No DAC chip inside?


#1

Apologies if this is too basic a query…but from the PerfectWave product info, it seem that this D/A processor does its D/A conversion in FPGA, as opposed to the more conventional approach of using OEM DAC chip (TI/BurrBrown, Cirrus, Wolfson, ESS, AD, etc.). Is this correct?

I think the only other major D/A processor manuf. that does this is Chord ???


#2

There is no conventional DAC chip in the DirectStream. The input is converted to DSD by the FPGA, the “DAC” proper is a passive filter of that DSD.


#3

Correct. The PW and DS do not use an OEM DAC chip but instead do the data conversion in software. This is how they can continue improving the product over time by issuing new firmware releases like Yale. I believe there are other manufacturers who are using software DACs instead of OEM chips besides Chord. dCS is one that comes to mind but I’m sure there are others. I have both Chord and dCS DACs and neither of them has issued as frequent firmware updates or improvements for their products as PSA.


#4

Thx for the feedback.

In software/firmware (FPGA), then, is PerfectWave D/A conversion done more like Sigma-Delta (I think that was in one of the PP slides, and is in line with DSD)?

I’m curious about digital filtering of PCM…

… is it done and how many times? Or is it a moot point because of the conversion to DSD?

Reason I ask: Non-oversampling (aka zero OS) has its share of fans. Many modern DIY projects, as well as a few current commercial products – like $$$ Zanden DAC or the new discrete-R2R TotalDAC – NOS/ZOS jobs.

I’d be curious if this could be (un)done in PW’s FPGA approach?


#5

There bare many acually but the catagory would need to be better difined

the mytek has done firmwares

But the most common are dap ,S

As in the u basso or hibino or the most common os the AK series astel series

For me the non chip dsd sounds good to me but psaudio wpth it’s new firmware has gotten closer than ever so far .


#6
13mh13 said In software/firmware (FPGA), then, is PerfectWave D/A conversion done more like Sigma-Delta (I think that was in one of the PP slides, and is in line with DSD)?
Yes DSD usually uses a Sigma Delta Modulator, and the DS uses a custom sigma delta modulator to convert the final result into a single bit representation as the output the FPGA.
I'm curious about digital filtering of PCM...

… is it done and how many times? Or is it a moot point because of the conversion to DSD?

Reason I ask: Non-oversampling (aka zero OS) has its share of fans. Many modern DIY projects, as well as a few current commercial products – like $$$ Zanden DAC or the new discrete-R2R TotalDAC – NOS/ZOS jobs.

I’d be curious if this could be (un)done in PW’s FPGA approach?

(BTW the TotalDAC has an optional output filter and the Zanden uses (a lot of) a chip that both does upsampling and has a reconstruction filter.)

In the DirectStream DAC 22.05k, 44.1k and 88.2k PCM get upsampled to 176.4k and 32k, 48k and 96k get upsampled to 192k. Then 1726.4k, 192k, 352.8k or DSD get upsampled to 22.5792MHz. But I don’t think this really addresses your point.

If no upsampling is done for 44.1k (or 48k) inputs some argue that the gross distortion of having no output reconstruction filter is less offensive than the effects of a brickwall reconstruction filter and the resultant phase distortion.

Which of these distortions sounds best depends on the listener, but the simple expedient of upsampling frees you from the bad choices and allows much simpler and less distorting output filters, in the case of DSD the filter can even be only a few dB / octave instead of hundreds of dB per octave and hence can be very benign.

The design of the DirectStream DAC is completely based on the approach of passively filtering a single bit digital output. Without upsampling the single bit outputs of the FPGA would only have a 6dB signal to noise ratio.


#7

To the original poster, you’ll never figure out the sound of the DS just by knowing all the specifics. It’s just something that has to be heard to be understood.


#8

Ted Smith wrote: "Yes DSD usually uses a Sigma Delta Modulator, and the DS uses a custom sigma delta modulator to convert the final result into a single bit representation as the output the FPGA. "

I know that, once upon a time, advanced digital-audio signal processing – e.g., dedicated-DSP-based digital filtering (Wadia ,Theta, et al.) – required one (or more) advanced DSP/FPGA chips (Motorola, etc). PMD had their own PMD100/200 chip (HDCD, ultimately sold to Microsoft, to be implemented in software, because by the late 90s, PC CPUs/etc. could do it). With all that said: I’m not sure how powerful (milli-watts) the FPGA processing (processor) has to be effectively implement DirectStream; but it would be great to enjoy DS in a portable device (HiFiMan or A&K-type DAP; or someday, a smartphone)*.

*(I have not heard DS so I can’t comment on its sound. From my own dabbling in DIY audio, I can acknowledge the fact that for effective D/A, power must be very clean and well-regulated; PCB layout needs also be optimal; etc. So merely implementing, e.g., a DirectStream app, will only take you so far.)


#9
Ted Smith said
In the DirectStream DAC 22.05k, 44.1k and 88.2k PCM get upsampled to 176.4k and 32k, 48k and 96k get upsampled to 192k. Then 1726.4k, 192k, 352.8k or DSD get upsampled to 22.5792MHz. But I don't think this really addresses your point.
Actually, it does address some of my curiosity.

I was thinking about digital filtering (aka oversampling or reconstruction filter). The term “reconstruction” filter is used loosely – some refer to it as the analog filter after the DAC (just before the output jack) (i.e., brick-wall filter, etc.). But I have seen the term “reconstruction filter” also apply to digital interpolation filters just before the DAC (e.g., old school: PMD100, SAA7220, DF1704, various models by NPC, etc; also new-school, as in the 8x filters built into modern DAC chips like PCM1792, etc.). I ran the confusing “reconstruction” nomenclature by John Atkinson, a while back, noting that perhaps Stereophile is in error when it refers to the device BEFORE the DAC (like PMD100) as “reconstruction” filter (again, when many sources I’ve encountered, use the same term for analog output (sometimes brick-wall) filter). JA said that reconstruction is also a correct term for interpolation:

Fig.4 Waveform reconstructed by perfect low-pass filter. Source: http://www.stereophile.com/reference/25/index.html

So that long, drawn-out comment brings me to your FPGA implementation in the PW-DS dac. While I can understand why this

… provides better audio from native DSD (all that clean, well-regulated power/decoupling, etc), I can’t understand why “DSDing” PCM would sound better. May be I’m too dense (!!) or maybe what’s happenin’, in the realm of audio math, is REALLY good implementation of reconstruction. (??)

OR: Maybe the improved sound quality is because the FPGA PW-DS implementation does not INTERPOLATE (fill in with guesstimate samples). This would incorporate some of the purported bennies of non-oversampling. (Alas, there is then anti-imaging issues).

But when the PCM is upsampled way, way up to DSD without interpolation, you get the benefits of both worlds. Is this kinda what you had in mind?


#10

Anytime you raise the sample rate you need a reconstruction filter to keep the images of the input from polluting the output with aliasing.

Anytime you lower the sample rate you need an antialiasing filter to keep the frequencies above the Nyquist frequency of the input from aliasing in the output.

When these are done digitally an obvious optimization is to upsample, use a combination filter which filters out everything above 1/2 of the lowest of the input sample rate and the output sample rate, and then downsample. In this case the upsampling and downsampling consist only of inserting zeros and deleting samples respectively.

Tho you may think of that digital filter as being there to get rid of the frequencies that might end up aliasing in the outputs, but that turns out to be exactly the same as a interpolation filter that “smooths” out the input filling in the zeros with the correct samples from the input. That information is available since any other signal would have frequencies above 1/2 of the input frequency and would therefor alias…

In the popular press the technical terms are often mixed up, but some text books also use different nomenclature than others. But there’s no real ambiguity, you need a filter to avoid aliasing/reconstruction images, to interpolate, to change sample rates, etc. The filter needs to filter out everything below 1/2 of the (input or output) sample rate.

No one is claiming that DSDing PCM is required to make PCM sound better, but simple physics says that achieving the component precision to do a multibit DAC is much harder than achieving the component precision for a single bit DAC. Also a single bit DAC is inherently linear and hence has no problems with missed codes, differential non-linearity, bigger errors when higher order bits change, etc. (Note that the chip that the Zanden dac uses achieves the component precision needed with DEM, i.e. by having a lot of sub DACs which aren’t quite as good, and then using them randomly to average out their precision problems.)

I implemented a DSD DAC and threw in PCM to DSD conversion as almost an afterthought (after all that’s only a software problem), but a passive output filter is all that’s needed for a DSD dac, so by converting PCM to DSD you can do a PCM dac with only a passive output fitler and some software.

As it turns out there’s so much more in the original PCM recordings than most of us knew. This isn’t new information that PCM to DSD created, it’s just information that most PCM dacs cover up with perhaps their non-linearities, perhaps their less than ideal power supplies or power supply bypassing, perhaps their less than ideal clocking, perhaps short cuts in the digital filters being used, etc. Also most PCM dacs (by necessity of using a DAC in a chip) suffer from having their digital supplies right next to their analog supplies at the DAC chip and their clocking right next to their power supplies, etc. all of which analog engineers are taught to avoid.


#11

Great post Ted.

I own two FPGA dacs yours and the Hugo. Both for me do good pcm and of course I feel yours does even better plus firmware rolls.

My point is by nature do both dacs using fpga chip sets do the same kind of up sampling tech for simple low pass at th end ?

Just a thought

al


#12

Ted: Thx for your thorough feedback.

Back, roughly 2003-2007, not sure you (anyone) recalls certain Toshiba CD/DVD players enjoyed some attention in modder/DIY circles. (I think some of the Oppo players used same base model).

Those players used an all-in-one Zoran processor (it handled video, too!). In the last months those units were made, you could pick them up for under $30 at Best Buy! Not sure how that DSP was configured WRT d/a conversion, but there must’ve been some audiophiles at Zoran:

Note, the differential output signal handling from the Zoran processing chip would have been classified as very esoteric in the days of the kilobuck Theta DAC.


#13

Thank you Ted for developing such a great device as the DSD dac. It’s giving me so much pleasure. After listening to many dacs and owning a Zanden, there is nothing comes close to the ambience, information and smoothness of the dsd. Its is also the first digital device I have heard that does not suffer when using the inbuilt volume control. Mark