La momentul in care am aflat de noua abordare a DAC-urilor de la Chord s-au aprins niste beculete (de fapt ei au inceput cu DAC64)
A fost primul DAC care nu folosea un chip DAC per se, denumirea tehnologiei aleasa de ei a fost The Pulse Array DAC.
sursa imaginii: computeraudiphile
Din imagini am putut sa deduc urmatoarele:
1. totul (USB si ce mai facea el) se petrecea intr-un singur procesor Xilinx Spartan
2. banuiesc semnalul (digital sau analog) intra in chipul mare din dreapta?
3. are 2 clock-uri 1 pt procesor si altul pt partea audio?
Intre timp se pare ca tehnologia “a prins” si apar alte implementare bazate pe acelasi principiu ca de exemplu PS Audio Direct Stream
Principiul de functionare este prezentat mai jos versus un DAC clasic.
1. USB intra direct in DB (Digital Board = Procesorul de la Xilinx)
2. PCM este transformat in DSD, apoi este se face upsampling x 10
3. este transformatin in DSD cu adancime de 1 bit (mai sus nu se povesteste adancimea streamului DSD)
4. intra in analog board si gata
Acum spicuiri din posturile lui Ted Smith (omul din spatele implementarii) postate pe threadul de pe computeraudiophile:
No upsampling is perfect, but synchronously upsampling keeping at least the same number of bits, etc and going to a higher sample rate is fairly benign. Upsampling all inputs to the same rate so everything can be handled the same way has obvious benefits.
Tho technically you do interesting mixing, filtering, etc. in the sigma delta modulator and so technically always use 1 bit between every block, there’s no downside at all to letting the single bit get wider to accommodate the operations (e.g. two bits for a sum) and then do the remodulation from the wider word. I’ve chosen a 30 bits since it’s the natural output size of my upsampler and 10 x DSD rate is the LCM of 192000 and 176400.
Tho people at times think that single bit is THE defining feature of DSD, it’s really it’s high rate and noise shaping. I do maintain that the simplest signal to convert from digital to analog is a one bit signal (i.e. all you need is a low pass filter) But there’s no reason to assume that you loose all DSD goodness if you allow the bit width to grow. The key point is to not loose info by capriciously lowering the sample rate or loose info by truncating the size with an appropriate re sigma delta modulation (analogous to using dither when truncating width in PCM.)
Upsampling proper is not magic. Perhaps choosing good sounding upsampling filters is.
Single bit is important exactly when you go to analog in that it allows the whole converter to be just a low pass filter.
atupi? The output of the DAC is transformer coupled and has plenty of available current behind it. To implement an optional 20dB attenuator we use a relay to put in 15 ohm resistor shunts to ground.
I keep the DSD hump down by using double rate DSD to give me room to relax the noise shaping. Since I’m “the final stop” I don’t need to have a S/N > 160dB (or whatever) over the audio band. I lessen the noise shaping (and the height of the hump) by trading off some theoretical headroom in the audio band than I’ll never reach in my hardware. I also use the higher sample rate to allow my passive low pass filter to not be as steep as it otherwise might have need to be.
I think I addressed this above, but the 28MHz comes from wanting to do integer synchronous sample rate conversion of both 176400 and 192000 input sample rates: it’s their least common multiple. I zero stuff 146 zeros / sample @ 192000 and 159 for 176400 then I lowpass filter with an IIR filter which uses wide samples internally. I chose 30 bits between it and the sigma delta modulator sort of arbitrarily: only keeping, say 24, seemed like a waste. I use an IIR because all of my experience says they sound more natural than FIR filters. I know what I’m doing so I can avoid the potential instabilities of IIR filters.
Q: I would like to know if the conversion PCM to DSD is a lossless process….
A:No, but the losses are almost all outside of the audio band. The digital DSD conversion process in the DirectStream has better S/N than the hardware. The benefits clearly outweigh the problems so I feel good about converting PCM to DSD for conversion to analog. The S/N rises to only about 100dB @ 60k, but that’s far from a real problem.
I don’t use clocking for the digital inputs at all. No PLL, FLL, etc. The technique isn’t that new and I don’t know why it isn’t used more often in audio products. But, for example, I believe that ESS also does it.
Anyway most S/PDIF receiver chips have a PLL and they recover both the clock and the data. Since the clock inherently has gobs of jitter, recovering it is more of a problem than a help. The trick is to treat the input as pure data. Just look at it with a FPGA (or whatever processor) fast enough so you never miss a transition and you can decode it like you might by eyeball on a scope. By that I mean you look at local features instead of using a derived clock and sampling when it tells you to. I know this sounds like gibberish till you get it. But once you do get it it’s obvious 
Perhaps a slide may help
I just “hook up the wires” from the inputs to the FPGA (literally in the case of a TOSLink receiver, thru a comparator for S/PDIF and AES/EBU) and let it find the bits, decode the samples and stuff them into a buffer.
atupi? I think of the video opamps as my switches. Perhaps it’s a question of semantics, but I’m not using them to amplify an analog signal. I bash them directly with the double rate DSD and then passively filter the outputs. Audio opamps don’t cut it for this purpose, we need linearity to above, say, 400MHz. The video opamps can drive quite a bit of current and I have balanced differential outputs so there are plenty of them as well. I burn a lot of voltage since the design has a factor of two headroom given that 0dBFS is -6dB of all ones or all zeros and further, the passive output burns another factor of two. The design uses the leakage inductance of the transformer as a part of the passive filter so we count on a certain load on the transformer to roll off (I mean filter ) the ultrasonics. Driving too low of a load will roll off the high end of the audio. But for example we implement the optional 20dB attenuator with 15 ohm shunts to ground.
EDIT 17 Apr 2014
Interesant. Ma intreb daca asta e modul de conversie folosit.
http://www.diyaudio.com/forums/class-d/237086-tpa3116d2-amp-274.html#post3895149
Recent Comments