A novel way to massively improve the SQ of computer audio streaming

Johnseye · October 10, 2017

11 minutes ago, ElviaCaprice said:

Roy could run this test. Bit Perfect to Blu to DAVE vs. HQP (768 upsample, chosen filters) to DAVE vs. All Three in combo.

Use the sCLK-EX modified server and any fixers appropriate or none.

Question would be if the low power server would choke on HQP. I wouldn't think so but with only 2GB of memory?

2GB is low. My server will use that on average.

unbalanced output · October 10, 2017

When you guys say that you are upsampling with low power computers, I assume you're using the -2s algorithms? My i7/gtx was not capable of upsampling to dsd256 with some of the single step filters.

austinpop · October 10, 2017

1 hour ago, unbalanced output said:

When I first read about Blu, that's the exact thought that came to my mind. Chord is funny at times. It is not entirely on topic, but a Chord representative actually wrote this on Head-Fi when it was suggested that an ultraRendu or SOTM Ultra would make very fine nice sources for the Mojo "and by the way the SOTM can even be paired with a master clock if you fancy one".

(and I quote)

At the risk of sounding arrogant the team that developed Poly are actually the best on the planet of course we use top notch well designed circuitry throughout and to suggest it’s comprmised because we aren’t clocking it externally is ridiculous. Most good engineers understand that external clocks may have a good long term timed signal say accurate to second in a year but they are often absolute rubbish because they have horrendous short term jitter due to there poorly designed frequency reduction and buffer circuitry. Far better to have good control of the accurate crystal clocking inside the design.

Sadly, this is the default response of most vendors, especially the non-technical sales types. They don't know what we're talking about. In this case, this guy is even further off the mark, because he's thinking in terms of word clock inputs. The idea that improving the clock quality on the system and USB/Ethernet data stream clocks could have an impact is foreign to them.

No - the breakthroughs will only come by engaging directly with the lead designers and lead engineers. Most of them have very strong opinions about this stuff. To his credit, Rob Watts at least has been willing to hear us explain our empirical findings, and due to Roy's efforts, he should soon be able to listen to the effect as well.

Over on the Ayre QX-5 thread, Charles Hansen did at least say they would "research" our findings, but I didn't sense a lot of interest there. Perhaps I'm wrong.

mozes · October 10, 2017

1 hour ago, austinpop said:

Sadly, this is the default response of most vendors, especially the non-technical sales types. They don't know what we're talking about. In this case, this guy is even further off the mark, because he's thinking in terms of word clock inputs. The idea that improving the clock quality on the system and USB/Ethernet data stream clocks could have an impact is foreign to them.

No - the breakthroughs will only come by engaging directly with the lead designers and lead engineers. Most of them have very strong opinions about this stuff. To his credit, Rob Watts at least has been willing to hear us explain our empirical findings, and due to Roy's efforts, he should soon be able to listen to the effect as well.

Over on the Ayre QX-5 thread, Charles Hansen did at least say they would "research" our findings, but I didn't sense a lot of interest there. Perhaps I'm wrong.

Let the manufacturers say what they want. We make our decisions based on what our ears tell us!

lmitche · October 11, 2017

4 hours ago, Johnseye said:

2GB is low. My server will use that on average.

My machine runs with 1.7 GB bytes of memory with Windows Roon and Hqplayer all running.

Johnseye · October 11, 2017

9 minutes ago, lmitche said:

My machine runs with 1.7 GB bytes of memory with Windows Roon and Hqplayer all running.

What did you do to trim it down? I use AO and disabled what additional services I knew weren't needed and am still at 2.3GB. Roon itself takes up 900MB with RAAT another 100MB. I'm also using 2016 Server, but would think that has less overhead. I have some trimming to do.

seeteeyou · October 11, 2017

4 hours ago, ElviaCaprice said:

Question would be if the low power server would choke on HQP. I wouldn't think so but with only 2GB of memory?

We do have that offline upsampling option

https://www.chisto.me/news/world/item/97124-offline-upsampling.html

CUDA offload didn't just reduce CPU utilization, that actually improved SQ when they're choosing Tesla K20 with both Double-Precision Performance and ECC bits enabled. Other audiophiles in Taiwan actually had 2 GPUs installed instead of only 1

http://www.myav.com.tw/bbs/showthread.php?threadid=20475707

http://www.myav.com.tw/bbs/showthread.php?threadid=20478306

Recently they also discovered that both Mixbus 4 and Mixbus 32C with CUDA would sound even better than HQPlayer

http://www.myav.com.tw/bbs/showthread.php?threadid=20480094

https://sites.fastspring.com/harrisonconsoles/product/mixbus4?coupon=LABORDAY17

https://sites.fastspring.com/harrisonconsoles/product/mixbus32c4?coupon=MB32C-UPDATE

They don't support *.DFF and *.DSF so we've gotta create DoPFLAC manually

http://harrisonconsoles.com/site/mixbus-info.html

https://sites.fastspring.com/harrisonconsoles/checkout/mixbus4demo

lmitche · October 11, 2017

39 minutes ago, Johnseye said:

What did you do to trim it down? I use AO and disabled what additional services I knew weren't needed and am still at 2.3GB. Roon itself takes up 900MB with RAAT another 100MB. I'm also using 2016 Server, but would think that has less overhead. I have some trimming to do.

I run the scripts in the windows optimization thread, and the strip windows command in AO. Also remove stuff from windows in appwiz.cpl like internet explorer. Use tightvnc instead of remote desktop. . . . . . And so on.

Johnseye · October 11, 2017

19 minutes ago, lmitche said:

I run the scripts in the windows optimization thread, and the strip windows command in AO. Also remove stuff from windows in appwiz.cpl like internet explorer. Use tightvnc instead of remote desktop. . . . . . And so on.

Would you mind pointing me to those scripts? I did a search but came up dry. Thanks.

seeteeyou · October 11, 2017

Strip Down & Repair Windows 10
http://images.computeraudiophile.com/graphics/2016/0622/audiophile-optimizer-setup-guide.pdf#page=47

Script to disable unnecessary services and apps
https://www.computeraudiophile.com/forums/topic/29167-script-to-disable-unnecessary-services-and-apps/

HQPlayer - Favorite settings?
https://www.computeraudiophile.com/forums/topic/26718-hqplayer-favorite-settings/

Windows 10 optimization script - A community effort?
https://www.computeraudiophile.com/forums/topic/24716-windows-10-optimization-script-a-community-effort/

Windows 8.1 Everyday Optimization Settings
https://www.computeraudiophile.com/forums/topic/23586-windows-81-everyday-optimization-settings/

Computer Audio Design OSX Audio Optimization Script
https://www.computeraudiophile.com/forums/topic/17574-computer-audio-design-osx-audio-optimization-script/

Windows 8 Audio Optimization Script
https://www.computeraudiophile.com/forums/topic/16037-windows-8-audio-optimization-script/

feelingears · October 11, 2017

Akin to what @austinpop said about assembling hardware (vs. software), I'm leery of HQP's apparent complexity and user unfriendliness despite (or because of) my work in software. But this is audio so one is always exploring or tempted to try more. So, begging everyone's pardon, but would someone please point me/others to a good "HQP for Dummies" post or set of instructions, anywhere? Something user-friendly and more "forgiving," if such a thing exists?

(Also, HQP is a one-man show is it not? I wish there were more "there, there" as a going concern.)

Thanks in advance.

@seeteeyou You just beat me to the punch, but I'll have to see if any of those speaks to HQP (and speaks in "Dummies" dialect).

AmusedToD · October 11, 2017

9 hours ago, austinpop said:

Sadly, this is the default response of most vendors, especially the non-technical sales types. They don't know what we're talking about. In this case, this guy is even further off the mark, because he's thinking in terms of word clock inputs. The idea that improving the clock quality on the system and USB/Ethernet data stream clocks could have an impact is foreign to them.

No - the breakthroughs will only come by engaging directly with the lead designers and lead engineers. Most of them have very strong opinions about this stuff. To his credit, Rob Watts at least has been willing to hear us explain our empirical findings, and due to Roy's efforts, he should soon be able to listen to the effect as well.

Over on the Ayre QX-5 thread, Charles Hansen did at least say they would "research" our findings, but I didn't sense a lot of interest there. Perhaps I'm wrong.

@austinpop, this is what Charles said to Audiostream during their review of the SOtM txUSBUltra:

”I asked Charles Hansen, co-founder and Research Director at Ayre Acoustics, why these USB enhancement devices seem to have an effect on a sophisticated well-designed DAC such as the Ayre Acoustics QX-5 Twenty?

"Please don't be misled by the word "clock". The QX-5 has an insanely good master clock that runs the DAC chip, giving the absolute best sound quality. However both USB and Ethernet inputs have their own separate clocks. The USB clock is a multiple of 12MHz and the Ethernet clock is a multiple of 25MHz - neither related to the digital audio master clock frequencies that are multiples of either 44.1kHz or 48kHz (two separate crystals used).
"Ayre has done everything possible to elicit the highest performance from each type of digital source. Yet the quality of the source still makes a difference. Therefore improving the quality of the source on any input will make some difference to the sound.

"You may recall that when Gordon Rankin (Wavelength Audio) developed asynchronous USB, he and I both thought that the source would no longer matter. However we were wrong. Not only does the source still make a difference, but it even matters if you are listening to 'flat' files or losslessly compressed files."

Read more at https://www.audiostream.com/content/sotm-tx-usbultra-usb-signal-regenerator#0m1GAl8uSWUXoZeK.99

romaz · October 11, 2017

I had a wonderful time at RMAF this past weekend. It was a genuine pleasure to have met the likes of @kennyb123 and @Always.Learning. Having benefited from their insightful posts over the past couple of years, both on CA and Head-Fi, it was great to finally chat in person. If I am ever in Seattle, we will have to catch a live event together! Of course, there was the man himself, @The Computer Audiophile, who was generous enough to buy everyone a round of drinks during happy hour on Saturday evening. I wish there would have been more time to chat as I wanted to hear more about your "dog" story during your drive home from CES earlier this year. You are a class act, Chris. Last but not least, it was my privilege to have spent some quality time with both @austinpop and @limniscate during dinner on Saturday night and for breakfast each morning. Just a couple of great guys with fascinating backgrounds and true music lovers. I couldn't ask for better company!

As RMAF was my 4th audio show this year (including CES, Munich, and the LA Audio Show), there wasn't much here that I hadn't already seen or heard but there was one consistent observation I was able to make and that was the very best sounding rooms were generally spinning vinyl or rolling tape. With very few exceptions, when digital was playing, the harshness and compression was painfully evident. In some rooms, it was bearable for only a few minutes, especially as this annoying glare was being amplified to very loud levels. As an example, within a few minutes of entering the Paradigm room where their nearly $20k top-of-the-line Personas were on display, I looked at Rajiv and we both knew what the other was thinking, that we had to get out of there in a hurry. It wasn't the speakers as I suspect they were very good. It was the digital front end. If there is one thing I have benefited greatly from with all the things we have learned on this thread, it is the reduction of that piercing glare that I once considered tolerable. Having gotten so accustomed to the lack of glare from the digital front end I have now, somehow, I have become all the more sensitive to bright and edgy electronics. There's simply no going back.

As life has gotten extremely busy and will get even busier, unfortunately, RMAF will represent my last audio indulgence for the indefinite future. This will also represent my last post here on CA. It has been my pleasure to have been a part of this wonderful community of audiophiles.

I will do my best to answer some of the questions that were directed at me.

With regards to the sCLK-EX in my server chassis, this clock board is clocking 4 items and not just 3 -- (1) the motherboard's system clock, (2) both of the motherboard's LAN ports, (3) the tX-USBexp, and (4) the tX-USBhubIN.

Regarding my choice of motherboard, I believe there are many very good boards out there. While I am very pleased with my DFI board, people should target the feature set they desire more than any particular brand. Having said that, a group of friends and I purchased a specific AsRock RACK motherboard some time ago that has the option of being powered via ATX or 12V. We powered this motherboard 3 ways: (1) directly from a 12V rail from my DR SR7, (2) from the 19V rail of my SR7 using an HDPlex DC-to-ATX converter, and (3) an EVGA Gold class 850 watt ATX PSU. With everything else kept constant, this motherboard sounded best when powered directly via a 12V rail from my DR SR7. The EVGA Gold ATX PSU sounded the worst by a very noticeable margin. This should not be taken as any kind of a sweeping statement against ATX PSUs as there are ATX PSUs that are undoubtedly much better than the EVGA we used but this is my strongest argument for using a low power motherboard. They are easier to power well and I have yet to hear anything that can do what an SR7 can.

Regarding oversampling, yes, absolutely, you should oversample and I have stated this preference from very early on in this thread. As far as more companies like Chord finally getting on the oversampling bandwagon that dCS started 20 years ago, this is an inaccurate statement. Rob Watts first postulated his "1 million TAPS" theory while at university in Cardiff during the early 1980s, well before FPGAs were invented. He designed his very first upsampling DAC in 1995 (22 years ago) and all of his DACs moving forward have been upsampling DACs. As his DACs have evolved, they have improved with regards to TAP-length but also filter design and he has now reached the "1 million TAP milestone" with his current M-scaler that he never thought would be reached in his lifetime. dCS lost their chief designer years ago and so their ring DAC technology has stagnated since. The latest dCS Vivaldi Upsampler ($22k) can only upsample to either DXD (24/358) or DSD256. Even HQPlayer can do better.

Does the DAVE, Blu2 or Hugo2 upsample to only 705.6 or 768kHz? Absolutely not. This would be just the beginning. DAVE, Blu2 or Hugo2 can accept any signal up to 768KHz PCM and DSD512 but then they upsample this signal much higher and through multiple WTA (Watts Transient Aligned) filter stages. As an example, when DAVE sees a 44.1 or 48kHz signal, the first WTA stage upsamples to 16FS (705.6 or 768kHz). Then the next WTA stage is at 256FS and finally, there is a third stage filter that upsamples all the way to 2048FS. This means a new filtered sample every 9.6nS. While there is no easy way to directly correlate PCM and DSD, this amounts to a degree of oversampling well beyond DSD2048. According to Rob, he upsamples to this degree for several reasons including to reduce the timing of transients uncertainty, to enable his noise shapers to work at 104MHz so that the noise shapers can reproduce depth correctly, and finally to allow no measurable noise floor modulation.

What is a TAP and what is its significance? Here is a good article to read:

http://www.the-ear.net/how-to/rob-watts-chord-mojo-tech

How is a TAP calculated?

Number of TAPS = (input sample rate frequency in Hz) * (delay time in seconds) * (oversampling rate) * 2

Are TAPS all that matter? Here is what Rob has to say:

"Real tap count is vitally important - and for sure increasing the tap count with everything else the same - algorithm and over sampling rate - will result in much better sound quality.

But there is much more too it than that, in that the algorithm is vitally important. The algorithm is the formula to calculate the coefficients for the filter, as an FIR filter is just audio data * coefficients added up to create the intermediate samples.

When I first started designing with FIR filters in the 90's I had 2048 taps on an FPGA. Now this used a closed form type digital filter - the coefficients were calculated directly, and the filter guaranteed that the original data was preserved. But I heard this filter, and it was so much better than before. Now I knew that to guarantee 16 bit accuracy for the filter for all the calculated interpolated samples I would need getting on for 1M taps, and that was quite impossible way back then. So I thought can the algorithm allow a better timing accuracy without needing 1M taps? Was there a short cut that would allow perfect performance without using impossible tap lengths?

Anyway I spent months thinking about it. I managed to figure out that there was an major measurable issue with filters - aliasing - and that removing this issue would help timing problems. But that would mean the original samples would get changed. And to me this was a very bad idea - it did not feel right, modifying the data. And I delayed trying this out.

Now when you design, you are governed by your thinking, so your ideas set how you go about things. The problem is, we all make assumptions - XYZ can't sound good because of ABC - and assumptions that are not based on hard and rigorous listening tests are very dangerous. And I did not like the idea of changing the data - but I knew that aliasing could be something really important. So I eventually went ahead and designed a (pretty much - better than 16 bits) guaranteed non-aliasing filter. And when I heard it, it was amazing. Eventually this became the first version of the WTA filter, and subsequent tests showed that it at least gave a boost of an order of magnitude to the performance - a 256 tap WTA sounded much better than a 2048 tap conventional filter.

But getting rid of aliasing was not the whole story, as increasing tap length does make huge differences, even when aliasing (which is vitally important in spite of the poor aliasing performance of MQA and other conventional filters) has been eliminated.

So the moral of the story? Don't let seemingly good ideas influence your decisions - test all assumptions with very careful listening tests. Moreover, audio is very complex, and things are never as simple as they might appear.

My mantra is "You know nothing Jon Snow..." and that is to remind me that there are very real limits to what I understand, and assumptions must be constantly tested with listening tests. And very big progress can be made by going down avenues that at first sight seem incapable of changing sound quality."

Regarding the perfect interpolation filter, here is what Rob has to say:

"For the reconstruction filter (the interpolation filter in the DAC) there is only one way of doing it, and that is using an ideal infinite response sinc function - that is if your intention is to perfectly recover the original bandwidth limited analogue signal. Any other type of filter will change the sound by adding uncertainty to the reconstruction of transient timing; and this uncertainty severely damages sound quality. Timing uncertainty due to small tap length filters add a softness, warmth and a bloated bass; some actually like that sound (I do not understand why as it sounds nothing like real un-amplified music) but for sure it is a distortion, or change, from the original.

Now theory has nothing to say about how the signal is bandwidth limited; it simply states there must be zero output above FS/2. Actually, I have designed two decimation filters; one that is linear phase but rings before the impulse; one that is non-linear phase (IIR) but has no pre-ringing. Both filters have attenuation at and above FS/2 of 300 dB, so both meets the requirements of bandwidth limiting in terms of aliasing and sampling theory. I will post blind recordings with the different filters so we can all hear the difference."

Is there an HQPlayer filter equivalent for Rob's WTA filters? Actually, all of Rob's DACs (including Mojo) upsample to 2048FS but they do so differently and so to say that you can match Rob's algorithms using one of HQPlayer's filters would be a gross overstatement.

With Mojo, you have: WTA 16FS > Linear interpolation to 2048FS > Low pass filter > Low pass filter > noise shaper

With DAVE, you have: WTA 16FS > WTA to 256FS > Linear interpolation to 2048FS > Low pass filter > Low pass filter > noise shaper

Rob found that the addition of the extra WTA filter to 256FS in DAVE got him closer to the ideal sinc impulse response and the closer this gets to ideal, the closer the DAC gets to reproducing the signal in between samples.

Regarding his WTA filters, they are proprietary and remain his IP (not Chord's). According to Rob:

"The WTA algorithm actually uses my own windowing function of the ideal sinc impulse response. I took an awful lot of time (man years) with this, both in listening tests and in trying to understand what was going on - the understanding being used to allow me to try listening to different things. At the end of the day, the algorithm is fine tuned by listening tests, but you need understanding in order to change the critical parameters - in short knowing what those parameters are."

Can a PC upsample to the degree that Rob's DACs upsample? Here is what Rob has to say:

"PC's are very restricted in what they can do for real time signals. You simply can't replicate the processing that Dave does in a PC - simply because PC processors are sequential serial devices with a very limited number of cores. When you are doing a doing a FIR filter (a tap) you need to read from memory the audio data; read from memory the coefficient data; multiply the numbers together;then read the accumulated data and add that to the previous multiplication; then save the result. Lots of things to do in sequence. With an FPGA you can do all of these things in parallel at once, so a single FIR tap can be accomplished within a single clock cycle (obviously pipelined) - you are not forced to do things in sequence.

With Dave I have 166 dsp cores running, plus FPGA fabric to do a considerable amount of further processing. You simply can't do that in a PC. To give you another example - converting DSD into DoP. You need a quad core processor to do this manipulation in real time - otherwise you get drop-outs - but in a FPGA I could do this simple operation thousands of times over, and at much faster rates than DSD256.

What some people do not understand is how capable FPGA's are and how widespread they are used - the backbone to the internet? FPGA's. Search engines? FPGA's. Why? because an FPGA is fantastic at doing fixed real time processing - it takes small die area, and can do complex operations with very low power. Mojo for example has 44 dsp cores, uses sophisticated filtering to 104 MHz, and noise shapes at this rate - but does all this whilst consuming only 0.45 W. There is no way any PC consuming huge amounts of power can do this.

Intel last year acquired Altera (an FPGA company) for $16.7 billion because they understand that the future of processing is with FPGA's

A second issue is not what you can do but how you can do it - it is not just about raw power, but how the filter algorithm is designed. I have put many thousands of hours and over twenty years improving and understanding how to make a transparent interpolation filter; and I am still learning things today.

And a third point is that a DAC is not simply a data processing machine but it has got crucial analogue parts too. If I dropped the WTA requirement, I would still need the same FPGA in order to do the noise shaping and other functions."

Over the past few years, upsampling to DSD has become a trend by DAC manufacturers and by adopters of HQPlayer. Here are Rob's comments on DSD (probably not a surprise to many):

"I was doing this 25 years ago in an attempt to solve the innate problems of DSD 256 (PDM 256). But I was using multiple, individually dithered noise shapers in an attempt to improve the noise shaper resolution. The benefits of doing this is that it reduces the jitter sensitivity, but does not eliminate correlated jitter problems. It's why in 1995 I invented pulse array as this eliminates correlated jitter (its a fixed switching activity scheme independent upon the output), improves (in the case of Dave) noise shaper resolution by a trillion times and eliminates noise floor modulation - something you can't do with DSD.

What was also amusing is that they seem to be using an 8 bit shift register to re-time the outputs. I too used to do this too in the early 90's, but quickly found that using discrete flip-flops gave much better measured and SQ performance. Its due to switching activity on-chip changing the propagation delay of the OP FF - so making signal correlated jitter much worse. Also, power draw on each FF modulates the devices internal power rail, creating distortion. Using discrete flip-flops eliminates the signal correlated jitter issue, and with appropriate low impedance power planes, one can eliminate the PSU induced distortion problems too.

This issue of correlated jitter problems can't be solved on-chip. A silicon chip design I worked on over ten years ago had this issue; on chip we had separate clock buffers, custom designed IO buffers, internal separate PSU paths, and all we could do was minimize the issue - it was still there on simulation, and still measurable in reality - although we reduced the problem by two orders of magnitude. But going discrete eliminates the issue. "

Here's another comment:

"There are actually two independent issues going on with DSD that limits the musicality - and they are interlinked problems.

The first issue is down to the resolving power of DSD. Now a DSD works by using a noise shaper, and a noise shaper is a feedback system. Indeed, you can think of an analogue amplifier as a first order noise shaper - so you have a subtraction input stage that compares the input to the output, followed by a gain stage that integrates the error. With a delta sigma noise shaper its exactly the same, but where the output stage is truncated to reduce the noise shaper output resolution so it can drive the OP - in the case of DSD its one bit, +1 or -1 op stage. But you use multiple gain stages connected together so you have n integrators - typically 5 for DSD. Now the number of integrators, together with the time constants will determine how much error correction you have within the system - and the time constants are primarily set by the over-sample rate of the noise shaper. Double the oversampling frequency and with a 5th order ideal system (i.e. one that does not employ resonators or other tricks to improve HF noise) it converges on a 30 dB improvement in distortion and noise.

So where does lack of resolution leave us? Well any signal that is below the noise floor of the noise shaper is completely lost - this is completely unlike PCM where an infinitely small signal is still encoded within the noise when using correct dithering. With DSD any signal below the noise shaper noise floor is lost for good. Now these small signals are essential for the cues that the brain uses to get the perception of sound stage depth - and depth perception is a major problem with audio - conventional high end audio is incapable of reproducing a sense of space in the same way one can perceive natural sounds. Now whilst optimising Hugo's noise shaper I noticed two things - once the noise shaper performance hit 200 dB performance (that is THD and noise being -200 dB in the audio bandwidth as measured using digital domain simulation) then it no longer got smoother. So in terms of warmth and smoothness, 200 dB is good enough. But this categorically did not apply to the perception of depth, where making further improvements improved the perception of how deep instruments were (assuming they are actually recorded with depth like a organ in a cathedral or off stage effects in Mahler 2 for example. Given the size of the FPGA and the 4e pulse array 2048FS DAC, I got the best depth I could obtain.

But with Dave, no such restriction on FPGA size applied, and I had a 20e pulse array DAC which innately has more resolution and allows smaller time constants for the integrator (so better performance). So I optimised it again, and kept on increasing the performance of the noise shaper - and the perception of depth kept on improving. After 3 months of optimising and redesigning the noise shaper I got to 360 dB performance - an extraordinary level, completely way beyond the performance of ordinary noise shapers. But what was curious was how easy it was to hear a 330 dB noise shaper against a 360 dB one - but only in terms of depth perception. My intellectual puzzle is whether this level of small signal accuracy is really needed, or whether these numbers are acting as a proxy for something else going on, perhaps within the analogue parts of the DAC - I am not sure on this point, something I will be researching. But for sure I have got the optimal performance from the noise shaper employed in Dave, and every DAC I have ever listened too shows similar behaviour.

The point I am making over this is that DSD noise shapers for DSD 64 is only capable of 120 dB performance - and that is some 10 thousand times worse than Hugo - and a trillion times worse than Dave. And every time I hear DSD I always get the same problem o perception of depth - it sounds completely flat with no real sense of depth. Now regular 16 bit red book categorically does not suffer from this problem - an infinitely small signal will be perfectly encoded in a properly dithered system - it will just be buried within the noise.

Now the second issue is timing. Now I am not talking about timing in terms of femtosecond clocks and other such nonsense - it always amuses me to see NOS DAC companies talking about femtosecond accuracy clocks when their lack of proper filtering generates hundreds of uS of timing problems on transients due to sampling reconstruction errors. What I am talking about is how accurately transients are timed against the original analogue signal in that the timing of transients is non-linear. Sometimes the transient will be at one point in time, other times delayed or advanced depending upon where the transient occurs against the sample time. In the case of PCM we have the timing errors of transients due to the lack of tap length in the FIR reconstruction filter. The mathematics is very clear cut - we need extremely long tap lengths to almost perfectly reconstruct the original timing of transients - and from listening tests I can hear a correlation between tap length and sound quality. With Dave I can still hear 100,000 taps increasing to 164,000 taps albeit I can now start to hear the law of diminishing returns. But we know for sure that increasing the tap length will mean that it would make absolutely no difference if it was sampled at 22 uS or 22 fS (assuming its a perfectly bandwidth limited signal). So red book is again limited on timing by the DAC not inherently within the format.

Unfortunately, DSD also has its timing non-linearity issues but they are different to PCM. This problem has never been talked about before, but its something I have been aware of for a long time, and its one reason I uniquely run my noise shapers at 2048FS. When a large signal transient occurs - lets say from -1 to +1 then the time delay for the signal is small as the signal gets through the integrators and OP quantizer almost immediately. But for small signals, it can't get through the quantizer, and so it takes some time for a small negative signal changing to a positive signal to work its way through the integrators. You see these effects on simulation, where the difference of a small transient to a large transient is several uS for DSD64.

Now the timing non linearity of uS is very audible and it affects the ability of the brain to perceive the starting and stopping of instruments. Indeed, the major surprise of Hugo was how well one can perceive that starting and stopping of notes - it was much better than I expected, and at the time I was perplexed where this ability was coming from. With Dave I managed to dig down into the problem, and some of the things I had done (for other reasons) had also improved the timing non-linearity. It turns out that the brain is much more sensitive that the order of 4 uS of timing errors (this number comes from the inter-aural delay resolution, its the accuracy the brain works to in measuring time from sounds hitting one ear against the other), and much smaller levels degrade the ability for the brain to perceive the starting and stopping of notes.

But timing accuracy has another important effect too - not only is it crucial to being able to perceive the starting and stopping of notes, its also used to perceive the timbre of an instrument - that is the initial transient is used by the brain to determine the timbre of an instrument and if timing of transients is non-linear, then we get compression in the perception of timbre. One of the surprising things I heard with Hugo was how easy it was to hear the starting and stopping of instruments, and how easy it was to perceive individual instruments timbre and sensation of power. And this made a profound improvement with musicality - I was enjoying music to a level I had never had before.

But the problem we have with DSD is that the timing of transients is non-linear with respect to signal level - and unlike PCM you are completely stuck as the error is on the recording and its impossible to remove. So when I hear DSD, it sounds flat in depth, and it has relatively poor ability to perceive the starting and stopping of notes (using Hugo/Dave against PCM). Acoustic guitar sounds quite pleasant, but there is a lack of focus when the string is initially struck - it sounds all unnaturally soft with an inability to properly perceive the starting and stopping. Also the timbre of the instrument is compressed, and its down to the substantial timing non-linearity with signal level.

Having emphasised the problems with delta-sigma or noise shaping you may think its better to use R2R DAC's instead. But they too have considerable timing errors too; making the timing of signals code independent is impossible. Also they have considerable low level non linearity problems too as its impossible to match the resistor values - much worse than DSD even - so again we are stuck with poor depth, perception of timing and timbre. Not only that they suffer from substantial noise floor modulation, giving a forced hard aggressive edge to them. Some listeners prefer that, and I won't argue with somebody else's taste - whatever works for you. But its not real and it not the sound I hear with live un-amplified instruments.

So to conclude; yes I agree, DSD is fundamentally flawed, and unlike PCM where the DAC is the fundamental limit, its in the format itself. And it is mostly limited by the format. Additionally, its very easy to underestimate how sensitive the brain is to extremely small errors, and these errors can have a profound effect on musicality."

Rob's quotes above were not placed here to suggest he is right and others are wrong. They were placed here so that people at least understand his perspective as a lot of misinformation is being spread. You can choose to agree or disagree. I suggest you perform your due diligence and let your ears decide and this is what I have tried to do. Do I agree with everything Rob says? No, I do not. For example, Rob believes all sources should sound the same with his DACs. It would appear Chord and their salespeople have bought into this philosophy also but my ears clearly tell me this is untrue and Rob will hear this for himself soon enough. I also wonder what would happen if Paul Hynes were to design a new PSU for my DAVE. Rob doesn't think it would make much difference but I have a gut feeling the difference would be significant, especially as I am using my DAVE to directly drive speakers.

With regards to combining HQPlayer upsampling and one of Rob's DACs, I have already reported earlier in this thread that I have done this with my DAVE in the past. Remember that when DAVE (or any of Rob's DACs) sees a signal, whether it be 16/44, PCM 705/768, or DSD512, it takes that signal and upsamples it further. Using my Windows workstation, I ran this exercise with HQPlayer again this evening because I was curious. My new Hugo2 just arrived today and so I tried upsampling from Redbook to both PCM 705.6kHz and DSD512. I did it also with Blu Mk2. Just like before, upsampling to DSD512 with either of these devices resulted in worse SQ. Upsampling to PCM 705.6kHz had better results and perhaps there is an ideal filter/dither combination I failed to try but with the ones I did try, SQ was very acceptable but compared to feeding either Hugo2 or Blu2 the original signal, SQ via HQPlayer was drier and less smooth. With Blu2 combined with either DAVE or Hugo2, the improvement was massive. HQPlayer does not come close to what the M-scaler in Blu2 can do with either Hugo2 or DAVE.

For those looking for a reasonably affordable alternative to upsampling with HQP, I suggest you give Hugo2 a try. Combined with even an unmodified NUC and an inexpensive endpoint like the ISO Regen or better yet, a tX-USBultra, I suspect it will give even the very best HQPlayer setups a run for their money. With the brief time I have had it, I am finding Hugo2 to be an excellent upsampling DAC and just like DAVE, it can directly drive high-efficiency speakers as well as headphones. I cannot overstate just how wonderfully resolving and transparent Hugo2 directly driving my Voxativs or Omegas sounds. Moreover, Hugo2 will combine with M-scaler to give you a full 1M taps, something that no other DAC except DAVE can currently provide and so as far as I am aware, when combined with M-scaler and when directly driving speakers, these are the 2 highest resolution and most transparent digital front ends in existence today regardless of price.

I wish you all success in your audiophile journeys. Signing out...

mozes · October 11, 2017

Thank you Roy for all your valuable contributions, wish you all the best!

lmitche · October 11, 2017

2 hours ago, romaz said:

I had a wonderful time at RMAF this past weekend. It was a genuine pleasure to have met the likes of @kennyb123 and @Always.Learning. Having benefited from their insightful posts over the past couple of years, both on CA and Head-Fi, it was great to finally chat in person. If I am ever in Seattle, we will have to catch a live event together! Of course, there was the man himself, @The Computer Audiophile, who was generous enough to buy everyone a round of drinks during happy hour on Saturday evening. I wish there would have been more time to chat as I wanted to hear more about your "dog" story during your drive home from CES earlier this year. You are a class act, Chris. Last but not least, it was my privilege to have spent some quality time with both @austinpop and @limniscate during dinner on Saturday night and for breakfast each morning. Just a couple of great guys with fascinating backgrounds and true music lovers. I couldn't ask for better company!

As RMAF was my 4th audio show this year (including CES, Munich, and the LA Audio Show), there wasn't much here that I hadn't already seen or heard but there was one consistent observation I was able to make and that was the very best sounding rooms were generally spinning vinyl or rolling tape. With very few exceptions, when digital was playing, the harshness and compression was painfully evident. In some rooms, it was bearable for only a few minutes, especially as this annoying glare was being amplified to very loud levels. As an example, within a few minutes of entering the Paradigm room where their nearly $20k top-of-the-line Personas were on display, I looked at Rajiv and we both knew what the other was thinking, that we had to get out of there in a hurry. It wasn't the speakers as I suspect they were very good. It was the digital front end. If there is one thing I have benefited greatly from with all the things we have learned on this thread, it is the reduction of that piercing glare that I once considered tolerable. Having gotten so accustomed to the lack of glare from the digital front end I have now, somehow, I have become all the more sensitive to bright and edgy electronics. There's simply no going back.

As life has gotten extremely busy and will get even busier, unfortunately, RMAF will represent my last audio indulgence for the indefinite future. This will also represent my last post here on CA. It has been my pleasure to have been a part of this wonderful community of audiophiles.

I will do my best to answer some of the questions that were directed at me.

With regards to the sCLK-EX in my server chassis, this clock board is clocking 4 items and not just 3 -- (1) the motherboard's system clock, (2) both of the motherboard's LAN ports, (3) the tX-USBexp, and (4) the tX-USBhubIN.

Regarding my choice of motherboard, I believe there are many very good boards out there. While I am very pleased with my DFI board, people should target the feature set they desire more than any particular brand. Having said that, a group of friends and I purchased a specific AsRock RACK motherboard some time ago that has the option of being powered via ATX or 12V. We powered this motherboard 3 ways: (1) directly from a 12V rail from my DR SR7, (2) from the 19V rail of my SR7 using an HDPlex DC-to-ATX converter, and (3) an EVGA Gold class 850 watt ATX PSU. With everything else kept constant, this motherboard sounded best when powered directly via a 12V rail from my DR SR7. The EVGA Gold ATX PSU sounded the worst by a very noticeable margin. This should not be taken as any kind of a sweeping statement against ATX PSUs as there are ATX PSUs that are undoubtedly much better than the EVGA we used but this is my strongest argument for using a low power motherboard. They are easier to power well and I have yet to hear anything that can do what an SR7 can.

Regarding oversampling, yes, absolutely, you should oversample and I have stated this preference from very early on in this thread. As far as more companies like Chord finally getting on the oversampling bandwagon that dCS started 20 years ago, this is an inaccurate statement. Rob Watts first postulated his "1 million TAPS" theory while at university in Cardiff during the early 1980s, well before FPGAs were invented. He designed his very first upsampling DAC in 1995 (22 years ago) and all of his DACs moving forward have been upsampling DACs. As his DACs have evolved, they have improved with regards to TAP-length but also filter design and he has now reached the "1 million TAP milestone" with his current M-scaler that he never thought would be reached in his lifetime. dCS lost their chief designer years ago and so their ring DAC technology has stagnated since. The latest dCS Vivaldi Upsampler ($22k) can only upsample to either DXD (24/358) or DSD256. Even HQPlayer can do better.

Does the DAVE, Blu2 or Hugo2 upsample to only 705.6 or 768kHz? Absolutely not. This would be just the beginning. DAVE, Blu2 or Hugo2 can accept any signal up to 768KHz PCM and DSD512 but then they upsample this signal much higher and through multiple WTA (Watts Transient Aligned) filter stages. As an example, when DAVE sees a 44.1 or 48kHz signal, the first WTA stage upsamples to 16FS (705.6 or 768kHz). Then the next WTA stage is at 256FS and finally, there is a third stage filter that upsamples all the way to 2048FS. This means a new filtered sample every 9.6nS. While there is no easy way to directly correlate PCM and DSD, this amounts to a degree of oversampling well beyond DSD2048. According to Rob, he upsamples to this degree for several reasons including to reduce the timing of transients uncertainty, to enable his noise shapers to work at 104MHz so that the noise shapers can reproduce depth correctly, and finally to allow no measurable noise floor modulation.

What is a TAP and what is its significance? Here is a good article to read:

http://www.the-ear.net/how-to/rob-watts-chord-mojo-tech

How is a TAP calculated?

Number of TAPS = (input sample rate frequency in Hz) * (delay time in seconds) * (oversampling rate) * 2

Are TAPS all that matter? Here is what Rob has to say:

"Real tap count is vitally important - and for sure increasing the tap count with everything else the same - algorithm and over sampling rate - will result in much better sound quality.

But there is much more too it than that, in that the algorithm is vitally important. The algorithm is the formula to calculate the coefficients for the filter, as an FIR filter is just audio data * coefficients added up to create the intermediate samples.

When I first started designing with FIR filters in the 90's I had 2048 taps on an FPGA. Now this used a closed form type digital filter - the coefficients were calculated directly, and the filter guaranteed that the original data was preserved. But I heard this filter, and it was so much better than before. Now I knew that to guarantee 16 bit accuracy for the filter for all the calculated interpolated samples I would need getting on for 1M taps, and that was quite impossible way back then. So I thought can the algorithm allow a better timing accuracy without needing 1M taps? Was there a short cut that would allow perfect performance without using impossible tap lengths?

Anyway I spent months thinking about it. I managed to figure out that there was an major measurable issue with filters - aliasing - and that removing this issue would help timing problems. But that would mean the original samples would get changed. And to me this was a very bad idea - it did not feel right, modifying the data. And I delayed trying this out.

Now when you design, you are governed by your thinking, so your ideas set how you go about things. The problem is, we all make assumptions - XYZ can't sound good because of ABC - and assumptions that are not based on hard and rigorous listening tests are very dangerous. And I did not like the idea of changing the data - but I knew that aliasing could be something really important. So I eventually went ahead and designed a (pretty much - better than 16 bits) guaranteed non-aliasing filter. And when I heard it, it was amazing. Eventually this became the first version of the WTA filter, and subsequent tests showed that it at least gave a boost of an order of magnitude to the performance - a 256 tap WTA sounded much better than a 2048 tap conventional filter.

But getting rid of aliasing was not the whole story, as increasing tap length does make huge differences, even when aliasing (which is vitally important in spite of the poor aliasing performance of MQA and other conventional filters) has been eliminated.

So the moral of the story? Don't let seemingly good ideas influence your decisions - test all assumptions with very careful listening tests. Moreover, audio is very complex, and things are never as simple as they might appear.

My mantra is "You know nothing Jon Snow..." and that is to remind me that there are very real limits to what I understand, and assumptions must be constantly tested with listening tests. And very big progress can be made by going down avenues that at first sight seem incapable of changing sound quality."

Regarding the perfect interpolation filter, here is what Rob has to say:

"For the reconstruction filter (the interpolation filter in the DAC) there is only one way of doing it, and that is using an ideal infinite response sinc function - that is if your intention is to perfectly recover the original bandwidth limited analogue signal. Any other type of filter will change the sound by adding uncertainty to the reconstruction of transient timing; and this uncertainty severely damages sound quality. Timing uncertainty due to small tap length filters add a softness, warmth and a bloated bass; some actually like that sound (I do not understand why as it sounds nothing like real un-amplified music) but for sure it is a distortion, or change, from the original.

Now theory has nothing to say about how the signal is bandwidth limited; it simply states there must be zero output above FS/2. Actually, I have designed two decimation filters; one that is linear phase but rings before the impulse; one that is non-linear phase (IIR) but has no pre-ringing. Both filters have attenuation at and above FS/2 of 300 dB, so both meets the requirements of bandwidth limiting in terms of aliasing and sampling theory. I will post blind recordings with the different filters so we can all hear the difference."

Is there an HQPlayer filter equivalent for Rob's WTA filters? Actually, all of Rob's DACs (including Mojo) upsample to 2048FS but they do so differently and so to say that you can match Rob's algorithms using one of HQPlayer's filters would be a gross overstatement.

With Mojo, you have: WTA 16FS > Linear interpolation to 2048FS > Low pass filter > Low pass filter > noise shaper

With DAVE, you have: WTA 16FS > WTA to 256FS > Linear interpolation to 2048FS > Low pass filter > Low pass filter > noise shaper

Rob found that the addition of the extra WTA filter to 256FS in DAVE got him closer to the ideal sinc impulse response and the closer this gets to ideal, the closer the DAC gets to reproducing the signal in between samples.

Regarding his WTA filters, they are proprietary and remain his IP (not Chord's). According to Rob:

"The WTA algorithm actually uses my own windowing function of the ideal sinc impulse response. I took an awful lot of time (man years) with this, both in listening tests and in trying to understand what was going on - the understanding being used to allow me to try listening to different things. At the end of the day, the algorithm is fine tuned by listening tests, but you need understanding in order to change the critical parameters - in short knowing what those parameters are."

Can a PC upsample to the degree that Rob's DACs upsample? Here is what Rob has to say:

"PC's are very restricted in what they can do for real time signals. You simply can't replicate the processing that Dave does in a PC - simply because PC processors are sequential serial devices with a very limited number of cores. When you are doing a doing a FIR filter (a tap) you need to read from memory the audio data; read from memory the coefficient data; multiply the numbers together;then read the accumulated data and add that to the previous multiplication; then save the result. Lots of things to do in sequence. With an FPGA you can do all of these things in parallel at once, so a single FIR tap can be accomplished within a single clock cycle (obviously pipelined) - you are not forced to do things in sequence.

With Dave I have 166 dsp cores running, plus FPGA fabric to do a considerable amount of further processing. You simply can't do that in a PC. To give you another example - converting DSD into DoP. You need a quad core processor to do this manipulation in real time - otherwise you get drop-outs - but in a FPGA I could do this simple operation thousands of times over, and at much faster rates than DSD256.

What some people do not understand is how capable FPGA's are and how widespread they are used - the backbone to the internet? FPGA's. Search engines? FPGA's. Why? because an FPGA is fantastic at doing fixed real time processing - it takes small die area, and can do complex operations with very low power. Mojo for example has 44 dsp cores, uses sophisticated filtering to 104 MHz, and noise shapes at this rate - but does all this whilst consuming only 0.45 W. There is no way any PC consuming huge amounts of power can do this.

Intel last year acquired Altera (an FPGA company) for $16.7 billion because they understand that the future of processing is with FPGA's

A second issue is not what you can do but how you can do it - it is not just about raw power, but how the filter algorithm is designed. I have put many thousands of hours and over twenty years improving and understanding how to make a transparent interpolation filter; and I am still learning things today.

And a third point is that a DAC is not simply a data processing machine but it has got crucial analogue parts too. If I dropped the WTA requirement, I would still need the same FPGA in order to do the noise shaping and other functions."

Over the past few years, upsampling to DSD has become a trend by DAC manufacturers and by adopters of HQPlayer. Here are Rob's comments on DSD (probably not a surprise to many):

"I was doing this 25 years ago in an attempt to solve the innate problems of DSD 256 (PDM 256). But I was using multiple, individually dithered noise shapers in an attempt to improve the noise shaper resolution. The benefits of doing this is that it reduces the jitter sensitivity, but does not eliminate correlated jitter problems. It's why in 1995 I invented pulse array as this eliminates correlated jitter (its a fixed switching activity scheme independent upon the output), improves (in the case of Dave) noise shaper resolution by a trillion times and eliminates noise floor modulation - something you can't do with DSD.

What was also amusing is that they seem to be using an 8 bit shift register to re-time the outputs. I too used to do this too in the early 90's, but quickly found that using discrete flip-flops gave much better measured and SQ performance. Its due to switching activity on-chip changing the propagation delay of the OP FF - so making signal correlated jitter much worse. Also, power draw on each FF modulates the devices internal power rail, creating distortion. Using discrete flip-flops eliminates the signal correlated jitter issue, and with appropriate low impedance power planes, one can eliminate the PSU induced distortion problems too.

This issue of correlated jitter problems can't be solved on-chip. A silicon chip design I worked on over ten years ago had this issue; on chip we had separate clock buffers, custom designed IO buffers, internal separate PSU paths, and all we could do was minimize the issue - it was still there on simulation, and still measurable in reality - although we reduced the problem by two orders of magnitude. But going discrete eliminates the issue. "

Here's another comment:

"There are actually two independent issues going on with DSD that limits the musicality - and they are interlinked problems.

The first issue is down to the resolving power of DSD. Now a DSD works by using a noise shaper, and a noise shaper is a feedback system. Indeed, you can think of an analogue amplifier as a first order noise shaper - so you have a subtraction input stage that compares the input to the output, followed by a gain stage that integrates the error. With a delta sigma noise shaper its exactly the same, but where the output stage is truncated to reduce the noise shaper output resolution so it can drive the OP - in the case of DSD its one bit, +1 or -1 op stage. But you use multiple gain stages connected together so you have n integrators - typically 5 for DSD. Now the number of integrators, together with the time constants will determine how much error correction you have within the system - and the time constants are primarily set by the over-sample rate of the noise shaper. Double the oversampling frequency and with a 5th order ideal system (i.e. one that does not employ resonators or other tricks to improve HF noise) it converges on a 30 dB improvement in distortion and noise.

So where does lack of resolution leave us? Well any signal that is below the noise floor of the noise shaper is completely lost - this is completely unlike PCM where an infinitely small signal is still encoded within the noise when using correct dithering. With DSD any signal below the noise shaper noise floor is lost for good. Now these small signals are essential for the cues that the brain uses to get the perception of sound stage depth - and depth perception is a major problem with audio - conventional high end audio is incapable of reproducing a sense of space in the same way one can perceive natural sounds. Now whilst optimising Hugo's noise shaper I noticed two things - once the noise shaper performance hit 200 dB performance (that is THD and noise being -200 dB in the audio bandwidth as measured using digital domain simulation) then it no longer got smoother. So in terms of warmth and smoothness, 200 dB is good enough. But this categorically did not apply to the perception of depth, where making further improvements improved the perception of how deep instruments were (assuming they are actually recorded with depth like a organ in a cathedral or off stage effects in Mahler 2 for example. Given the size of the FPGA and the 4e pulse array 2048FS DAC, I got the best depth I could obtain.

But with Dave, no such restriction on FPGA size applied, and I had a 20e pulse array DAC which innately has more resolution and allows smaller time constants for the integrator (so better performance). So I optimised it again, and kept on increasing the performance of the noise shaper - and the perception of depth kept on improving. After 3 months of optimising and redesigning the noise shaper I got to 360 dB performance - an extraordinary level, completely way beyond the performance of ordinary noise shapers. But what was curious was how easy it was to hear a 330 dB noise shaper against a 360 dB one - but only in terms of depth perception. My intellectual puzzle is whether this level of small signal accuracy is really needed, or whether these numbers are acting as a proxy for something else going on, perhaps within the analogue parts of the DAC - I am not sure on this point, something I will be researching. But for sure I have got the optimal performance from the noise shaper employed in Dave, and every DAC I have ever listened too shows similar behaviour.

The point I am making over this is that DSD noise shapers for DSD 64 is only capable of 120 dB performance - and that is some 10 thousand times worse than Hugo - and a trillion times worse than Dave. And every time I hear DSD I always get the same problem o perception of depth - it sounds completely flat with no real sense of depth. Now regular 16 bit red book categorically does not suffer from this problem - an infinitely small signal will be perfectly encoded in a properly dithered system - it will just be buried within the noise.

Now the second issue is timing. Now I am not talking about timing in terms of femtosecond clocks and other such nonsense - it always amuses me to see NOS DAC companies talking about femtosecond accuracy clocks when their lack of proper filtering generates hundreds of uS of timing problems on transients due to sampling reconstruction errors. What I am talking about is how accurately transients are timed against the original analogue signal in that the timing of transients is non-linear. Sometimes the transient will be at one point in time, other times delayed or advanced depending upon where the transient occurs against the sample time. In the case of PCM we have the timing errors of transients due to the lack of tap length in the FIR reconstruction filter. The mathematics is very clear cut - we need extremely long tap lengths to almost perfectly reconstruct the original timing of transients - and from listening tests I can hear a correlation between tap length and sound quality. With Dave I can still hear 100,000 taps increasing to 164,000 taps albeit I can now start to hear the law of diminishing returns. But we know for sure that increasing the tap length will mean that it would make absolutely no difference if it was sampled at 22 uS or 22 fS (assuming its a perfectly bandwidth limited signal). So red book is again limited on timing by the DAC not inherently within the format.

Unfortunately, DSD also has its timing non-linearity issues but they are different to PCM. This problem has never been talked about before, but its something I have been aware of for a long time, and its one reason I uniquely run my noise shapers at 2048FS. When a large signal transient occurs - lets say from -1 to +1 then the time delay for the signal is small as the signal gets through the integrators and OP quantizer almost immediately. But for small signals, it can't get through the quantizer, and so it takes some time for a small negative signal changing to a positive signal to work its way through the integrators. You see these effects on simulation, where the difference of a small transient to a large transient is several uS for DSD64.

Now the timing non linearity of uS is very audible and it affects the ability of the brain to perceive the starting and stopping of instruments. Indeed, the major surprise of Hugo was how well one can perceive that starting and stopping of notes - it was much better than I expected, and at the time I was perplexed where this ability was coming from. With Dave I managed to dig down into the problem, and some of the things I had done (for other reasons) had also improved the timing non-linearity. It turns out that the brain is much more sensitive that the order of 4 uS of timing errors (this number comes from the inter-aural delay resolution, its the accuracy the brain works to in measuring time from sounds hitting one ear against the other), and much smaller levels degrade the ability for the brain to perceive the starting and stopping of notes.

But timing accuracy has another important effect too - not only is it crucial to being able to perceive the starting and stopping of notes, its also used to perceive the timbre of an instrument - that is the initial transient is used by the brain to determine the timbre of an instrument and if timing of transients is non-linear, then we get compression in the perception of timbre. One of the surprising things I heard with Hugo was how easy it was to hear the starting and stopping of instruments, and how easy it was to perceive individual instruments timbre and sensation of power. And this made a profound improvement with musicality - I was enjoying music to a level I had never had before.

But the problem we have with DSD is that the timing of transients is non-linear with respect to signal level - and unlike PCM you are completely stuck as the error is on the recording and its impossible to remove. So when I hear DSD, it sounds flat in depth, and it has relatively poor ability to perceive the starting and stopping of notes (using Hugo/Dave against PCM). Acoustic guitar sounds quite pleasant, but there is a lack of focus when the string is initially struck - it sounds all unnaturally soft with an inability to properly perceive the starting and stopping. Also the timbre of the instrument is compressed, and its down to the substantial timing non-linearity with signal level.

Having emphasised the problems with delta-sigma or noise shaping you may think its better to use R2R DAC's instead. But they too have considerable timing errors too; making the timing of signals code independent is impossible. Also they have considerable low level non linearity problems too as its impossible to match the resistor values - much worse than DSD even - so again we are stuck with poor depth, perception of timing and timbre. Not only that they suffer from substantial noise floor modulation, giving a forced hard aggressive edge to them. Some listeners prefer that, and I won't argue with somebody else's taste - whatever works for you. But its not real and it not the sound I hear with live un-amplified instruments.

So to conclude; yes I agree, DSD is fundamentally flawed, and unlike PCM where the DAC is the fundamental limit, its in the format itself. And it is mostly limited by the format. Additionally, its very easy to underestimate how sensitive the brain is to extremely small errors, and these errors can have a profound effect on musicality."

Rob's quotes above were not placed here to suggest he is right and others are wrong. They were placed here so that people at least understand his perspective as a lot of misinformation is being spread. You can choose to agree or disagree. I suggest you perform your due diligence and let your ears decide and this is what I have tried to do. Do I agree with everything Rob says? No, I do not. For example, Rob believes all sources should sound the same with his DACs. It would appear Chord and their salespeople have bought into this philosophy also but my ears clearly tell me this is untrue and Rob will hear this for himself soon enough. I also wonder what would happen if Paul Hynes were to design a new PSU for my DAVE. Rob doesn't think it would make much difference but I have a gut feeling the difference would be significant, especially as I am using my DAVE to directly drive speakers.

With regards to combining HQPlayer upsampling and one of Rob's DACs, I have already reported earlier in this thread that I have done this with my DAVE in the past. Remember that when DAVE (or any of Rob's DACs) sees a signal, whether it be 16/44, PCM 705/768, or DSD512, it takes that signal and upsamples it further. Using my Windows workstation, I ran this exercise with HQPlayer again this evening because I was curious. My new Hugo2 just arrived today and so I tried upsampling from Redbook to both PCM 705.6kHz and DSD512. I did it also with Blu Mk2. Just like before, upsampling to DSD512 with either of these devices resulted in worse SQ. Upsampling to PCM 705.6kHz had better results and perhaps there is an ideal filter/dither combination I failed to try but with the ones I did try, SQ was very acceptable but compared to feeding either Hugo2 or Blu2 the original signal, SQ via HQPlayer was drier and less smooth. With Blu2 combined with either DAVE or Hugo2, the improvement was massive. HQPlayer does not come close to what the M-scaler in Blu2 can do with either Hugo2 or DAVE.

For those looking for a reasonably affordable alternative to upsampling with HQP, I suggest you give Hugo2 a try. Combined with even an unmodified NUC and an inexpensive endpoint like the ISO Regen or better yet, a tX-USBultra, I suspect it will give even the very best HQPlayer setups a run for their money. With the brief time I have had it, I am finding Hugo2 to be an excellent upsampling DAC and just like DAVE, it can directly drive high-efficiency speakers as well as headphones. I cannot overstate just how wonderfully resolving and transparent Hugo2 directly driving my Voxativs or Omegas sounds. Moreover, Hugo2 will combine with M-scaler to give you a full 1M taps, something that no other DAC except DAVE can currently provide and so as far as I am aware, when combined with M-scaler and when directly driving speakers, these are the 2 highest resolution and most transparent digital front ends in existence today regardless of price.

I wish you all success in your audiophile journeys. Signing out...

Here it is, the longest post in the history of CA and 3/4 of it is marketing material from Chord. Roy, you should know better. I am disappointed.

auricgoldfinger · October 11, 2017

8 hours ago, feelingears said:

Akin to what @austinpop said about assembling hardware (vs. software), I'm leery of HQP's apparent complexity and user unfriendliness despite (or because of) my work in software. But this is audio so one is always exploring or tempted to try more. So, begging everyone's pardon, but would someone please point me/others to a good "HQP for Dummies" post or set of instructions, anywhere? Something user-friendly and more "forgiving," if such a thing exists?

(Also, HQP is a one-man show is it not? I wish there were more "there, there" as a going concern.)

Thanks in advance.

@seeteeyou You just beat me to the punch, but I'll have to see if any of those speaks to HQP (and speaks in "Dummies" dialect).

Start here, and then follow up by reading the manual starting on p. 8 to get more detail on filter and sampling settings.

afrancois · October 11, 2017

I received my 3 sheets of 3M AB5100SHF-210X297 today and used them around the sMS-200 ultra, MC3+ USB and even under some of my LPS-1’s with very good results.

I’m so satisfied that I immediately ordered extra 5 sheets. I still got a lot of equipment that could potentially benefit from this miracle paper. DAC’s, power amplifiers and headphone amplifier.

Perhaps not everybody will benefit as much as I do. Most of my equipment is in the garage and it seems a rather noisy environment.

BigGuy · October 11, 2017

11 minutes ago, afrancois said:

I received my 3 sheets of 3M AB5100SHF-210X297 today and used them around the sMS-200 ultra, MC3+ USB and even under some of my LPS-1’s with very good results.

I’m so satisfied that I immediately ordered extra 5 sheets. I still got a lot of equipment that could potentially benefit from this miracle paper. DAC’s, power amplifiers and headphone amplifier.

Perhaps not everybody will benefit as much as I do. Most of my equipment is in the garage and it seems a rather noisy environment.

From what I see, 3M sells sheets for specific applications. Looks like AB5100 is for EMI rather than RFI. Not questioning your observations but would not a garage be more likely to be RFI environ?

limniscate · October 11, 2017

5 hours ago, lmitche said:

Here it is, the longest post in the history of CA and 3/4 of it is marketing material from Chord. Roy, you should know better. I am disappointed.

Have you actually heard the DAVE with the Blu mk.2? I didn't really think the DAVE was that good, but the DAVE with the Blu mk.2 is pretty astounding.

afrancois · October 11, 2017

27 minutes ago, BigGuy said:

From what I see, 3M sells sheets for specific applications. Looks like AB5100 is for EMI rather than RFI. Not questioning your observations but would not a garage be more likely to be RFI environ?

I know a bit confusing, but this is what Wiki says:

Electromagnetic interference (EMI), also called radio-frequency interference (RFI) when in the radio frequency spectrum,
is a disturbance generated by an external source that affects an electrical circuit by electromagnetic induction,
electrostatic coupling, or conduction.

And another explanation:

EMI (Electromagnetic Interference) is also called RFI (Radio Frequency Interference). Although the terms EMI and RFI are often used interchangeably, EMI is actually any frequency of electrical noise, whereas RFI is a specific subset of electrical noise on the EMI spectrum. RFI is a disturbance that affects an electrical circuit due to either electromagnetic conduction or electromagnetic radiation emitted from an external source. The disturbance may interrupt, obstruct, degrade or limit the effective performance of the circuit. The source may be any object, artificial or natural, that carries rapidly changing electrical currents, such as an electrical circuit or the Sun. There are two types of RFI. Conducted RFI is unwanted high frequencies that ride on the AC wave form. Radiated RFI is emitted through the air. There are many pieces of equipment that can generate RFI, variable frequency drives included.

unbalanced output · October 11, 2017

7 hours ago, romaz said:

"PC's are very restricted in what they can do for real time signals. You simply can't replicate the processing that Dave does in a PC - simply because PC processors are sequential serial devices with a very limited number of cores. When you are doing a doing a FIR filter (a tap) you need to read from memory the audio data; read from memory the coefficient data; multiply the numbers together;then read the accumulated data and add that to the previous multiplication; then save the result. Lots of things to do in sequence. With an FPGA you can do all of these things in parallel at once, so a single FIR tap can be accomplished within a single clock cycle (obviously pipelined) - you are not forced to do things in sequence.

That's what a beginner programmer would do - FIR filters can actually be implemented in efficient ways even without considering paralellisation. That said, FPGAs are neat...

7 hours ago, romaz said:

But the problem we have with DSD is that the timing of transients is non-linear with respect to signal level - and unlike PCM you are completely stuck as the error is on the recording and its impossible to remove. So when I hear DSD, it sounds flat in depth, and it has relatively poor ability to perceive the starting and stopping of notes (using Hugo/Dave against PCM). Acoustic guitar sounds quite pleasant, but there is a lack of focus when the string is initially struck - it sounds all unnaturally soft with an inability to properly perceive the starting and stopping. Also the timbre of the instrument is compressed, and its down to the substantial timing non-linearity with signal level.

I totally agree that DSD sounds flat... on the Mojo. In reality, it sounds very very similar if not the same as PCM. Switch to a native DSD DAC like the Lampizator and two or three new dimensions immediately jump from the speakers.

Johnseye · October 11, 2017

1 hour ago, afrancois said:

I received my 3 sheets of 3M AB5100SHF-210X297 today and used them around the sMS-200 ultra, MC3+ USB and even under some of my LPS-1’s with very good results.

I’m so satisfied that I immediately ordered extra 5 sheets. I still got a lot of equipment that could potentially benefit from this miracle paper. DAC’s, power amplifiers and headphone amplifier.

Perhaps not everybody will benefit as much as I do. Most of my equipment is in the garage and it seems a rather noisy environment.

Glad to hear you had positive results. As I mentioned my results were similarly impressive and my equipment is in a dedicated listening room with the only equipment causing noise being the listening equipment itself and maybe some canned lighting if its even on or impactful.

austinpop · October 11, 2017

6 hours ago, lmitche said:

Here it is, the longest post in the history of CA and 3/4 of it is marketing material from Chord. Roy, you should know better. I am disappointed.

Larry, I'm sorry you took that away from Roy's post. Knowing him a friend, there is no way that was even remotely his intent. He genuinely enjoys helping others, and I think his body of contributions here speaks volumes to validate that.

No - what I was terribly disappointed to hear was this:

8 hours ago, romaz said:

This will also represent my last post here on CA.

He will be sorely missed.

Johnseye · October 11, 2017

1 minute ago, austinpop said:

No - what I was terribly disappointed to hear was this:

He will be sorely missed.

Agreed. What a wealth of knowledge, not to mention deep pockets to help with testing He's also a very helpful person, providing detailed explanations for tests conducted or questions answered. Hopefully he pops in on occasion to share in any new discoveries he makes. This thread has inspired many with revisited ideas like the EMI sheets, to new experiments with motherboard clocking.

Cheers Roy.

AmusedToD · October 11, 2017

Good luck @romaz, you will be utterly missed!

A novel way to massively improve the SQ of computer audio streaming

Recommended Posts

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Link to comment

Create an account or sign in to comment

Create an account

Sign in