for my own amusement, education, and eventual use. I am not designing a product to satisfy the wants and desires of anybody else.


The above shows the typical closed-mindness of most diyers and audio designers.



I have nothing more to say to you. I learned long ago that you can’t win an argument with a fool.

Fool was a synonim of genius, so I take that as a compliment.
Obviously, I dont see the point to keep following this thread or your project.

Telstar
01-01-2009, 18:28
I know more about the workings of the PCM1716 than the engineers at TI/Burr-Brown. The PCM1716 datasheet is only accurate for chips made before RoHS compliance and in DIP package. Some of my product upgrades are based on my discoveries of undocumented features in the 1716.

So it wouldn't surprise me if the 1704 and its datasheet suffers from the same problem.

I have used two different dacs built around pcm1704 and both feed i2s to the chips, so the proof is there.
Besides, it accepts up to 210khz.

Marco
01-01-2009, 18:28
Hi Telstar,

It's good to see you contributing again. I hope you weren't offended by Tam's rather ignorant comments :)

Marco.

StanleyB
02-01-2009, 09:59
I have used two different dacs built around pcm1704 and both feed i2s to the chips, so the proof is there.
Besides, it accepts up to 210khz.
In 2005/2006 I made the first TC-7510 with non RoHS DIP chips. Then I did a new prototype with non-RohS surface mount chips. After that TI supplied me with the RoHS chips and I produced the MK1. The sound was nothing like the prototype. After a lot of exchanges with the TI technical people, it turned out the new chip was different as I had been telling them!

leo
02-01-2009, 11:27
Leo, how would you fancy being in charge of the D.I.Y room? We'll give you a custom title of your choice :)

You can organise this section of the forum and lay it out the way you want, create an FAQ, categorise it into a reference area for projects/technical data - do whatever you like :smoking:

Let me know what you think.

Marco.

Marco,

Thank you for the kind offer, I sent you a email, not sure if you got it

Leo

Marco
02-01-2009, 15:58
Hi Leo,

I've got it but not had a chance to reply yet.

When I get a chance I'll outline what I had in mind and you can mull it over :)

Marco.