Rob Watts
Member of the Trade: Chord Electronics
- Joined
- Apr 1, 2014
- Posts
- 3,144
- Likes
- 12,424
I promised that after Can-Jam Singapore I would publish my slides, so here goes. Apologies in advance if you find it too technical - but we are dealing with an immensely complex subject, and I have tried my best to simplify it as much as possible.
So the opening slide.
So here I am talking about the WTA filter, and I have assumed that the reader is aware of the WTA filter and why the recovery of timing of transients is important - so I assume you know already that more taps gives better accuracy, and much better sound, and that moving from 8 FS (output at 352/384 kHz) to 16 FS (output at 705/768 kHz) is seen as a benefit in SQ due to the improved time resolution of the output.
The FPGA is jam packed - I use all of the available DSP cores, all of the memory, and 99.8% of the logic too. The downside to all this is now the FPGA uses a lot of power, around 800 mW.
The output from the WTA 1 is now passed to another WTA filter (WTA 2) that takes us from 16 FS to 256 FS (output at 11.289.6/12.288 MHz) so now the WTA is working to a resolution of 88 nS.
The filter option is actually a big departure for me - I go for what is technically correct (assuming it sounds best), and don't give options. Now the HF filter is a technically correct option - as using the HF filter with HD files can reduce HF noise from the recording, which is not music but noise shaper ADC distortion and noise - if this gets into the analogue parts it can cause more noise floor modulation, so removing it will make it sound smoother. So having the HF filter is a technically valid option. But the 256 FS filter is always the more accurate option, and will recover the timing much more accurately than not using it. So why did I add this as an option?
Several reasons. I thought it would be cool for people to actually hear the effect of the WTA going from 16 FS to 256 FS. Normally, I do lots of listening tests, and so build up knowledge to allow better designs and future improvements, so I thought it would be good for one to hear the effects of 16 FS to 256 FS. What you hear is an immediate change in the ability to perceive the starting and stopping of notes. This quality is very different to the usual WTA benefits (better timbre, pitch, instrument separation and focus etc) in that being able to perceive the starting edge of a signal (the initial pluck of a string or how the piano sounds when instantly hitting the key) all depends upon timing accuracy going from uS down to tens of nS - so I thought it would be interesting to actually hear what I am talking about directly. Now this filter will be called the incisive option, as calling it Hugo is a bit confusing. Its incisive because you can now perceive the starting and stopping of notes much more clearly - and when the brain can't perceive the leading edges, then it becomes a blur and things sound soft.
The second reason for the filter options is that the incisive revealing nature of the filter does make it sound brighter. Now it is absolutely technically more accurate; it only sounds brighter because the brain can now more accurately perceive the starting and stopping of notes, and the starting and stopping transients have a lot of high frequency energy. When the brain can't perceive something, it simply ignores it, so it then sounds unnaturally soft, in that this is not truly transparent. But sometimes when you have say a bad bright recording, or say hard headphones, having a filter that allows you to hear high frequency energy may be a bit too much. But for sure you are using an aberration to hide another problem. So my advice is this; if you use the 16FS option (orange or red) all the time, then consider getting a warmer set of headphones, or trying out EQ. Normally you should be using white or green - I run with green all the time as its useful with 192 recordings.
So this is fairly straightforward, and with conventional DAC's you can easily measure noise floor modulation. I will be showing the measurements shortly.
OK this is a quick summary of the difficulties involved an having a DAC that has no measurable noise floor modulation, and there is a great deal more besides, as noise floor modulation occurs from a myriad of problems within DAC's.
So this is why I have to filter DSD sources - as without it we would get large amounts of noise floor modulation, and other problems - gurgle noise, distortion etc.
With this filter I wanted top match the abilities of the DSD+ mode in Dave, and I believe I have - I know get that level of sound quality. But to do that I had to have incredible levels of filter attenuation - so with much better than 200 dB, I ensure that THD and noise from the analogue is entirely limited by the DSD 64 digital performance - there is no added THD or significant noise that is not present on the file.
So this is employing the knowledge gained from Dave with noise shapers, so now I am getting much better perception of depth and detail resolution. Incidentally, I have now designed and listened to dozens of noise shapers and every time I got the same result as Dave - better noise shaper performance gives better depth, and the smallest small signal error is audible in terms of depth truncation.
This is the same cross-feed on the original Hugo - indeed its the only piece of code coming from Hugo! But there was an interesting story with it. When developing products you go through many stages - and formal listening tests plays a major part. But then there is the part where you are listening for pleasure, and at these times it is valuable because you can see the scale of where the sound quality is, and its only by listening without consciously testing that one can asses musicality or the ability to get emotional with music. And that is the real reason to be doing all this.
So last December one of the prototypes was ready for listening, and I took it on many flights. I was listening to music, and bells were being played - and the depth was so convincing I thought it was in the plane, and not on the recording. This was a major surprise, as getting headphones to portray good depth has been a major issue. It turned out that the cross-feed setting was crucial - with it off, the sound-stage collapsed, and with it on, I got decent depth from headphones.
Measurements are vitally important, as if you are concerned with making truly transparent devices - and that is the only way ultimately for musicality - then measurements plays a vital part. Now its true that a good sounding device can sound better than a better measured one; but for sure, when you improve the measurements, (now matter how small the measured change is) and with all other things being equal, you will perceive an improvement in SQ.
Additionally - many manufacturers make bold claims - sometimes with the best of intentions (they genuinely believe it is better), sometimes with the intent to deceive at worst or at best to extract cash from you - and then from the measurements you can see that they are talking nonsense. Moreover, I can tell from a suite of measurements pretty much how something will sound, and moreover what is the intent and capability of the designer or design team. So although it is highly fashionable to talk about not caring about doing measurements, to me it is throwing out the baby with the bath water.
So I made some claims about noise floor modulation, and here it proves it - absolutely none, and zero an-harmonic products too.
Now I should add I often see so called measurements of my products on other threads using poor quality test equipment, and hence "proving" that xyz is not that good after all. So I should add that Hugo 2 outperforms test equipment easily. A measurement is not objective reality, it is just data subject to error done at a particular time with particular test equipment. And my DAC's require the absolute state of the art test gear - in this case the APX555. This is the only test equipment that is capable of measuring noise floor modulation, as ADC noise floor modulation is way bigger than Hugo 2 - and the APX 555 uses a special technique with 4 ADC's to overcome the ADC limitation.
We can see also the extraordinary low THD - this is only beaten by Dave.
Now one of the features I used was the second order analogue noise shaper OP stage that first appeared with Dave. The benefit of this technique is that it eliminates crossover distortion, as high frequency distortion does not significantly increase with a 33 ohm load. in the past adding a load of 33 ohms would harden up the sound - now it makes no difference whatsoever.
This shows how isolated the DAC and amp is from the power supply and each channel.
So proof again that is immune from jitter - no artifacts at all. The only thing you can see is a tiny residual at 11 and 13 kHz these are artifacts from my APX555.
I am very happy with these measurements; I am confident that no other non Chord DAC at any price comes close to this level of performance.
Rob
So the opening slide.
So here I am talking about the WTA filter, and I have assumed that the reader is aware of the WTA filter and why the recovery of timing of transients is important - so I assume you know already that more taps gives better accuracy, and much better sound, and that moving from 8 FS (output at 352/384 kHz) to 16 FS (output at 705/768 kHz) is seen as a benefit in SQ due to the improved time resolution of the output.
The FPGA is jam packed - I use all of the available DSP cores, all of the memory, and 99.8% of the logic too. The downside to all this is now the FPGA uses a lot of power, around 800 mW.
The output from the WTA 1 is now passed to another WTA filter (WTA 2) that takes us from 16 FS to 256 FS (output at 11.289.6/12.288 MHz) so now the WTA is working to a resolution of 88 nS.
The filter option is actually a big departure for me - I go for what is technically correct (assuming it sounds best), and don't give options. Now the HF filter is a technically correct option - as using the HF filter with HD files can reduce HF noise from the recording, which is not music but noise shaper ADC distortion and noise - if this gets into the analogue parts it can cause more noise floor modulation, so removing it will make it sound smoother. So having the HF filter is a technically valid option. But the 256 FS filter is always the more accurate option, and will recover the timing much more accurately than not using it. So why did I add this as an option?
Several reasons. I thought it would be cool for people to actually hear the effect of the WTA going from 16 FS to 256 FS. Normally, I do lots of listening tests, and so build up knowledge to allow better designs and future improvements, so I thought it would be good for one to hear the effects of 16 FS to 256 FS. What you hear is an immediate change in the ability to perceive the starting and stopping of notes. This quality is very different to the usual WTA benefits (better timbre, pitch, instrument separation and focus etc) in that being able to perceive the starting edge of a signal (the initial pluck of a string or how the piano sounds when instantly hitting the key) all depends upon timing accuracy going from uS down to tens of nS - so I thought it would be interesting to actually hear what I am talking about directly. Now this filter will be called the incisive option, as calling it Hugo is a bit confusing. Its incisive because you can now perceive the starting and stopping of notes much more clearly - and when the brain can't perceive the leading edges, then it becomes a blur and things sound soft.
The second reason for the filter options is that the incisive revealing nature of the filter does make it sound brighter. Now it is absolutely technically more accurate; it only sounds brighter because the brain can now more accurately perceive the starting and stopping of notes, and the starting and stopping transients have a lot of high frequency energy. When the brain can't perceive something, it simply ignores it, so it then sounds unnaturally soft, in that this is not truly transparent. But sometimes when you have say a bad bright recording, or say hard headphones, having a filter that allows you to hear high frequency energy may be a bit too much. But for sure you are using an aberration to hide another problem. So my advice is this; if you use the 16FS option (orange or red) all the time, then consider getting a warmer set of headphones, or trying out EQ. Normally you should be using white or green - I run with green all the time as its useful with 192 recordings.
So this is fairly straightforward, and with conventional DAC's you can easily measure noise floor modulation. I will be showing the measurements shortly.
OK this is a quick summary of the difficulties involved an having a DAC that has no measurable noise floor modulation, and there is a great deal more besides, as noise floor modulation occurs from a myriad of problems within DAC's.
So this is why I have to filter DSD sources - as without it we would get large amounts of noise floor modulation, and other problems - gurgle noise, distortion etc.
With this filter I wanted top match the abilities of the DSD+ mode in Dave, and I believe I have - I know get that level of sound quality. But to do that I had to have incredible levels of filter attenuation - so with much better than 200 dB, I ensure that THD and noise from the analogue is entirely limited by the DSD 64 digital performance - there is no added THD or significant noise that is not present on the file.
So this is employing the knowledge gained from Dave with noise shapers, so now I am getting much better perception of depth and detail resolution. Incidentally, I have now designed and listened to dozens of noise shapers and every time I got the same result as Dave - better noise shaper performance gives better depth, and the smallest small signal error is audible in terms of depth truncation.
This is the same cross-feed on the original Hugo - indeed its the only piece of code coming from Hugo! But there was an interesting story with it. When developing products you go through many stages - and formal listening tests plays a major part. But then there is the part where you are listening for pleasure, and at these times it is valuable because you can see the scale of where the sound quality is, and its only by listening without consciously testing that one can asses musicality or the ability to get emotional with music. And that is the real reason to be doing all this.
So last December one of the prototypes was ready for listening, and I took it on many flights. I was listening to music, and bells were being played - and the depth was so convincing I thought it was in the plane, and not on the recording. This was a major surprise, as getting headphones to portray good depth has been a major issue. It turned out that the cross-feed setting was crucial - with it off, the sound-stage collapsed, and with it on, I got decent depth from headphones.
Measurements are vitally important, as if you are concerned with making truly transparent devices - and that is the only way ultimately for musicality - then measurements plays a vital part. Now its true that a good sounding device can sound better than a better measured one; but for sure, when you improve the measurements, (now matter how small the measured change is) and with all other things being equal, you will perceive an improvement in SQ.
Additionally - many manufacturers make bold claims - sometimes with the best of intentions (they genuinely believe it is better), sometimes with the intent to deceive at worst or at best to extract cash from you - and then from the measurements you can see that they are talking nonsense. Moreover, I can tell from a suite of measurements pretty much how something will sound, and moreover what is the intent and capability of the designer or design team. So although it is highly fashionable to talk about not caring about doing measurements, to me it is throwing out the baby with the bath water.
So I made some claims about noise floor modulation, and here it proves it - absolutely none, and zero an-harmonic products too.
Now I should add I often see so called measurements of my products on other threads using poor quality test equipment, and hence "proving" that xyz is not that good after all. So I should add that Hugo 2 outperforms test equipment easily. A measurement is not objective reality, it is just data subject to error done at a particular time with particular test equipment. And my DAC's require the absolute state of the art test gear - in this case the APX555. This is the only test equipment that is capable of measuring noise floor modulation, as ADC noise floor modulation is way bigger than Hugo 2 - and the APX 555 uses a special technique with 4 ADC's to overcome the ADC limitation.
We can see also the extraordinary low THD - this is only beaten by Dave.
Now one of the features I used was the second order analogue noise shaper OP stage that first appeared with Dave. The benefit of this technique is that it eliminates crossover distortion, as high frequency distortion does not significantly increase with a 33 ohm load. in the past adding a load of 33 ohms would harden up the sound - now it makes no difference whatsoever.
This shows how isolated the DAC and amp is from the power supply and each channel.
So proof again that is immune from jitter - no artifacts at all. The only thing you can see is a tiny residual at 11 and 13 kHz these are artifacts from my APX555.
I am very happy with these measurements; I am confident that no other non Chord DAC at any price comes close to this level of performance.
Rob
